text
stringlengths 4
2.78M
| meta
dict |
|---|---|
---
abstract: 'The present paper has two objectives. On one side, we develop and test numerically divergence free Virtual Elements in three dimensions, for variable “polynomial” order. These are the natural extension of the two-dimensional divergence free VEM elements, with some modification that allows for a better computational efficiency. We test the element’s performance both for the Stokes and (diffusion dominated) Navier-Stokes equation. The second, and perhaps main, motivation is to show that our scheme, also in three dimensions, enjoys an underlying discrete Stokes complex structure. We build a pair of virtual discrete spaces based on general polytopal partitions, the first one being scalar and the second one being vector valued, such that when coupled with our velocity and pressure spaces, yield a discrete Stokes complex.'
author:
- 'L. Beirão da Veiga [^1]'
- 'F. Dassi [^2]'
- 'G. Vacca [^3]'
bibliography:
- 'biblio.bib'
title: ' **The Stokes complex for Virtual Elements in three dimensions**'
---
Introduction
============
The Virtual Element Method (VEM) was introduced in [@volley; @autostoppisti] as a generalization of the Finite Element Method (FEM) allowing for general polytopal meshes. Nowadays the VEM technology has reached a good level of success; among the many papers we here limit ourselves in citing a few sample works [@GPDM:2015; @BPP:2017; @BB:2017; @dassi-mascotto:2018; @BBDMRbis:2018; @MPP:2018; @AMV:2018; @AHHW:2018]. It was soon recognized that the flexibility of VEM allows to build elements that hold peculiar advantages also on more standard grids. One main example is that of “divergence-free” Virtual Elements for Stokes-type problems, initiated in [@BLV:2017; @ABMV:2014] and further developed in [@BLV:2018; @vacca:2018]. An advantage of the proposed family of Virtual Elements is that, without the need of a high minimal polynomial degree as it happens in conforming FEM, it is able to yield a discrete divergence-free (conforming) velocity solution, which can be an interesting asset as explored for Finite Elements in [@guzman-neilan:2014; @john-linke-merdon-neilan-rebholz:2017; @neilan-sap:2016; @linke-merdon:2016; @guzaman-neilan:2018]. For a wider look in the literature, other VEM for Stokes-type problems can be found in [@LLC:2017; @CGM:2016; @CGS:2018; @GMS:2018; @CW:2018; @dassi-vacca:2018] while different polygonal methods for the same problem in [@LVY:2014; @CFQ:2017; @BdPD:2018; @dPK:2018].
The present paper has two objectives. On one side, we develop and test numerically for the first time the divergence free Virtual Elements in three dimensions (for variable “polynomial” order $k$). These are the natural extension of the two-dimensional VEM elements of [@BLV:2017; @BLV:2018], with some modification that allows for a better computational efficiency. We first test the element’s performance for the Stokes and (diffusion dominated) Navier-Stokes equation for different kind of meshes (such as Voronoi, but also cubes and tetrahedra) and then show a specific test that underlines the divergence free property (in the spirit of [@linke-merdon:2016; @BLV:2018]).
The second, and perhaps main, motivation is to show that our scheme, also in three dimensions, enjoys an underlying discrete Stokes complex structure. That is, a discrete structure of the kind $${{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
W_h \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
{\boldsymbol{\Sigma}}_h \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
{\boldsymbol{V}}_h \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
Q_h \, \xrightarrow[]{ \quad 0 \quad }
0$$ where the image of each operator exactly corresponds to the kernel of the following one, thus mimicking the continuous complex $${{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
H^1(\Omega) \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
\boldsymbol{\Sigma}(\Omega) \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
[H^1(\Omega)]^3 \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
L^2(\Omega) \, \xrightarrow[]{ \quad 0 \quad }
0 \,,$$ with $\boldsymbol{\Sigma}(\Omega)$ denoting functions of $L^2(\Omega)$ with ${\bf curl}$ in $H^1(\Omega)$ [@girault-raviart:book; @amrouche-et-al:1998]. Discrete Stokes complexes has been extensively studied in the literature of Finite Elements since the presence of an underlying complex implies a series of interesting advantages (such as the divergence free property), in addition to guaranteeing that the discrete scheme is able to correcly mimic the structure of the problem under study [@demkowicz-et-al:2000; @demkowicz-buffa:2005; @arnold-falk-winter:2006; @arnold-falk-winther:2006bis; @arnold-falk-winter:2010; @buffa-rivas-sangalli-velasquez:2011; @falk-neilan:2013; @evans-hughes:2013; @neilan:2015]. This motivation is therefore mainly theoretical in nature, but it serves the important purpose of giving a deeper foundation to our method. We therefore build a pair of virtual discrete spaces based on general polytopal partitions of $\Omega$, the first one $W_h$ which is conforming in $H^1(\Omega)$ and the second one ${\boldsymbol{\Sigma}}_h$ conforming in $\boldsymbol{\Sigma}(\Omega)$, such that, when coupled with our velocity and pressure spaces, yield a discrete Stokes complex. We also build a set of carefully chosen associated degrees of freedom. This construction was already developed in two dimensions in [@BMV:2018], but here things are more involved due to the much more complex nature of the curl operator in 3D when compared to 2D. In this respect we must underline that, to the best of the authors knowledge, no Stokes exact complex of the type above exists for conforming Finite Elements in three dimensions. There exist FEM for different (more regular) Stokes complexes, but at the price of developing cumbersome elements with a large minimal polynomial degree (we refer to [@john-linke-merdon-neilan-rebholz:2017] for an overview) or using a subdivision of the element [@christiansen:2018]. We finally note that our construction holds for a general “polynomial” order $k \ge 2$. The paper is organized as follows. After introducing some notation and preliminaries in Section \[sec:notations\], the Virtual Element spaces and the associated degrees of freedom are deployed in Section \[sec:spaces\]. In Section \[sec:derham\] we prove that the introduced spaces constitute an exact complex. In Section \[sec:VEM-ns\] we describe the discrete problem, together with the associated projectors and bilinear forms. In Section \[sec:num\_test\] we provide the numerical tests. Finally, in the appendix we prove a useful lemma.
Notations and preliminaries {#sec:notations}
===========================
In the present section we introduce some basic tools and notations useful in the construction and theoretical analysis of Virtual Element Methods.
Throughout the paper, we will follow the usual notation for Sobolev spaces and norms [@Adams:1975]. Hence, for an open bounded domain $\omega$, the norms in the spaces $W^s_p(\omega)$ and $L^p(\omega)$ are denoted by $\|{\cdot}\|_{W^s_p(\omega)}$ and $\|{\cdot}\|_{L^p(\omega)}$ respectively. Norm and seminorm in $H^{s}(\omega)$ are denoted respectively by $\|{\cdot}\|_{s,\omega}$ and $|{\cdot}|_{s,\omega}$, while $(\cdot,\cdot)_{\omega}$ and $\|\cdot\|_{\omega}$ denote the $L^2$-inner product and the $L^2$-norm (the subscript $\omega$ may be omitted when $\omega$ is the whole computational domain $\Omega$).
Basic notations and mesh assumptions {#sub:mesh}
------------------------------------
From now on, we will denote with $P$ a general polyhedron having $\ell_V$ vertexes $V$, $\ell_e$ edges $e$ and $\ell_f$ faces $f$.\
For each polyhedron $P$, each face $f$ of $P$ and each edge $e$ of $f$ we denote with:
- ${\boldsymbol{n}}_P^f$ (resp. ${\boldsymbol{n}}_P$) the unit outward normal vector to $f$ (resp. to $\partial P$),
- ${\boldsymbol{n}}_f^e$ (resp. ${\boldsymbol{n}}_f$) the unit vector in the plane of $f$ that is normal to the edge $e$ (resp. to $\partial f$) and outward with respect to $f$,
- ${\boldsymbol{t}}_f^e$ (resp. ${\boldsymbol{t}}_f$) the unit vector in the plane of $f$ tangent to $e$ (resp. to $\partial f$) counterclockwise with respect to ${\boldsymbol{n}}_P^f$,
- $\boldsymbol{\tau}_1^f$ and $\boldsymbol{\tau}_2^f$ two orthogonal unit vectors lying on $f$ and such that $\boldsymbol{\tau}_1^f \wedge \boldsymbol{\tau}_2^f = {\boldsymbol{n}}_P^f$,
- ${\boldsymbol{t}}_e$ a unit vector tangent to the edge $e$.
Notice that the vectors ${\boldsymbol{t}}_f^e$, ${\boldsymbol{t}}_f$, $\boldsymbol{\tau}_1^f$ and $\boldsymbol{\tau}_2^f$ actually depend on $P$ (we do not write such dependence explicitly for lightening the notations).
In the following ${\mathcal{O}}$ will denote a general geometrical entity (element, face, edge) having diameter $h_{{\mathcal{O}}}$.
Let $\Omega$ be the computational domain that we assume to be a contractible polyhedron (i.e. simply connected polyhedron with boundary $\partial \Omega$ which consists of one connected component), with Lipschitz boundary. Let $\set{\Omega_h}_h$ be a sequence of decompositions of $\Omega$ into general polyhedral elements $P$ where $
h := \sup_{P \in \Omega_h} h_P
$.
We suppose that for all $h$, each element $P$ in $\Omega_h$ is a contractible polyhedron that fulfils the following assumptions:
$\mathbf{(A1)}$
: $P$ is star-shaped with respect to a ball $B_P$ of radius $ \geq\, {\varrho}\, h_P$,
$\mathbf{(A2)}$
: every face $f$ of $P$ is star-shaped with respect to a disk $B_f$ of radius $ \geq\, {\varrho}\, h_P$,
$\mathbf{(A3)}$
: every edge $e$ in $P$ satisfies $ h_e \geq {\varrho}\, h_P$,
where ${\varrho}$ is a uniform positive constant. We remark that the hypotheses $\mathbf{(A1)}$, $\mathbf{(A2)}$ and $\mathbf{(A3)}$, though not too restrictive in many practical cases, can be further relaxed, as investigated in [@BLR:2017; @brenner-guan-sung:2017; @brenner-sung:2018; @cao-chen:2018].
The total number of vertexes, edges, faces and elements in the decomposition $\Omega_h$ are denoted respectively with $L_V$, $L_e$, $L_f$, $L_P$.
For any mesh object ${\mathcal{O}}$ and for $n \in {{\mathbb}{N}}$ we introduce the spaces:
- ${{\mathbb}{P}}_n({\mathcal{O}})$ the polynomials on ${\mathcal{O}}$ of degree $\leq n$ (with the extended notation ${{\mathbb}{P}}_{-1}({\mathcal{O}})=\{0\}$),
- $\widehat{{{\mathbb}{P}}}_{n \setminus m}({\mathcal{O}}) :=
{{\mathbb}{P}}_{n}({\mathcal{O}}) \setminus {{\mathbb}{P}}_{m}({\mathcal{O}})$ for $m \leq n$, denotes the polynomials in ${{\mathbb}{P}}_{n}({\mathcal{O}})$ with monomials of degree strictly greater than $m$.
Moreover for any mesh object ${\mathcal{O}}$ of dimension $d$ we define $$\label{eq:poly_dim}
\pi_{n, d} := \dim({{\mathbb}{P}}_{n}({\mathcal{O}})) = \dim({{\mathbb}{P}}_{n}({{\mathbb}{R}}^d)) \,,$$ and thus $\dim(\widehat{{{\mathbb}{P}}}_{n \setminus m}({\mathcal{O}})) = \pi_{n, d} -\pi_{m, d}$.
In the following the symbol $\lesssim$ will denote a bound up to a generic positive constant, independent of the mesh size $h$, but which may depend on $\Omega$, on the “polynomial” order $k$ and on the shape constant ${\varrho}$ in assumptions $\mathbf{(A1)}$, $\mathbf{(A2)}$ and $\mathbf{(A3)}$.
Vector calculus & de Rham complexes {#sub:vector_calculus}
-----------------------------------
Here below we fix some additional notation of the multivariable calculus.\
**Three dimensional operators.** In three dimensions we denote with ${\boldsymbol{x}}= (x_1, \, x_2, \, x_3)$ the independent variable. With a usual notation the symbols ${\nabla}$ and $\Delta$ denote the gradient and Laplacian for scalar functions, while ${\boldsymbol{\Delta}}$, ${\boldsymbol{\nabla}}$, ${\boldsymbol{{\varepsilon}}}$, ${{\rm div}}$ and ${\boldsymbol{{\rm curl}}}$ denote the vector Laplacian, the gradient and the symmetric gradient operator, the divergence and the curl operator for vector fields. Note that on each polyhedron $P$ the following useful polynomial decompositions hold $$\begin{aligned}
\label{eq:poly_decomposition_3D_grad}
[{{\mathbb}{P}}_n(P)]^3 &= \nabla({{\mathbb}{P}}_{n+1}(P)) \oplus ({\boldsymbol{x}}\wedge [{{\mathbb}{P}}_{n-1}(P)]^3)\,,
\\
\label{eq:poly_decomposition_3D_curl}
[{{\mathbb}{P}}_n(P)]^3 &= {\boldsymbol{{\rm curl}}}({{\mathbb}{P}}_{n+1}(P)) \oplus {\boldsymbol{x}}\,{{\mathbb}{P}}_{n-1}(P) \,.\end{aligned}$$ **Tangential operators.** Let $f$ be a face of a polyhedron $P$, we denote with ${\boldsymbol{x}}_f:= ({x_{f}}_1, \, {x_{f}}_2)$ the independent variable (i.e. a local coordinate system on $f$ associated with the axes $\boldsymbol{\tau}_f^1$ and $\boldsymbol{\tau}_f^2$). The tangential differential operators are denoted by a subscript $f$. Therefore the symbols ${\nabla}_f$ and $\Delta_f$ denote the gradient and Laplacian for scalar functions, while ${\boldsymbol{\Delta}}_f$, ${\boldsymbol{\nabla}}_f$, and ${{\rm div}}_f$ denote the vector Laplacian, the gradient operator and the divergence for vector fields on $f$ (with respect to the coordinate ${\boldsymbol{x}}_f$). Furthermore for a scalar function ${\varphi}$ and a vector field ${\boldsymbol{v}}:= (v_1, \, v_2)$ we set $${\boldsymbol{{\rm rot}}}_f \, {\varphi}:= \left( \frac{\partial {\varphi}}{\partial {x_{f}}_2}, \, -\frac{\partial {\varphi}}{\partial {x_{f}}_1} \right)
\qquad \text{and} \qquad
{{\rm rot}}_f \, {\boldsymbol{v}}:= \frac{\partial v_2}{\partial {x_{f}}_1} - \frac{\partial v_1}{\partial {x_{f}}_2} \,.$$ The following 2-d polynomial decompositions hold $$\begin{aligned}
[{{\mathbb}{P}}_n(f)]^2 &= \nabla_f({{\mathbb}{P}}_{n+1}(f)) \oplus {\boldsymbol{x}}_f^{\perp} \,{{\mathbb}{P}}_{n-1}(f) \,,
\\
[{{\mathbb}{P}}_n(f)]^2 &= {\boldsymbol{{\rm rot}}}_f({{\mathbb}{P}}_{n+1}(f)) \oplus {\boldsymbol{x}}_f \,{{\mathbb}{P}}_{n-1}(f) \,,\end{aligned}$$ where ${\boldsymbol{x}}_f^{\perp}:= ({x_{f}}_2, \, -{x_{f}}_1)$.
Given a 3-d vector valued function ${\boldsymbol{v}}$ defined in $P$, the tangential component ${\boldsymbol{v}}_{f}$ of ${\boldsymbol{v}}$ with respect to the face $f$ is defined by $${\boldsymbol{v}}_{f} := {\boldsymbol{v}}- ({\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f){\boldsymbol{n}}_P^f \,.$$ Noticing that ${\boldsymbol{v}}_{f}$ is a 3-d vector field tangent to $f$, with a slight abuse of notations we define the 2-d vector field ${\boldsymbol{v}}_{\tau}$ on $\partial P$, such that on each face $f$ its restriction to the face $f$ satisfies $${\boldsymbol{v}}_{\tau}({\boldsymbol{x}}_f) := {\boldsymbol{v}}_{f}({\boldsymbol{x}}) \,.$$ The 3-d function ${\boldsymbol{v}}$ and its 2-d tangential restriction ${\boldsymbol{v}}_{\tau}$ are related by the ${\boldsymbol{{\rm curl}}}$-${{\rm rot}}$ compatibility condition $$\label{eq:compatibility_CC_rr}
{\boldsymbol{{\rm curl}}}\, {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f = {{\rm rot}}_f \, {\boldsymbol{v}}_{\tau} \qquad \text{on any $f \in \partial P$.}$$ Moreover the Gauss theorem ensures the following ${{\rm rot}}$-tangent component relation $$\label{eq:compatibility_rr_tf}
\int_f {{\rm rot}}_f \, {\boldsymbol{v}}_{\tau} \,{\rm d}f = \int_{\partial f} {\boldsymbol{v}}\cdot {\boldsymbol{t}}_f \, {\rm d}s \qquad \text{for any $f \in \partial P$.}$$ Finally, for any scalar function $v$ defined in $P$, we denote with $v_{\tau}$ the scalar function defined in $\partial P$ such that $$v_{\tau}({\boldsymbol{x}}_f) := v({\boldsymbol{x}})_{|f} \qquad \text{on each face $f \in \partial P$.}$$
On a generic mesh object ${\mathcal{O}}$ with geometrical dimension $d$, on a face $f$ and on a polyhedron $P$ we define following the functional spaces: $$\begin{aligned}
L^2_0({\mathcal{O}}) &:= \{ v \in L^2({\mathcal{O}}) \quad \text{s.t.} \quad \int_{{\mathcal{O}}} v \, {\rm d}{\mathcal{O}}= 0 \}\\
{\boldsymbol{Z}}({\mathcal{O}}) &: \{ {\boldsymbol{v}}\in [H^1({\mathcal{O}})]^2 \quad \text{s.t.} \quad {{\rm div}}\,{\boldsymbol{v}}= 0 \quad \text{in ${\mathcal{O}}$} \} \\
{\boldsymbol H}({{\rm div}}, \, {\mathcal{O}}) &:= \{ {\boldsymbol{v}}\in [L^2({\mathcal{O}})]^d \quad \text{with} \quad {{\rm div}}\,{\boldsymbol{v}}\in L^2({\mathcal{O}}) \}
\\
{\boldsymbol H}({{\rm rot}}, \, f) &:= \{ {\boldsymbol{v}}\in [L^2(f)]^2 \quad \text{with} \quad {{\rm rot}}_f \,{\boldsymbol{v}}\in L^2(f) \}
\\
{\boldsymbol H}({\boldsymbol{{\rm curl}}}, \, P) &:= \{ {\boldsymbol{v}}\in [L^2(P)]^3 \quad \text{with} \quad {\boldsymbol{{\rm curl}}}\,{\boldsymbol{v}}\in [L^2(P)]^3 \}
\\
\boldsymbol{\Sigma}(P) &:= \{ {\boldsymbol{v}}\in [L^2(P)]^3 \quad \text{with} \quad {\boldsymbol{{\rm curl}}}\,{\boldsymbol{v}}\in [H^1(P)]^3 \}
\\
\boldsymbol{\Psi}(P) &:= \{{\boldsymbol{v}}\in H({{\rm div}}, \, P) \cap H({\boldsymbol{{\rm curl}}}, \, P)
\quad \text{s.t.} \quad {{\rm div}}\, {\boldsymbol{v}}\in H^1(P) \,, \quad
{\boldsymbol{{\rm curl}}}\, {\boldsymbol{v}}\in [H^1(P)]^3
\} \end{aligned}$$ with the “homogeneous counterparts” $$\begin{aligned}
{\boldsymbol{Z}}_0({\mathcal{O}}) &:= \{ {\boldsymbol{v}}\in {\boldsymbol{Z}}({\mathcal{O}}) \quad \text{s.t.} \quad {\boldsymbol{v}}= {\boldsymbol 0} \quad \text{on $\partial {\mathcal{O}}$} \} \\
{\boldsymbol H}_0({{\rm div}}, \, {\mathcal{O}}) &:= \{ {\boldsymbol{v}}\in H({{\rm div}}, \, {\mathcal{O}}) \quad \text{s.t.} \quad {\boldsymbol{v}}\cdot {\boldsymbol{n}}_{{\mathcal{O}}} = 0 \quad \text{on $\partial {\mathcal{O}}$} \}
\\
{\boldsymbol H}_0({{\rm rot}}, \, f) &:= \{ {\boldsymbol{v}}\in H({{\rm rot}}, \, f) \quad \text{s.t.} \quad {\boldsymbol{v}}\cdot {\boldsymbol{t}}_{f} = 0 \quad \text{on $\partial f$} \}
\\
{\boldsymbol H}_0({\boldsymbol{{\rm curl}}}, \, P) &:= \{ {\boldsymbol{v}}\in H({\boldsymbol{{\rm curl}}}, \, P) \quad \text{s.t.} \quad {\boldsymbol{v}}_{\tau} = {\boldsymbol 0} \quad \text{on $\partial P$} \}
\\
\boldsymbol{\Sigma}_0(P) &:= \{ {\boldsymbol{v}}\in \boldsymbol{\Sigma}(P) \quad \text{s.t.} \quad
{\boldsymbol{v}}_{\tau} = {\boldsymbol 0} \quad \text{and} \quad {\boldsymbol{{\rm curl}}}\, {\boldsymbol{v}}= \mathbf{0}
\quad \text{on $\partial P$} \}
\\
\boldsymbol{\Psi}_0(P) &:= \{{\boldsymbol{v}}\in \boldsymbol{\Psi}(P)
\quad \text{s.t.} \quad
\int_{\partial P} {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P \,{\rm d}f = 0 \,, \quad
{\boldsymbol{v}}_{\tau} = {\boldsymbol 0} \quad \text{and} \quad {\boldsymbol{{\rm curl}}}\, {\boldsymbol{v}}= \mathbf{0}
\, \, \text{on $\partial P$}
\} \,.\end{aligned}$$
\[rem:tau\_wedge\] Notice that for each face $f \in \partial P$, the vector fields ${\boldsymbol{v}}_f$ and ${\boldsymbol{v}}\wedge {\boldsymbol{n}}_P^f$ are different. In fact both lie in the plane of the face $f$, but ${\boldsymbol{v}}_f$ is $\pi/2$-rotation in $f$ (with respect to the axes $\boldsymbol{\tau}_1$ and $\boldsymbol{\tau}_2$) of ${\boldsymbol{v}}\wedge {\boldsymbol{n}}_P^f$. However ${\boldsymbol{v}}_f = {\boldsymbol 0}$ if and only if ${\boldsymbol{v}}\wedge {\boldsymbol{n}}_P^f = {\boldsymbol 0}$. For that reason in the definition of $\boldsymbol{\Psi}_0(P)$, we consider a slightly different, but sustantially equivalent, set of homogeneous boundary conditions to that considered in literature [@girault-raviart:book; @bendali-dominguez-gallic:1985; @amrouche-et-al:1998].
Recalling that a sequence is exact if the image of each operator coincides with the kernel of the following one, and that $P$ is contractible, from and it is easy to check that the following sequence is exact [@BBMR:2016]: $$\label{eq:poly_exact}
{{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
{{\mathbb}{P}}_{n+2}(P) \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
[{{\mathbb}{P}}_{n+1}(P)]^3 \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
[{{\mathbb}{P}}_{n}(P)]^3 \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
{{\mathbb}{P}}_{n-1}(P) \, \xrightarrow[]{ \quad 0 \quad }\,
0$$ where $i$ denotes the mapping that to every real number $r$ associates the constant function identically equal to $r$ and $0$ is the mapping that to every function associates the number $0$.
The three dimensional de Rham complex with minimal regularity (in a contractible domain $\Omega$) is provided by [@arnold-falk-winter:2006; @demkowicz-et-al:2000] $${{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
H^1(\Omega) \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
{\boldsymbol H}({\boldsymbol{{\rm curl}}}, \, \Omega) \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
{\boldsymbol H}({{\rm div}}, \, \Omega) \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
L^2(\Omega) \, \xrightarrow[]{ \quad 0 \quad }\,
0 \,.$$ In this paper we consider the de Rham sub-complex with enhanced smoothness [@evans-hughes:2013] $$\label{eq:exact}
{{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
H^1(\Omega) \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
\boldsymbol{\Sigma}(\Omega) \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
[H^1(\Omega)]^3 \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
L^2(\Omega) \, \xrightarrow[]{ \quad 0 \quad }
0 \,,$$ that is suitable for the Stokes (Navier–Stokes) problem. Therefore our goal is to construct conforming (with respect to the decomposition $\Omega_h$) virtual element spaces $$\label{eq:virtual_spaces}
W_h \subseteq H^1(\Omega) \,, \qquad
{\boldsymbol{\Sigma}}_h \subseteq {\boldsymbol{\Sigma}}(\Omega) \,, \qquad
{\boldsymbol{V}}_h \subseteq [H^1(\Omega)]^3 \,, \qquad
Q_h \subseteq L^2(\Omega)$$ that mimic the complex , i.e. are such that $$\label{eq:virtual_exact}
{{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
W_h \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
{\boldsymbol{\Sigma}}_h \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
{\boldsymbol{V}}_h \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
Q_h \, \xrightarrow[]{ \quad 0 \quad }
0$$ is an exact sub-complex of . To the best of our knowledge, no conforming finite elements sub-complex of exists (see for instance [@john-linke-merdon-neilan-rebholz:2017]).
The virtual element spaces {#sec:spaces}
==========================
The present section is devoted to the construction of conforming virtual element spaces that compose the virtual sub-complex . As we will see, the space $W_h$ consists of the lowest degree three dimensional nodal VEM space [@projectors; @BDR:2017], whereas the spaces ${\boldsymbol{V}}_h$ and $Q_h$ (that are the spaces actually used in the discretization of the problem) are the three dimensional counterparts of the inf-sup stable couple of spaces introduced in [@vacca:2018; @BLV:2018]. Therefore the main novelty of the present section is in the construction of the ${\boldsymbol{\Sigma}}$-conforming space ${\boldsymbol{\Sigma}}_h$. In order to facilitate the reading, we present the spaces in the reverse order, from right to left in the sequence . In particular, in accordance with , the space ${\boldsymbol{\Sigma}}_h$ will be careful designed to fit ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\Sigma}}_h \subseteq {\boldsymbol{V}}_h$. We stress that the readers mainly interested on the virtual elements approximation of the three dimensional Navier–Stokes equation (and not on the virtual de Rham sequence) can skip Subsection \[sub:Sg\_h\], Subsection \[sub:W\_h\] and Section \[sec:derham\].
One essential idea in the VEM construction is to define suitable (computable) polynomial projections. For any $n \in {{\mathbb}{N}}$ and each polyhedron/face ${\mathcal{O}}$ we introduce the following polynomial projections:
- the $\boldsymbol{L^2}$**-projection** $\Pi_n^{0, {\mathcal{O}}} \colon L^2({\mathcal{O}}) \to {{\mathbb}{P}}_n({\mathcal{O}})$, defined for any $v \in L^2({\mathcal{O}})$ by $$\label{eq:P0_k^E}
\int_{{\mathcal{O}}} q_n (v - \, {\Pi}_{n}^{0, {\mathcal{O}}} v) \, {\rm d} {\mathcal{O}}= 0 \qquad \text{for all $q_n \in {{\mathbb}{P}}_n({\mathcal{O}})$,}$$ with obvious extension for vector functions $\Pi_n^{0, {\mathcal{O}}} \colon [L^2({\mathcal{O}})]^3 \to [{{\mathbb}{P}}_n({\mathcal{O}})]^3$, and tensor functions $\boldsymbol{\Pi}_{n}^{0, {\mathcal{O}}} \colon [L^2({\mathcal{O}})]^{3 \times 3} \to [{{\mathbb}{P}}_{n}({\mathcal{O}})]^{3 \times 3}$,
- the $\boldsymbol{H^1}$**-seminorm projection** ${\Pi}_{n}^{\nabla,{\mathcal{O}}} \colon H^1({\mathcal{O}}) \to {{\mathbb}{P}}_n({\mathcal{O}})$, defined for any $v \in H^1({\mathcal{O}})$ by $$\label{eq:Pn_k^E}
\left\{
\begin{aligned}
& \int_{{\mathcal{O}}} {\nabla}\,q_n \cdot {\nabla}( v - \, {\Pi}_{n}^{\nabla,{\mathcal{O}}} v) \, {\rm d} {\mathcal{O}}= 0 \qquad \text{for all $q_n \in {{\mathbb}{P}}_n({\mathcal{O}})$,} \\
& \int_{\partial {\mathcal{O}}} (v - \, {\Pi}_{n}^{\nabla, {\mathcal{O}}} v) \,{\rm d}\sigma = 0 \, ,
\end{aligned}
\right.$$ with obvious extension for vector functions $\Pi_n^{\nabla, {\mathcal{O}}} \colon [H^1({\mathcal{O}})]^3 \to [{{\mathbb}{P}}_n({\mathcal{O}})]^3$.
Let $k \geq 2$ be the polynomial degree of accuracy of the method. We recall that, in standard finite element fashion, the virtual element spaces are defined element-wise and then are assembled in such a way the global regularity requirements are satisfied.
Scalar $L^2$-conforming space {#sub:Q_h}
-----------------------------
We start our construction with the rightmost discrete space $Q_h$ in . Since we are not requiring any smoothness on $Q_h$, the local space $Q_h(P)$ is simply defined by $$Q_h(P) := {{\mathbb}{P}}_{k-1}(P) \,,$$ having dimension (cf. ) $\dim(Q_h(P)) = \pi_{k-1, 3}$. The corresponding DoFs are chosen, defining for each $q \in Q_h(P)$ the following linear operators
- $\mathbf{D}_{Q}$: the moments up to order $k-1$ of $q$, i.e., $$\int_P q \, p_{k-1} \,{\rm d}P \qquad \text{for any $p_{k-1} \in {{\mathbb}{P}}_{k-1}(P)$.}$$
The global space is given by $$\label{eq:Q_h}
Q_h := \{ q \in L^2(\Omega) \quad \text{s.t.} \quad q_{|P} \in Q_h(P) \quad \text{for all $P \in \Omega_h$} \} \,.$$ It is straightforward to see that the dimension of $Q_h$ is $$\label{eq:dim_Q_h}
\dim(Q_h) = \pi_{k-1, 3} \,L_P \,.$$
Vector $H^1$-conforming VEM space {#sub:V_h}
---------------------------------
The subsequent space in the de Rham complex is the vector-valued $H^1$-conforming virtual element space ${\boldsymbol{V}}_h$. The construction of ${\boldsymbol{V}}_h$ has to combine two main ingredients:
- to define a 3-d version of the space [@BLV:2018] that fits the conformity requirement (that definition follows the guidelines of Appendix of reference [@BLV:2017]);
- “to play” with the enhanced technique [@projectors] in order to achieve the computability of the polynomial projections stated in Proposition \[prp:proj\].
We first consider on each face $f$ of the element $P$, the face space $$\label{eq:Bhat_h^n}
\begin{aligned}
\widehat{{{\mathbb}{B}}}_k(f):= \biggl\{
v \in H^1(f) \, \, \, \, \text{s.t.} \, \, \, \,
& v_{|\partial f} \in C^0(\partial f) \,,
\quad
{v}_{|e} \in {{\mathbb}{P}}_k(e) \quad \text{for all $e \in \partial f$,}
\\
& \Delta_f \, v \in {{\mathbb}{P}}_{k+1}(f) \,,
\\
& \left( v - \Pi_k^{\nabla, f} v , \, \widehat{p}_{k+1} \right)_f = 0\quad
\text{for all $\widehat{p}_{k+1} \in \widehat{{{\mathbb}{P}}}_{k+1 \setminus k-2}(f)$}
\biggr \}
\end{aligned}$$ and the boundary space $$\label{eq:Bhatf}
\widehat{{{\mathbb}{B}}}_k(\partial P):= \left\{ v \in C^0(\partial P) \quad \text{such that} \quad
v_{|f} \in \widehat{{{\mathbb}{B}}}_k(f)
\quad \text{for any $f \in \partial P$}
\right\}$$ that is a modification of the standard boundary nodal VEM [@BBMR:2016]. Indeed the “super-enhanced” constraints (the last line in the definition ) are needed to exactly compute the polynomial projection $\Pi_{k+1}^{0, f}$ (see Proposition \[prp:proj\]).
On the polyhedron $P$ we define the virtual element space ${\boldsymbol{V}}_h(P)$ $$\label{eq:V_h^P}
\begin{aligned}
{\boldsymbol{V}}_h(P) := \biggl\{
{\boldsymbol{v}}\in [H^1(P)]^3 \, \, \, \text{s.t.} \, \, \,
& {\boldsymbol{v}}_{|\partial P} \in [\widehat{{{\mathbb}{B}}}_k(\partial P)]^3 \,,
\\
& \biggl\{
\begin{aligned}
& \boldsymbol{\Delta} {\boldsymbol{v}}+ \nabla s \in {\boldsymbol{x}}\wedge [{{\mathbb}{P}}_{k-1}(P)]^3, \\
& {\rm div} \, {\boldsymbol{v}}\in {{\mathbb}{P}}_{k-1}(P),
\end{aligned}
\biggr. \quad \text{ for some $s \in L^2_0(P)$}
\\
\biggl.
& \left( {\boldsymbol{v}}- \Pi_k^{\nabla, P} {\boldsymbol{v}}, \, {\boldsymbol{x}}\wedge \widehat{{\boldsymbol{p}}}_{k-1} \right)_P = 0 \quad
\text{for all $\widehat{{\boldsymbol{p}}}_{k-1} \in [\widehat{{{\mathbb}{P}}}_{k-1 \setminus k-3}(P)]^3$}
\biggr \}.
\end{aligned}$$ The definition above is the 3-d counterpart of the virtual elements [@BLV:2018], in particular we remark that the enhancing constraints (the last line in ) are necessary to achieve the computability of the $L^2$-projection $\Pi_k^{0, P}$(see Proposition \[prp:proj\]). Moreover, notice that the space ${\boldsymbol{V}}_h(P)$ contains $[{{\mathbb}{P}}_k(P)]^3$ and this will guarantee the good approximation property of the space (cf. Theorem \[thm:u\]).
\[prp:V\_h\^E\_dofs\] The dimension of ${\boldsymbol{V}}_h(P)$ is given by $$\dim({\boldsymbol{V}}_h(P)) =
3 \, \ell_V + 3 \, (k-1)\, \ell_e + 3 \, \pi_{k-2, 2} \, \ell_f + 3 \, \pi_{k-2, 3} \,.$$ Moreover, the following linear operators $\mathbf{D}_{{\boldsymbol{V}}}$, split into five subsets constitute a set of DoFs for ${\boldsymbol{V}}_h(P)$:
- $\mathbf{D^1}_{{\boldsymbol{V}}}$: the values of ${\boldsymbol{v}}$ at the vertexes of the polyhedron $P$,
- $\mathbf{D^2}_{{\boldsymbol{V}}}$: the values of ${\boldsymbol{v}}$ at $k-1$ distinct points of every edge $e$ of the polyhedron $P$,
- $\mathbf{D^3}_{{\boldsymbol{V}}}$: the face moments of ${\boldsymbol{v}}$ (split into normal and tangential components) $$\int_f ({\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f) \, p_{k-2} \, {\rm d}f \,, \qquad
\int_f ({\boldsymbol{v}}\cdot \boldsymbol{\tau}_1^f) \, p_{k-2} \, {\rm d}f \,, \qquad
\int_f ({\boldsymbol{v}}\cdot \boldsymbol{\tau}_2^f) \, p_{k-2} \, {\rm d}f \,,$$ for all $p_{k-2} \in {{\mathbb}{P}}_{k-2}(f)$,
- $\mathbf{D^4}_{{\boldsymbol{V}}}$: the volume moments of ${\boldsymbol{v}}$ $$\int_P {\boldsymbol{v}}\cdot ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-3})\, {\rm d}P \qquad
\text{for all ${\boldsymbol{p}}_{k-3} \in [{{\mathbb}{P}}_{k-3}(P)]^3$,}$$
- $\mathbf{D^5}_{{\boldsymbol{V}}}$: the volume moments of ${\rm div} \,{\boldsymbol{v}}$ $$\int_P ({\rm div} \,{\boldsymbol{v}}) \, \widehat{p}_{k-1} \, {\rm d}P \qquad \text{for all $\widehat{p}_{k-1} \in \widehat{{{\mathbb}{P}}}_{k-1 \setminus 0}(P)$.}$$
We only sketch the proof since it follows the guidelines of Proposition 3.1 in [@vacca:2018] for the analogous 2-d space. First of all, recalling and polynomial decomposition , simple computations yield $$\label{eq:D_V}
\begin{gathered}
\texttt{number} (\mathbf{D^1}_{{\boldsymbol{V}}}) = 3 \, \ell_V \,,
\qquad
\texttt{number} (\mathbf{D^2}_{{\boldsymbol{V}}}) = 3 \, (k-1) \,\ell_e \,,
\qquad
\texttt{number} (\mathbf{D^3}_{{\boldsymbol{V}}}) = 3 \, \pi_{k-2, 2} \, \ell_f \,,
\\
\texttt{number} (\mathbf{D^4}_{{\boldsymbol{V}}}) = 3 \, \pi_{k-2, 3} - \pi_{k-1, 3} + 1 \,,
\qquad
\texttt{number} (\mathbf{D^5}_{{\boldsymbol{V}}}) = \pi_{k-1, 3} -1 \,,
\end{gathered}$$ and therefore $$\texttt{number} (\mathbf{D}_{{\boldsymbol{V}}}) = 3 \, \ell_V + 3 \, (k-1)\, \ell_e + 3 \, \pi_{k-2, 2} \, \ell_f + 3 \, \pi_{k-2, 3} \,.$$ Now employing Proposition 2 and Remark 5 in [@projectors], it can be shown that the DoFs $\mathbf{D^1}_{{\boldsymbol{V}}}$, $\mathbf{D^2}_{{\boldsymbol{V}}}$, $\mathbf{D^3}_{{\boldsymbol{V}}}$ are unisolvent for the space $[\widehat{{{\mathbb}{B}}}_k(\partial P)]^3$. Therefore it holds that $$\label{eq:dofs3}
\dim([\widehat{{{\mathbb}{B}}}_k(\partial P)]^3) =
3 \, \ell_V + 3 \, (k - 1) \, \ell_e + 3 \, \pi_{k-2, 2} \, \ell_f \,,$$ which in turn implies (recalling ) $$\dim({\boldsymbol{V}}_h(P)) \geq \texttt{number} (\mathbf{D}_{{\boldsymbol{V}}})
\,.$$ Now the result follows by proving that $\mathbf{D}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = {\boldsymbol 0}$ implies that ${\boldsymbol{v}}$ is identically zero, that can be shown first works on $\partial P$ and then inside $P$. As a consequence the linear operators $\mathbf{D}_{{\boldsymbol{V}}}$ are unisolvent for ${\boldsymbol{V}}_h$ and in particular $\dim({\boldsymbol{V}}_h(P)) = \texttt{number} (\mathbf{D}_{{\boldsymbol{V}}})$.
The global space ${\boldsymbol{V}}_h$ is defined by gluing the local spaces with the obvious associated sets of global DoFs: $$\label{eq:V_h}
{\boldsymbol{V}}_h := \{{\boldsymbol{v}}\in [H^1(\Omega)]^3 \quad \text{s.t.} \quad {\boldsymbol{v}}_{|P} \in {\boldsymbol{V}}_h(P) \} \,.$$ The dimension of ${\boldsymbol{V}}_h$ is given by $$\label{eq:dim_V_h}
\dim({\boldsymbol{V}}_h) =
3 \, L_V + 3 \, (k-1)\, L_e + 3 \, \pi_{k-2, 2} \, L_f + 3 \, \pi_{k-2, 3} \, L_P \,.$$ We also consider the discrete local kernel $${\boldsymbol{Z}}_h(P) := \left \{
{\boldsymbol{v}}\in {\boldsymbol{V}}_h(P) \quad \text{s.t.} \quad \int_{P} {{\rm div}}\, {\boldsymbol{v}}\, q \,{\rm dP}= 0 \quad \text{for all $q \in Q_h(P)$}
\right\}\,,$$ and the corresponding global version $$\label{eq:Z_h}
{\boldsymbol{Z}}_h := \left\{ {\boldsymbol{v}}\in {\boldsymbol{V}}_h \quad \text{s.t.} \quad \int_{\Omega} {{\rm div}}\, {\boldsymbol{v}}\, q \,{\rm d\Omega}= 0 \quad \text{for all $q \in Q_h$} \right\} \,.$$ A crucial observation is that, extending to the 3-d case the result in [@BLV:2017], the proposed discrete spaces and are such that $
{{\rm div}}\, {\boldsymbol{V}}_h \subseteq Q_h
$. As a consequence the considerable kernel inclusion holds $$\label{eq:kernel_inclusion}
{\boldsymbol{Z}}_h \subseteq {\boldsymbol{Z}}\,.$$ The inclusion here above and explicit computations (cf. ) yield that $$\dim({\boldsymbol{Z}}_h) =
3 \, L_V + 3 \, (k-1)\, L_e + 3 \, \pi_{k-2, 2} \, L_f + (3 \, \pi_{k-2, 3} - \pi_{k-1, 3}) \, N_P \,.$$ The notable property leads to a series of important advantages, as explored in [@linke-merdon:2016; @john-linke-merdon-neilan-rebholz:2017; @BLV:2018].
\[rm:projectionPD\] In the third line of Definition the $H^1$-seminorm projection $\Pi^{\nabla, P}_k$ can be actually replaced by any polynomial projection $\Pi^{P}_k$ that is computable on the basis of the DoFs $\mathbf{D}_{{\boldsymbol{V}}}$ (in the sense of Proposition \[prp:proj\]). This change clearly propagates throughout the rest of the analysis (see Definitions and ). An analogous observation holds also for the operator $\Pi^{\nabla, f}_k$ in the third line of Definition . The present remark allows to make use of computationally cheaper projections, as done in the numerical tests of Section \[sec:num\_test\].
Vector ${\boldsymbol{\Sigma}}$-conforming VEM space {#sub:Sg_h}
---------------------------------------------------
In the present subsection we consider the construction of the ${\boldsymbol{\Sigma}}$-conforming virtual space ${\boldsymbol{\Sigma}}_h$ in . As mentioned before, this brick constitutes the main novelty in the foundation of the virtual de Rham sequence . The core ideas in building such space are the following:
- the space is careful designed to satisfy ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\Sigma}}_h = {\boldsymbol{Z}}_h$;
- the DoFs are conveniently chosen in order to have a direct correspondence between the ${\boldsymbol{{\rm curl}}}$ of the Lagrange-type basis functions of ${\boldsymbol{\Sigma}}_h$ and the Lagrange basis functions of ${\boldsymbol{V}}_h$;
- the boundary space and the boundary DoFs are picked in accordance with the global conformity requirements ensuing from the regularity of the space ${\boldsymbol{\Sigma}}$.
We start by introducing on each face $f \in \partial P$ the face space $$\label{eq:Sf}
\begin{aligned}
\boldsymbol{S}_k(f) := \biggl\{
{\boldsymbol{\sigma}}\in {\boldsymbol H}({{\rm div}}_f, \, f) \cap {\boldsymbol H}({{\rm rot}}_f, \, f)
\quad \text{s.t.} \quad
& ({{\boldsymbol{\sigma}}} \cdot {\boldsymbol{t}}_e)_{|e} \in {{\mathbb}{P}}_0(e) &
\text{$\forall e \in \partial f$,}
\\
& {{\rm div}}_f \, {\boldsymbol{\sigma}}= 0 \,,
\\
& {{\rm rot}}_f \, {\boldsymbol{\sigma}}\in \widehat{{{\mathbb}{B}}}_k(f)
& & \quad \biggr\} \,,
\end{aligned}$$ and the boundary space $$\label{eq:S}
\begin{aligned}
{\boldsymbol S}_k(\partial P):= \biggl\{
{\boldsymbol{\sigma}}\in [L^2(\partial P)]^3 \quad \text{s.t.} \quad
&{\boldsymbol{\sigma}}_{\tau} \in {\boldsymbol S}_k(f)
&\text{for any $f \in \partial P$,}
\\
& ({{\boldsymbol{\sigma}}_{f_1}} \cdot {\boldsymbol{t}}_e)_{|e} = ({{\boldsymbol{\sigma}}_{f_2}} \cdot {\boldsymbol{t}}_e)_{|e}
&\forall
e \subseteq \partial f_1 \cap \partial f_2 \,, &\quad f_1, f_2 \in \partial P
\biggr \} \,.
\end{aligned}$$ Concerning the differential problem in definition , we recall that on simply connected polygon $f$, given two sufficiently regular functions ${\boldsymbol g}$ and ${\boldsymbol h}$ defined on $f$ and a sufficiently regular function $\omega$ defined on $\partial f$, the problem $$\label{eq:hdivhrot}
\left \{
\begin{aligned}
& \text{find ${\boldsymbol{\sigma}}\in {\boldsymbol H}({{\rm div}}_f, \, f) \cap {\boldsymbol H}({{\rm rot}}_f, \, f)$ s.t.}
\\
&
\begin{aligned}
& {{\rm div}}_f \, {\boldsymbol{\sigma}}= {\boldsymbol g} \qquad &\text{in $f$,} \\
& {{\rm rot}}_f \, {\boldsymbol{\sigma}}= {\boldsymbol h} \qquad &\text{in $f$,} \\
& {\boldsymbol{\sigma}}\cdot {\boldsymbol{t}}_f = \omega \qquad &\text{on $\partial f$,}
\end{aligned}
\end{aligned}
\right .$$ is well posed if and only if, in accordance with , the following holds $$\label{eq:compatibility_rr_tf_prb}
\int_{f} {\boldsymbol h} \, {\rm d}f = \int_{\partial f} \omega \, {\rm d}s \,.$$ On the polyhedron $P$ we define the virtual space: $$\label{eq:Sg_h^P}
\begin{aligned}
{\boldsymbol{\Sigma}}_h(P) := \biggl\{
&{\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}(P) \quad \text{s.t.} \quad
{\boldsymbol{{\varphi}}}_{|\partial P} \in {\boldsymbol S}_k(\partial P) \, ,
\quad
\int_{\partial P} {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P \, {\rm d}f = 0 \,,
\biggr.
\\
& \qquad ({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{|\partial P} \in [\widehat{{{\mathbb}{B}}}_k(\partial P)]^3 \,,
\\
& \qquad
\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P =
\int_P \widetilde{{\boldsymbol{p}}}_{k-1} \cdot {\boldsymbol{\psi}}\, {\rm d}P,
\qquad
\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,}
\\
& \qquad \qquad \qquad \qquad \qquad
\text{for some $\widetilde{{\boldsymbol{p}}}_{k-1} \in [{{\mathbb}{P}}_{k-1}(P)]^3 \cap {\boldsymbol{Z}}(P)$
,}
\\
\biggl.
& \qquad \left( {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}- \Pi_k^{\nabla, P} {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}, \, {\boldsymbol{x}}\wedge \widehat{{\boldsymbol{p}}}_{k-1} \right)_P = 0\quad
\text{$\forall \, \widehat{{\boldsymbol{p}}}_{k-1} \in [\widehat{{{\mathbb}{P}}}_{k-1 \setminus k-3}(P)]^3$}
\biggr \}.
\end{aligned}$$ We stress that the variational problem stated in is coupled with the non homogeneous version of the boundary conditions in [@girault-raviart:book; @amrouche-et-al:1998]. In fact, in order to force ${\boldsymbol{\Sigma}}$-conforming regularity, for any function ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h(P)$, we need to prescribe ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ and ${\boldsymbol{{\varphi}}}_{\tau}$ on $\partial P$. We address the well-posedness of the biharmonic problem in definition in the Appendix.
Note that, in accordance with the target ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\Sigma}}_h \subseteq {\boldsymbol{V}}_h$, the second and the last line in definition are the ${\boldsymbol{{\rm curl}}}$ version of the first and last line in definition . Whereas we will see that ${\boldsymbol{{\rm curl}}}$ of the solutions of the biharmonic problem in are solutions to the Stokes problem in (see Proposition \[prp:inclusion2\]).
\[prp:Sg\_h\] The dimension of ${\boldsymbol{\Sigma}}_h(P)$ is given by $$\dim({\boldsymbol{\Sigma}}_h(P)) =
3 \, \ell_V + (3 k-2)\, \ell_e + (3 \, \pi_{k-2, 2} -1)\, \ell_f + 3 \, \pi_{k-2, 3} - \pi_{k-1, 3} + 1 \,.$$
Moreover, the following linear operators $\mathbf{D}_{{\boldsymbol{\Sigma}}}$, split into five subsets constitute a set of DoFs for ${\boldsymbol{\Sigma}}_h(P)$:
- $\mathbf{D^1}_{{\boldsymbol{\Sigma}}}$: the values of ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ at the vertexes of the polyhedron $P$,
- $\mathbf{D^2}_{{\boldsymbol{\Sigma}}}$: the values of ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ at $k-1$ distinct points of every edge $e$ of the polyhedron $P$,
- $\mathbf{D^3}_{{\boldsymbol{\Sigma}}}$: the face moments of ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ (split into normal and tangential components) $$\begin{gathered}
\int_f ({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P^f) \, \widehat{p}_{k-2} \, {\rm d}f \,, \qquad
\text{for all $\widehat{p}_{k-2} \in \widehat{{{\mathbb}{P}}}_{k-2 \setminus 0}(f)$,}
\\
\int_f ({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\cdot \boldsymbol{\tau}_1^f) \, p_{k-2} \, {\rm d}f \,, \qquad
\int_f ({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\cdot \boldsymbol{\tau}_2^f) \, p_{k-2} \, {\rm d}f \,,
\qquad \text{for all $p_{k-2} \in {{\mathbb}{P}}_{k-2}(f)$},\end{gathered}$$
- $\mathbf{D^4}_{{\boldsymbol{\Sigma}}}$: the volume moments of ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ $$\int_P {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\cdot ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-3})\, {\rm d}P \qquad
\text{for all ${\boldsymbol{p}}_{k-3} \in [{{\mathbb}{P}}_{k-3}(P)]^3$,}$$
- $\mathbf{D^5}_{{\boldsymbol{\Sigma}}}$: the edge mean value of ${\boldsymbol{{\varphi}}}\cdot {\boldsymbol{t}}_e$, i.e. $$\frac{1}{|e|} \, \int_e {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{t}}_e \,{\rm d}s \,.$$
We start the proof counting the number of the linear operators $\mathbf{D}_{{\boldsymbol{\Sigma}}}$. Using similar computations as in we have: $$\begin{gathered}
\texttt{number} (\mathbf{D^1}_{{\boldsymbol{\Sigma}}}) = 3 \, \ell_V \,,
\qquad
\texttt{number} (\mathbf{D^2}_{{\boldsymbol{\Sigma}}}) = 3 \, (k-1) \,\ell_e \,,
\qquad
\texttt{number} (\mathbf{D^3}_{{\boldsymbol{\Sigma}}}) = (3 \, \pi_{k-2, 2} -1) \, \ell_f \,,
\\
\texttt{number} (\mathbf{D^4}_{{\boldsymbol{\Sigma}}}) = 3 \, \pi_{k-2, 3} - \pi_{k-1, 3} + 1 \,,
\qquad
\texttt{number} (\mathbf{D^5}_{{\boldsymbol{\Sigma}}}) = \ell_e\,,
\end{gathered}$$ and thus $$\label{eq:numberDSg}
\texttt{number} (\mathbf{D}_{{\boldsymbol{\Sigma}}}) = 3 \, \ell_V + (3 k-2)\, \ell_e + (3 \, \pi_{k-2, 2} -1)\, \ell_f + 3 \, \pi_{k-2, 3} - \pi_{k-1, 3} + 1 \,.$$ For sake of simplicity, we prove that $\mathbf{D}_{{\boldsymbol{\Sigma}}}$ constitutes a set of DoFs for the non-enhanced space associated with ${\boldsymbol{\Sigma}}_h(P)$, i.e. the space $\widetilde{{\boldsymbol{\Sigma}}}_h(P)$ obtained by dropping the last line in (the enhanced constraints) and by taking in the biharmonic system $\widetilde{{\boldsymbol{p}}}_{k-3} \in [{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P)$, i.e. the space $$\begin{aligned}
\widetilde{{\boldsymbol{\Sigma}}}_h(P) := \biggl\{
&{\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}(P) \quad \text{s.t.} \quad
{\boldsymbol{{\varphi}}}_{|\partial P} \in {\boldsymbol S}_k(\partial P) \, ,
\quad
\int_{\partial P} {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P \, {\rm d}f = 0 \,,
\biggr.
\\
& \qquad ({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{|\partial P} \in [\widehat{{{\mathbb}{B}}}_k(\partial P)]^3\,,
\\
& \qquad
\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P =
\int_P \widetilde{{\boldsymbol{p}}}_{k-3} \cdot {\boldsymbol{\psi}}\, {\rm d}P,
\qquad
\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,}
\\
& \qquad \qquad \qquad \qquad \qquad
\text{for some $\widetilde{{\boldsymbol{p}}}_{k-3} \in [{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P)$}
\biggr \}\,.
\end{aligned}$$ Once the proof for $\widetilde{{\boldsymbol{\Sigma}}}_h(P)$ is given, the extension to the original space ${\boldsymbol{\Sigma}}_h(P)$ easily follows by employing standard techniques for VEM enhanced spaces (see [@projectors] and Proposition 5.1 in [@BMV:2018]).
Employing Theorem \[thm:wellposed\], given
- $\widetilde{{\boldsymbol{p}}}_{k-3} \in [{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P)$,
- ${\boldsymbol g} \in [\widehat{{{\mathbb}{B}}}_k(\partial P)]^3$,
- ${\boldsymbol h} \in {\boldsymbol S}_k(\partial P)$ satisfying the compatibility condition (cf. ) $$\label{eq:dofs0}
{\boldsymbol g} \cdot {\boldsymbol{n}}_P^f = {{\rm rot}}_f \, {\boldsymbol h}_{\tau} \quad \text{on any $f \in \partial P$,}$$
there exists a unique function ${\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}(P)$ such that $$\left \{
\begin{aligned}
&\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P =
\int_P \widetilde{{\boldsymbol{p}}}_{k-3} \cdot {\boldsymbol{\psi}}\, {\rm d}P,
\qquad
&\text{for all ${\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,}
\\
&\int_{\partial P} {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P \,{\rm d}f = 0 \,,
\\
&{\boldsymbol{{\varphi}}}_{\tau} = {\boldsymbol h}_{\tau} \qquad &\text{on $\partial P$,}
\\
&{\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\boldsymbol g} \qquad &\text{on $\partial P$.}
\end{aligned}
\right .$$ Therefore $$\label{eq:dofs1}
\begin{split}
\dim(\widetilde{{\boldsymbol{\Sigma}}}_h(P))
&= \dim([{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P)) +
\dim([\widehat{{{\mathbb}{B}}}_k(\partial P)]^3) +
\dim({\boldsymbol S}_k(\partial P))
- \dim(\widehat{{{\mathbb}{B}}}_k(f)) \, \ell_f
\end{split}$$ where the last term $(-\dim(\widehat{{{\mathbb}{B}}}_k(f)) \, \ell_f)$ ensues from the compatibility condition . We calculate the addenda in the right hand side of . Regarding the first term in , we preliminary note that the following characterization ensues from the exact sequence and polynomial decomposition $$\label{eq:z_CC}
[{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P) = {\boldsymbol{{\rm curl}}}\, \left( [{{\mathbb}{P}}_{k-2}(P)]^3\right) =
{\boldsymbol{{\rm curl}}}\left( {\boldsymbol{x}}\wedge [{{\mathbb}{P}}_{k-3}(P)]^3 \right) \,.$$ Employing again the exact sequence , ${\boldsymbol{{\rm curl}}}$ restricted to $\left( {\boldsymbol{x}}\wedge [{{\mathbb}{P}}_{k-3}(P)]^3 \right)$ is actually an isomorphism, therefore from and follows that $$\label{eq:dofs2}
\dim([{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P)) =
\dim \left( {\boldsymbol{x}}\wedge [{{\mathbb}{P}}_{k-3}(P)]^3\right)=
3 \, \pi_{k-2, 3} - \pi_{k-1, 3} + 1 \,.$$ From definitions and and since problem is well-posed, direct computations yield $$\label{eq:dofs4}
\dim({\boldsymbol S}_k(\partial P)) = \ell_e + (\dim(\widehat{{{\mathbb}{B}}}_k(f)) - 1) \,
\ell_f$$ where the $-1$ in the formula above is due to the compatibility condition .
Collecting , and in (compare with ) we get $$\dim(\widetilde{{\boldsymbol{\Sigma}}}_h(P)) = \texttt{number} (\mathbf{D}_{{\boldsymbol{\Sigma}}}) \,.$$ Having proved that $\texttt{number} (\mathbf{D}_{{\boldsymbol{\Sigma}}})$ is equal to $\dim(\widetilde{{\boldsymbol{\Sigma}}}_h(P))$, in order to validate that the linear operators $\mathbf{D}_{{\boldsymbol{\Sigma}}}$ constitute a set of DoFs for $\widetilde{{\boldsymbol{\Sigma}}}_h(P)$ we have to check that they are unisolvent. Let ${\boldsymbol{{\varphi}}}\in \widetilde{{\boldsymbol{\Sigma}}}_h(P)$ such that $\mathbf{D}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$, we need to show that ${\boldsymbol{{\varphi}}}$ is identically zero. It is straightforward that $\mathbf{D^1}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = \mathbf{D^2}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$ implies $$\label{eq:unisolvency1}
({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{|\partial f} = {\boldsymbol 0} \qquad \text{for any $f \in \partial P$.}$$ Recalling the well known results for nodal boundary spaces [@BBMR:2016], it is quite obvious to check that $\mathbf{D^3}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$ implies $$\label{eq:unisolvency2}
({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{f} = {\boldsymbol 0} \qquad \text{for any $f \in \partial P$.}$$ In order to get also the normal component of $({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{|f}$ equal to zero, based on $\mathbf{D^3}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$, it is sufficient to observe that the compatibility conditions and give $$\label{eq:unisolvency2a}
\begin{split}
\int_f {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P^f \, {\rm d}f &=
\int_{f} {{\rm rot}}_f \, {\boldsymbol{{\varphi}}}_{\tau} \,{\rm d}f =
\int_{\partial f} {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{t}}_f \,{\rm d}s
\\
&=
\sum_{e \in \partial f} \int_{ e} {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{t}}_f^e \,{\rm d}s =
\sum_{e \in \partial f} |e| \, \mathbf{D^5}_{{\boldsymbol{\Sigma}}, e}({\boldsymbol{{\varphi}}}) \, {\boldsymbol{t}}_e \cdot {\boldsymbol{t}}_f^e
\qquad \text{for any $f \in \partial P$.}
\end{split}$$ that is equal to 0 since $\mathbf{D^5}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$. Therefore we have proved that $$\label{eq:unisolvency3}
{\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\boldsymbol 0} \qquad \text{on $\partial P$.}$$ Moreover, being $\mathbf{D^5}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$, from and and definition for any $f \in \partial P$ we infer $$({\boldsymbol{{\varphi}}}_f \cdot {\boldsymbol{t}}_e)_{|e} = 0 \quad \text{for any $e \in \partial f$,}
\qquad
{{\rm div}}_f \, {\boldsymbol{{\varphi}}}_{\tau} = 0 \quad \text{and} \quad
{{\rm rot}}_f \, {\boldsymbol{{\varphi}}}_{\tau} = 0 \quad \text{on $f$,}$$ and thus, being well-posed, we obtain $$\label{eq:unisolvency4}
{\boldsymbol{{\varphi}}}_{\tau} = {\boldsymbol 0} \qquad \text{for any $f \in \partial P$.}$$ Finally, by definition of $\widetilde{{\boldsymbol{\Sigma}}}_h(P)$, there exists $\widetilde{{\boldsymbol{p}}}_{k-3} \in [{{\mathbb}{P}}_{k-3}(P)]^3 \cap {\boldsymbol{Z}}(P)$ such that $$\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P =
\int_P \widetilde{{\boldsymbol{p}}}_{k-3} \cdot {\boldsymbol{\psi}}\, {\rm d}P
\qquad
\text{for all ${\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$.}$$ Therefore, being ${\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}_0(P)$ (cf. and ) we infer $$\label{eq:unisolvency6}
\begin{aligned}
\|{\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\|_{0, P}^2 & =
\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\, {\rm d}P =
\int_P \widetilde{{\boldsymbol{p}}}_{k-3} \cdot {\boldsymbol{{\varphi}}}\, {\rm d}P
\\
& = \int_P {\boldsymbol{{\rm curl}}}\, ({\boldsymbol{x}}\wedge {\boldsymbol q}_{k-3}) \cdot {\boldsymbol{{\varphi}}}\, {\rm d}P
\quad & \text{(characterization \eqref{eq:z_CC})}
\\
& = \int_P {\boldsymbol{x}}\wedge {\boldsymbol q}_{k-3} \cdot {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\, {\rm d}P
\quad & \text{(integration by parts + \eqref{eq:unisolvency4})}
\end{aligned}$$ and thus, since $\mathbf{D^4}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = {\boldsymbol 0}$, we obtain $
\|{\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\|_{0, P}^2 = 0
$. Now the proof follows by the fact that $\|{\boldsymbol{\Delta}}\,\cdot\|_{0, P}$ is a norm on $\boldsymbol{\Psi}_0(P)$ (see Lemma 5.2 in [@girault-raviart:book] and ).
\[rm:Sg\_h\] A careful inspection of Theorem \[thm:wellposed\] (see also Remark 5.1 in [@girault-raviart:book] and [@bendali-dominguez-gallic:1985]) reveals that the space admits the equivalent formulation $$\label{eq:Sg_h^Pnew}
\begin{aligned}
{\boldsymbol{\Sigma}}_h(P) := \biggl\{
&{\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}(P) \quad \text{s.t.} \quad
{\boldsymbol{{\varphi}}}_{|\partial P} \in {\boldsymbol S}_k(\partial P) \,,
\quad
({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{|\partial P} \in [\widehat{{{\mathbb}{B}}}_k(\partial P)]^3 \,,
\\
& \qquad
{\boldsymbol{\Delta}}^2 \, {\boldsymbol{{\varphi}}}\in [{{\mathbb}{P}}_{k-1}(P)]^3 \cap {\boldsymbol{Z}}(P) \,,
\\
& \qquad
{{\rm div}}\, {\boldsymbol{{\varphi}}}= 0 \,,
\\
\biggl.
& \qquad
\left( {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}- \Pi_k^{\nabla, P} {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}, \, {\boldsymbol{x}}\wedge \widehat{{\boldsymbol{p}}}_{k-1} \right)_P = 0\quad
\text{$\forall \, \widehat{{\boldsymbol{p}}}_{k-1} \in [\widehat{{{\mathbb}{P}}}_{k-1 \setminus k-3}(P)]^3$}
\biggr \} \,.
\end{aligned}$$
The global space ${\boldsymbol{\Sigma}}_h$ is defined by collecting the local spaces ${\boldsymbol{\Sigma}}_h(P)$, i.e. $$\label{eq:Sg_h}
{\boldsymbol{\Sigma}}_h := \{{\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}(\Omega) \quad \text{s.t.} \quad {\boldsymbol{{\varphi}}}_{|P} \in {\boldsymbol{\Sigma}}_h(P) \} \,.$$ The global set of DoFs is the global counterpart of $\mathbf{D}_{{\boldsymbol{\Sigma}}}$, in particular the choice of DoFs $\mathbf{D}_{{\boldsymbol{\Sigma}}}$ establishes the conforming property ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\Sigma}}_h \subseteq [H^1(\Omega)]^3$. The dimension of ${\boldsymbol{\Sigma}}_h$ is given by $$\dim({\boldsymbol{\Sigma}}_h) =
3 \, L_V + (3 k-2)\, L_e + (3 \, \pi_{k-2, 2} -1) \, L_f + (3 \, \pi_{k-2, 3} - \pi_{k-1, 3} + 1) \,L_P \,.$$
Scalar $H^1$-conforming VEM space {#sub:W_h}
---------------------------------
In the present section we briefly define the $H^1$-conforming space $W_h$ in the virtual complex . The space $W_h$ consists of low order nodal VEM [@BBMR:2016].
We first introduce the low order boundary space $$\label{eq:B_h^n}
{{\mathbb}{B}}_1(f):= \{
v \in H^1(f) \quad \text{s.t.} \quad
v_{|\partial f} \in C(\partial f) \,, \quad
{v}_{|e} \in {{\mathbb}{P}}_1(e) \quad \text{for all $e \in \partial f$}\,, \quad
\Delta_f \, v = 0
\}\,,$$ and then we consider the VEM space on the polyhedron $P$ $$\label{eq:W_h^P}
W_h(P) :=
\{
v \in H^1(P)
\, \, \, \text{s.t.} \, \, \,
v_{|\partial P} \in C^0(\partial P)\,, \, \, \,
v_{|f} \in {{\mathbb}{B}}_1(f) \quad \text{for any $f \in \partial P$,} \, \, \,
\Delta \, v = 0
\}\,,$$ with the associated set of DoFs:
- $\mathbf{D}_{W}$: the values of $v$ at the vertexes of the polyhedron $P$.
It is straightforward to see that the dimension of $W_h(P)$ is given by $
\dim(W_h(P)) = \ell_V
$. The global space is obtained by collecting the local spaces $$\label{eq:W_h}
W_h :=
\{
v \in H^1(\Omega)
\quad \text{s.t.} \quad
v_{|P} \in W_h(P) \quad \text{for all $P \in \Omega_h$}
\}$$ with the obvious associated DoFs. The dimension of $W_h$ thus is given by $$\dim(W_h) = L_V \,.$$
The virtual elements de Rham sequence {#sec:derham}
=====================================
The aim of the present section is to show that the set of virtual spaces introduced in Section \[sec:spaces\] realizes the exact sequence .
\[thm:exact\] The sequence constitutes an exact complex.
The theorem follows by Proposition \[prp:inclusion1\], Proposition \[prp:inclusion2\] and Proposition \[prp:inclusion3\], here below, stating that the image of each operator in coincides with the kernel of the following one.
\[prp:inclusion1\] Let $W_h$ and ${\boldsymbol{\Sigma}}_h$ be the spaces defined in and respectively. Then $$\nabla\, W_h = \{ {\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h \quad \text{s.t.} \quad {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\bf 0} \quad \text{in $\Omega$}\} \,.$$
Essentially we need to prove that
- : for every $w \in W_h$, $\nabla \, w \in {\boldsymbol{\Sigma}}_h$ and ${\boldsymbol{{\rm curl}}}(\nabla \, w) = {\bf 0}$,
- : for every ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h$ with ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\bf 0}$, there exists $w \in W_h$ such that $\nabla \, w = {\boldsymbol{{\varphi}}}$.
For what concerns the inclusion ${\boldsymbol {(i1)}}$, every $w \in W_h$ clearly satisfies $$\nabla \, w \in [L^2(\Omega)]^3
\quad \text{and} \quad
{\boldsymbol{{\rm curl}}}(\nabla \, w) = {\bf 0} \in [H^1(\Omega)]^3\,.$$ Therefore we need to verify that $(\nabla \, w)_{|P} \in {\boldsymbol{\Sigma}}_h(P)$ for any $P \in \Omega_h$. Notic0e that the tangential component of $\nabla \, w$ verifies $$\label{eq:grad_tan}
(\nabla \, w)_{\tau} = \nabla_f \, w_{\tau}
\qquad \text{ on each face $f \in \partial P$.}$$ From definition and , for any $f \in \partial P$ we infer $$\begin{aligned}
& {{\rm div}}_f ( \nabla_f \, w_{\tau} ) = 0
\qquad \text{in $f$,}
\qquad \qquad &\text{($\Delta_f \, w_{\tau} = 0$)}
\\
& {{\rm rot}}_f ( \nabla_f \, w_{\tau} ) = 0
\qquad \text{in $f$,}
\qquad \qquad &\text{(vector calculs identity)}
\\
& (\nabla \, w \cdot {\boldsymbol{t}}_e)_{|e} = \frac{\partial w}{\partial {\boldsymbol{t}}_e} \in {{\mathbb}{P}}_0(e)
\qquad \text{$\forall \, e \in \partial f$,}
\qquad \qquad &\text{($w_{|e} \in {{\mathbb}{P}}_1(e)$)}\end{aligned}$$ that, recalling , implies $(\nabla \, w)_{\tau} \in {\boldsymbol S}_k(f)$. Moreover $w \in C^0(\partial P)$ entails $$[(\nabla \, w)_{f_1} \cdot {\boldsymbol{t}}_e]_{|e} = [(\nabla \, w)_{f_2} \cdot {\boldsymbol{t}}_e]_{|e}
\quad \text{for any $e \subseteq \partial f_1 \cap f_2$,}$$ and thus (cf. definition ) $$\label{eq:grad_sk}
(\nabla \, w)_{|\partial P} \in {\boldsymbol S}_k(\partial P) \,.$$ Furthermore definition implies $$\label{eq:grad_sg}
\begin{aligned}
&\int_{\partial P} \nabla \, w \cdot {\boldsymbol{n}}_P \,{\rm d}f =
\int_P \Delta \, w \, {\rm d}P = 0 \,,
&
\qquad &\text{(div. thereom + $\Delta \, w = 0$)}
\\
&{\boldsymbol{{\rm curl}}}( \nabla \, w) = {\bf 0}
&\text{in $\overline{P}$,}
\qquad &\text{(vector calculus identity)}
\\
& {\boldsymbol{\Delta}}(\nabla \, w) = {\boldsymbol{\nabla}}(\Delta \, w) = {\bf 0}
&\text{in $P$.}
\qquad &\text{($\Delta \, w = 0$)}
\end{aligned}$$ Collecting and in definition , we easily obtain $\boldsymbol{(i1)}$.
We prove now the property $\boldsymbol{(i2)}$. Consider ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h$ such that ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\bf 0}$. Since is an exact sequence, there exists unique (up to constant) $\widetilde{w} \in H^1(\Omega)$ such that $\nabla \, \widetilde{w} ={\boldsymbol{{\varphi}}}$. Therefore for any face $f$ in the decomposition $\Omega_h$, the tangential component of $\nabla \, \widetilde{w}$ satisfies (cf. definition ) $$(\nabla \, \widetilde{w})_{\tau} =
\nabla_f \, \widetilde{w}_{\tau} = {\boldsymbol{{\varphi}}}_{\tau} \in [L^2(f)]^2 \quad \text{on $f$.}$$ Hence on each face $f$ the function $\widetilde{w}$ fulfils $$\label{eq:gradwt}
\left \{
\begin{aligned}
& (\nabla \, \widetilde{w} \cdot {\boldsymbol{t}}_e)_{|e} = ({\boldsymbol{{\varphi}}}\cdot {\boldsymbol{t}}_e)_{|e} \in {{\mathbb}{P}}_0(e) \quad
&\text{on any $e \in \partial f$,}
\\
& \widetilde{w}_{\tau} \in H^1(f)
\quad
&\text{in $f$.}
\end{aligned}
\right .$$ From it follows that $\widetilde{w}$ restricted to the mesh skeleton is continuous and piecewise linear. Thus the function $\widetilde{w}$ is well defined (single valued) on the vertexes of the decomposition $\Omega_h$ and $\mathbf{D}_{{\boldsymbol{V}}}(\widetilde{w})$ makes sense. Let now $w \in W_h$ be the interpolant function of $\widetilde{w}$ in the sense of DoFs, i.e. the function uniquely determined by $$\label{eq:gradwt1}
\mathbf{D}_{{\boldsymbol{V}}}(w) = \mathbf{D}_{{\boldsymbol{V}}}(\widetilde{w}) \,.$$ Inclusion $\boldsymbol{(i1)}$ guarantees that $\nabla \, w \in {\boldsymbol{\Sigma}}_h$. Hence, by Proposition \[prp:Sg\_h\], $w$ realizes $\boldsymbol{(i2)}$ if and only if $\mathbf{D}_{{\boldsymbol{\Sigma}}}(\nabla \, w) = \mathbf{D}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}})$. Being ${\boldsymbol{{\rm curl}}}(\nabla \, w) = {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\bf 0}$, this reduce to verify that $$\mathbf{D^5}_{{\boldsymbol{\Sigma}}}(\nabla \, w) = \mathbf{D^5}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) \,.$$ For any edge $e$ in the decomposition $\Omega_h$, we denote with $\nu_2$ and $\nu_1$ the two endpoints of $e$, with ${\boldsymbol{t}}_e$ pointing from $\nu_1$ to $\nu_2$. Therefore, from and , we infer $$\begin{split}
\mathbf{D^5}_{{\boldsymbol{\Sigma}}, e}(\nabla \, w) &=
\frac{1}{|e|} \int_e \nabla \, w \cdot {\boldsymbol{t}}_e \,{\rm d}s =
\mathbf{D}_{{\boldsymbol{V}}, \nu_2}(w) - \mathbf{D}_{{\boldsymbol{V}}, \nu_1}(w) =
\mathbf{D}_{{\boldsymbol{V}}, \nu_2}(\widetilde{w}) - \mathbf{D}_{{\boldsymbol{V}}, \nu_1}(\widetilde{w})
\\
&=
\frac{1}{|e|} \int_e \nabla \, \widetilde{w} \cdot {\boldsymbol{t}}_e \,{\rm d}s
= \frac{1}{|e|} \int_e {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{t}}_e \,{\rm d}s
= \mathbf{D^5}_{{\boldsymbol{\Sigma}}, e}({\boldsymbol{{\varphi}}}) \,,
\end{split}$$ that concludes the proof.
\[prp:inclusion2\] Let ${\boldsymbol{\Sigma}}_h$ and ${\boldsymbol{Z}}_h$ be the spaces defined in and respectively. Then $${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\Sigma}}_h = {\boldsymbol{Z}}_h \,.$$
The proof follows by showing the following points:
- : for every ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h$, ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\in {\boldsymbol{Z}}_h$,
- : for every ${\boldsymbol{v}}\in {\boldsymbol{Z}}_h$ there exists ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h$ such that ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\boldsymbol{v}}$.
Let us analyse the inclusion ${\boldsymbol {(i1)}}$. Let ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h$, clearly ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\in {\boldsymbol{Z}}(\Omega)$. Therefore we need to verify that $({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}})_{|P} \in {\boldsymbol{V}}_h(P)$ for any $P \in \Omega_h$. It is evident that the second and the last line in definition correspond respectively to the ${\boldsymbol{{\rm curl}}}$ version of the first and last line in definition . Hence it remains to show that ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ is the velocity solution of the Stokes problem associated with definition on each element $P$. A careful inspection of the biharmonic problem in definition , imply that the following are equivalent $$\begin{aligned}
&\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P \widetilde{{\boldsymbol{p}}}_{k-1} \cdot {\boldsymbol{\psi}}\, {\rm d}P
\quad &\text{$\forall{\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,}
\quad &\text{(by definition \eqref{eq:Sg_h^P})}
\\
&\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P {\boldsymbol{{\rm curl}}}({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-1}) \cdot {\boldsymbol{\psi}}\, {\rm d}P
\quad &\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,}
\quad &\text{(characterization \eqref{eq:z_CC})}
\\
&\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-1}) \cdot {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}\, {\rm d}P
\quad &\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$.}
\quad &\text{(int. by parts + b.c)}
\end{aligned}$$ In particular the last equation is still valid considering all ${\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)\cap {\boldsymbol{Z}}(P)$. Therefore, using the identity ${\boldsymbol{\Delta}}= -{\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\rm curl}}}+ {\boldsymbol{\nabla}}\, {{\rm div}}$ and an integration by parts (coupled with the homogeneous boundary condition ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}= 0$ on $\partial P$), it can be proved that the following are equivalent $$\label{eq:bih-sto}
\begin{aligned}
&\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-1}) \cdot {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}\, {\rm d}P
\quad &\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)\cap {\boldsymbol{Z}}(P)$,}
\quad &
\\
&\int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot (- {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}) \, {\rm d}P
= \int_P ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-1}) \cdot {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}\, {\rm d}P
\quad &\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P) \cap {\boldsymbol{Z}}(P)$,}
\quad &
\\
&\int_P -{\boldsymbol{\Delta}}({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}) \cdot {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-1}) \cdot {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}\, {\rm d}P
\quad &\text{$\forall {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P) \cap {\boldsymbol{Z}}(P)$.}
\quad &
\end{aligned}$$ Exploiting Lemma 5.1 in [@girault-raviart:book], for every ${\boldsymbol{z}}\in {\boldsymbol{Z}}_0(P)$ there exists ${\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P) \cap {\boldsymbol{Z}}(P)$ such that ${\boldsymbol{z}}= {\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}$. Therefore the last equation in is equivalent to $$\int_P {\boldsymbol{\nabla}}({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}) : {\boldsymbol{\nabla}}{\boldsymbol{z}}\, {\rm d}P
= \int_P ({\boldsymbol{x}}\wedge {\boldsymbol{p}}_{k-1}) \cdot {\boldsymbol{z}}\, {\rm d}P
\quad \text{for all ${\boldsymbol{z}}\in {\boldsymbol{Z}}_0(P)$,}$$ and therefore ${\boldsymbol{v}}= {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}$ is the velocity solution of a Stokes problem as in Definition . That concludes the proof for $\boldsymbol{(i1)}$.
We focus now on $\boldsymbol{(i2)}$. Let ${\boldsymbol{v}}\in {\boldsymbol{Z}}_h \subseteq {\boldsymbol{Z}}$, then from Corollary 3.3 in [@girault-raviart:book] there exists $\widetilde{{\boldsymbol{{\varphi}}}} \in [H^2(\Omega)]^3 \cap {\boldsymbol{Z}}(\Omega)$, such that ${\boldsymbol{{\rm curl}}}\, \widetilde{{\boldsymbol{{\varphi}}}} = {\boldsymbol{v}}$. Notice that, being $\widetilde{{\boldsymbol{{\varphi}}}} \in [H^2(\Omega)]^3$ and ${\boldsymbol{{\rm curl}}}\, \widetilde{{\boldsymbol{{\varphi}}}} \in {\boldsymbol{V}}_h(P)$ for any $P$ in $\Omega_h$, it makes sense to compute $\mathbf{D}_{{\boldsymbol{\Sigma}}}(\widetilde{{\boldsymbol{{\varphi}}}})$.
Let us consider the interpolant ${\boldsymbol{{\varphi}}}\in {\boldsymbol{\Sigma}}_h$ of $\widetilde{{\boldsymbol{{\varphi}}}}$ in the sense of DoFs, i.e. the function uniquely determined by (cf. Proposition \[prp:Sg\_h\]) $$\label{eq:dofsg}
\mathbf{D}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) = \mathbf{D}_{{\boldsymbol{\Sigma}}}(\widetilde{{\boldsymbol{{\varphi}}}}) \,.$$ Property $\boldsymbol{(i1)}$ ensures ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\in {\boldsymbol{Z}}_h$. Therefore, employing Proposition \[prp:V\_h\^E\_dofs\], ${\boldsymbol{{\varphi}}}$ realizes $\boldsymbol{(i2)}$ if and only if $\mathbf{D}_{{\boldsymbol{V}}}({\boldsymbol{{\rm curl}}}\,{\boldsymbol{{\varphi}}}) = \mathbf{D}_{{\boldsymbol{V}}}({\boldsymbol{v}})$. Is it straightforward to check that $$\mathbf{D^5}_{{\boldsymbol{V}}}({\boldsymbol{{\rm curl}}}\,{\boldsymbol{{\varphi}}}) = \mathbf{D^5}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = {\bf 0} \,,$$ $$\mathbf{D^{\boldsymbol i}}_{{\boldsymbol{V}}}({\boldsymbol{{\rm curl}}}\,{\boldsymbol{{\varphi}}}) =
\mathbf{D^{\boldsymbol i}}_{{\boldsymbol{\Sigma}}}({\boldsymbol{{\varphi}}}) =
\mathbf{D^{\boldsymbol i}}_{{\boldsymbol{\Sigma}}}(\widetilde{{\boldsymbol{{\varphi}}}}) =
\mathbf{D^{\boldsymbol i}}_{{\boldsymbol{V}}}({\boldsymbol{v}})
\quad \text{for $\boldsymbol{i=1,2,3, 4}$,}$$ except the face moment (that is slightly more subtle) $$\int_f {\boldsymbol{{\rm curl}}}\,{\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P^f \, {\rm d}f
\quad \text{and} \quad
\int_f {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f \, {\rm d}f
\qquad \text{for any face $f$.}$$ In order to show that the two quantities above are equal we exploit the same computations in and $$\begin{split}
\int_f {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P^f \,{\rm d}f &=
\sum_{e \in \partial f} |e| \, \mathbf{D^5}_{{\boldsymbol{\Sigma}}, e}({\boldsymbol{{\varphi}}}) \, {\boldsymbol{t}}_e \cdot {\boldsymbol{t}}_f^e
=
\sum_{e \in \partial f} |e| \, \mathbf{D^5}_{{\boldsymbol{\Sigma}}, e}(\widetilde{{\boldsymbol{{\varphi}}}}) \, {\boldsymbol{t}}_e \cdot {\boldsymbol{t}}_f^e
\\
&=
\int_f {\boldsymbol{{\rm curl}}}\,\widetilde{{\boldsymbol{{\varphi}}}} \cdot {\boldsymbol{n}}_P^f \,{\rm d}f
= \int_f {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f \,{\rm d}f \,.
\end{split}$$ This ends the proof.
\[prp:inclusion3\] Let ${\boldsymbol{V}}_h$ and $Q_h$ be the spaces defined in and respectively. Then $${{\rm div}}\, {\boldsymbol{V}}_h = Q_h \,.$$
We follow same strategy adopted in the previous propositions and show that
- : for every ${\boldsymbol{v}}\in {\boldsymbol{V}}_h$, ${{\rm div}}\, {\boldsymbol{v}}\in Q_h$,
- : for every $q \in Q_h$ there exists ${\boldsymbol{v}}\in {\boldsymbol{V}}_h$ such that ${{\rm div}}\, {\boldsymbol{v}}= q$.
The inclusion ${\boldsymbol {(i1)}}$ is trivial. Regarding the point ${\boldsymbol {(i2)}}$, since is an exact sequence, for any $q \in Q_h$ there exists $\widetilde{{\boldsymbol{v}}} \in [H^1(\Omega)]^3$ such that ${{\rm div}}\,\widetilde{{\boldsymbol{v}}} = q$. Now let ${\boldsymbol{v}}\in {\boldsymbol{V}}_h$ the function uniquely determined by (cf. Proposition \[prp:V\_h\^E\_dofs\]) $$\label{eq:dofsv}
\begin{gathered}
\mathbf{D^1}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = \mathbf{D^2}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = \mathbf{D^4}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = {\bf 0} \,,
\\
\mathbf{D^3}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = {\bf 0} \quad \text{except the face moments}
\quad
\int_f {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f \, {\rm d}f =
\int_f \widetilde{{\boldsymbol{v}}} \cdot {\boldsymbol{n}}_P^f \, {\rm d}f \,,
\\
\mathbf{D^5}_{{\boldsymbol{V}}}({\boldsymbol{v}}) = \mathbf{D^5}_{{\boldsymbol{V}}}(\widetilde{{\boldsymbol{v}}}) \,.
\end{gathered}$$ Notice that being $\widetilde{{\boldsymbol{v}}} \in [H^1(\Omega)]^3$ the face moments in and $\mathbf{D^5}_{{\boldsymbol{V}}}(\widetilde{{\boldsymbol{v}}})$ are actually well defined. Therefore for any $P \in \Omega_h$ we infer $$\label{eq:div1}
\int_P ({{\rm div}}\, {\boldsymbol{v}}) \, \widehat{{\boldsymbol{p}}}_{k-1} \, {\rm d}P =
\int_P ({{\rm div}}\, \widetilde{{\boldsymbol{v}}}) \, \widehat{{\boldsymbol{p}}}_{k-1} \, {\rm d}P =
\int_P q \, \widehat{{\boldsymbol{p}}}_{k-1} \, {\rm d}P
\qquad
\text{for all $\widehat{{\boldsymbol{p}}}_{k-1} \in \widehat{{{\mathbb}{P}}}_{k-1 \setminus 0}(P)$.}$$ Moreover employing the divergence theorem, implies $$\label{eq:div2}
\begin{split}
\int_P {{\rm div}}\, {\boldsymbol{v}}\, {\rm d}P &=
\int_{\partial P} {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P \, {\rm d}f =
\sum_{f \in \partial P} \int_f {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P^f \, {\rm d}f
\\
&=
\sum_{f \in \partial P} \int_f \widetilde{{\boldsymbol{v}}} \cdot {\boldsymbol{n}}_P^f \, {\rm d}f =
\int_{\partial P} \widetilde{{\boldsymbol{v}}} \cdot {\boldsymbol{n}}_P \, {\rm d}f =
\int_P {{\rm div}}\, \widetilde{{\boldsymbol{v}}} \, {\rm d}P =
\int_P q \, {\rm d}P \,.
\end{split}$$ Notice that and coincide with $\mathbf{D}_Q({{\rm div}}\, {\boldsymbol{v}}) = \mathbf{D}_Q(q)$ that coupled with ${{\rm div}}\, {\boldsymbol{v}}\in Q_h$ (from $\boldsymbol{(i1)}$) concludes the proof.
Virtual Elements for the 3-d Navier–Stokes equation {#sec:VEM-ns}
===================================================
We consider the steady Navier–Stokes equation on a polyhedral domain $\Omega \subseteq {{\mathbb}{R}}^3$ with homogeneous Dirichlet boundary conditions:
$$\label{eq:ns continuous}
\left\{
\begin{aligned}
& \text{find $({\boldsymbol{u}}, \, p) \in [H^1_0(\Omega)]^3 \times L^2_0(\Omega)$, such that} \\
& \nu \, a({\boldsymbol{u}}, \,{\boldsymbol{v}}) + c({\boldsymbol{u}}; \, {\boldsymbol{u}}, {\boldsymbol{v}}) + b({\boldsymbol{v}}, p) = ({\boldsymbol{f}}, \, {\boldsymbol{v}}) \qquad & \text{for all ${\boldsymbol{v}}\in [H^1_0(\Omega)]^3$,} \\
& b({\boldsymbol{u}}, \, q) = 0 \qquad & \text{for all $q \in Q$,}
\end{aligned}
\right.$$
where $\nu > 0$ represents the viscosity, ${\boldsymbol{f}}\in [L^2(\Omega)]^3$ is the external force and $$\begin{aligned}
\label{eq:a-cont}
a({\boldsymbol{u}},\, {\boldsymbol{v}}) &:= \int_{\Omega} {\boldsymbol{{\varepsilon}}}( {\boldsymbol{u}}) : {\boldsymbol{{\varepsilon}}}( {\boldsymbol{v}}) \, {\rm d}\Omega
\quad
&\text{for all ${\boldsymbol{u}}$, ${\boldsymbol{v}}\in [H^1(\Omega)]^3$,}
\\
\label{eq:c-cont}
c({\boldsymbol{w}}; \, {\boldsymbol{u}},\, {\boldsymbol{v}}) &:= \int_{\Omega} [({\boldsymbol{\nabla}}\, {\boldsymbol{u}}) \, {\boldsymbol{w}}] \cdot {\boldsymbol{v}}\, {\rm d}\Omega
\quad
&\text{for all ${\boldsymbol{w}}$, ${\boldsymbol{u}}$, ${\boldsymbol{v}}\in [H^1(\Omega)]^3$,}
\\
\label{eq:b-cont}
b({\boldsymbol{u}},\, q) &:= \int_{\Omega} {{\rm div}}\, {\boldsymbol{u}}\, q \, {\rm d}\Omega
\quad
&\text{for all ${\boldsymbol{u}}\in [H^1(\Omega)]^3$ and $q \in L^2(\Omega)$.}\end{aligned}$$ For sake of simplicity we here consider Dirichlet homogeneous boundary conditions, different boundary conditions can be treated as well.
It is well known [@quarteroni-valli:book] that in the diffusion dominated regime $$\mathbf{(H)} \qquad \gamma := \frac{\|{\boldsymbol{f}}\|_{-1}}{\nu^2} \ll 1$$ the Navier–Stokes equation has a unique solution $({\boldsymbol{u}}, \, p)$ with $$|{\boldsymbol{u}}|_1 \leq \frac{\|{\boldsymbol{f}}\|_{-1}}{\nu} \,.$$ Moreover Problem can be formulated in the equivalent kernel form: $$\left\{
\begin{aligned}
& \text{find ${\boldsymbol{u}}\in {\boldsymbol{Z}}_0(\Omega)$, such that} \\
& \nu \, a({\boldsymbol{u}}, \,{\boldsymbol{v}}) + c({\boldsymbol{u}}; \, {\boldsymbol{u}}, {\boldsymbol{v}}) = ({\boldsymbol{f}}, \, {\boldsymbol{v}}) \qquad & \text{for all ${\boldsymbol{v}}\in {\boldsymbol{Z}}_0(\Omega)$.}
\end{aligned}
\right.$$
Discrete forms and load term approximation {#sub:forms}
------------------------------------------
In this subsection we briefly describe the construction of a discrete version of the bilinear form $a(\cdot, \cdot)$ given in and trilinear form $c(\cdot; \cdot, \cdot)$ given in . We can follow in a rather slavish way the procedure initially introduced in [@volley] for the laplace problem and further developed in [@BLV:2018] for flow problems. First, we decompose into local contributions the bilinear form $a(\cdot, \cdot)$ and the trilinear form $c(\cdot; \cdot, \cdot)$ by considering $$a({\boldsymbol{u}}, \, {\boldsymbol{v}}) =: \sum_{P \in \Omega_h} a^P({\boldsymbol{u}}, \, {\boldsymbol{v}}) \,,
\qquad
c({\boldsymbol{w}}; \, {\boldsymbol{u}}, {\boldsymbol{v}}) =: \sum_{P \in \Omega_h} c^P({\boldsymbol{w}}; \, {\boldsymbol{u}}, {\boldsymbol{v}}) \,,$$ for all ${\boldsymbol{w}}$, ${\boldsymbol{u}}$, ${\boldsymbol{v}}\in [H^1(\Omega)]^3$.
As usual in VEM framework the discrete counterpart of the continuous forms above are defined starting from the polynomial projections defined in and . The following proposition extends to the 3-d case the result for the bi-dimensional spaces [@BLV:2017; @vacca:2018].
\[prp:proj\] Let $\widehat{{{\mathbb}{B}}}_k(f)$ and ${\boldsymbol{V}}_h(P)$ be the spaces defined in and respectively. The DoFs $\mathbf{D}_{{\boldsymbol{V}}}$ allow us to compute exactly the face projections $$\Pi_{k}^{\nabla, f} \colon [\widehat{{{\mathbb}{B}}}_k(f)]^3 \to [{{\mathbb}{P}}_k(f)]^3 \,,
\qquad
\Pi_{k+1}^{0, f} \colon [\widehat{{{\mathbb}{B}}}_k(f)]^3 \to [{{\mathbb}{P}}_{k+1}(f)]^3 \,,$$ for any $f \in \partial P$, and the element projections $$\begin{aligned}
\Pi_{k}^{\nabla, P} &\colon {\boldsymbol{V}}_h(P) \to [{{\mathbb}{P}}_k(P)]^3 \,,
\\
{\boldsymbol \Pi}_{k-1}^{0, P} &\colon {\boldsymbol{\nabla}}( {\boldsymbol{V}}_h(P) ) \to [{{\mathbb}{P}}_{k-1}(P)]^{3 \times 3} \,,
\\
\Pi_{k}^{0, P} &\colon {\boldsymbol{V}}_h(P) \to [{{\mathbb}{P}}_k(P)]^3 \,,
\end{aligned}$$ in the sense that, given any ${\boldsymbol{v}}_h \in {\boldsymbol{V}}_h(P)$, we are able to compute the polynomials $$\Pi_{k}^{\nabla, f} {\boldsymbol{v}}_h \,, \qquad
\Pi_{k+1}^{0, f} {\boldsymbol{v}}_h \,, \qquad
\Pi_{k}^{\nabla, P} {\boldsymbol{v}}_h \,, \qquad
{\boldsymbol \Pi}_{k}^{0, P} ({\boldsymbol{\nabla}}\, {\boldsymbol{v}}_h) \,, \qquad
\Pi_{k}^{0, P} {\boldsymbol{v}}_h \,,$$ using only, as unique information, the DoFs values $\mathbf{D}_{{\boldsymbol{V}}}$ of ${\boldsymbol{v}}_h$.
The computability of the face projections is a direct application of Remark 5 in [@projectors]. Concerning the element projections we here limit to prove the last item, the first two follow analogous techniques.
By definition of $L^2$-projection , in order the determine, for any ${\boldsymbol{v}}\in {\boldsymbol{V}}_h(P)$, the polynomial $\Pi_k^{0, P} {\boldsymbol{v}}$ we need to compute $$\int_P {\boldsymbol{v}}\cdot {\boldsymbol{p}}_k \, {\rm d}P
\qquad
\text{for all ${\boldsymbol{p}}_k \in [{{\mathbb}{P}}_k(P)]^3$.}$$ From polynomial decomposition we can write $${\boldsymbol{p}}_k = \nabla \, \widehat{p}_{k+1} + {\boldsymbol{x}}\wedge \widehat{\boldsymbol q}_{k-1}
+ {\boldsymbol{x}}\wedge{\boldsymbol q}_{k-3}$$ for some $\widehat{p}_{k+1} \in \widehat{{{\mathbb}{P}}}_{k+1 \setminus 0}(P)$, $\widehat{\boldsymbol q}_{k-1} \in [\widehat{{{\mathbb}{P}}}_{k-1 \setminus k-3}(P)]^3$, ${\boldsymbol q}_{k-3} \in [{{{\mathbb}{P}}}_{k-3}(P)]^3$. Thus $$\begin{aligned}
\int_P {\boldsymbol{v}}\cdot {\boldsymbol{p}}_k \, {\rm d}P = &
\int_P {\boldsymbol{v}}\cdot \left(\nabla \, \widehat{p}_{k+1} + {\boldsymbol{x}}\wedge \widehat{\boldsymbol q}_{k-1}
+ {\boldsymbol{x}}\wedge {\boldsymbol q}_{k-3} \right) \,{\rm d}P
\\
= &
\int_P \Pi_k^{\nabla, P}{\boldsymbol{v}}\cdot ({\boldsymbol{x}}\wedge \widehat{\boldsymbol q}_{k-1} ) \,{\rm d}P
+ \int_P {\boldsymbol{v}}\cdot ({\boldsymbol{x}}\wedge {\boldsymbol q}_{k-3} ) \,{\rm d}P \,+
\quad & \text{(enhancing def. \eqref{eq:V_h^P})}
\\
& - \int_P ({{\rm div}}\, {\boldsymbol{v}}) \, \widehat{p}_{k+1} \,{\rm d}P
+ \int_{\partial P} {\boldsymbol{v}}\cdot {\boldsymbol{n}}_P \, \widehat{p}_{k+1} \, {\rm d}f
\quad & \text{(integration by parts)}
\\
= &
\int_P \Pi_k^{\nabla, P}{\boldsymbol{v}}\cdot ({\boldsymbol{x}}\wedge \widehat{\boldsymbol q}_{k-1} ) \,{\rm d}P
+ \int_P {\boldsymbol{v}}\cdot ({\boldsymbol{x}}\wedge {\boldsymbol q}_{k-3} ) \,{\rm d}P \,+
\\
& - \int_P ({{\rm div}}\, {\boldsymbol{v}}) \, \widehat{p}_{k+1} \,{\rm d}P
+ \sum_{f \in \partial P}
\int_{f} (\Pi_{k+1}^{0, f} {\boldsymbol{v}}) \cdot {\boldsymbol{n}}_P^f \, \widehat{p}_{k+1} \, {\rm d}f
\quad & \text{(by def. \eqref{eq:P0_k^E})}
\end{aligned}$$ The first and the last integrals are computable being $\Pi_k^{\nabla, P}{\boldsymbol{v}}$ and $\Pi_{k+1}^{0, f}$ computable. The second addend corresponds to the DoFs $\mathbf{D^4}_{{\boldsymbol{V}}}$. For the third addend we observe that, since ${{\rm div}}\, {\boldsymbol{v}}$ is a polynomial of degree less or equal than $k-1$ we can exactly reconstruct its value from the DoFs $\mathbf{D^5}_{{\boldsymbol{V}}}$ and the normal face moments in $\mathbf{D^3}_{{\boldsymbol{V}}}$.
On the basis of the projections above, following a standard procedure in the VEM framework, we define the computable (in the sense of Proposition \[prp:proj\]) discrete local forms and the approximated right hand side $$\begin{aligned}
\label{eq:ahP}
a_h^P({\boldsymbol{u}}, \, {\boldsymbol{v}}) &:=
\int_P
\left({\boldsymbol \Pi}_{k-1}^{0, P} {\boldsymbol{{\varepsilon}}}({\boldsymbol{u}}) \right) :
\left({\boldsymbol \Pi}_{k-1}^{0, P} {\boldsymbol{{\varepsilon}}}({\boldsymbol{v}}) \right) \, {\rm d}P
+ \mathcal{S}^P \left( (I - \Pi_k^{{\nabla}, P} ) {\boldsymbol{u}}, \, (I - \Pi_k^{{\nabla}, P} ) {\boldsymbol{v}}\right) \,,
\\
\label{eq:chP}
c_h^P({\boldsymbol{w}}; \, {\boldsymbol{u}}, {\boldsymbol{v}}) &:=
\int_P \left[ \left(\boldsymbol{\Pi}_{k-1}^{0, P} {\boldsymbol{\nabla}}\, {\boldsymbol{u}}\right) \, \Pi_k^{0, P} {\boldsymbol{w}}\right] \cdot \Pi_k^{0, P} {\boldsymbol{v}}\,{\rm d}P \,,
\\
\label{eq:fhP}
({\boldsymbol{f}}_h, \, {\boldsymbol{v}})_P &:= \int_P \Pi_k^{0, P} {\boldsymbol{f}}\cdot {\boldsymbol{v}}\, {\rm d}P \,,\end{aligned}$$ for all ${\boldsymbol{w}}$, ${\boldsymbol{u}}$, ${\boldsymbol{v}}\in {\boldsymbol{V}}_h(P)$, where clearly $${\boldsymbol \Pi}_{k-1}^{0, P} {\boldsymbol{{\varepsilon}}}({\boldsymbol{v}}) =
\frac{{\boldsymbol \Pi}_{k-1}^{0, P} {\boldsymbol{\nabla}}\, {\boldsymbol{v}}+
({\boldsymbol \Pi}_{k-1}^{0, P} {\boldsymbol{\nabla}}\, {\boldsymbol{v}})^{\rm T}
}{2}$$ and the symmetric stabilizing form $\mathcal{S}^P \colon {\boldsymbol{V}}_h(P) \times {\boldsymbol{V}}_h(P) \to {{\mathbb}{R}}$ satisfies $$|{\boldsymbol{v}}|_{1, P}^2 \lesssim \mathcal{S}^P({\boldsymbol{v}}, \, {\boldsymbol{v}}) \lesssim |{\boldsymbol{v}}|_{1, P}^2
\qquad \text{for all ${\boldsymbol{v}}\in {\rm Ker}(\Pi_k^{{\nabla}, P})$.}$$ The condition above essentially requires the stabilizing term $\mathcal{S}^P({\boldsymbol{v}}, \, {\boldsymbol{v}})$ to scale as $|{\boldsymbol{v}}|_{1, P}^2$. For instance, a standard choice for the stabilization is the $D$-recipe stabilization introduced in [@BDR:2017].
\[rm:projectionPDbis\] The $H^1$-seminorm projection $\Pi^{\nabla, P}_k$ in the stabilization term of Definition can be replaced by any polynomial projection $\Pi^{P}_k$ that is computable on the basis of the DoFs $\mathbf{D}_{{\boldsymbol{V}}}$ (in the sense of Proposition \[prp:proj\]). A possible choice will be explored in Section \[sec:num\_test\].
The global virtual forms and the global approximated right-hand side are defined by simply summing the local contributions: $$\label{eq:formh}
a_h({\boldsymbol{u}}, \, {\boldsymbol{v}}) := \sum_{P \in \Omega_h} a_h^P({\boldsymbol{u}}, \,{\boldsymbol{v}})\,,
\quad
c_h({\boldsymbol{w}}; \, {\boldsymbol{u}}, \, {\boldsymbol{v}}) := \sum_{P \in \Omega_h} c_h^P({\boldsymbol{w}}; \, {\boldsymbol{u}}, \, {\boldsymbol{v}})\,,
\quad
({\boldsymbol{f}}_h, \, {\boldsymbol{v}}) := \sum_{P \in \Omega_h} ({\boldsymbol{f}}_h, \,{\boldsymbol{v}})_P \,,$$ for all ${\boldsymbol{w}}$, ${\boldsymbol{u}}$, ${\boldsymbol{v}}\in {\boldsymbol{V}}_h$.
The discrete problem {#seb:discrete-problme}
---------------------
Referring to the discrete spaces , , the discrete forms and the approximated load term and the ${{\rm div}}$ form , the virtual element approximation of the Navier–Stokes equation is given by $$\label{eq:ns virtual}
\left\{
\begin{aligned}
& \text{find $({\boldsymbol{u}}_h, \, p_h) \in {\boldsymbol{V}}_{h, 0} \times Q_{h, 0}$, such that} \\
& \nu \, a_h({\boldsymbol{u}}_h, \, {\boldsymbol{v}}_h) + c_h({\boldsymbol{u}}_h; \, {\boldsymbol{u}}_h, {\boldsymbol{v}}_h) + b({\boldsymbol{v}}_h, \, p_h) = ({\boldsymbol{f}}_h, \, {\boldsymbol{v}}_h) \qquad & \text{for all ${\boldsymbol{v}}_h \in {\boldsymbol{V}}_{h, 0}$,} \\
& b({\boldsymbol{u}}_h, \, q_h) = 0 \qquad & \text{for all $q_h \in Q_{h, 0}$,}
\end{aligned}
\right.$$ where ${\boldsymbol{V}}_{h, 0} := {\boldsymbol{V}}_h \cap [H^1_0(\Omega)]^3$ and $Q_{h, 0} := Q_h \cap L^2_0(\Omega)$. Recalling the kernel inclusion , Problem can be also formulated in the equivalent kernel form $$\label{eq:nsvirtual ker}
\left\{
\begin{aligned}
& \text{find ${\boldsymbol{u}}_h \in {\boldsymbol{Z}}_{h, 0}$, such that} \\
& \nu \, a_h({\boldsymbol{u}}_h, \, {\boldsymbol{v}}_h) + c_h({\boldsymbol{u}}_h; \, {\boldsymbol{u}}_h, {\boldsymbol{v}}_h) = ({\boldsymbol{f}}_h, {\boldsymbol{v}}_h) \qquad & \text{for all ${\boldsymbol{v}}_h \in {\boldsymbol{Z}}_{h, 0}$,}
\end{aligned}
\right.$$ with the obvious notation ${\boldsymbol{Z}}_{h, 0} := {\boldsymbol{Z}}_h \cap [H^1_0(\Omega)]^3$.
Combining the arguments in [@BLV:2017; @BLV:2018; @brenner-sung:2018] it is possible to show that the virtual space ${\boldsymbol{V}}_h$ has an optimal interpolation order of accuracy with respect to the degree $k$, and that the couple of spaces $({\boldsymbol{V}}_h, \, Q_h)$ is inf-sup stable [@boffi-brezzi-fortin:book]. The following existence and convergence theorem extends the analogous result for the bi-dimensional case [@BLV:2018].
\[thm:u\] Under the assumptions $\mathbf{(A1)}$, $\mathbf{(A2)}$, $\mathbf{(A3)}$ and and $\mathbf{(H)}$, let $({\boldsymbol{u}}, \, p) \in [H^1_0(\Omega)]^3 \times L^2_0(\Omega)$ be the solution of Problem and $({\boldsymbol{u}}_h, \, p_h) \in {\boldsymbol{V}}_{h, 0} \times Q_{h, 0}$ be the (unique) solution of Problem . Assuming moreover ${\boldsymbol{u}}, {\boldsymbol{f}}\in [H^{s+1}(\Omega)]^3$ and $p \in H^s(\Omega)$, $0 < s \leq k$, then $$\begin{gathered}
\label{eq:thm:u}
| {\boldsymbol{u}}- {\boldsymbol{u}}_h |_{1} \lesssim \, h^{s} \, \mathcal{F}({\boldsymbol{u}}; \, \nu, \gamma) + \, h^{s+2} \, \mathcal{H}({\boldsymbol{f}}; \nu) \,,
\\
\label{eq:p-est}
\|p - p_h\|_0 \lesssim \, h^{s} \, |p|_{s} + h^s \, \mathcal{K}({\boldsymbol{u}}; \nu, \gamma) + h^{s+2} \, |{\boldsymbol{f}}|_{s+1} \end{gathered}$$ for suitable functions $\mathcal{F}$, $\mathcal{H}$, $\mathcal{K}$ independent of $h$.
Note that, as a consequence of the important property , there is no direct dependence of the velocity error on the pressure solution.
\[rm:curlformulation\] Since Proposition \[prp:inclusion2\] yields an explicit characterization of ${\boldsymbol{Z}}_h$ as ${\boldsymbol{{\rm curl}}}\, {\boldsymbol{\Sigma}}_h$, one could follow and build an equivalent ${\boldsymbol{{\rm curl}}}$ (discrete) formulation (see for instance Problem (77) in [@BMV:2018]). Such approach is less appealing in 3-d since the ${\boldsymbol{{\rm curl}}}$ operator has a non trivial kernel and thus some stabilization or additional Lagrange multiplier would be needed in the formulation. Moreover this approach does not seem to be competitive in terms of number of DoFs with the reduced version of the method (see Subsection \[sub:reduced\]). As a consequence, we do not explore any scheme resulting from the ${\boldsymbol{{\rm curl}}}$ formulation.
Reduced spaces and reduced scheme {#sub:reduced}
---------------------------------
In the present section we briefly show that Problem is somehow equivalent to a suitable reduced problem entangling relevant fewer DoFs, especially for large $k$. This reduction is analogous to its two-dimensional counterpart in Section 5 in [@BLV:2017] and Section 5.2 in [@vacca:2018].
The core idea is that $\mathbf{D^5}_{{\boldsymbol{V}}}({\boldsymbol{u}}_h) = {\bf 0}$, where ${\boldsymbol{u}}_h$ denotes the solution of , and therefore such degrees of freedom (and also the associated pressures) can be trivially eliminated from the system. Hence on each polygon $P$, let us define the reduced local spaces: $$\begin{aligned}
\widetilde{{\boldsymbol{V}}}_h(P) := \biggl\{
{\boldsymbol{v}}\in [H^1(P)]^3 \, \, \, \text{s.t.} \, \, \,
& {\boldsymbol{v}}_{|\partial P} \in [\widehat{{{\mathbb}{B}}}_k(\partial P)]^3
\\
& \biggl\{
\begin{aligned}
& \boldsymbol{\Delta} {\boldsymbol{v}}+ \nabla s \in {\boldsymbol{x}}\wedge [{{\mathbb}{P}}_{k-1}(P)]^3, \\
& {\rm div} \, {\boldsymbol{v}}\in {{\mathbb}{P}}_{0}(P),
\end{aligned}
\biggr. \quad \text{ for some $s \in L^2_0(P)$}
\\
\biggl.
& \left( {\boldsymbol{v}}- \Pi_k^{\nabla, P} {\boldsymbol{v}}, \, {\boldsymbol{x}}\wedge \widehat{{\boldsymbol{p}}}_{k-1} \right)_P = 0 \quad
\text{for all $\widehat{{\boldsymbol{p}}}_{k-1} \in [\widehat{{{\mathbb}{P}}}_{k-1 \setminus k-3}(P)]^3$}
\biggr \}
\end{aligned}$$ and $$\widetilde{Q}_h(P) := {{\mathbb}{P}}_{0}(P) \,.$$ Exploiting the same tools of Proposition \[prp:V\_h\^E\_dofs\] it can be proved that the linear operators $\widetilde{\mathbf{D}}_{{\boldsymbol{V}}}$, split into four subsets, defined by $$\mathbf{\widetilde{D}^{\boldsymbol{i}}}_{{\boldsymbol{V}}} =
\mathbf{D^{\boldsymbol{i}}}_{{{\boldsymbol{V}}}}
\qquad \text{for $\boldsymbol{i = 1, 2, 3, 4}$,}$$ constitute a set of DoFs for $\widetilde{{\boldsymbol{V}}}_h(P)$. Concerning the space $\widetilde{Q}_h(P)$, it is straightforward to see that $\dim(\widetilde{Q}_h(P)) = 1$ with unique DoF $\widetilde{\mathbf{D}}_{Q}$ defined by $
\widetilde{\mathbf{D}}_{Q}(q) := \int_P q \,{\rm d}P
$. The global spaces $\widetilde{{\boldsymbol{V}}}_h$ and $\widetilde{Q}_h$ are obtained in the standard fashion by gluing the local spaces: $$\begin{aligned}
\label{eq:Vr_h}
\widetilde{{\boldsymbol{V}}}_h &:= \{{\boldsymbol{v}}\in [H^1(\Omega)]^3 \quad \text{s.t.} \quad {\boldsymbol{v}}_{|P} \in \widetilde{{\boldsymbol{V}}}_h(P) \} \,,
\\
\label{eq:Qr_h}
\widetilde{Q}_h &:= \{q \in L^2(\Omega) \quad \text{s.t.} \quad q_{|P} \in \widetilde{Q}_h(P) \} \,.\end{aligned}$$ We remark that by construction ${\boldsymbol{Z}}_h \subseteq \widetilde{{\boldsymbol{V}}}_h$, therefore employing Proposition \[prp:inclusion1\] and Proposition \[prp:inclusion2\] we can state the following result.
\[prp:virtualr\_exact\] Referring to , , and , the sequence $${{\mathbb}{R}}\, \xrightarrow[]{ \,\quad \text{{$i$}} \quad \, } \,
W_h \, \xrightarrow[]{ \quad \text{{${\nabla}$}} \quad }\,
{\boldsymbol{\Sigma}}_h \, \xrightarrow[]{ \, \, \, \text{{${\boldsymbol{{\rm curl}}}$}} \, \, \, }\,
\widetilde{{\boldsymbol{V}}}_h \, \xrightarrow[]{ \, \, \, \, \text{{${{\rm div}}$}} \, \, \, \, }\,
\widetilde{Q}_h \, \xrightarrow[]{ \quad 0 \quad }
0$$ is an exact sub-complex of .
Referring to , and , we consider the reduced problem: $$\label{eq:ns virtualr}
\left\{
\begin{aligned}
& \text{find $(\widetilde{{\boldsymbol{u}}}_h, \, \widetilde{p}_h) \in \widetilde{{\boldsymbol{V}}}_{h, 0} \times \widetilde{Q}_{h, 0}$, such that} \\
& \nu \, a_h(\widetilde{{\boldsymbol{u}}}_h, \, {\boldsymbol{v}}_h) + c_h(\widetilde{{\boldsymbol{u}}}_h; \, \widetilde{{\boldsymbol{u}}}_h, {\boldsymbol{v}}_h) + b({\boldsymbol{v}}_h, \, \widetilde{p}_h) = ({\boldsymbol{f}}_h, \, {\boldsymbol{v}}_h) \qquad & \text{for all ${\boldsymbol{v}}_h \in \widetilde{{\boldsymbol{V}}}_{h, 0}$,} \\
& b(\widetilde{{\boldsymbol{u}}}_h, \, q_h) = 0 \qquad & \text{for all $q_h \in \widetilde{Q}_{h, 0}$,}
\end{aligned}
\right.$$ where $\widetilde{{\boldsymbol{V}}}_{h, 0} := \widetilde{{\boldsymbol{V}}}_h \cap [H^1_0(\Omega)]^3$ and $\widetilde{Q}_{h, 0} := \widetilde{Q}_h \cap L^2_0(\Omega)$. It is trivial to check that the reduced scheme has $ (2 \, \pi_{k-1, 3} - 2) \, N_P$ degrees of freedom less when compared with the original one .
The following proposition is easy to check and states the relation between Problem and the reduced Problem .
\[prp:equivalent\] Let $({\boldsymbol{u}}_h, \, p_h) \in {\boldsymbol{V}}_h \times Q_h$ and $(\widetilde{{\boldsymbol{u}}}_h, \, \widetilde{p}_h) \in \widetilde{{\boldsymbol{V}}}_h \times \widetilde{Q}_h$ be the solution of Problem and Problem respectively. Then $$\widetilde{{\boldsymbol{u}}}_h = {\boldsymbol{u}}_h \qquad \text{and} \qquad
\widetilde{p}_h = \Pi_0^{0, P} p_h \quad \text{in every $P \in \Omega_h$.}$$
Numerical validation {#sec:num_test}
====================
In this section we numerically verify the proposed discretization scheme. Before dealing with such examples, we briefly describe an alternative (computationally cheaper) projection adopted in the implementation of the method. Then we outline the polyhedral meshes and the error norms used in the analysis.
An alternative DoFs-based projection {#sub:dofs-proj}
------------------------------------
In the light of Remark \[rm:projectionPD\] and Remark \[rm:projectionPDbis\], the aim of the present subsection is to exhibit an alternative projection to be used in the place of the standard $H^1$-seminorm projection $\Pi^{\nabla, P}_k$ in and that will turn out to be very easy to implement. An analogous alternative projection could also be used to substitute $\Pi^{\nabla, f}_k$ in .
For any element $P \in \Omega_h$, let $\texttt{NDoF} := \dim({\boldsymbol{V}}_h(P))$. Then referring to Proposition \[prp:V\_h\^E\_dofs\] we set $\mathbf{D}_{{\boldsymbol{V}}} := \{\mathbf{D}_{{\boldsymbol{V}}, i}\}_{i=1}^{\texttt{NDoF}}$, and we denote with $\mathcal{D} \colon{\boldsymbol{V}}_h(P) \to {{\mathbb}{R}}^{\texttt{NDoF}}$ the linear operator defined for all ${\boldsymbol{v}}\in {\boldsymbol{V}}_h(P)$ by $$\left(\mathcal{D} \, {\boldsymbol{v}}\right)_i = \mathbf{D}_{{\boldsymbol{V}}, i} ({\boldsymbol{v}}) \qquad \text{for $i=1, \dots, \texttt{NDoF}$,}$$ i.e. $\mathcal{D} \, {\boldsymbol{v}}$ is the vector containing the degree of freedom values $\mathbf{D}_{{\boldsymbol{V}}}$ associated to ${\boldsymbol{v}}$. We consider:
- the **DoFs-projection** ${\Pi}_{n}^{\mathcal{D},P} \colon {\boldsymbol{V}}_h(P) \to [{{\mathbb}{P}}_n(P)]^3$ defined for any ${\boldsymbol{v}}\in {\boldsymbol{V}}_h(P)$ by $$\label{eq:PD_k^E}
\left(\mathcal{D}\, {\boldsymbol q}_n \,, \mathcal{D}\, ({\boldsymbol{v}}- {\Pi}_{n}^{\mathcal{D},P} \, {\boldsymbol{v}}) \right)_{{{\mathbb}{R}}^{\texttt{NDoF}}} = 0
\qquad \text{for all ${\boldsymbol q}_n \in [{{\mathbb}{P}}_n(P)]^3$.}$$
Notice that ${\Pi}_{n}^{\mathcal{D},P}$ is a special case of the serendipity projection introduced in [@BBMR:2016:serendipity]. Although the projection ${\Pi}_{n}^{\mathcal{D},P}$ may seem awkward on paper, it is quite simple and cheap to implement on the computer (since it is nothing but an euclidean projection with respect to the degree of freedom vectors). Indeed, it can be checked that the matrix formulation $\boldsymbol{\Pi}_{n}^{\mathcal{D},P}$ of the operator ${\Pi}_{n}^{\mathcal{D},P}$ acting from ${\boldsymbol{V}}_h(P)$ to ${\boldsymbol{V}}_h(P)$ (containing $[{{\mathbb}{P}}_n(P)]^3$) with respect to the basis $\boldsymbol{\mathcal{V}}$ (cf. [@autostoppisti], formula (3.18)) is $$\boldsymbol{\Pi}_{n}^{\mathcal{D},P} = D \, (D^T D)^{-1} D^T \in {{\mathbb}{R}}^{\texttt{NDoF} \times \texttt{NDoF}}\,,$$ where $D \in {{\mathbb}{R}}^{\texttt{NDoF} \times 3\pi_{n, 3}}$ is the matrix defined by (cf. [@autostoppisti], formula (3.17)) $$D_{i,\alpha} := \mathbf{D}_{{\boldsymbol{V}}, i} ({\boldsymbol m}_{\alpha})
\quad \text{for $i=1, \dots, \texttt{NDoF}$ and
$\alpha =1, \dots, 3\pi_{n, 3}$,}$$ where using standard VEM notation, ${\boldsymbol m}_{\alpha}$ denotes the scaled monomial $${\boldsymbol m}_{\alpha} :=
\left(
\left(\frac{{\boldsymbol{x}}- {\boldsymbol{x}}_B}{h_P} \right)^{\boldsymbol{\alpha_1}}, \quad
\left(\frac{{\boldsymbol{x}}- {\boldsymbol{x}}_B}{h_P} \right)^{\boldsymbol{\alpha_2}}, \quad
\left(\frac{{\boldsymbol{x}}- {\boldsymbol{x}}_B}{h_P} \right)^{\boldsymbol{\alpha_3}}
\right)^T$$ with ${\boldsymbol{x}}_B$ barycenter of the polyhedron $P$, and ${\boldsymbol{\alpha_1}}$, ${\boldsymbol{\alpha_2}}$ and ${\boldsymbol{\alpha_3}}$ suitable multi-indexes.
Meshes and error norms
----------------------
We consider the standard $[0,\,1]^3$ cube as domain $\Omega$ and we make four different discretizations of such domain:
a) **Structured** refers to meshes composed by structured cubes inside the domain, Figure \[fig:meshes\] (a).
b) **Tetra** is a constrained Delaunay tetrahedralization of $\Omega$, Figure \[fig:meshes\] (b).
c) **CVT** refers to a Centroidal Voronoi Tessellation, i.e., a Voronoi tessellation where the control points coincide with the centroid of the cells they define, Figure \[fig:meshes\] (c).
d) **Random** is a Voronoi diagram of a point set randomly displayed inside the domain $\Omega$, Figure \[fig:meshes\] (d).
-------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------
![Adopted mesh types: (a) structured, (b) tetra, (c) CVT and (d) random.[]{data-label="fig:meshes"}](cube "fig:"){width="37.00000%"} ![Adopted mesh types: (a) structured, (b) tetra, (c) CVT and (d) random.[]{data-label="fig:meshes"}](tetra "fig:"){width="37.00000%"}
(a) (b)
![Adopted mesh types: (a) structured, (b) tetra, (c) CVT and (d) random.[]{data-label="fig:meshes"}](voro "fig:"){width="37.00000%"} ![Adopted mesh types: (a) structured, (b) tetra, (c) CVT and (d) random.[]{data-label="fig:meshes"}](rand "fig:"){width="37.00000%"}
(c) (d)
-------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------
We would like to underline that the last type of mesh will severely test the robustness of the proposed method. Indeed, **Random** meshes are characterized by elements whose faces can be very small and distorted, see the detail in Figure \[fig:meshes\] (d).
The tetrahedral meshes are generated via `tetgen` [@si2015tetgen], while the last two are obtained by exploiting the c++ library `voro++` [@voroPlusPlus]. To analyze the error convergence rate, we make, for each family, a sequence of four meshes with decreasing size. For each mesh we define the mesh-size as $$h := \frac{1}{L_P}\sum_{P\in\Omega_h} h_P\,.$$
Let $({\boldsymbol{u}},p)$ and $({\boldsymbol{u}}_h,p_h)$ be the continuous and discrete VEM solution of the Stokes (or Navier-Stokes problem) under study. To evaluate how this discrete solution is close to the exact one, we use the following error measures, that make use of the local projection described in Proposition \[prp:proj\]:
- **$H^1$–velocity error**: $$e_{H^1}^{{\boldsymbol{u}}} := \sqrt{\sum_{P\in\Omega_h}\big|\big| \nabla{\boldsymbol{u}}- {\boldsymbol \Pi}_{k-1}^{0, P}\,\nabla{\boldsymbol{u}}_h \big|\big|^2_{L^2(P)}}\,,$$ the theoretical expected convergence rate is $h^k$ (cf. );
- **$L^2$–pressure error**: $$e_{L^2}^p := \sqrt{\sum_{P\in\Omega_h}\big|\big|p - p_h \big|\big|^2_{L^2(P)}}\,,$$ the expected rate is $h^{k}$ (cf. ).
Numerical tests
---------------
In this subsection we consider three different tests. In the first two examples, we numerically verify the theoretical trend of all the errors for a Stokes and Navier–Stokes problem. Finally, we propose two benchmark examples for the Stokes equation with the property of having the velocity solution in the discrete space ${\boldsymbol{V}}_h$. It is well known that classical mixed finite element methods lead in this situations to significant velocity errors, stemming from the velocity/pressure coupling in the error estimates. This effect is greatly reduced by the presented methods (cf. Theorem \[thm:u\], estimate ).
#### Example 1 (Stokes problem).
In this section we solve the Stokes problem on the unit cube $[0\,,1]^3$, the discreted version being as in but without the trilinear form $c_h(\cdot\,;\cdot,\cdot)$. We consider Neumann homogeneous boundary conditions on the faces associated with the planes $x=0$ and $x=1$. The load term and the Dirichlet boundary conditions on the remaining faces are chosen in such a way that the exact solution is $${\boldsymbol{u}}(x,\,y,\,z):=
\left(
\begin{array}{r}
\phantom{-2}\sin(\pi x)\,\cos(\pi y)\,\cos(\pi z)\\
\phantom{-2}\cos(\pi x)\,\sin(\pi y)\,\cos(\pi z)\\
{-2}\cos(\pi x)\,\cos(\pi y)\,\sin(\pi z)
\end{array}
\right)$$ and $$p(x,\,y,\,z):= -\pi\,\cos(\pi x)\,\cos(\pi y)\,\cos(\pi z)\,.$$
We consider the **Structured**, **CVT** and **Random** meshes. In Figure \[fig:ese1Conv\] we show the behaviour of the errors $e_{H^1}^{{\boldsymbol{u}}}$ and $e_{L^2}^p$. The slope of such errors are the expected ones, $O(h^k)$ see Theorem \[thm:u\]. Moreover, for each approximation degree $k$ the convergence lines associated with different meshes are close to each other and this represents a numerical evidence that the proposed method is robust with respect to the adopted meshes.
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Example 1 Stokes problem: convergence lines for **Structured**, **CVT** and **Random** meshes and degrees $k=2,3$ and 4.[]{data-label="fig:ese1Conv"}](ese1H1.pdf "fig:"){height="37.00000%"} ![Example 1 Stokes problem: convergence lines for **Structured**, **CVT** and **Random** meshes and degrees $k=2,3$ and 4.[]{data-label="fig:ese1Conv"}](ese1PL2.pdf "fig:"){height="37.00000%"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
#### Example 2 (Navier–Stokes problem).
In this subsection we consider the Navier–Stokes problem described in Equation with Dirichlet boundary conditions. We consider the same discretization of the unit cube of the previuos example, i.e. the set of meshes **Structured**, **CVT** and **Random**. We define the right hand side and the boundary conditions in such a way that the exact solution is $${\boldsymbol{u}}(x,\,y,\,z):=
\left(
\begin{array}{r}
\phantom{-2}\sin(\pi x)\,\cos(\pi y)\,\cos(\pi z)\\
\phantom{-2}\cos(\pi x)\,\sin(\pi y)\,\cos(\pi z)\\
{-2}\cos(\pi x)\,\cos(\pi y)\,\sin(\pi z)
\end{array}
\right)$$ and $$p(x,\,y,\,z):= \sin(2\pi x)\,\sin(2\pi y)\,\sin(2 \pi z)\,.$$
We solve the nonlinear problem by using standard Newton-Rapson iterations with a stopping criterion based on the displacement convergence test error with a tolerance $\texttt{tol=1e-10}$, i.e. until $||{\mathbf{x}}_n-{\mathbf{x}}_{n+1}||<\texttt{tol}\,||{\mathbf{x}}_n||$ where ${\mathbf{x}}_n$ refers to the solution at the $n$-step. In Figure \[fig:ese2Conv\] we show the convergence lines of the $H^1$ error on the velocity and the $L^2$ error on the pressure, respectively. In all these cases we have the predicted trend: $h^{k}$ for the velocity and $h^k$ for the pressure, see Theorem \[thm:u\]. Moreover, also in this case the lines are close to each other varying the mesh discretization, expecially for the velocity solution. Note that, for the pressure solution and random meshes, higher order case $k=3$, there seems to be a loss of accuracy at the second step. We believe this is due to difficulties related to the Newton convergence iterates (the associated linear system getting quite badly conditioned) since random meshes have a very bad geometry and we are reaching near the memory limit of our platform. We where unable to run a further step due to memory limits. Improving this aspect, possibly by exploring more advanced solvers or changing the adopted virtual element basis [@dassi-mascotto:2018], is beyond the scope of the present paper.
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Example 2 Navier–Stokes problem: convergence line for **Structured**, **CVT** and **Random** meshes and degrees $k=2$ and 3.[]{data-label="fig:ese2Conv"}](ese3H1.pdf "fig:"){height="37.00000%"} ![Example 2 Navier–Stokes problem: convergence line for **Structured**, **CVT** and **Random** meshes and degrees $k=2$ and 3.[]{data-label="fig:ese2Conv"}](ese3PL2.pdf "fig:"){height="37.00000%"}
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
#### Example 3 (Benchmark Problems).
In this paragraph, inspired by [@benchmark], we consider a particular example to numerically show the an advantage of the proposed method. It is well known that the error on the velocity field of standard inf-sup stable elements for the Stokes equation is pressure dependent [@boffi-brezzi-fortin:book]. Consequently, the accuracy of the discrete solution ${\boldsymbol{u}}_h$ is affected by the discrete pressure error. As already shown for the two-dimensional case in [@BLV:2017], also in the three-dimensional case we do not have such dependency on the error, i.e. the error on the discrete velocity field ${\boldsymbol{u}}_h$ does not depend on the pressure, but *only* on the velocity ${\boldsymbol{u}}$ and on the load term ${\boldsymbol{f}}$ (see Theorem \[thm:u\], estimate ). Note that the present method, although div-free, is not pressure-robust in the sense of [@benchmark] since the error on the velocities is indirectly affected by the pressure through the loading approximation term [@BLV:2017]. Nevertheless it is still much better then the inf-sup stable element in this respect, as the accuracy of the load approximation (being a known quantity) can be easily improved.
To numerically verify such property we consider two Stokes problems where the exact velocity field is contained in $\textbf{V}_h$ $${\boldsymbol{u}}(x,\,y,\,z):=
\left(
\begin{array}{c}
k\,x\,z^{k-1}\\
k\,y\,z^{k-1}\\
(2-k)\,x^k+(2-k)\,y^k-2\,z^k
\end{array}
\right)\,,
$$ where $k$ is the VEM approximation degree, but we vary the solution on the pressure. More specifically we will consider these two pressure solutions: a polynomial pressure $$p_1(x,\,y,\,z) := x^k\,y+y^k\,z+z^k\,x-\frac{3}{2(k+1)}\,,$$ and an analytic pressure $$p_2(x,\,y,\,z) := \sin(2\pi x)\,\sin(2\pi y)\,\sin(2\pi z)\,.$$
Note that in both cases, since $p_i\not\in Q_h$ for $i=1,2$, a standard inf-sup stable element of analogous polynomial degree would obtain $O(h^k)$ error for the velocities in the $H^1$ norm even if ${\boldsymbol{u}}\in\textbf{V}_h$. In the first case, the velocity is a polynomial vector field of degree $k$, while the pressure is a polynomial of degree $k$ and the load term ${\boldsymbol{f}}$ is a polynomial of degree $k$. In such configuration the presented VEM scheme yields the exact solution up to machine precision for the velocity field. Indeed, the velocity virtual element space contains polynomials of degree $k$ and, since the load term is a polynomial of degree $k$, the term $\mathcal{H}({\boldsymbol{f}},\nu)$ in Equation is close to the machine precision, i.e. we approximate exactly the load term ${\boldsymbol{f}}$ (cf. Definition ), so the error on ${\boldsymbol{u}}$ is close to the machine precision.
In table of Figure \[fig:ese3Poly\] left, we collect the errors $e_{H^1}^{{\boldsymbol{u}}}$ only for the coarsest meshes composed by 27 and 68 elements for the **Structured** and **Tetra** meshes, respectively.
[|c|c|c|]{}\
$k$ &**Structured** &**Tetra**\
`2` &`1.0576e-13` &`7.2075e-13`\
`3` &`2.7333e-13` &`1.1927e-12`\
`4` &`1.5266e-12` &`2.2718e-10`\
![Example 3 Benchmark problem: the values of the errors $e_{H^1}^{{\boldsymbol{u}}}$ for the coarsest meshes of **Structured** and **Tetra** meshes, left, pressure convergence lines, right, for a Stokes problem where we consider $p_1$ as pressure.[]{data-label="fig:ese3Poly"}](ese2PolyPressL2.pdf){width="100.00000%"}
In the second case the velocity is still a polynomial of degree $k$, but, since the pressure is a sinusoidal function, now the right hand side ${\boldsymbol{f}}$ is not a polynomial. Even if the velocity virtual element space contains the polynomial of degree $k$, the error is affected by the term $\mathcal{H}({\boldsymbol{f}},\nu)$ in Equation , i.e. it is affected by the polynomial approximation we make of the load term so the expected error for $e_{H^1}^{{\boldsymbol{u}}}$ is $h^{k+2}$, which is still much better than $O(h^k)$.
In Figure \[fig:ese3Sin\] we show the convergence lines for both $e_{H^1}^{{\boldsymbol{u}}}$ and $e_{L^2}^{p}$. The error trends are the expected ones: we get $O(h^4)$, $O(h^5)$ and $O(h^6)$ for degrees $k=2,3$ and 4, respectively, while we get $O(h^k)$ for the pressure. In the last step of the $H^1$ norm error, the error is higher than expected (this behaviour is due to machine algebra effects since we are in a range of very small errors).
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Example 3 Benchmark: convergence lines for a Stokes problem with **Structured** and **Tetra** meshes where we consider a sinusoidal pressure function, $p_2$.[]{data-label="fig:ese3Sin"}](ese2SinVelH1.pdf "fig:"){height="37.00000%"} ![Example 3 Benchmark: convergence lines for a Stokes problem with **Structured** and **Tetra** meshes where we consider a sinusoidal pressure function, $p_2$.[]{data-label="fig:ese3Sin"}](ese2SinPressL2.pdf "fig:"){height="37.00000%"}
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Appendix {#appendix .unnumbered}
========
The aim of this appendix is addressing the well-posedness of the biharmonic problem with the non homogeneous boundary conditions stated in Definition . Indeed, although in the literature one can find many references for the homogeneous case [@bendali-dominguez-gallic:1985; @amrouche-et-al:1998; @girault-raviart:book], to the authors best knowledge the extension to the non homogeneous case is labeled as feasible but never explicited. For completeness, we here provide the details.
We first recall that the space ${\boldsymbol \Psi}(P)$ is provided with the norm [@girault-raviart:book]: $$\|{\boldsymbol{\psi}}\|_{{\boldsymbol \Psi}(P)}^2 :=
\|{\boldsymbol{\psi}}\|_{0, P}^2 + \|{\boldsymbol{{\rm curl}}}\, {\boldsymbol{\psi}}\|_{1, P}^2 + \|{{\rm div}}\, {\boldsymbol{\psi}}\|_{1, P}^2 \,.$$ Moreover, if $P$ is a contractible polyhedron the following bounds hold (Lemma 5.2 [@girault-raviart:book]) $$\label{eq:norm_equivalence}
\|{\boldsymbol{\psi}}\|_{0, P}^2 + \|{\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\|_{0, P}^2
\lesssim
\|{\boldsymbol{\psi}}\|_{{\boldsymbol \Psi}(P)}^2
\lesssim
\|{\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\|_{0, P}^2
\qquad \text{for all $ {\boldsymbol{\psi}}\in {\boldsymbol \Psi}_0(P)$.}$$
We start our analysis by recalling the following result concerning the case of homogeneous boundary conditions (see Lemma 5.1 [@girault-raviart:book]).
\[lm:girault-raviart\] Let $P$ be a contractible polyhedron and let ${\boldsymbol F} \colon {\boldsymbol \Psi}_0(P) \to {{\mathbb}{R}}$ be a given continuous functional. The biharmonic problem coupled with homogeneous boundary conditions $$\left \{
\begin{aligned}
& \text{find ${\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}_0(P)$, such that} \\
& \int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= {\boldsymbol F}({\boldsymbol{\psi}})
\qquad & \text{for all $ {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,}
\end{aligned}
\right .$$ has a unique solution ${\boldsymbol{{\varphi}}}$.
The next theorem extends the well-posedness result of the previous lemma to the case of inhomogeneous boundary conditions.
\[thm:wellposed\] Let $P$ be a contractible polyhedron and let
- ${\boldsymbol h} \in [L^2(\partial P)]^3$ such that for any $f$, $f_1$, $f_2 \in \partial P$ and for any $e \subseteq f_1 \cap f_2$ $${\boldsymbol h}_{\tau} \in H({{\rm div}}_f, \, f) \cap H({{\rm rot}}_f, \, f)
\qquad \text{and} \qquad
({\boldsymbol h}_{f_1} \cdot {\boldsymbol{t}}_e)_{|e} = ({\boldsymbol h}_{f_2} \cdot {\boldsymbol{t}}_e)_{|e} \,,$$
- ${\boldsymbol g} \in [H^{1/2}(\partial P)]^3$ such that $$\label{eq:NS_bs_compatibility}
{\boldsymbol g} \cdot {\boldsymbol{n}}_P^f = {{\rm rot}}_f \, {\boldsymbol h}_{\tau} \qquad \text{for any $f \in \partial P$,}$$
- ${\boldsymbol f} \in [L^2(P)]^3 \cap {\boldsymbol{Z}}(P)$.
The biharmonic problem coupled with the non homogeneous boundary conditions $$\label{eq:NS_bs}
\left \{
\begin{aligned}
& \text{find ${\boldsymbol{{\varphi}}}\in \boldsymbol{\Psi}(P)$, such that} \\
& \int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P {\boldsymbol f} \cdot {\boldsymbol{\psi}}\, {\rm d}P
\qquad & \text{for all $ {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$,} \\
& \int_{\partial P} {\boldsymbol{{\varphi}}}\cdot {\boldsymbol{n}}_P \,{\rm d}f = 0 \,, \\
& {\boldsymbol{{\varphi}}}_{\tau} = {\boldsymbol h}_{\tau} \qquad & \text{on $\partial P$,} \\
& {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}= {\boldsymbol g} \qquad & \text{on $\partial P$,}
\end{aligned}
\right .$$ has a unique solution ${\boldsymbol{{\varphi}}}$.
Let us consider the following auxiliary problem $$\label{eq:NS_bs_partial}
\left \{
\begin{aligned}
& \text{find ${\boldsymbol{{\varphi}}}^{\partial} \in \boldsymbol{\Psi}(P)$, such that} \\
&
\begin{aligned}
& \int_{\partial P} {\boldsymbol{{\varphi}}}^{\partial} \cdot {\boldsymbol{n}}_P \, {\rm d}f = 0 \,, \\
& {\boldsymbol{{\varphi}}}_{\tau}^{\partial} = {\boldsymbol h}_{\tau} \qquad & \text{on $\partial P$,} \\
& {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}^{\partial} = {\boldsymbol g} \qquad & \text{on $\partial P$.}
\end{aligned}
\end{aligned}
\right .$$ We construct by hand a suitable ${\boldsymbol{{\varphi}}}^{\partial}$ that satisfies .
Let us consider the Stokes-type problem defined on $P$ $$\label{eq:NS_bs_stokes}
\left\{
\begin{aligned}
& \text{find $({\boldsymbol{u}}, p) \in [H^1(P)]^3 \times L^2_0(P)$, such that} \\
&
\begin{aligned}
& - {\boldsymbol{\Delta}}\, {\boldsymbol{u}}+ \nabla p = {\boldsymbol 0}
\qquad & \text{in $P$}
\\
& {{\rm div}}\, {\boldsymbol{u}}= 0
\qquad & \text{in $P$}
\\
& {\boldsymbol{u}}= {\boldsymbol g}
\qquad & \text{on $\partial P$}
\end{aligned}
\end{aligned}
\right.$$ then by Theorem 3.4 [@girault-raviart:book], there exists a vector potential ${\boldsymbol{{\varphi}}}^{\boldsymbol g}$ (possibly not unique) satisfying $$\label{eq:NS_bs_ffig}
\left\{
\begin{aligned}
& \text{${\boldsymbol{{\varphi}}}^{\boldsymbol g} \in {\boldsymbol \Psi}(P)$, such that} \\
&
\begin{aligned}
& {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol g} = {\boldsymbol{u}}\qquad & \text{in $P$,}
\\
& {{\rm div}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol g} = 0
\qquad & \text{in $P$.}
\end{aligned}
\end{aligned}
\right.$$ Moreover implies that $-{\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol g} = {\boldsymbol{{\rm curl}}}\, {\boldsymbol{u}}$, thus the following stability estimate holds [@boffi-brezzi-fortin:book] $$\label{eq:NS_bs_ffig_stability}
\|{\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol g}\|_{0, P}
= \|{\boldsymbol{{\rm curl}}}\, {\boldsymbol{u}}\|_{0, P}
\leq \| {\boldsymbol{u}}\|_{1, P}
\lesssim |{\boldsymbol g}|_{1/2, \partial P} \,.$$
Notice that from , , and , on each face $f \in \partial P$, we infer $${{\rm rot}}_f ({\boldsymbol h} - {\boldsymbol{{\varphi}}}^{\boldsymbol g})_{\tau} =
{{\rm rot}}_f \, {\boldsymbol h}_{\tau} - {{\rm rot}}_f \, {\boldsymbol{{\varphi}}}^{\boldsymbol g}_{\tau} =
{\boldsymbol g} \cdot {\boldsymbol{n}}_P^f - ({\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol g})_{|f}\cdot {\boldsymbol{n}}_P^f = 0 \,.$$ Therefore it can be shown that there exists $\zeta \in H^1(\partial P)$ such that $$\label{eq:NS_bs_face}
{\nabla}_f \, \zeta_{\tau} = ({\boldsymbol h} - {\boldsymbol{{\varphi}}}^{\boldsymbol g})_{\tau}
\qquad \text{on any $f \in \partial P$.}$$ Now we consider the elliptic problem $$\left \{
\begin{aligned}
& \Delta \, \omega = 0 \qquad \text{in $P$,} \\
& \omega = \zeta \qquad \text{on $\partial P$.}
\end{aligned}
\right .$$ We observe that ${\boldsymbol{{\varphi}}}^{\boldsymbol h} := {\nabla}\, \omega$ satisfies, also recalling , $$\label{eq:NS_bs_ffih}
\left\{
\begin{aligned}
& \text{${\boldsymbol{{\varphi}}}^{\boldsymbol h} \in {\boldsymbol \Psi}(P)$, such that} \\
&
\begin{aligned}
& {\boldsymbol{{\rm curl}}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol h} = {\boldsymbol 0}
\qquad & \text{in $P$,}
\\
& {{\rm div}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol h} = 0
\qquad & \text{in $P$,}
\\
& {\boldsymbol{{\varphi}}}^{\boldsymbol h}_{\tau} = {\boldsymbol h} - {\boldsymbol{{\varphi}}}_{\tau}^{\boldsymbol g}
\qquad & \text{on $\partial P$.}
\end{aligned}
\end{aligned}
\right.$$ From it holds that $$\label{eq:NS_bs_ffih_stability}
{\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}^{\boldsymbol h} = {\boldsymbol 0} \,.$$ By construction ${\boldsymbol{{\varphi}}}^{\partial} := {\boldsymbol{{\varphi}}}^{\boldsymbol g} + {\boldsymbol{{\varphi}}}^{\boldsymbol h}$ satisfies and from and it holds that $$\label{eq:NS_bs_ffip_stability}
\| {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}^{\partial}\|_{0, P} \lesssim |{\boldsymbol g}|_{1/2, \partial P} \,.$$ We consider now the homogeneous auxiliary problem $$\label{eq:NS_bs_0}
\left \{
\begin{aligned}
& \text{find ${\boldsymbol{{\varphi}}}^{\rm hom} \in \boldsymbol{\Psi}_0(P)$, such that} \\
& \int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}^{\rm hom}\cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
= \int_P {\boldsymbol f} \cdot {\boldsymbol{\psi}}\, {\rm d}P
- \int_P {\boldsymbol{\Delta}}\, {\boldsymbol{{\varphi}}}^{\partial} \cdot {\boldsymbol{\Delta}}\, {\boldsymbol{\psi}}\, {\rm d}P
\qquad & \text{for all $ {\boldsymbol{\psi}}\in \boldsymbol{\Psi}_0(P)$.}
\end{aligned}
\right .$$ Being ${\boldsymbol f} \in [L^2(P)]^3$, from , and Lemma \[lm:girault-raviart\], Problem has a unique solution ${\boldsymbol{{\varphi}}}^{\rm hom} \in \boldsymbol{\Psi}_0(P)$. It is straightforward to see that ${\boldsymbol{{\varphi}}}:= {\boldsymbol{{\varphi}}}^{\rm hom} + {\boldsymbol{{\varphi}}}^{\partial}$ is a solution to Problem . The uniqueness easily follows from the norm equivalence .
Acknowledgements {#acknowledgements .unnumbered}
================
The authors were partially supported by the European Research Council through the H2020 Consolidator Grant (grant no. 681162) CAVE, Challenges and Advancements in Virtual Elements. This support is gratefully acknowledged.
[^1]: [email protected]
[^2]: [email protected]
[^3]: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The bulk of the extragalactic background between 10 keV and 10 GeV is likely to be explained by the emission of Seyfert galaxies, type Ia supernovae, and blazars. However, as revealed by the INTEGRAL satellite, the bulge of our galaxy is an intense source of a 511 keV gamma-ray line, indicating the production of a large number of positrons that annihilate. The origin of the latter is debated, and they could be produced, in particular, by the ($S$- or $P$-wave) annihilation of light Dark Matter particles into $\,e^+e^-$. In any case, the cumulated effect of similar sources at all redshifts could lead to a new background of hard $X$-ray and soft gamma-ray photons. On the basis of the hierarchical model of galaxy formation, we compute analytically the SNIa contribution to the background, and add it to Seyfert and blazars emission models. Confronting these expected contributions to observation, we find that any extra contribution to this unresolved background around 511 keV should be lower than about 4 keV cm$^{-2}\!$ s$^{-1}\!$ sr$^{-1}$. We also estimate analytically the extragalactic background due to Dark Matter annihilation, increasing the accuracy of the earlier computations. Indeed, we take into account the large positron escape fraction from low mass dark matter halos, unable to confine a dense and magnetized interstellar medium. Our new background estimate turns out to be one order of magnitude [*lower*]{} so that the hypothesis of a light Dark Matter candidate remains compatible with the observed extragalactic background for a wider range of particle masses and cross-sections.'
author:
- Yann Rasera
- Romain Teyssier
- Patrick Sizun
- Michel Cassé
- Pierre Fayet
- Bertrand Cordier
- Jacques Paul
date: 'March 9, 2006'
title: 'Soft gamma-ray background and light Dark Matter annihilation\'
---
Introduction
============
The cosmic gamma-ray background (CGB) between $10$ keV and $10$ GeV has been measured by several gamma-ray satellites (HEAO, SMM, COMPTEL and EGRET) [@Zdziarski96; @Strong04]. Below 100 keV, it is believed that the main contribution comes from Seyfert galaxies [@Zdziarski96][^1]. Above 10 MeV, a simple model for blazars reproduces both the amplitude and the slope of the data [@Comastri99]. In the intermediate energy range, however, another type of sources is needed, since blazar spectra show a clear break near 10 MeV and the cosmological gamma-ray background from Seyfert galaxies falls off above about 100 keV (see Fig. \[contrib\]). As discussed by several authors [@Watanabe99; @Strigari05; @Ruiz-Lapuente01], type Ia supernovae could make a significant contribution in this energy range, which we shall evaluate in Section 2.
![Diffuse background spectrum as a function of photon energy inspired by Fig. 4 of @Strigari05 (we remove the error bars for sake of visibility). The crosses (HEAO), stars (COMPTEL) and diamonds (EGRET) correspond to the observations [@Zdziarski96; @Strong04]. At low energy, Seyfert galaxies (dashed line) are the main contributors [@Zdziarski96]. At intermediate energy, Type Ia supernovae (continuous line), as calculated in this article, dominate. At high energy, blazars (dot-dashed line) explain the observed cosmological gamma-ray background [@Comastri99]. Altogether the sum of the three contributions (dotted line) is a factor of 2 below the observations near 511 keV (indicated by an arrow). An additional contribution from light Dark Matter particles, of up to about 4 keV cm$^{-2}$ s$^{-1}$ sr$^{-1}$, is not excluded.[]{data-label="contrib"}](1.ps){width="\hsize"}
Furthermore, the recent observation, by the INTEGRAL satellite, of a 511 keV diffuse emission line from the galactic bulge[@Knodlseder05] shows that electron-positron annihilation are taking place there with a very large rate $\simeq \,1.5\ 10^{43}$ s$^{-1}$. Such a high rate is difficult, if not impossible, to impute to astrophysical objects, and the source of these positrons in the bulge is subject to intense debate. This emission from the bulge of the Milky Way could be the signature of light Dark Matter particles annihilating into $e^+e^-$ [@boehmfayet; @fayet04], the positrons eventually annihilating with electrons encountered in the interstellar medium [@boehmetal]. In any case, if one extrapolates this diffuse emission to all other galaxies in the Universe, the integrated flux could make a significant additional contribution to the hard $X$-ray and soft gamma-ray background.
Since the potential implications of such an hypothesis for particle physics and cosmology are very important, we want to test here its validity or at least its consistency, as far as the cosmological gamma-ray background is concerned. Using a recent model of galaxy formation [@Rasera05], we compute self-consistently the gamma-ray background coming from both type-Ia supernovae (adding Seyfert-galaxies and blazars), and annihilating positrons from light Dark Matter halos. We follow in this respect the earlier work of Ahn & Komatsu [@Ahn05], who were the first to compute this background, assuming, like them, however, that positrons annihilate “on the spot” in all Dark Matter halos in the Universe. These authors explored various scenarios for the internal structure of Dark Matter halos and analyzed how these various models affect the amplitude of the gamma-ray background.
In the present paper, we would like to go one step further, and explicitly take into account the role of baryons in the process of positron confinement and annihilation. This process is indeed possible only if the parent halo contains enough baryons (and therefore also electrons) to host a dense, magnetized, interstellar medium. This is a necessary condition for the “on the spot” approximation to be valid. Since low mass halos are unable to host enough baryons in a cold and magnetized disc [@Rasera05], the escaping positron mean free path increases dramatically. To compute the positron escape fraction and their propagation in the expanding background, a complex diffusion study would be necessary. As a first order approximation we shall assume that the positron escape fraction goes from zero to one below the critical mass for a Dark Matter halo to host a galaxy, as computed in [@Rasera05], and that these escaping positrons never annihilate.
The outline of this article is as follows. In Section 2, we estimate the contribution of SNIa to the gamma-ray background, adding it to that of Seyfert galaxies and blazars (at lower and higher energies, respectively). The difference with the observed spectrum provides an upper limit on any additional contribution such as the one due to annihilating positrons, that could come from light Dark Matter particle annihilation. In Section 3, we calculate the diffuse cosmological background induced by all Dark Matter halos in the Universe, taking into account that positrons cannot annihilate in small mass halos, and compare it to the previous calculation performed by Ahn & Komatsu [@Ahn05]. In Section 4, we present the gamma-ray background constraints on the annihilation cross-sections and the masses of light Dark Matter candidates. We summarize our main conclusions in Section 5. In an Appendix, we estimate for calibration purpose, and confront with SPI/INTEGRAL observations, the 511 keV emission from the galactic bulge. We consider various annihilation cross-sections (depending on whether they are $S$- or $P$- wave dominated) and the corresponding halo Dark Matter density profiles.
Diffuse gamma-ray background from SNIa {#Diffuse}
======================================
The type Ia supernovae contribution to the gamma-ray background depends primarily on the star formation history in the Universe, which, in this paper, is derived from a new self-consistent model of galaxy formation [@Rasera05]. This analytical model predicts the cosmological evolution of the four main baryon phases in the Universe: diffuse intergalactic gas, hot gas, cold gas in galaxies and stars. These theoretical predictions were validated with high-resolution cosmological simulations using the RAMSES [@Teyssier02] and GADGET codes [@Springel03b]. They also reproduce the observed amount of cold gas in the Damped Lyman-Alpha systems [@Somerville01; @Pei99] and the observed Cosmic Star Formation Rate [@Hughes98; @Steidel99; @Flores99; @Glazebrook99; @Yan99; @Massarotti01; @Giavalisco03; @Dahlen04] (see Fig. \[snr\]), which is of prime interest here. The model particularly emphasizes the important cosmological role of the minimal mass for a halo to host galaxies, $M_{\rm min}(z)$.
We shall use it to, first, compute the SNIa gamma-ray background, and, also, to evaluate the new background that could be attributed to the annihilation of positrons, possibly generated in annihilating Dark Matter halos. This internal consistency allows us to perform a fair comparison between the two types of gamma-ray sources.
![Comoving supernova rates as a function of redshift from our analytical model. The dashed line shows the SNII rates and corresponds to the star formation rate multiplied by the fraction of SNII per unit of stellar mass formed ($\epsilon_{SNII} \simeq 0.007$ M$_\odot^{-1}$). The dot-dashed line represents the SNIa rate used for computing the diffuse gamma-ray background. Both rates are compatible with the observations from @Dahlen04 (symbols) and with the upper limit from @Strigari05 (dashed region).[]{data-label="snr"}](2.ps){width="\hsize"}
In the general case, the background intensity $I_\nu$ is given by $$I_{\nu}\ =\ \frac{c}{4\pi}\ \int^{t_H}_0 j_{\,\nu}(\,\nu(1+z)\,,\,z\,)\ \,dt\ ,
\label{sum}$$ with $j_{\nu}(\nu,z)$ the comoving emissivity at redshift $z$, or time $t$.
The gamma-ray lines from SNIa result from the explosive synthesis of radioactive $^{56}$Ni nuclei, decaying successively into $^{56}$Co and $^{56}$Fe. The comoving emissivity from SNIa may be expressed as $$j_{\nu}(\nu,z)\ \simeq\ h\nu\ \dot{\rho}_*(t-t_{SN})\ \epsilon_{SN}\
\frac{M_{ej}}{A_{\rm Ni}\,m_p}
\ S_{SN}(\nu)\ ,$$ with $\ \dot{\rho}_*(t-t_{SN})\,$ the comoving star formation rate at time $t-t_{SN}$ from our analytical model (see Fig. \[snr\]), $t_{SN}$ denoting the average delay between star formation and SNIa explosion. $\epsilon_{SN}$ is the number of SNIa per unit of stellar mass formed, $M_{\rm ej}$ the mass of Nickel produced and $S_{SN}(\nu)$ the average spectrum per Nickel nucleus. We take $\,t_{SN}\simeq 2.5\,$ Gyr, $\epsilon_{SN}=1.4\times10^{-3}$ M$_\odot^{-1}$ (so that the resulting SNIa rates be within the 2$\sigma$ error bars of observed rates [@Dahlen04]), $M_{\rm ej}\simeq 0.5$ M$_{\odot}$, and the spectrum $S(\nu)$ as computed in [@Nomoto84].
The resulting extragalactic background spectrum from SNIa is presented in Fig. \[contrib\]. It shows a bump in the range 300 keV up to 3 MeV, at a level which turns out to be close to the predicted contributions from Seyfert galaxies and blazars. This contribution, in agreement with the SNIa contribution from @Strigari05, is slightly higher than their preferred model [^2], because our star formation history is slightly more efficient, as suggested by recent observations [@Rasera05].
Altogether the resulting evaluation from known astrophysical sources reproduces reasonably well the observed extragalactic background below $100$ keV and above $3$ MeV. However, in the range from $100~$keV to $3$ MeV the three contributions fall short of explaining the bulk of the Cosmic Gamma-Ray background (as emphasized by @Strigari05). Particularly, in the $100$ keV-$511$ keV range of interest in this article, the sum of the three contributions appears to be lower with a difference of the order of 4 keV cm$^{-2}$ s$^{-1}$ sr$^{-1}$. This sets an upper limit on a possible Dark Matter annihilation signal.
Diffuse background from cosmological halos
==========================================
Diffuse background
------------------
The diffuse background is simply the sum of the redshifted emissions from positron annihilation in all cosmological halos, in principle at all redshifts (Eq. \[sum\]). The comoving emissivity can be computed by summing up the individual halo emissivities $$\label{comoving}
j_{\nu}(\nu,z)\ = \,\int_{M_{\rm min}}^\infty \!M\ \frac{dN}{d\,\ln M \,dV}\
\frac{L_{\nu}(M,z)}{M}\ \,d\,\ln M\,$$ where $M_{\rm min}$ is the minimal mass for emitting halos. The luminosity per halo integrated up to the radius of the halo $R_{200}$ is $$\begin{aligned}
L_{\nu}(M,z)&=&\int^{R_{200}}_0 P_{\nu}(r)\ 4 \pi \,r^2 \,dr\ , \end{aligned}$$ and $N(M,z)$ is the Press-Schechter [@Press74] mass function for cosmological halos. Considering positron annihilation “on the spot”, the volume emissivity is given by $$\begin{aligned}
\label{emissivity}
P_{\nu}(r)\ =\
\frac{1}{2}\ \,S_{\rm pos}(\nu)\ \,\rho^{\ 2}_X(r) \ \,
\frac{\left< \sigma \,v_{\rm rel}\,(r) \right>}{m_X^{\ 2}}\ \,, \end{aligned}$$ with $\rho_X(r)$ the Dark Matter mass density profile, $\left< \sigma
\,v_{\rm rel}\,(r) \right>$ is the annihilation cross-section, the factor $\frac{1}{2}$ being present only in the case of self-conjugate Dark Matter particles [^3].
Positronium annihilation introduces a specific emission spectrum $S_{\rm pos}(\nu)$, with 25% of the energy injected in the 511 keV line, and the remaining 75% spread over a 3$\gamma$ continuum.
Dark matter density profile
---------------------------
The mass distribution in each Dark Matter halo is in fact quite uncertain. @Ahn05 have explored a wide range of halo density profile parameters. In this paper, we restrict ourselves to Dark Matter distribution parameters at face value, as suggested by $N$- body simulations, based on the following general fitting formula $$\rho_X(r)\propto \frac{1}{x^{\gamma} (1+x^{\alpha})^{\frac{\beta-\gamma}{\alpha}}}\ ,$$ where $x=r/r_{s}$ with $r_s$ the scaling radius corresponding to the concentration parameter $c=R_{200}/r_s$ (typically between $4$ and $40$ depending on halo mass and redshift). $\gamma$, $\alpha$ and $\beta$ control the slope respectively for small ($r<r_s$), intermediate ($r\simeq r_s$) and large radii ($r>r_s$).
The concentration parameter defines whether halos are rather peaked ($c
\simeq 40$) or shallow ($c \simeq 4$). Here again the mean value as a function of redshift and halo mass is given by a fit on cosmological simulations [@Bullock01], $$\label{c}
c=max(4,4 \frac{1+z_c}{1+z})$$ with $z_c$ the collapse redshift given by $M_*(z_c)=10^{-2} M$, $M_*$ being the non-linear mass at a redshift $z$. This formula is valid only for halos greater than $\simeq 10^6$ M$_{\odot}$. The behaviour of the concentration parameter for smaller halo masses, unresolved by numerical simulation, is totally unknown.
The slope at the center of Dark Matter halos is thought to be between $\alpha=1$ and $\alpha=1.5$. We therefore consider two extreme dark matter profiles as given by Navarro, Frenk and White [@Navarro97] ($\alpha=1$, $\beta=3$ and $\gamma=1$) and Moore [@Moore99] ($\alpha=1.5$, $\beta=3$ and $\gamma=1.5$). Note however that both density profiles saturate at very low radius $R_{min}$ when $n(R_{min})\left< \sigma \,v_{\rm rel}\,(r) \right>=1/t_H$ due to self-annihilation ($n(R_{min})=\rho_X(R_{min})/m_X$ is the dark matter numeric density and $t_H$ is the age of the universe).
Annihilation cross-section
--------------------------
The relic abundance of Dark Matter particles depends on their decoupling temperature, which is a fraction of their mass (i.e. $T_{F\!}= \frac{m_X}{x_F}$ with $\,x_F\simeq\,$ 16 to 20 depending on $m_X$) and is, roughly, inversely proportional to their annihilation cross-section $\,\left< \sigma\,v_{\rm
rel}/c \right>_F\,$ at freeze-out. The values required for a correct abundance, corresponding to $\,\Omega_{\rm dm}\simeq 23\ \% $, are then, for such light particles, of the order of a few (up to $\approx$ 10) picobarns (corresponding to $\left<\sigma\,v _{\rm rel}\right>_F
\approx 10^{-25}$ cm$^3$ s$^{-1}$) [@boehmfayet; @fayet04], depending on whether they are self-conjugate or not, and on the possible velocity-dependence of their annihilation cross-section at freeze-out.
Such values are in any case rather large compared to ordinary weak-interaction cross-sections, especially when dealing with light particles. This necessitates an unusual, more powerful, annihilation mechanism, that could result from the exchanges of a new light neutral gauge boson $U$, or, in the case of spin-0 Dark Matter particles, from the exchanges of new heavy (e.g. mirror) fermions [@boehmfayet; @fayet04; @boehmetal; @Umirror].
A rather large annihilation cross-section could lead to an excessive continuum of gamma-ray photons at various energies (depending on $m_X$). Cross-sections which behave, at least to a large extent, proportionally to $v^2$ ($P$-wave annihilation), may therefore be preferred [@boehmes], especially at lower $\,m_X$. The residual annihilation of Dark Matter particles in bulges of spiral galaxies or in ellipticals would then include a suppression factor that could be, in the pure $P$-wave case, as strong as $\,v_{\rm
halo}^{\,2}/v_F^{\,2}\,\approx\,10^{-5}$. Furthermore, and independently of the above argument, lighter Dark Matter masses $m_X$ tend to be preferred, to avoid excessive gamma-ray production as compared to $\,e^+$ production, in our galaxy [@fayet04; @beacom].
We shall therefore consider annihilation cross-sections parametrized as $\,\sigma \, v_{\rm rel} \,\simeq\, a+b\,v^2$, with $\left< \sigma
\,v_{\rm rel} \right>_F \ \approx 10^{-25}$ cm$^3$ s$^{-1}$ at freeze-out (for a self-conjugate particle – or twice this value, for a non self-conjugate one). And explore in particular, for low-velocity halo particles, the two extreme situations $\left< \sigma
v_{\rm rel} \right>\ \simeq \,a$ ($S$-wave) and $\,\simeq \,bv^2$ ($P$-wave annihilation). The resulting emission profiles deduced from a given Dark Matter profile $\rho_X$ are computed from $\,\left< \sigma v_{\rm rel} \right>\ \rho_X^{\ 2}/m_X^{\ 2}\,$ (cf. Eq. \[emissivity\] for a self-conjugate particle [^4]). See also the Appendix for further comments.
Note that for the pure $P$-wave cross-section, the emissivity now depends on the Dark Matter 3D velocity dispersion [@asc]. We therefore compute $\sigma^2_{3D}$ as a function of the radius by solving the Jeans equation for a NFW or Moore potential. The resulting emission profiles turns out to be less peaked than for a pure $S$-wave cross-section.
Role of the baryons
-------------------
![Redshift evolution of the minimal halo mass $M_{\rm min}$ below which cold disk gas cannot form (see [@Rasera05] for details). We use a reionization redshift $z_r \simeq \,20$ as suggested by WMAP.[]{data-label="tf"}](3.ps){width="\hsize"}
This rather standard approach has been applied to compute the soft gamma-ray background in [@Ahn05], integrating individual halo emissivity over the Press & Schechter distribution (given by Eq. \[comoving\]), using as lower bound of the integration interval the maximum between the Dark Matter free-streaming and the Dark Matter Jeans masses. This leads to a minimal mass ($M_{\rm min}$) equal to a fraction of solar mass. As a consequence the comoving gamma-ray emissivity is dominated by the cumulated emission of numerous small mass halos (if one uses the concentration parameters given by Eq. \[c\] [^5]).
However, as we have already discussed, small mass halos are unable to retain gas and annihilate positrons, and therefore cannot contribute to the gamma-ray background in which we are interested. We take into account the crucial role of the baryons (and associated electrons, and magnetic fields) in confining and annihilating the Dark Matter positrons. This trapping can be achieved only if both the density of the interstellar medium and the galactic magnetic field are sufficient. Indeed cosmological simulations of galaxy formation in the hierarchical framework of structure formation show that baryons cannot collapse and form high-density centrifugally-supported gas discs in halos having a mass lower than a minimal value $M_{\rm min} \,\approx\,10^7-10^{11}$ h$^{-1}$ M$_\odot$.
This critical mass threshold is a key ingredient of the current galaxy formation theory. Gnedin showed that the fraction of baryons decreases strongly in halos smaller than the so-called filtering mass[@Gnedin00], as a consequence of the non-zero temperature of the intergalactic medium which prevents gas from collapsing into too small Dark Matter halos. @Hoeft04 also showed that the halo mass must be greater than the minimal cooling mass; if not, the fraction of baryons is high but galaxies cannot form because cooling is inefficient. The resulting minimal halo mass $M_{\rm min}$ for galaxy formation is then the maximum between the minimal cooling mass and the filtering mass, as computed in [@Rasera05]. The evolution of this minimal mass with redshift is shown in Fig. \[tf\]. It is of course much larger than the one used in [@Ahn05].
As computing accurately the escape fraction of positrons as a function of halo mass is beyond the scope of this paper, we shall consider here, for simplicity, that below $M_{\rm min}$, essentially all positrons escape the halos (and we neglect their contribution to the background), while above $M_{\rm min}$, confinement is supposed to be efficient and all positrons are taken to contribute. Furthermore, the contribution from diffuse baryons in the Universe (baryons which are not in collapsed halos) is negligible in this range of wavelength because the annihilation time scale is larger than the age of the Universe[^6]. Note that, while the “on the spot” approximation, used by Ahn & Komatsu [@Ahn05] for the whole mass range, leads, in our opinion, to an overestimation of the diffuse background, our approach, though more accurate, should lead to an underestimation of the background level.
![Distribution of comoving positron emissivity as a function of halo masses for three different redshifts: $z=0$ (continuous line), $z=6$ (dashed line) and $z=12$ (dot-dashed line). For each redshift, the vertical line indicates the minimal mass for a halo to host a galaxy and therefore to be able to annihilate positrons into gamma-rays. The bold lines delineate the regions, to the right of the vertical lines, where positrons are converted into photons. It follows that only about 10 % of the released positrons are actually converted into gamma-rays.[]{data-label="npointfrac"}](4.ps){width="\hsize"}
Fig. \[npointfrac\] illustrates the distribution of comoving positron emissivity as a function of halo masses for different redshifts. The comoving emissivity is dominated by small mass halos. However, only a small fraction of the halos contains enough gas and magnetic field for the production of gamma-rays: altogether, only a fraction of about 10 % of the emitted positrons are converted into photons. As a consequence, our computation of the gamma-ray background leads to a result about 10 times smaller than the one evaluated by @Ahn05.
More precisely, we have computed the soft gamma-ray background for direct annihilation with a $S$-wave cross-section, a NFW Dark Matter density profile and a particle masse $m_X=20$ MeV in the two different cases : dark matter-based minimal mass as in [@Ahn05] and baryon-based minimal mass as in the present paper. Fig. \[gamma\_sigmav\_ahn\] illustrates the decrease by a factor of 10 using our new approach. Thus, if we would like to reach the same level of background, we would have to divide the particle mass $m_X$ by a factor of $\simeq \sqrt{10}$.
Note that the spectral shape is also modified. The spectrum declines at low energy because it corresponds to high redshift where halos greater than the minimal mass for galaxy formation becomes rare. Note that in @Ahn05, identical results were recovered for the same Dark Matter particle mass, using rather extreme values for the halo concentration parameter $c$. Recall that here we consider halo concentration parameters only at face value, as predicted by $N$-body simulations, but we take into account baryon physics as part of the annihilation mechanism.
![Cosmic gamma ray spectrum produced by NFW profiles with $m_X=20$ MeV, a $S$-wave cross-section and, considering that no positronium is formed at all (for the purpose of comparison with earlier results). The upper dot-dashed curve was computed using the dark matter-based minimal mass (as in @Ahn05). On the contrary, for the lower continuous curve we took into account baryon physics, introducing a minimal mass for a halo to host a galaxy. The crosses and stars are the observational data.[]{data-label="gamma_sigmav_ahn"}](5.ps){width="\hsize"}
Calibration on Milky Way
------------------------
As we have seen in the previous sections, the diffuse gamma-ray background depends on three main quantities. The first is the annihilation cross-section: we are going to explore two extreme cases: $S$-wave and $P$-wave. The second ingredient is the dark matter mass density profile: we are going to test peaked distributions (Moore, $c=15$) and shallow ones (NFW, $c=5$). The last unknown quantity is the dark matter particle mass $m_X$. In this Section, we are going to fix the latter (for a given set of annihilation cross-section and density profile) using the constraint set by the detected galactic signal.
The line emission at 511 keV detected by INTEGRAL from the galactic center region is at a level of $10^{-3}$ ph cm$^{-2}$ s$^{-1}$ [@Knodlseder05]. Several types of astrophysical sources have been considered as potential candidates to explain this emission. However, SNIa fall short sustaining the high positron injection rate [@Casse04; @Prantzos04]. Hypernovae [@Schanne05] and the related gamma-ray bursts [@Casse04; @Schanne05; @Parizot05; @Bertone04] are in a better position, but since the number of massive stars is about ten times larger in the disk than in the bulge, hot spots of 511 keV emission should show up in the disk plane[^7], which is not the case. Low mass X-ray binaries have also been suggested [@Prantzos04], but no 511 keV emission has been observed from these objects. We therefore consider the hypothesis that light Dark Matter particles annihilate into electron-positron pairs, mainly in the galactic center region where the Dark Matter density is at a maximum[@boehmfayet; @fayet04; @boehmetal; @asc]. The resulting low-energy positrons are confined by the magnetic field of the bulge, where they are progressively slowed down by ionisation losses. A large fraction (0.93) forms positronium with ambient electrons and annihilate into two (25% of probability) or three $\gamma$ photons [@Guessoum05]. Positronium formation plays therefore an important role because it decreases by a factor of about 3 the intensity of the $511$ keV line.
The flux of gamma rays (from a direction making an angle $\theta$ with the direction of the galactic center) is given by the integral of the emissivity along the line of sight $$\begin{aligned}
F_{\nu}(\theta)\ =\ \frac{1}{4\pi}\ \int_{\rm los} P_{\nu}(r)\ dl\ , \end{aligned}$$ where $P_{\nu}(r)$ is the volume emissivity. We assume that essentially no annihilation can take place outside the stellar bulge, due to a lack of gas (corresponding roughly to an angle $\theta_{\rm bulge}=16^\circ$).
The resulting profiles are convolved with the INTEGRAL/SPI Point Spread Function (PSF). This method allows a fair comparison between observations and models, and depends only weakly of the poorly known size of the gaseous bulge. For each couple of cross-section - dark matter profile, we have computed which mass $m_X$ fits best the observed level of 511 keV emission. The results are summarized in Table \[tablecalib\].
$m_{\rm X}$ (MeV) $S$-Wave $P$-wave
------------------- ------ ---------- ----------
c=15 1500 2.4
Moore c=10 900 1.2
c= 5 440 0.44
c=15 190 0.42
NFW c=10 110 0.20
c= 5 45 0.060
: This Table summarizes, for different cases, which dark matter particle mass (in MeV) is (or would be) required to reproduce the level of the INTEGRAL signal from the galactic bulge. We explore different cross-sections ($S$-wave or $P$-wave), different inner slopes (Moore or NFW) and different concentration parameters ($c=5-10-15$). (A result smaller than $\frac{1}{2}$ MeV indicates that the signal cannot be reproduced with the annihilation cross-section and density profile considered.)[]{data-label="tablecalib"}
As expected, the most peaked profiles (such as Moore profile) and the most concentrated ones ($c=15$) require the largest values of $m_X$, since for a given mass density the number density, and hence the annihilation rate decreases when the Dark Matter particle mass increases. It is worth noting that there is a factor of $10^2-10^3$ for the mass (corresponding to $10^4-10^6$ in flux) in favor of the $S$-wave cross-section compared to the $P$-wave one. Both the NFW $S$-wave case and the Moore $P$-wave reproduce the total flux of the bulge $511$ keV emission with reasonable Dark Matter particle mass of the order of $m_X \simeq 100$ MeV and $m_X \simeq 1$ MeV, respectively. On the opposite, the NFW $P$-wave case would require masses ($m_X<0.42$ MeV) so small that they are unable to produce 511 keV photons. And the Moore $S$-wave case required such high masses ($m_X>440$ MeV) that they would lead to an excessive bremsstrahlung emission of soft gamma-ray photons [@beacom].
Among the specific cases considered two models (NFW with $S$-wave and Moore with $P$-wave) are therefore favored by the galactic signal. If one considers only the total emission from the bulge, neither the $S$-wave nor the $P$-wave cross-section can be excluded. If one considers the [*emission profile*]{} however, a different conclusion could be drawn: according to a recent article [@asc] the [*shape*]{} of the emission profile could be used to exclude the $P$-wave scenario. We address this interesting question in the Appendix, in which we conclude that, to our opinion, both scenarios cannot be discriminated yet.
Results
-------
![Diffuse background spectrum. Crosses: HEAO data; stars: COMPTEL observations. The continuous line is the sum of the contributions from Seyferts, SNIa and blazars. The dot-dot-dot-dashed line and the dashed line represent the revised positron contribution to the cosmological gamma-ray background, for a $S$-wave cross-section with a NFW profile, and a $P$-wave cross-section associated to a Moore profile, respectively.[]{data-label="fondgamma"}](6.ps){width="\hsize"}
Using the calibration on the Milky Way, we are now able to compute the diffuse background for our two best models: $S$-wave cross-section, NFW profile with a $100$ MeV Dark Matter mass; and $P$-wave cross-section, Moore profile with a $1$ MeV Dark Matter mass. These are used here as specific benchmarks for the the purposes of our analysis, many other intermediate situations being obviously also possible.
As shown in Fig. \[fondgamma\], the background predictions in the two models are at the same level with a slightly different spectral signature. The main conclusion is that the level of the predicted background is more than a factor of 100 below the observed background. Calibrating on the Milky Way, the relative smallness of the obtained results for the cosmic gamma-ray background shows that the light Dark Matter annihilation hypothesis is by far not ruled-out by the current soft gamma-ray extragalactic background constraint.
Note that this conclusion can be applied more generally to other positron sources since the $511$ keV emission from the Milky Way is quite weak compared to the observed background intensity. In order to make this background large, one has to assume that other galaxies have much higher positron production rates than the Milky Way. Indeed, in this case, the Milky Way would not be representative from other halos of the same mass.
Constraints on dark matter candidates
=====================================
![Constraints on the dark matter candidate in the $a-m_X$ plane, with $a$ the velocity-independent part in the annihilation cross-section $\,\sigma\,v_{\rm rel}\,$ and $m_X$ the mass of the dark matter particle. $b$ is fixed by the relic density requirement, so that $\left<\sigma\,v_{\rm rel} \right>_F \simeq \ <a+b\,v^2>_F\
\approx 10^{-25}$ cm$^3$ s$^{-1}$ at freeze-out. The upper line $a\simeq 10^{-25}$ cm$^3$ s$^{-1}$ corresponds to a purely $S$-wave cross-section, and the lower part of the diagram to a $P$-wave dominated one ($a$ being negligible). As $m_X$ cannot be too large, in order not to overproduce gamma and radio continuum from the galactic center, we limit ourselves, conservatively, to the interval 0.511 to 100 MeV (the actually-allowed mass interval could in fact be significantly smaller [@beacom; @Casse05; @Beacom05; @Fayet06] depending on how seriously these other constraints are taken).\
The grey region is the one compatible with the galactic constraint, based on the total level of emission, using the (Moore or NFW) dark-matter distributions of Table I (its upper part corresponds to a Milky-Way emission that would-be $S$-wave-dominated, with $ a$ behaving like $1/m_X^{\ 2}$).\
The dot-dashed line is associated with a dark-matter induced background (for a NFW profile) that would correspond to the missing $4$ keV cm$^{-2}$ s$^{-1}$ sr$^{-1}$, the top left-hand corner above this line being excluded on the basis of the cosmic background data (same for the dashed line with Moore profiles). []{data-label="contraintes"}](7.ps){width="\hsize"}
The main objective of this paper is the computation of the Dark Matter induced background in the 100 keV-511 keV energy range *taking into account the important role of baryons*. Using standard concentration parameters for the Dark Matter halo profiles (as given by N body simulations), the net result of this exercice is a decrease by a factor of 10 of the level of the background emission compared to the precedent computation in [@Ahn05].
Calibration of $\left< \sigma \,v_{\rm rel}\,(r) \right>$ and $m_X^{\
2}$ on the INTEGRAL signal is however uncertain. Indeed, it depends strongly on the Milky-Way Dark Matter profile which is not well known. The grey region in Fig. \[contraintes\] shows the range of allowed parameters corresponding to this calibration. To obtain this domain we have considered a reasonable range of Dark Matter profiles from peaked and concentrated ones (Moore profile and $c=15$) to less peaked and less concentrated ones (NFW and $c=5$). Then we have found for each profile which cross-section and particle mass reproduce the level of the galactic emission. As a result, each mass and cross-section in the grey region could reproduce the level of the bulge emission with a reasonable Dark Matter profile. The corresponding background is below the observed background (see the previous Section). As a consequence, no light Dark Matter candidate is ruled out by the diffuse background constraint and one has to invoke another gamma-ray source in order to explain the missing 4 keV cm$^{-2}$ s$^{-1}$ sr$^{-1}$.
If we consider now that the Milky-Way Dark Matter halo is totally different from other halos of similar mass in the Universe, we could relax the previous calibration and obtain independent constraints on the mass and the cross-section (following in that sense the strategy used in [@Ahn05] on the observed background only). In the present paper, constraints are of course less stringent than in [@Ahn05], since baryon physics has led us to decrease the level of the emission by a factor of 10. As shown in Fig. \[contraintes\], constraints exclude only the upper left hand corner, corresponding to low-mass Dark Matter particles and $S$-wave dominated cross-section. Interestingly, for halos with a Moore profile, the Dark Matter candidate mass and cross-section required to reach the level of the observed background are not so far from the region favored by the galactic constraints.
Conclusions
===========
Having estimated the SNIa contribution in the 100 keV-10 MeV energy range, we have found that an unexplained gamma-ray background emission at most of the order of 4 keV cm$^{-2}$ s$^{-1}$ sr$^{-1}$ remains. As proposed in [@boehmetal], the strong 511 keV emission from the galactic center detected by the INTEGRAL satellite could be explained by light Dark Matter annihilation, and we have verified that the observed emission profile can be reproduced, both for $S$ and $P$-wave annihilation cross-sections. Using the hierarchical model of structure formation, we have computed the corresponding gamma-ray background, and found it to be compatible with current observational bounds, [*if one takes into account the minimal halo mass for galaxy formation*]{}. The new positron-generated (Dark Matter-induced) extragalactic background is in fact overwhelmed by other emissions from SNIa, Seyferts, and blazars. The exclusion of small mass halos as (redshifted) 511 keV photon sources leads to an order of magnitude decrease of the extragalactic flux around 500 keV as compared to earlier studies. The spectral shape of the extragalactic background is also modified in the sense that the number of Dark Matter halos capable of hosting gas rich galaxies decrease very strongly with increasing redshift.
[*Acknowledgements:*]{} The authors would like to thank the anonymous referee for his helpful remarks that have greatly improved the quality of the paper.
[35]{}
natexlab\#1[\#1]{}
bibnamefont
\#1[\#1]{}
bibfnamefont
\#1[\#1]{}
citenamefont
\#1[\#1]{}
url
\#1[`#1`]{}
urlprefix
[ ]{}, [MNRAS]{} **[281]{}**, [L9+]{} ([1996]{}).
[ ]{}, [ ]{}, [ ]{}, **[613]{}**, [956]{} ([2004]{}).
[ ]{}, **[625]{}**, [89]{} ([2005]{}).
[ ]{}, [Astrophysical Letters Com.]{} **[39]{}**, [181]{} ([1999]{}).
[ ]{}, **[516]{}**, [285]{} ([1999]{}).
[ ]{}, [astro-ph 0502150]{} ([2005]{}).
[ ]{}, [ ]{}, [ ]{}, **[549]{}**, [483]{} ([2001]{}).
J. Knödlseder *et al.*, Astron. Astrophys. [**411**]{}, L457 (2003); P. Jean *et al.*, Astron. Astrophys. [**407**]{}, L55 (2003); J. Knödlseder *et al.*, Astron. Astrophys. [**441**]{}, L513 (2003);
C. Bo$\!e$hm and P. Fayet, Nucl. Phys. B [**683**]{}, 219 (2004). [ ]{}, **[70]{}**, [023514]{} ([2004]{}).
C. Bo$\!e$hm [*et al.*]{}, [*Phys. Rev. Lett.*]{} [**92**]{}, 101301 (2004); C. Bo$\!e$hm, P. Fayet and J. Silk, Phys. Rev. D [**69**]{}, (2004) 101302 (2004).
[ ]{} [ ]{}, [Astron. Astrophys.]{} **[385]{}**, [1]{} ([2006]{}).
[ ]{} [ ]{}, **[71]{}**, [021303]{} ([2005]{}); K. Ahn and E. Komatsu, Phys. Rev. D [**72**]{}, 061301 (2005); K. Ahn, E. Komatsu and P. Höflich, Phys. Rev. D [**71**]{}, 121301 (2005).
[ ]{}, [Astron. Astrophys.]{} **[385]{}**, [337]{} ([2002]{}).
[ ]{} [ ]{}, [MNRAS]{} **[339]{}**, [312]{} ([2003]{}).
[ ]{}, [ ]{}, [ ]{}, [MNRAS]{} **[320]{}**, [504]{} ([2001]{}).
[ ]{}, [ ]{}, [ ]{}, **[522]{}**, [604]{} ([1999]{}).
[ ]{}, , **[394]{}**, [241]{} ([1998]{}).
[ ]{}, , **[519]{}**, [1]{} ([1999]{}).
[ ]{}, , **[517]{}**, [148]{} ([1999]{}).
[ ]{}, , [MNRAS]{} **[306]{}**, [843]{} ([1999]{}).
[ ]{}, , **[519]{}**, [L47]{} ([1999]{}).
[ ]{}, [ ]{}, [ ]{}, **[559]{}**, [L105]{} ([2001]{}).
[ ]{}, , **[600]{}**, [L103]{} ([2004]{}).
[ ]{}, , **[613]{}**, [189]{} ([2004]{}).
[ ]{}, [ ]{}, [ ]{}, **[286]{}**, [644]{} ([1984]{}).
[ ]{} [ ]{}, **[187]{}**, [425]{} ([1974]{}).
[ ]{}, , [MNRAS]{} **[321]{}**, [559]{} ([2001]{}).
[ ]{}, [ ]{}, [ ]{}, **[490]{}**, [493]{} ([1997]{}).
[ ]{}, , [MNRAS]{} **[310]{}**, [1147]{} ([1999]{}).
P. Fayet, Nucl. Phys. B [**187**]{}, 184 (1981); B [**347**]{}, 743 (1990); Phys. Lett. B [**142**]{}, 263 (1984).
C. Bo$\!e$hm, T. Ensslin and J. Silk, J. Phys. G [**30**]{}, 279 (2004).
J. Beacom, N. Bell and G. Bertone, Phys. Rev. Lett. [**94**]{}, 171301 (2005).
C. Bo$\!e$hm and Y. Ascasibar, Phys. Rev. D [**70**]{} (2004) 115013; Y. Ascasibar [*et al.*]{}, astro-ph/0507142 (2005).
[ ]{}, **[542]{}**, [535]{} ([2000]{}).
[ ]{}, , [Baryons in Dark Matter Halos]{} ([2004]{}).
[ ]{}, , **[602]{}**, [L17]{} ([2004]{}).
[ ]{}, [astro-ph/0404501]{} ([2004]{}).
[ ]{}, , *[35th COSPAR Scientific Assembly]{}* ([2005]{}), pp. [2307-+]{}.
[ ]{}, , [Astron. Astrophys.]{} **[432]{}**, [889]{} ([2005]{}).
G. Bertone *et al.*, astro-ph/0405005 (2004).
[ ]{}, [ ]{}, [ ]{}, [Astron. Astrophys.]{} **[436]{}**, [171]{} ([2005]{}).
A. C. Robin, *et al.*, Astron. Astrophys. 409, 523 (2003)
R. Launhardt, R. Zylka, and P. G. Mezger, Astron. Astrophys. 384, 112 (2002)
M. Cassé and P. Fayet, astro-ph/0510490
J.F. Beacom and H. Yuksel, astro-ph/0512411
P. Fayet, D. Hooper and G. Sigl, hep-ph/0602169
511 keV emission from the galactic bulge {#mw}
========================================
The computation of the Milky-Way emission profile $F(\theta)$ is not the main goal of this paper, since we only used the integrated emission (and not the shape) for calibration purposes. However, it is essential to compare carefully predicted profiles to the observed one (see the recent paper [@asc]). In this Appendix, we would like to outline some interesting issues concerning this point. Using the same hypothesis as in Section 2, we consider the three different cases of Table \[tablecalib\]: a NFW profile with $c=10$ and a $S$-wave cross-section; a NFW profile with $c=15$ and a $P$-wave cross-section; and finally, a Moore profile with $c=10$ and a $P$-wave cross-section.
![ INTEGRAL/SPI overall counting rate in the 511 keV line as a function of the angular distance to the galactic center. Points with error bars show the *instrumental background substracted* count rate of the SPI camera in the 506 to 516 keV band, averaged over the first year of data. Lines indicate the expected counting rates for three different situations: $S$-wave annihilation cross-section with a NFW Dark Matter profile (continuous line); or $P$-wave cross-section with a NFW (dashed line) or Moore (dotted line) profile, as specified in the text. The figure does not show the actual 511 keV emission profiles, but the results of their *convolution with the spectrometer’s response*, the Point Spread Function (PSF) – which is a crude but more rigourous way in order to compare data with models. The lower dot-dashed line shows the rate (not compatible with the data) that would correspond to a pointlike source. For the calibration, we choose the Dark Matter particle mass so that the observed integral of the signal over the inner 16 degrees is equal to the one from the theoretical profile.[]{data-label="fluxmwtot"}](8.ps){width="\hsize"}
In the pure $S$-wave case with an essentially constant , we have chosen to consider and test, with the above NFW distribution, a Dark Matter particle with mass $m_X =110$ MeV$/c^2$. Let us note however that, given our hypothesis, the same emission profile would have been obtained from the same $\,\rho_X$ but with a mass of $\,\simeq 11$ (or 1.1) MeV$/c^2$ only, and an $a$ term that would be $\,\simeq 100$ (or 10$^4$) times smaller. The corresponding ($S+P$-wave) cross-section would then be [*$P$-wave dominated at freeze-out*]{}, while appearing as [*$S$-wave-dominated*]{} for low-velocity annihilation [*in the galactic center*]{}.
In the pure $P$-wave case on the other hand, for which the annihilation cross-section in the galactic center should be lower (i.e. typically $\,\approx 10^{-30}$ cm$^3$ s$^{-1}$), $m_X$ should in general be taken relatively small, to get (with a correct relic density) a sufficiently intense gamma-ray line. In practice we test $m_X$ about 0.5 (with the NFW profile) and 1.2 MeV$/c^2$ (with the Moore profile).
At first sight, all three tested profiles seem compatible with the observations, within the precision of the analysis. Attempting to discriminate between them (or with analogous ones) would require a careful chi-squared analysis and is beyond the scope of this article, as we are mainly interested here in the total observed intensity and global morphology of the 511 keV emission of the galactic bulge.
If the cross-section for Dark Matter annihilation in halos is velocity-independent ($S$-wave or effectively $S$-wave annihilation), the emissivity of a NFW Dark Matter halo scales near the center as $P_{\nu}\propto r^{-2}$. Convolving this emission with the SPI Point Spread Function (PSF), we obtain the profile presented as a continuous line in Fig. \[fluxmwtot\]. We emphasize here that the profiles shown are the profiles after convolution by the PSF.
If, however, this cross-section is $S$-wave suppressed, the emissivity now depends on the Dark Matter 3D velocity dispersion [@asc]. We then compute $\sigma^2_{3D}$ as a function of the radius by solving the Jeans equation, at first for a NFW potential. The resulting emission profile (shown after convolution as a dashed line in Fig. \[fluxmwtot\]) also turns out to fit the data, although it is [*less peaked*]{}, as the velocities increase with $r$, within the region of interest. Furthermore, as mentioned earlier, to get in this case the appropriate intensity for the 511 keV line we need to consider both rather small values of $m_X$ (about $0.5$ MeV$/c^2$) and somewhat extreme parameters for the Milky -Way Dark Matter halo (choosing $c\,\simeq\,15$).
These restrictions may be avoided to some extent, however, with a [*steeper profile*]{} such as the Moore profile [@Moore99], which enhances the rate of Dark Matter annihilation, especially near the center of the galaxy, so that the resulting emission profile (shown after convolution as a dotted line in Fig. \[fluxmwtot\]) gets now [*more peaked*]{}. With such profiles, $P$-wave annihilation with standard Milky-Way parameters appear to be compatible with the data, even for less small values of $\,m_X$. Furthermore, note that [*if Dark Matter is subject to the stellar gravitational potential that dominates the central region of the galaxy*]{} [@robinetal], with a radial density profile declining roughly as $\,r^{-2}$ [@launhardtetal], the $S$- and $P$-wave cases would be essentially indistinguishable.
To conclude this Appendix, we have verified that one can reproduce the photon flux and distribution observed by INTEGRAL, both for $S$- and $P$ wave cross-sections, with standard Dark Matter profiles $\rho_X$ and appropriate mass $m_X$. Again, in this approach, attempting to further discriminate between emission profiles associated with $S$- or $P$-wave annihilation appears as difficult, given the width of the PSF function and the variety of the Dark Matter profiles, gravitational potential profiles and gas density profiles which may be considered.
[^1]: It is worth noting, in addition, that recent deep INTEGRAL observations [@Krivonos05] of the Coma region suggest that the cosmic ray background above 20 keV cannot be explained in terms of obscured Seyfert galaxies.
[^2]: Our model is close to the upper bound of Fig. \[snr\] in @Strigari05
[^3]: For [*non self-conjugate*]{} Dark Matter particles Eq. (\[emissivity\]) gets replaced by $$\vspace{-2mm}
\ \ \ \ P_{\nu}(r)\ =\ S_{\rm pos}(\nu)\ \ \rho_X(r)
\ \rho_{\bar X} (r)\ \ \frac{\left< \sigma \,v_{\rm rel}\,(r)\right>}
{m_X^{\ 2}}\ \ ,$$ and one can generally assume equal densities for Dark Matter particles and antiparticles, so that $\,\rho_X(r) =\rho_{\bar X}
(r)\,=\,\frac{1}{2} \ \rho_{{\rm tot}\,(X+\bar X)}\, (r)\ $.
[^4]: For [*non self-conjugate*]{} Dark Matter particles, the $\,\frac{1}{2}\ \rho_X^{\,2}$ in Eq. \[emissivity\] gets replaced by $\,\rho_X\,\rho_{\bar X}$ or simply $\,\frac{1}{4}\ \rho_{{\rm tot}\, (X+\bar X)}^{\ \,2}$ while the annihilation cross-section gets doubled, so that the expected emissivity and resulting emission profile remain the same.
[^5]: Moreover the concentration parameters of these small halos are quite uncertain because they are smaller than the numerical resolution of cosmological simulations.
[^6]: Indeed, direct annihilation of positrons having escaped out of small halos as well as annihilation through positronium in the intergalactic medium are negligible (for $z<50$) due to the very low density of the latter (about $2 \ 10^{-7}\ (1+z)^3$ cm$^{-3}$).
[^7]: Except if positrons escape rapidly from the thin disk, where the gas is concentrated.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We introduce and study the adiabatic dynamics of free-fermion models subject to a local Lindblad bath and in the presence of a time-dependent Hamiltonian. The merit of these models is that they can be solved exactly, and will help us to study the interplay between non-adiabatic transitions and dissipation in many-body quantum systems. After the adiabatic evolution, we evaluate the excess energy (average value of the Hamiltonian) as a measure of the deviation from reaching the target final ground state. We compute the excess energy in a variety of different situations, where the nature of the bath and the Hamiltonian is modified. We find a robust evidence of the fact that an optimal working time for the quantum annealing protocol emerges as a result of the competition between the non-adiabatic effects and the dissipative processes. We compare these results with matrix-product-operator simulations of an Ising system and show that the phenomenology we found applies also for this more realistic case.'
author:
- Maximilian Keck
- Simone Montangero
- 'Giuseppe E. Santoro'
- Rosario Fazio
- Davide Rossini
title: 'Dissipation in adiabatic quantum computers: Lessons from an exactly solvable model'
---
Introduction
============
The recent experimental advances in the field of quantum technologies have drastically enhanced our capability to control the quantum coherent dynamics of many-body systems in a variety of physical systems, ranging from atomic and molecular optics, to trapped ions, and cavity/circuit quantum electrodynamics. These progresses have made real the possibility to experimentally realise quantum simulators [@NatPhys_2012] as well as the implementation of the first quantum algorithms [@Corcoles_2015; @Chow_2014].
Together with the progresses in implementing quantum gates and concatenate them, i.e., by realising standard circuit computation [@NielsenChuang], recently adiabatic quantum computation (AQC) [@Fahri_2001] and quantum annealing [@Santoro_2002] have received a tremendous boost thanks to the experiments performed with D-Wave machines [@Denchev_2016; @Lanting_2014; @King_2016; @Venturelli_2015]. The strategy underlying adiabatic quantum computation [@Fahri_2001; @Santoro_2002] is based on the fact that any quantum algorithm can be formulated in terms of identifying the global minimum (ground state) of a given function (Hamiltonian) over a set of many local minima. On the experimental side, quantum effects were seen to survive on eight [@Johnson_2011; @Boixo_2013], sixteen [@Dickson_2013] and even in more than one hundred qubits [@Boixo_2014]. Whether these machines hold already at present the so called “quantum supremacy” or not is still under debate [@Ronnow_2014; @Boixo_2016]. However, it is clearly very important to understand the actual mode of operation of these adiabatic computers, in order to understand the limit of their performances and to push it forward.
A key problem in this framework is to understand the role of dissipation and decoherence on adiabatic quantum computers. This question amounts to understanding the key features that control the adiabatic evolution of a many-body open quantum system.
Let us first state the general problem. Suppose to be able to follow the quantum dynamics of an appropriate time-dependent Hamiltonian: $$H(t) = [1-f(t)] \, H_{\rm in} + f(t) \, H_{\rm fin}, \quad t \in [t_{\rm in}, t_{\rm fin}],
\label{eq:QAnnealing}$$ $f(t)$ being a generic function of time, with $f(t_{\rm in})=0$ and $f(t_{\rm fin}) = 1$. The Hamiltonian $H_{\rm in}$ sets the initial condition as its ground state, while the sought solution to the problem is entailed into the ground state of $H_{\rm fin}$. If the control time is much larger than the typical inverse gap between the ground state and the first excited state, the system will adiabatically follow its instantaneous ground state $|\psi_0(t)\rangle$. Reaching the ground state of $H_{\rm fin}$ by adiabatic evolution is the way AQC works [@Fahri_2001; @Santoro_2002]. As long as the evolution is unitary, the only source of errors is due to excitations generated by non-adiabatic effects in the dynamical evolution. The complexity of the adiabatic algorithms is reflected in the scaling of the minimum gap with the number of qubits. In general, AQC also requires a special form of $f(t)$ to gain a speedup as compared to classical algorithms, see Ref. [@Roland_2002] for a prominent example of an adiabatic Grover search and Ref. [@Torrentegui_2013] for a review. Moreover, optimal controlled ramps can provide additional speedups [@QOC_2011]. An alternative protocol, which is more general than Eq. , would incorporate an extra (possibly non-linear) term, such as $f(t)[1-f(t)]H_{\rm E}$, with $H_E$ being properly chosen Hamiltonian. This may have beneficial effects on the minimum gap of the system, and therefore greatly enhance the AQC performance [@non-linearAQC].
It is however clear that, especially for long annealing times, another important source of defects is related to incoherent fluctuations induced by finite temperature, or more in general by the unavoidable coupling of the system to some external environment. In this case the quantum state of the system will be mixed, described by a density matrix $\rho(t)$, satisfying a dynamical equation of the form: $$\frac{\partial \rho}{\partial t} = - \frac{i}{\hbar} [H(t), \rho] + \mathbb{D}[\rho] .
\label{eq:master}$$ The first term on the right-hand side describes the coherent unitary time evolution, which is ruled by a many-body time-varying Hamiltonian $H(t)$, according to the quantum annealing protocol . The second term on the right-hand side of Eq. accounts for the coupling to the environment, and its form will depend on the nature of noise and dissipation as well as the form of the coupling between the many-body system and the external bath. The dissipator is a completely positive, trace preserving map, that – in general – drives the system into a fixed point, that is a steady state or a steady-state manifold (if the fixed point is not unique). As such, it is inducing a decay in the system towards the steady state. Note that the differential form of Eq. generally requires a Markovian bath. Understanding the effect of dissipation on AQC amounts to quantifying, in some way, how the state of the system deviates from the final ground state, because of the presence of the extra term $\mathbb{D}[\rho]$ in Eq. .
Does the presence of the environment facilitate the reaching of the final ground state or is it detrimental for AQC? It is clear that there cannot be a unique answer to this question: the deviations from the unitary case may depend strongly on the form of $\mathbb{D}[\rho]$ in relation to the type of evolution imposed by $H(t)$. This variety of possible answers is reflected in the wide spectrum of cases already considered in the literature. Within the plethora of possible scenarios, it is however important to establish some general trends that may serve as guidelines in going deeper in this formidable problem.
This type of analysis was first performed [@Patane_2008] in the context of the Kibble-Zurek (KZ) mechanism for defect formation [@Kibble; @Zurek]. The KZ mechanism is related to the fact that, when crossing a gapless critical point of $H(t)$, no matter how slow the variation of the Hamiltonian in time is, the adiabatic theorem is violated and a finite density of defects will be produced. More than thirty years ago, Kibble put forward a scaling argument aimed at predicting the size and the number of such defects [@Kibble], while the mechanism yielding the correct scaling was found later by Zurek [@Zurek], roughly dividing the dynamics in either adiabatic or impulsive, according to the distance from the critical point. The KZ mechanism has been tested in a variety of quantum toy models at zero temperature, including ordered and disordered systems, as well as for crossing isolated or extended critical regions (see, e.g., Ref. [@Dziarmaga_rev] for a review). When the annealing velocity is progressively increased, a crossover behaviour sets in between the KZ scaling and the generation of excitations due to faster quenches, where the dynamics can be described by the underlying classical model [@Silvi_2016].
The proliferation of defects due to Landau-Zener transitions is intimately related to the occurrence of errors in the AQC. While, in the unitary case, the number of defects decreases on increasing annealing time, the environment will be dominant for long annealing times. In this regime, one expects a defect formation which is almost independent on the annealing protocol. This picture was confirmed and detailed in Ref. [@Patane_2008], where the scaling in the crossover between the KZ-dominated regime and the environment-dominated regime was also found. In the presence of spatially correlated noise, additional intermediate regimes emerge due to the comparison of the correlation length of the noise and the correlation length of the system [@Nalbach_2015].
The role of temperature and external noise was further considered in the context of AQC in several papers, showing that in some cases it may be beneficial in reaching the target ground state. Work has been done on the comparison between AQC and classical approaches using thermal hopping [@Boixo_2016b], on the use of quantum diffusion, showing better performance than closed-system quantum annealing [@Smelyanskiy_2017] as well as on the crucial role of noise-induced thermalisation in AQC that can outperform simulated annealing [@Kechedzhi_2016]. Further, the relation of thermally-assisted tunneling to quantum Monte Carlo has been studied in Ref. [@Jiang_2017]. In addition to this, the effect of noise of a thermal environment [@Albash_2015] and decoherence [@Albash_2012] have been studied as well.
It should be kept in mind that the adiabatic dynamics of dissipative many-body system is linked to the understanding of the Landau-Zener problem of a two-level system coupled to an environment, an extensive studied problem in many different areas [@Ao_1989; @Wubs_2006; @Ashhab_2006; @Zueco_2012; @Javanbakht_2015; @Arceci_2017].
The problem of describing the adiabatic dynamics of a many-body open quantum system is a formidable problem and approximations are necessary. It is however of fundamental importance to have some non-trivial examples where the outcomes of the analysis are not hampered by any approximation. The aim of this work is to present some simple, yet non-trivial, examples where the adiabatic dynamics can be analysed in full details. Most importantly, the phenomenology that we will extract is related to the dynamics of certain spin systems that are very close to the relevant implementations of AQC. This means that the exactly solvable models that we consider here may be used as a very useful benchmark to test important approximations in more complex cases.
The bath we will deal with is Markovian. This means that the dissipative term in Eq. can be written in the Lindblad form $$\mathbb{D} [\rho] = \sum_n \kappa_n \Big( L_n \rho L^\dagger_n - \tfrac12 \{ \rho, L^\dagger_n L_n \} \Big) ,
\label{eq:Lindbladian}$$ where $L_n$ are suitable local Lindblad operators that describe the environment (to be defined later) and $\kappa_n$ are the corresponding couplings, which have to be positive for a Markovian Lindblad master equation. The choice of local Lindblad operators that we are going to study does not lead to a thermal state in the steady state. We will dwell in more details on this point later.
We will consider quadratic fermionic models whose Lindblad dynamics can be worked out analytically [@Prosen_2008; @Eisler_2011; @Horstmann_2013], for different types of local system-bath coupling. The dynamics of this class of models can be studied exactly, and it will help us to clarify several features of the interplay between non-adiabatic effects and incoherent transitions due to the external bath. In order to understand to which extent the results we find can be applied to more realistic cases, later we will consider a spin-$1/2$ one-dimensional Ising model. In such case, for the incoherent spin decay/pumping, the master equation cannot be mapped into local fermionic operators, therefore we will resort to a numerical study based on a matrix-product-operator (MPO) representation of the density matrix [@Verstraete_2004; @Zwolak_2004]. As we will discuss in more details in the rest of the paper, the overall phenomenology remains unchanged, thus reinforcing the fact that the exactly solvable models introduced here can be very useful benchmarks. We should remind that, in general, the thermodynamic properties of low-dimensional systems can be strongly affected by the dimensionality: for example, thermal fluctuations wash out quantum fluctuations of finite-temperature systems in one dimension, but not anymore in two dimensions. Furthermore, methods that are very powerful in one dimension, as the Jordan-Wigner transformation, are not applicable in higher dimensions. However, in the special case of free fermions, the system’s thermodynamics is not expected to change with the dimensionality.
In all the situations we have addressed, we find a robust evidence of the fact that an optimal working time for the quantum annealing protocol emerges as a result of the competition between the non-adiabatic effects and the dissipative processes. There is an optimal time for which the final state is closest to the true final ground state. The scaling of such optimal time (and that of the corresponding generated defects) can be accurately predicted by assuming that the number of defects produced during the time evolution is a sum of the two contributions due to non-adiabaticity and dissipation/decoherence.
For larger working times it may happen that, depending on the type of system-bath coupling, an overshooting point sets in, where the density of the generated defects is larger than that for an infinitely slow annealing, which would adiabatically drive the system through the instantaneous steady state. While this kind of behaviour cannot appear in the unitary scenario, where the defects production is monotonic non-increasing with the annealing speed, in the system-bath scenario this can emerge even for small systems, being eventually related to the spectral structure of the Liouvillian and not necessarily to many-body characteristics.
The paper is organised as follows. In Sec. \[sec:Model\] we define the Hamiltonian models, the various dissipation schemes, and the annealing problem under investigation. We then study the departure from the instantaneous ground state, in the presence of dissipative processes which may cause decay or dephasing, both for a translationally invariant free-fermion model \[sec:TraslInv\], and for an Ising spin chain (Sec. \[sec:Ising\]). We end with a discussion of our findings and with the concluding remarks in Sec. \[sec:Conclusions\]. Technical details on the calculations for the quadratic fermionic model are provided in the Appendix.
Adiabatic dynamics with local dissipation: From Ising systems to fermion chains {#sec:Model}
===============================================================================
One of the simplest (and exactly solvable) models exhibiting a quantum phase transition is the one-dimensional Ising chain [@Sachdev]. This is defined by the Hamiltonian $$\label{eq:Ham_Ising}
H(t) = - J \sum_{n} \sigma^{x}_{n}\sigma^{x}_{n+1} - \Gamma(t) \sum_{n} \sigma^{z}_{n},$$ where $\sigma^{\alpha}_{n} (\alpha = x,y,z)$ are the spin-$1/2$ Pauli operators for the $n$-th spin of the chain, while $J$ and $\Gamma(t)$ are respectively the coupling strength between neighbouring spins and the transverse magnetic field. We assume periodic boundary conditions, in such a way as to preserve translational invariance of the model. We will also work in units of $\hbar = 1$ and set the energy scale by fixing $J=1$.
It is possible to realise quantum annealing in the Ising model by tuning the parameter $\Gamma(t)$ in time according to a linear ramping, as for example: $$\Gamma(t) = - t / \tau, \quad \text{for} \ \ t \in (-\infty,0] ,
\label{eq:annealing}$$ where $\tau$ is related to the ramping speed. The choice ensures that, during the annealing procedure, the system will encounter a critical point in which the ground-state energy gap closes and defects will start to appear. Duality arguments [@Fisher_1995] show that the phase transition occurs at $\Gamma_{\rm c}=1$. The system is driven from a paramagnetic phase, where all the spins are aligned along the field direction $z$ \[i.e., the ground state of the initial Hamiltonian $H(t_{\rm in}) \propto -\sum_n \sigma^z_n$, since $\Gamma(t_{\rm in}) = +\infty$\], to a doubly-degenerate ferromagnetic phase, where the spins are all pointing along the coupling direction $x$ \[i.e., the ground state of the final Hamiltonian $H(t_{\rm fin}) = -\sum_n \sigma^x_n \sigma^x_{n+1}$, since $\Gamma(t_{\rm fin}) = 0$\].
Several papers have already addressed the KZ scaling of defects with $\tau$ in the paradigmatic Ising model, both for the clean system [@Zurek_2005; @Dziarmaga_2005] and for the disordered system [@Dziarmaga_2006; @Caneva_2007]. Here we are going to add the effect of the coupling to an external environment, modelled through a master equation of the form in Eq. .
To retain analytic solvability of the full open-system problem [@Prosen_2008; @Eisler_2011], we will however start from a mapped version of the Ising chain into a free-fermion model. The latter can be achieved by employing a Jordan-Wigner transformation (JWT), which maps the spin operators in terms of spinless fermions: $$\sigma_n^- = \exp \Big( i \pi \sum_{m<n} c^\dagger_m c_m \Big) c_n ,
\label{eq:JWT}$$ where $\sigma_n^\pm = \tfrac12 (\sigma_n^x \pm i \sigma_n^y)$, while $c_n$ ($c_n^\dagger$) denotes the fermionic annihilation (creation) operator on site $n$, obeying the anticommutation relations $\{ c^\dagger_m ,c_n \} = \delta_{m,n}$, $\{ c_m, c_n \} = 0$. The resulting Hamiltonian for an Ising chain of length $L$ is quadratic in such operators and reads: $$\label{eq:Ham_quadratic}
H(t) = \sum_{m,n} \Big[ c^{\dagger}_{m} \, A_{m,n}(t) \, c_{n}
+ \tfrac{1}{2} (c^{\dagger}_{m} \, B_{m,n} \, c^\dagger_{n} + \text{H.c.}) \Big],$$ where $A, B$ respectively are a symmetric and an antisymmetric $L \times L$ matrix whose sole non-zero elements are $A_{n,n}(t) = - \Gamma(t), A_{n,n+1} = A_{n+1,n} = - J/2 ,
B_{n,n+1} = - B_{n+1,n} = - J/2$. To enforce periodic boundary conditions, the following matrix elements are non-zero as well: $A_{1,L} = A_{L,1} = (-1)^{N_{\rm F}} J/2$ and $B_{L,1} = - B_{1,L} = (-1)^{N_{\rm F}} J/2$, where $(-1)^{N_{\rm F}}$ denotes the parity of the number of fermions $N_{\rm F} = \sum_{n} c^{\dagger}_n c_n$, which commutes with $H$. The Hamiltonian can be exactly diagonalized using a Bogolibuov transformation [@LSM_1961; @Young_1997].
In the following we will completely relax the requirement on fermion-parity dependent boundary conditions, and we simply assume that anti-periodic boundary conditions for the fermions are always enforced. This assumption is perfectly justified for a purely coherent evolution, where the fermion parity is conserved and the initial ground state has an even number of fermions. The reason for enforcing the requirement even when considering the open-system adiabatic dynamics is that, as explained below, our Lindblad operators change the fermion parity, and it would be impossible to solve the problem by using a parity-dependent boundary conditions. More in detail, we will study below the adiabatic dynamics of $H(t)$ under the action of three different types of memoryless [*local*]{} environments. We model them in such a way that the Lindbladian is a sum of terms that act uniformly ($\kappa_n = \kappa, \; \forall n$) on each site $n$ of the chain: $$\begin{aligned}
i) \;\; & L_{n}^{\rm (1)} = c^{\dagger}_{n} && \mbox{pumping mechanism} , \label{eq:L_pump}\\
ii) \;\; & L_{n}^{\rm (2)} = c_{n} && \mbox{decaying mechanism} , \label{eq:L_decay}\\
iii) \;\; & L_{n}^{\rm (3)} = c^{\dagger}_{n} c_{n} && \mbox{dephasing mechanism} . \label{eq:L_dephas}\end{aligned}$$ Note that while the fermionic dephasing environment can be directly mapped into a dephasing environment of spins, since there is direct local mapping $c^{\dagger}_{n} c_{n} \to \tfrac12 (\sigma^{z}_{n}+1)$, the same is not true for the decay and pumping. Indeed, when mapping from a spin operator $\sigma^{-}_{n}$ ($\sigma^+_n$) to a fermionic operator $c_{n}$ ($c^\dagger_n$), the Jordan-Wigner transformation includes a non-local operator (string). We should stress that the naming of these Lindblad operators has been chosen with respect to their action on a system with a local, diagonal Hamiltonian, and not with respect to their actual effect on the system that we study. The choice of these specific Lindblad operators is motivated by the possibility to study, in an essentially exact way, the competition between the unitary dynamics and dissipative effects, focusing on features that do not depend qualitatively on the form of the coupling to the environment.
In the next sections we will first address analytically the fermionic model with dissipation provided by Lindblad terms as in -, without making any reference to the mapping with spins. The effect of the non-local part of the JWT will be only discussed in Sec. \[sec:Ising\], where we will study numerically the Ising model with a spin-decay mechanism provided by $L_n^{(1a)} = \sigma_n^-$ that generalises Eq. to spins. We restate that this choice of local Lindblad operators does not lead to thermalisation in the steady state, as for this purpose non-local terms would be required. However they cover a special interest since in several experimental implementations, such as circuit-QED, cold-atom settings or trapped ions, this kind of local damping is the relevant one.
In order to quantify the loss of adiabaticity during the annealing protocol, originating both from the closure of the Hamiltonian gap and from the dissipative processes, we are going to study the excess energy $\varepsilon$ per site, at the end of the annealing. The excess energy at a given time $t$ expresses the difference between the instantaneous energy during the annealing, $E(t) = {\rm Tr} \big[ H(t) \, \rho(t) \big]$, where $\rho(t)$ is the solution of the master equation at time $t$, and the ground-state energy $E_0(t) = {\rm Tr} \big[ H(t) |\psi_0(t)\rangle \langle \psi_0(t)| \big]$ of the instantaneous Hamiltonian system described by $H(t)$: $$\label{eq:excess_energy}
\varepsilon(t) = \tfrac{1}{L} \Big\{ {\rm Tr} [ H(t) \, \rho(t) ]
- \langle \psi_0(t)|H(t)|\psi_0(t) \rangle \Big\}.$$ Using the aforementioned Bogoliubov transformation, the second term of Eq. can be computed straightforwardly, while the first term $E(t)$ is non trivial (see the Appendix). For the Ising spin system we will resort to a fully numerical MPO approach. We point out that a related quantity of interest is the density of defects $\mathcal{N} \equiv \tfrac{1}{2L} \sum_{n} \langle 1-\sigma^{x}_{n}\sigma^{x}_{n+1} \rangle$, which, in the case of ordered chains and at the end of the annealing, is equivalent to the excess energy $\varepsilon(0)$, apart from trivial constants.
Free-fermionic system {#sec:TraslInv}
=====================
We first analyse a fermionic system described by the Hamiltonian in Eq. , where anti-periodic boundary conditions are imposed. A Fourier transform drastically helps in the diagonalisation of the unitary problem, since the different momentum modes decouple (see App. \[app:TraslInv\_U\]). Note that for the sake of simplicity, here we only consider one-dimensional systems, but our analysis of fermions can be easily extended to larger dimensionalities, since a larger dimension will affect calculations only by changing the Brillouin zone.
Let us concentrate on the case in which each lattice site is coupled to some external bath through a pumping mechanism as in Eq. . The master equation during the annealing protocol can be easily integrated via a straightforward generalisation of the time-dependent Bogoliubov method already employed by Dziarmaga [@Dziarmaga_2005], as detailed in App. \[app:TraslInv\_Decay\]. The crucial point resides in the fact that, as for the Hamiltonian, the dissipative part of the Lindbladian with $L^{(1)}_n=c^{\dagger}_n$ does not mix the various modes at different momenta, once a Fourier transform has been employed. As a consequence, the density matrix at time $t$ factorizes into different contributions for the various modes: $$\rho(t) = \bigotimes_k \rho_k(t).
\label{eq:RhoT_k}$$ The relevant Hilbert space for each positive momentum $k$ has dimension 4, and thus the Liouvillian dynamics can be easily followed inside it. We recall that, for the unitary Schrödinger dynamics, a further decomposition into independent $2 \times 2$ problems was possible, due to the additional conservation of the fermionic parity (which is now violated by the dissipative decaying terms).
![Final excess energy as a function of the annealing time, for the free-fermion model coupled to an environment which induces a pumping mechanism, as in Eq. : $L_n^{(1)} = c^{\dagger}_n$. The various data sets denote different values of the dissipative coupling $\kappa$, as listed in the legend. Here we simulated the annealing protocol of Eq. for chains of $L=10^3$ sites. Black squares denote data for $\kappa=0$, which obey a power-law behaviour for $\tau > 1$ with the KZ scaling exponent $\gamma = 0.5$ (dashed line).[]{data-label="fig:ExcT_FreeF-Pump"}](ExcT_FreeF-Pump.pdf){width="\columnwidth"}
The excess energy per site $\varepsilon$ during the annealing protocol is thus obtained via a numerical integration of the linearized Liouville equations for each $k$ mode . For numerical convenience, we restricted the initial point of the annealing procedure to $t_{\rm in} = -5 \tau$, and checked that the results are not appreciably affected by this choice [@Caneva_2007]. We studied systems up to $L=10^3$ sites and annealing times up to $\tau = 10^3$; a fourth-order Runge-Kutta integration procedure with time step $dt = 10^{-2}$ has been employed.
Fig. \[fig:ExcT\_FreeF-Pump\] shows the behaviour of the excess energy at the end of the annealing protocol, $\varepsilon(0)$, for various values of the dissipation strength $\kappa$, as a function of the annealing time $\tau$. In the absence of dissipation ($\kappa=0$), we recover the Kibble-Zurek (KZ) scaling [@Zurek_2005; @Dziarmaga_2005] $$\varepsilon(\tau) \sim 1/ \tau^\gamma \quad \mbox{with} \quad \gamma=1/2 ,$$ which can be obtained by the knowledge of the Ising critical exponents associated to the phase transition at $\Gamma_{\rm c}=1$ across which the system is driven. A finite dissipation $\kappa>0$ induces a competition between the KZ mechanism of defect generation due to the crossing of a gapless point (which is progressively reduced, with increasing annealing time $\tau$), and the production of defects generated by the incoherent driving itself. Such competition clearly emerges in Fig. \[fig:ExcT\_FreeF-Pump\] as a non-monotonic behaviour, which generates an optimal working point for the annealing procedure in the presence of dissipation.
Let us now have a closer look at the non-monotonicity, and focus on the optimal (minimal) value $\varepsilon_{\rm opt}$ reached by the excess energy, and on the corresponding annealing time $\tau_{\rm opt}$. Figure \[fig:OptT\_FreeF-Pump\] displays how such quantities depend on $\kappa$. Our numerical data nicely agree with a power-law behaviour over more than two decades of $\kappa$ values, such that $\varepsilon_{\rm opt} \sim \kappa^{1/3}$ and $\tau_{\rm opt} \sim \kappa^{-2/3}$. Below we show that this behaviour can be easily predicted by assuming that the KZ production of defects is totally independent of that generated by the dissipation. The above mentioned competition is thus explained in terms of an incoherent summation of the two (independent) contributions.
![Optimal excess energy $\varepsilon_{\rm opt}$ (orange squares) and corresponding annealing time $\tau_{\rm opt}$ (violet diamonds), as a function of the dissipation strength $\kappa$. Numerical data (symbols) are obtained using the same parameters as in Fig. \[fig:ExcT\_FreeF-Pump\], and nicely follow a power-law behaviour (dashed lines) with slope $1/3$ and $-2/3$, respectively.[]{data-label="fig:OptT_FreeF-Pump"}](OptT_FreeF-Pump.pdf){width="\columnwidth"}
Scaling of the optimal point {#sec:ScalOpt}
----------------------------
We start from the observation that, after the annealing procedure, the final state of the closed system can be easily written as a Bogoliubov state where excitations are provided by pairs of quasiparticles with equal and opposite momenta [@Dziarmaga_2005]: $$|\psi(t_{\rm fin})\rangle = \prod_{k>0} (\alpha_{k} + \beta_{k} \gamma^{\dagger}_{k}\gamma^{\dagger}_{-k}) |0\rangle.$$ Here $|0\rangle$ indicates the Bogoliubov vacuum corresponding to the final ground state of $H(0)$, $\alpha_k$ and $\beta_k$ are complex amplitudes, while the momentum $k$ can take $L/2$ positive values from 0 to $\pi$ (see App. \[app:TraslInv\_U\] for details).
In the dissipative case, we will not only have those doubly excited states $|1_k, 1_{-k}\rangle = \gamma_k^\dagger \gamma_{-k}^\dagger |0\rangle$, but also singly excited states such as $|1_{k}\rangle = \gamma_k^\dagger |0\rangle$ and $|1_{-k}\rangle = \gamma_{-k}^\dagger |0\rangle$, which represent further sources of defects. Indeed, by using the Bogoliubov transformation, we can rotate the Master equation in this frame. This allows us to write down the dynamical equation for $\langle1_{k}|\rho_{k}|1_{k}\rangle$. We find $$\tfrac{\mathrm{d}}{\mathrm{d}t} \langle1_{k}|\rho_{k}|1_{k}\rangle
= \kappa \, \langle0|\rho_{k}|0\rangle \, f(\Gamma,k),$$ with $$\label{eq:scaling_function}
[f(\Gamma,k)]^{-1} = 1 \! + \!
\bigg( \dfrac{\Gamma - \cos k + \sqrt{1 \! + \! \Gamma^2 \! - \! 2\Gamma \cos k}}{\sin^2 k} \bigg)^2$$ for the specific choice of $L_{n} = c^{\dagger}_{n}$. In the adiabatic regime where the KZ scaling argument holds and for small dissipation, the density of defects is much smaller than $1$, so that $\langle0|\rho_{k}|0\rangle$ can be approximated by its initial value $1$. Note that, since the density of defects $\mathcal{N}$ is written as $\mathcal{N} = \sum_{k} \gamma^{\dagger}_{k}\gamma_{k}$ in the Bogoliubov basis, excitations of the form $|1_{k}\rangle$ only contribute to the positive values of $k$, while excitations due to coherent dynamics $|1_{k} 1_{-k}\rangle$ contribute to both, $k$ and $-k$. Following this, the incoherent part of density of defects can be estimated according to $$\mathcal{N}_{\text{inc}} = \dfrac{\kappa}{L} \sum_{k>0}\int_{-\infty}^{0} \mathrm{d}t\ f[\Gamma(t),k]
= \dfrac{1}{2} \kappa \, \tau,$$ where the last equality has been obtained after a change of variables from $t$ to $\Gamma(t) = -t/\tau$, and observing that the summation over $k>0$ after the integral over $\Gamma$ yields a constant factor $L/2$.
Assuming now that the mechanisms of defect generation due to KZ and due to dissipation are unrelated [@Dutta_2016], we have: $$\mathcal{N} \sim \mathcal{N}_{\text{KZ}} + \mathcal{N}_\text{inc}
= \dfrac{1}{2\pi \sqrt{2}} \, \tau^{-1/2} + \dfrac{1}{2} \kappa \, \tau.$$ From this expression for the total density of defects, the optimal annealing time minimizing the defects production can be thus estimated by the condition $\partial_\tau \mathcal{N}(\tau) |_{\tau_{\rm opt}} = 0$. A direct calculation gives $$\tau_{\rm opt} = \left( \dfrac{1}{2\pi \sqrt{2}}\right)^{2/3} \kappa^{-2/3},
\label{eq:TauOpt}$$ with a corresponding density of defects $$\mathcal{N}_{\rm opt} = \mathcal{N}(\tau_{\rm opt}) = \dfrac{3}{2} \left( \dfrac{1}{2\pi \sqrt{2}}\right)^{2/3} \kappa^{1/3} .
\label{eq:NOpt}$$
The predictions given by these equations are in nice agreement with our numerical data shown in Fig. \[fig:OptT\_FreeF-Pump\], keeping in mind that $\varepsilon = 2 \mathcal{N}$.
To further highlight the role of the dissipation during the annealing procedure, we have also analysed the excess energy at the end of the annealing, after subtracting the corresponding excess energy in the absence of dissipation: $$\Delta(\tau) = \varepsilon(\kappa,\tau) - \varepsilon(\kappa=0,\tau).
\label{eq:Delta_def}$$ Note that, in order to properly define the quantity $\Delta$, we have manifested in Eq. the $\kappa$-dependence of $\varepsilon$. After rescaling such quantity as $\Delta(\tau) \to \Delta(\tau) / \kappa$, we observe a fairly good data collapse with $\tau$, as plotted in Fig. \[fig:Rscl\_FreeF-Pump\]. In addition, our data obey a linear scaling as a function of the annealing time except for deviations induced by bigger values of $\kappa$ (rather than by longer annealing times $\tau$) in the regime where the excess energy is nearly saturated to its maximal value (see also Fig. \[fig:ExcT\_FreeF-Pump\]). We have checked that the behaviour of $\varepsilon(\tau) - \varepsilon(+\infty)$ towards saturation decays with a power law as $\sim \tau^{-1}$, which is in accordance with Ref. [@CamposVenuti].
![Final excess-energy difference $\Delta$ as a function of $\tau$, once rescaled by $\kappa$. The various data sets stand for different values of $\kappa$, and correspond to those of Fig. \[fig:ExcT\_FreeF-Pump\], where the same colour code has been used. A straight line indicating the linear scaling in the annealing time $\tau$ is shown in black.[]{data-label="fig:Rscl_FreeF-Pump"}](RsclK_FreeF-Pump){width="\columnwidth"}
The observations made above point toward a substantial independence of the role played by the dissipation, with respect to the KZ mechanism. The incoherent coupling to the external bath acts uniformly and irrespective of the adiabaticity condition ruled by the ground-state energy gap.
Interplay between pumping and decay
-----------------------------------
![Same plot as in Fig. \[fig:ExcT\_FreeF-Pump\], but for a free-fermion model coupled to an environment which induces a decay mechanism, as in Eq. : $L_n^{(2)} = c_n$. We observe the same initial trend as for the pumping mechanism, however for longer annealing times we observe an overshooting before the saturation sets in.[]{data-label="fig:ExcT_FreeF-Dec"}](ExcT_FreeF-Dec.pdf){width="\columnwidth"}
Here we study the interplay between pumping and decaying mechanism and the question whether the steady state of a system subject to both mechanisms is thermal or not. For this, first we focus on the annealing protocol in the presence of a uniform incoherent decay mechanism only, induced by the Lindblad operators $L_n^{(2)} = c_n$. The behaviour of the final excess energy $\varepsilon(\tau)$ as a function of the annealing time is shown in Fig. \[fig:ExcT\_FreeF-Dec\] for different values of the dissipation strength $\kappa$. As one can see from the figure, at relatively small annealing times the trend is qualitatively analogous to that obtained for the incoherent pumping (see Fig. \[fig:ExcT\_FreeF-Pump\]). The non-monotonic behaviour of $\varepsilon(\tau)$ reveals the presence of an optimal working point, where the number of defects is minimal. However for larger times $\tau$ we also recognize the appearance of an overshooting point, where the energy defects become larger than those reached for an infinitely slow annealing. Here as well, we have checked that the behaviour of $\varepsilon(\tau) - \varepsilon(+\infty)$, after such overshooting point, decays with a power law as $\sim \tau^{-1}$, and again a linear scaling with $\kappa$ [@CamposVenuti].
To better highlight the overshooting behaviour, let us recall that, contrary to the incoherent pumping mechanism, the incoherent decay will drastically affect the completely filled ground state of the initial Hamiltonian at $\Gamma(t_{\rm in}) = +\infty$, since it would tend to empty the system and thereby increasing the energy in the system. Consequently in the limit $\tau \to \infty$, where we can assume the system always to be in the instantaneous steady state, its energy will be $E(t) = 2 \Gamma(t) > 0$, so it will approach its final value $E(t_{\rm fin})=0$ from above. Since for $1 \ll \tau < \infty$ we know that the dynamics approximately follows this open adiabatic dynamics, it is reasonable to expect that its instantaneous energy will follow a similar trend, in particular it will approach its final value from above as well, and the corrections due to finite $\tau$ will result in the observed overshooting.
In Fig. \[fig:OptT\_MaxT\_FreeF-Dec\] (upper panel) we analysed the minimum excess energy that is reached at the optimal working point, and the corresponding annealing time. Their behaviour with $\kappa$ again follows a power law which is similar to the pumping case, as discussed in Sec. \[sec:ScalOpt\]. Note that the argument leading to the scaling predictions for a pumping environment holds as well for a decaying environment, only the function in Eq. changes. However, this does not influence the scaling behaviour discussed here, but only the pre-factors. We also stress that, in the decay case, the integral involved in this calculation strongly depends on the value $-5 \tau$ used to replace the initial value of the field by $-\infty$, which is reasonable since we have seen that this environment creates defects already long before the quantum critical region is reached. As a consequence, the scaling behaviour of $\varepsilon_{\rm opt}$ and the corresponding $\tau$ behaves in accordance with Eqs. -.
![Excess energy (orange squares) and corresponding annealing time (violet diamonds) as a function of the dissipation strength $\kappa$, in the presence of an incoherent decay mechanism, both for the optimal working point (upper panel) and for the overshooting point (lower panel). Numerical data (symbols) are obtained using the same parameters as in Fig. \[fig:ExcT\_FreeF-Dec\], and agree with a power-law behaviour (dashed lines) with slopes $1/3$ and $-0.2/3$ (upper), and a constant value as well as a slope of $-1$ (lower).[]{data-label="fig:OptT_MaxT_FreeF-Dec"}](OptT_MaxT_FreeF-Dec.pdf){width="\columnwidth"}
In the lower panel of Fig. \[fig:OptT\_MaxT\_FreeF-Dec\] we have repeated a similar analysis for the maximum excess energy at the overshooting point and the corresponding annealing time, as a function of the dissipation strength. We observe that such annealing time $\tau_{\rm max}$ scales linearly with $\kappa$, while the change of the maximum excess energy $\varepsilon_{\rm max}$ is relatively small, since it varies by less than $10 \%$ over almost two orders of magnitude.
For a better understanding of the overshooting, in Fig. \[fig:InExcT\_FreeF-Dec\] we show the instantaneous excess energies for different annealing times during the protocol. For very small annealing times we see that the instantaneous steady-state energy is far away from the actual dynamics and no overshooting takes place. For long annealing times, the excess energy increases hugely in the beginning and then follows the (open) adiabatic dynamics, while the behaviour is similar for intermediate annealing times, but not as drastic. As a consequence, there is an intermediate regime where the annealing time $\tau$ is big enough such that an overshooting can take place, and the final excess energy $\varepsilon(\tau)$ will be bigger than in the infinite-time limit $\varepsilon(\infty)$.
![Instantaneous excess energy as a function of the external field $\Gamma(t)$, for various annealing times $\tau$ and fixed dissipation strength $\kappa=0.1$ of the decaying environment. One observes that for long annealing times ($\tau =1000$) the system follows the instantaneous steady state, which has an energy $2 \Gamma(t)$, and at the end it saturates toward $\varepsilon(\infty) = 1$. Intermediate times show the same trend but are not following the open adiabatic dynamics as closely, thus resulting in a higher final excess energy $\varepsilon(\tau) > \varepsilon(\infty)$. For very short annealing times ($\tau = 1$), the influence of the dissipation is much smaller and the final excess energy is smaller than $\varepsilon(\infty)$.[]{data-label="fig:InExcT_FreeF-Dec"}](InExcT_FreeF-Dec.pdf){width="\columnwidth"}
To underline the difference between the two kinds of dissipation (pumping/decay), in Fig. \[fig:InExcT\_FreeF-DecPump\] we plotted the instantaneous excess energy for the same parameters ($\tau = 10^3$, $\kappa=10^{-1}$), but different type of dissipation. We observe that, as stated above, in the pumping case the excess energy mostly increases in the last fifth of the protocol, which is close to the quantum critical point. The decaying scenario shows a completely different behaviour, rather following adiabatically the instantaneous steady state of the system.
![Instantaneous excess energy during the annealing protocol as a function of the external field $\Gamma(t)$ for the two different types of dissipation. $\Gamma_c=1$ locates the critical point where the Ising-like quantum phase transition at zero temperature occurs. Here we fixed $\kappa=0.1$.[]{data-label="fig:InExcT_FreeF-DecPump"}](InExcT_FreeF-DecPump.pdf){width="\columnwidth"}
![Final excess energy as a function of the annealing time coupled to an environment which induces both a pumping as well as a decaying mechanism. The various data sets denote different values of the ratio between the two $\eta$ as listed in the legend. Here we simulated the annealing protocol of Eq. for chains of $L=10^3$ sites.[]{data-label="fig:ExcT_FreeF-DecPump"}](ExcT_FreeF-DecPump.pdf){width="\columnwidth"}
Now, we turn our attention to the interplay between pumping and decay and how the overshooting observed for pure decay is influenced. For this we study the final excess energy as a function of the ratio between pumping and decaying, $\eta = \kappa_{\rm pump} / \kappa_{\rm decay}$. In Fig. \[fig:ExcT\_FreeF-DecPump\] we show the results for values of $\eta$ ranging from $0$ (no pumping), where we observe the biggest overshooting, to $1$, where the overshooting is completely disappeared. Note that the optimal working point does not change much with varying $\eta$ since, for the given parameters, the contribution by the pumping mechanism in this regime are far smaller than the one by the decay. The scaling of the maximum value of the overshooting as a function of $\eta$ is shown in Fig. \[fig:MaxT\_FreeF-DecPump\]. An explanation for the diminishing overshooting can be given when looking at the dependence of the instantaneous steady-state energy during the protocol: if $\eta$ is smaller than $1$, it decreases from an initial positive value to $0$ linearly, such that an overshooting is possible. For $\eta = 1$, the instantaneous steady-state energy is constant equal to $0$ such that an overshooting due to adiabatic dynamics is prevented. For $\eta > 1$, the instantaneous steady-state energy approaches $0$ linearly from below, again preventing an overshooting.
![Maximum value of the excess energy at the overshooting point during the annealing protocol, as obtained from the data in Fig. \[fig:ExcT\_FreeF-DecPump\], as a function of the ratio $\eta$.[]{data-label="fig:MaxT_FreeF-DecPump"}](MaxT_FreeF-DecPump.pdf){width="\columnwidth"}
Finally, we comment on the issue of thermalisation: A single qubit subjected to both incoherent pumping ($L^{(1)} = c^{\dagger}$) and decay ($L^{(2)} = c$) processes would relax to a thermal state whose inverse temperature $\beta$ is related to the ratio of the strengths of the two Lindblad operators. Since in our case the translational invariant quadratic Hamiltonian $H(t)$ factorizes into many Hamiltonians (each one describing a mode of pseudo-momentum $k$) whose Hilbert spaces are, each of them, essentially two-dimensional, this might raise the question if our system shows thermalisation as well. Indeed, the steady state of each of these modes can be approximated by a thermal state with very high fidelity ($>98\%$) for the complete range of physical relevant coupling strengths. However, the corresponding inverse temperature $\beta$ ($k_{\rm B} = 1$) of each mode depends on $k$, and therefore the complete steady state is not well approximated by a thermal state of a single parameter $\beta$.
Dephasing {#sec:Dephasing}
---------
Up to now all the discussion was based on a system-bath coupling scheme which induces a decay/pumping mechanism. There is however a complementary effect of decoherence, where the dissipation can generate pure dephasing. This can be easily obtained through diagonal Lindblad terms $L_n^{(3)} = c^\dagger_n c_n$ (which are proportional to the onsite fermionic number operator), as in Eq. . As detailed in App. \[app:TrInv\_deph\], despite the translational invariance, in such case the solution to the master equation cannot be trivially written in a tensor structure as that in Eq. . As a matter of fact, the Lindbladian $\mathbb{D}[\rho]$ now transforms into a non-local object, where the different momentum modes are now coupled together. Therefore it is more suitable to solve a close set of $4L$ differential linear equations for the relevant two-point correlators [@Eisler_2011], see Eq. . By employing a fourth-order Runge-Kutta integration procedure of those equations, with time step $dt = 10^{-2}$, we were able to reach annealing times up to $\tau = 10^3$.
![Final excess energy $\varepsilon(\tau)$ (upper panel) and rescaled difference $\Delta (\tau) / \kappa$ (lower panel) as a function of the annealing time $\tau$, in the free-fermion model coupled to a dephasing environment $L^{(3)}_n = c^\dagger_n c_n$. Here we simulated the annealing protocol of Eq. for chains of $L=501$ sites. The other parameters are set as in Fig. \[fig:ExcT\_FreeF-Pump\].[]{data-label="fig:ExcT_RsclK_FreeF-Deph"}](ExcT_FreeF-Deph.pdf "fig:"){width="\columnwidth"} ![Final excess energy $\varepsilon(\tau)$ (upper panel) and rescaled difference $\Delta (\tau) / \kappa$ (lower panel) as a function of the annealing time $\tau$, in the free-fermion model coupled to a dephasing environment $L^{(3)}_n = c^\dagger_n c_n$. Here we simulated the annealing protocol of Eq. for chains of $L=501$ sites. The other parameters are set as in Fig. \[fig:ExcT\_FreeF-Pump\].[]{data-label="fig:ExcT_RsclK_FreeF-Deph"}](RsclK_FreeF-Deph.pdf "fig:"){width="\columnwidth"}
The main results of our analysis are summarized in Fig. \[fig:ExcT\_RsclK\_FreeF-Deph\], where we plot (upper panel) the excess energy $\varepsilon(\tau)$ at the end of the annealing, as a function of the annealing time $\tau$. Comparing these data with those of Fig. \[fig:ExcT\_FreeF-Pump\], we immediately recognize a qualitatively analogous trend as for the pumping mechanism. In particular, the non-monotonic behaviour again reveals a competing effect between the KZ mechanism and the incoherent dephasing. Quantitative differences are barely visible on the scale of the two figures. We observed a slight worsening of the annealing protocol, for the same value of $\tau$, the excess energy being slightly larger than that of the previous case. As we did previously, we also analysed the excess-energy difference $\Delta(\tau)$ rescaled by $\kappa$ (bottom panel). Its scaling with $\tau$ is completely analogous to that in Fig. \[fig:Rscl\_FreeF-Pump\], with data growing linearly with the annealing time, and eventually deviating for sufficiently large values of $\kappa$ and $\tau$.
Finally we recall that the argument of Sec. \[sec:ScalOpt\] for determining the scaling of the optimal working point for the annealing protocol as a function of $\kappa$ holds also in this case. Indeed the corresponding data (with the same power-laws), shown in Fig. \[fig:OptT\_FreeF-Deph\], are closely similar to those of Fig. \[fig:OptT\_FreeF-Pump\]).
![Optimal excess energy (orange squares) and corresponding annealing time (violet diamonds) as a function of the dissipation strength. Dashed lines denote power-laws with slopes $1/3$ and $-2/3$, respectively for $\varepsilon_{\rm opt}$ and for $\tau_{\rm opt}$. Data are taken from Fig. \[fig:ExcT\_RsclK\_FreeF-Deph\], and refer to the free-fermion model with a dephasing environment.[]{data-label="fig:OptT_FreeF-Deph"}](OptT_FreeF-Deph){width="\columnwidth"}
Summarizing the results of our analysis on the quantum annealing in a translationally invariant free-fermion model interacting with a local environment, the emerging scenario for the different types of dissipation is the following. For all the three incoherent mechanisms we observe a competition which leads to the onset of an optimal working point for the annealing procedure at a given $\tau_{\rm opt}$ rate. On the other side, for larger values of $\tau$ an overshooting point appears only in the presence of a decay mechanism, due to the fact that the instantaneous energy approaches the steady-state value $\varepsilon(\tau)=1$ from below (while the opposite happens for the pumping and for the dephasing). Finally, we analysed how the final excess energy approaches the $\tau \to \infty$ limit: while for pumping and decay we observed a behaviour $|\varepsilon(\tau) - \varepsilon(\infty)| \sim \tau^{-1}$, for dephasing we found $|\varepsilon(\tau)-\varepsilon(\infty)| \sim \exp(-\tau)$.
Ising chain {#sec:Ising}
===========
Let us now go back to the spin-1/2 language and discuss the effects of the coupling to an external bath on the quantum annealing of the Ising chain, Eq. . We first notice that dephasing can be induced by a Lindblad term $L_n^{(3a)} = \sigma^z_n$, which is readily mapped into the local fermionic operator $(2 c^\dagger_n c_n -1)$, through the JWT of Eq. . In such case, one would thus recover the dephasing mechanism for free fermions (we refer to Sec. \[sec:Dephasing\] for details). On the other hand, incoherent pumping/decay would be induced by $L_n^{(1a)} = \sigma^+_n$ and $L_n^{(2a)} = \sigma^-_n$, respectively; in that case, when mapping into fermions, the appearance of the JW string operator forbids an analytic treatment as the one discussed previously. Let us thus concentrate on the latter scenario.
We employ a numerical method based on an efficient approximation of the many-body density matrix in terms of a MPO [@Verstraete_2004; @Zwolak_2004]. We expect this to be valid whenever the amount of correlations in the system is sufficiently small to satisfy an area-law scaling for the bipartite entanglement in the operator space. The time evolution is performed by means of the time-evolving block decimation (TEBD) algorithm, after a Trotter decomposition of the Liouvillian superoperator in the right-hand side of Eq. . In our simulations of the annealing protocol for the Ising model we considered systems up to $L=20$ sites, using MPOs with a bond link $m \approx 250$ and adopting a typical Trotter step $dt=10^{-2}$. We adopted the same time dependence of the field $\Gamma(t)$ as in Eq. , where for practical convenience we started from $t_{\rm in} = -3\tau$, and verified that (on the scales of the figures shown below) the results are not affected by this choice. As detailed below, we found an emerging physical scenario which is consistent to that previously discussed in Sec. \[sec:TraslInv\], already for small sizes $L \gtrsim 10$.
![Final excess energy as a function of the annealing time, for the Ising chain coupled to an environment through Lindblad operators inducing a decay mechanism $L_n^{(2a)}= \sigma^-_n$. The various data sets denote different values of the dissipative coupling $\kappa$, as listed in the legend. Filled symbols and continuous lines refer to chains with $L=20$ sites, while empty symbols with dashed lines are for $L=10$.[]{data-label="fig:ExcT_Ising"}](ExcT_Ising-Dec){width="\columnwidth"}
Our numerical results for the annealing in the presence of incoherent decay, showing the final excess energy as a function of $\tau$ and for various dissipation strengths $\kappa$, are summarized in Fig. \[fig:ExcT\_Ising\]. Despite the KZ power-law scaling cannot be seen for limited system sizes (not even in the absence of dissipation), the non-monotonicity of the various curves for $\kappa \neq 0$ clearly emerges as a result of the open-system dynamics. We ascribe this behaviour to the emerging picture described in Sec. \[sec:TraslInv\], where we discussed much longer systems of free fermions. Indeed, in Fig. \[OptT\_Ising\] we repeated the same analysis for the scaling of the optimal working time $\tau_{\rm opt}$ and of the corresponding optimal excess energy $\varepsilon_{\rm opt}$ with the dissipation strength, finding a similar power-law behaviour. The exponents do agree within $20\%$ of relative difference. We point out that we were not able to fully resolve the overshooting behaviour in this case, since it would require longer annealing times. However this is already visible in Fig. \[fig:ExcT\_Ising\], for the curve corresponding to $\kappa= 0.1$. Moreover, we also checked that the scaling with $\tau$ of the final excess-energy difference $\Delta(\tau)$ is again linear for sufficiently small values of $\kappa$ and $\tau$, as for the fermionic model.
![Optimal excess energy (orange squares) and corresponding annealing time (violet diamonds) as a function of the dissipation strength. Numerical data (symbols) are obtained using the same parameters as in Fig. \[fig:ExcT\_Ising\], for $L=10$, and fairly agree with power laws (dashed lines) of slopes $1/3$ and $-2/3$, respectively.[]{data-label="OptT_Ising"}](OptT_Ising-Dec){width="\columnwidth"}
Conclusions {#sec:Conclusions}
===========
We presented an extensive study of the adiabatic dynamics of free-fermion models, being driven across their quantum critical point, within an open-system approach using local Lindblad operators. Using the excess energy, we quantified the deviations from adiabatic dynamics of the ground state, showing a competition between the unitary dynamics following a KZ mechanism and incoherent defect generation due to dissipation. While being local, the studied environment covers a wide range of possible sources of dissipation —varying from decay and pumping, separately or simultaneously, to dephasing— and at the same time showing a consistent behaviour for all of them: The competition between the two processes leading to an optimal working point. This can be modelled by the Ansatz of independent processes, which brings to a scaling behaviour that predicts the observed optimal working point in a fairly accurate way. For larger annealing times, we highlighted the possibility to observe an overshooting point, where defects become larger than those reached for an infinitely slow annealing. This effect is intrinsically due to the coupling with an external bath, which drives the system toward the steady state according to the Liouvillian dynamics of the master equation.
Furthermore, we studied the one dimensional Ising chain, which is closely related to the free-fermion models, by means of a matrix-product-operator technique, where we found the same behaviour for small system sizes as well, suggesting a generic nature of the observed phenomena. Within the framework of free-fermion models, a generalization to higher dimensions is straightforward.
We thank S. Barbarino and D. Venturelli for fruitful discussions. We acknowledge the CINECA award under the ISCRA initiative, for the availability of high performance computing resources and support. SM gratefully acknowledges the support of the DFG via a Heisenberg fellowship. This research was partly supported by the EU projects QUIC and RYSQ, by SFB/TRR21 and by CRP-QSYNC.
Unitary dynamics {#app:TraslInv_U}
================
Here we provide technical details concerning the dynamics of the free-fermion model in Eq. , where periodic boundary conditions are imposed. Namely, $$H(t) = - \sum_{n=1}^L \Big\{ \big( c^\dagger_n c_{n+1} + c^\dagger_n c^\dagger_{n+1} + {\rm H.c.} \big)
+ 2 \Gamma(t) c^\dagger_n c_n \Big\},
\label{eq:HamTrInv_A}$$ with $c_{L+1} = -c_1$ for the positive parity sector, while $c_{L+1} = c_1$ for the negative parity sector. Here we have implicitly set the coupling strength to one.
The annealing procedure of Eq. in this context has been already studied in Ref. [@Dziarmaga_2005]. The approach consists in employing a Fourier transform of the type $$c_{n} = \dfrac{^{-i\pi/4}}{\sqrt{L}} \sum_{k} c_{k} e^{ikn} ,
\label{eq:Fourier}$$ where the operators $c_{k}^{(\dagger)}$ satisfy canonical anticommutation relations for fermions as well, and the index $k$ takes values (assuming $L$ to be even, without loss of generality) $k = \pm 1 \tfrac{\pi}{L}, \pm 3 \tfrac{\pi}{L}, \dots , \pm (L-1) \tfrac{\pi}{L}$. The resulting Hamiltonian in Fourier space takes the form $$H \! = \! \sum_{k} \!\! \Big\{ 2c^{\dagger}_{k}c_{k} \big[ -\Gamma(t) - \cos k \big] \!
+ \sin k (c^{\dagger}_{k} c^{\dagger}_{-k} + c_{-k}c_{k}) \! \Big\} .
\label{eq:HamTraslInv}$$ Since $H$ conserves the fermionic parity, the global Hilbert space ${\cal H}$ can be written as a direct sum over different $k$ subspaces: ${\cal H} = \oplus_{k>0} {\cal H}_k$, and indeed $H=\sum_{k>0} H_k$ as in Eq. . Each subspace at fixed $k>0$ is built from the two states $\{ |0\rangle , |1_{k}, 1_{-k}\rangle \}$ or $\{ |1_k\rangle , |1_{-k}\rangle \}$, depending on the parity number (even or odd, respectively). The ground state is found in the even parity sector, as one can see from the diagonalization of $H_k$.
The Hamiltonian $H_k$ at a given fixed time $t$, as extrapolated from Eq. , can be readily diagonalized by means of a Bogoliubov transformation [@LSM_1961; @Young_1997] $$c_{k} = u_{k}(t) \gamma_{k} + v^{*}_{-k}(t) \gamma^{\dagger}_{-k} ,
\label{eq:Bogol}$$ so that the ground state is annihilated by all the quasiparticle operators $\gamma_k$. The system dynamics once the parameter $\Gamma(t)$ is varied can thus be found by employing the time-dependent Bogoliubov method [@Barouch_1970], which makes the Ansatz that the instantaneous system wave function $|\psi(t)\rangle$ is annihilated by a set of quasiparticle operators $\tilde \gamma_{k(H)}$, in the Heisenberg representation, which are defined though the transformation $c_{k(H)} = u_{k}(t) \tilde \gamma_{k(H)} + v^{*}_{-k}(t) \tilde \gamma^{\dagger}_{-k(H)}$. This Ansatz satisfies the Heisenberg equation $$\tfrac{\mathrm{d}}{\mathrm{d}t} c_{k(H)} = i [ H(t), c_{k(H)} ],$$ with the constraint $\tilde \gamma_{k(H)}|\psi(t)\rangle = 0$, provided the coefficients $u_k(t)$ and $v_k(t)$ obey the time-dependent Bogoliubov-de Gennes equations $$\begin{aligned}
i \tfrac{\mathrm{d}}{\mathrm{d}t} u_{k} = - 2 u_{k} [\Gamma(t) + \cos k] + 2 v_{k} \sin k ,\\
i \tfrac{\mathrm{d}}{\mathrm{d}t} v_{k} = + 2 v_{k} [\Gamma(t) + \cos k] + 2 u_{k} \sin k .
\end{aligned}$$ These equations can be integrated starting from the initial condition $\Gamma(-\infty)= + \infty$, and thus mapping them into a Landau-Zener problem [@Dziarmaga_2005].
Fermionic decay bath {#app:TraslInv_Decay}
====================
Let us now describe the superimposed action of an environment that induces decay in the system, so the Lindblad operator on each site $n$ is given by $L_{n} = c_{n}$. It is important to stress that, when applying the Fourier transform on the dissipative part of the master equation , this does not mix different modes: $$\mathbb{D} [\rho] \equiv \sum_k \mathbb{D}_k [\rho]
= \kappa \, \Big( \sum_{k} c_k \, \rho \, c^{\dagger}_{k} - \tfrac{1}{2} \{ \rho, c^{\dagger}_{k} c_{k} \} \Big) .$$ The reason resides in the fact that each term contains two fermionic operators $c_n^{(\dagger)}$, such as for example, $\sum_n c_n \, \rho \, c_n^\dagger \to \sum_n \tfrac{1}{L} \sum_{k,k'} c_k \, \rho \, c_{k'}^\dagger e^{-i(k-k')n}$, and thus the exponential factor, once summed over $n$, gives a Kronecker delta $\delta_{k,k'}$.
As a consequence, the density matrix factorizes into $\rho(t) = \otimes_{k>0} \rho_{k}(t)$, and we can decouple the problem into the same $k$ modes as for the non-dissipative case. Notice however that the dissipation part violates parity conservation of fermions, therefore here the different subspaces for a given $k>0$ are built up from the four states $\{ |0\rangle, |1_{k}\rangle, |1_{-k}\rangle, |1_{k}, 1_{-k}\rangle \}$, and not simply from two states. The Hamiltonian in this basis can be explicitly written as $H = \sum_{k>0} H_{k}$, where $$H_{k} = \begin{bmatrix}
0 & 0 & 0 & 2 \sin k\\
0 & -2 (\Gamma + \cos k) & 0 & 0\\
0 & 0 & -2 (\Gamma + \cos k) & 0\\
2 \sin k & 0 & 0 & -4(\Gamma + \cos k)
\end{bmatrix},
\nonumber$$ with $$c_{k} = \begin{bmatrix}
0 & 1 & 0 & 0\\
0 & 0 & 0 & 0\\
0 & 0 & 0 & 1\\
0 & 0 & 0 & 0
\end{bmatrix}
\quad \mbox{and} \quad
c_{-k} = \begin{bmatrix}
0 & 0 & 1 & 0\\
0 & 0 & 0 & 1\\
0 & 0 & 0 & 0\\
0 & 0 & 0 & 0
\end{bmatrix}.
\nonumber$$
As a matter of fact, solving the full quantum dynamics of $\rho(t)$ translates into solving $L/2$ Lindblad equations of dimension 4 for $\rho_k(t)$. In the vectorized form, they can be written as the following linear differential equations with dimension 16 ($k>0$): $$\begin{aligned}
\nonumber
\tfrac{{\rm d}}{{\rm d}t} |\rho_k\rangle\rangle = &
\Big\{ i \big( \mathbb{1} \otimes H_k - H_k \otimes \mathbb{1} \big) \\
- & \tfrac{\kappa}{2} \Big[ \mathbb{1} \otimes \big( c_k^\dagger c_k + c_{-k}^\dagger c_{-k} \big)
+ \big( c_k^\dagger c_k + c_{-k}^\dagger c_{-k} \big) \otimes \mathbb{1} \Big] \nonumber \\
+ & \kappa \big( c_k \otimes c_k + c_{-k} \otimes c_{-k} \big)\Big\} |\rho_k\rangle\rangle .
\label{eq:vectorMEQ}\end{aligned}$$ In practice, for every linear operator $W= \sum_{m,n} W_{mn} |m\rangle \langle n|$ acting on the four-dimensional Hilbert space ${\cal H}_k$ which is spanned by the basis $\{|m\rangle\}_{m = 1,\ldots,4}$, we associate a vector in the 16-dimensional superoperator space ${\cal H}_k \otimes {\cal H}_k$, which is spanned by the basis $\{ |m\rangle \otimes |n\rangle \}_{m,n=1,\ldots,4}$, using the convention $$W_{m,n} \to |W\rangle\rangle_{\phi},\quad \phi = m+(n-1)\, L ,
\label{eq:vectorization_rho}$$ with $|W\rangle\rangle \equiv \sum_{m,n} W_{mn} |m \rangle |n\rangle$. In this way we have that $|W_1 W_2 W_3 \rangle \rangle = (W_1 \otimes W_3^{\rm T}) |W_2\rangle\rangle$, where ${}^\text{T}$ denotes the transpose operation. A vectorization of the master equation for $\rho_k(t)$ using this rule, the fact that $H_k = H_k^{\rm T} = H_k^\dagger$, and that $c_k^\dagger = c_k^{\rm T}$, we finally arrive at Eq. . The excess energy is then readily obtained, since $\rho(t) = \otimes_{k>0} \rho_{k}(t)$.
Fermionic dephasing bath {#app:TrInv_deph}
========================
In the case of dephasing Lindblad operators $L_{n} = c^{\dagger}_{n}c_{n}$ on each site, the Fourier transform applied to $\mathbb{D}[\rho]$ turns out to yield a non-local object, since each term now contains four fermionic operators, and thus it is not possible to decouple the different $k$ modes. Therefore, this kind of dissipation scheme cannot be directly embedded into the Dziarmaga formalism [@Dziarmaga_2005] described above.
In the following, it is more convenient to reduce our study to two-point correlators, since all the relevant quantities for our purposes (such as the excess energy , can be expressed in terms of those correlators. This drastically simplifies the analysis into a closed set of differential equations which scale linearly (or at most quadratically, for the non-homogeneous case) with $L$ [@Eisler_2011]. We define $$\begin{aligned}
F_{m,n} \equiv \langle c^{\dagger}_{m}c_{n} \rangle, & \qquad
G_{m,n} \equiv \langle c_{m}c^{\dagger}_{n} \rangle , \\
I_{m,n} \equiv \langle c^{\dagger}_{m}c^{\dagger}_{n} \rangle, & \qquad
K_{m,n} \equiv \langle c_{m}c_{n} \rangle.
\end{aligned}
\label{eq:correlFGIK}$$ Using anticommutation relations for fermions and the fact that $(c_m c_n)^\dagger = c_n^\dagger c_m^\dagger$, we have $G_{m,n} = \delta_{m,n} - F_{n,m}$ and also $K_{m,n}^* = I_{n,m}$.
Here we adopt the Heisenberg representation, where the dynamics is described by means of an adjoint Lindblad master equation for a given observable $O$: $$\begin{aligned}
\label{eq:adj_Master_eq}
\frac{\mathrm{d}}{\mathrm{d}t} O & = & i [H,O] + \tilde{\mathbb D}[O], \quad \mbox{ where} \\
\tilde{\mathbb D}[O] & = &
\frac{1}{2} \sum_{n=1}^L \kappa_n \Big( L^{\dagger}_{n} \, [O,L_{n}] - [O,L^{\dagger}_{n}] \, L_{n} \Big) .
\label{eq:adj_Diss}\end{aligned}$$ Since in this appendix we deal with the homogeneous cases, we set $\kappa_n=\kappa$.
We first note that $$[c_{m},H] = 2 \sum_{n=1}^{L} A_{m,n} c_{n} + B_{m,n} c^{\dagger}_{n} ,
\label{eq:c_H}$$ where the matrices $A$ and $B$ have been defined in Eq. , with $J_n = h_n = 1$. Moreover, specializing to the dephasing bath $L_{n} = c^{\dagger}_{n} c_{n}$ with uniform couplings $\kappa_n = \kappa$, we have $$\begin{aligned}
\tilde{\mathbb D}[c^{\dagger}_{m}c_{n}] &= - \kappa \, c^{\dagger}_{m}c_{n} (1-\delta_{m,n}),\\
\tilde{\mathbb D}[c_{m}c^{\dagger}_{n}] &= - \kappa \, c_{m}c^{\dagger}_{n} (1-\delta_{m,n}),\\
\tilde{\mathbb D}[c^{\dagger}_{m}c^{\dagger}_{n}] &= - \kappa \, c^{\dagger}_{m}c^{\dagger}_{n} (1+\delta_{m,n}),\\
\tilde{\mathbb D}[c_{m}c_{n}] &= - \kappa \, c_{m}c_{n} (1+\delta_{m,n}).
\end{aligned}
\label{eq:Diss_dephasing}$$ The adjoint Lindblad master equation for the Hamiltonian and the dephasing bath, referred to the operator $c^\dagger_m c_n$, reads: $$\begin{aligned}
\dfrac{\mathrm{d}}{\mathrm{d}t} c^{\dagger}_{m}c_{n} & =
i [H,c^{\dagger}_{m}]c_{n} + i c^{\dagger}_{m}[H,c_{n}] + \tilde{\mathbb D}[c^{\dagger}_{m}c_{n}] \nonumber \\
& = 2i \sum_{j} \big( A_{m,j}c^{\dagger}_{j}c_{n} + B_{m,j}c_{j}c_{n} \nonumber \\
&- \! A_{n,j} c^{\dagger}_{m}c_{j} \! - \! B_{n,j} c^{\dagger}_{m}c^{\dagger}_{j} \big) \!
- \! \kappa c^{\dagger}_{m}c_{n} (1 \! - \! \delta_{m,n})
\label{eq:derF}\end{aligned}$$ and correspondingly, for the other two-point operators, $$\begin{aligned}
\dfrac{\mathrm{d}}{\mathrm{d}t} c_{m}c^{\dagger}_{n} & =
2i \sum_{j} \big( -A_{m,j}c_{j}c^{\dagger}_{n} - B_{m,j}c^{\dagger}_{j}c^{\dagger}_{n} \nonumber \\
& + \! A_{n,j} c_{m}c^{\dagger}_{j} \! + \! B_{n,j} c_{m}c_{j} \big) \!
- \! \kappa c_{m}c ^{\dagger}_{n} (1 \! - \! \delta_{m,n}) , \! \label{eq:derG} \\
%
\dfrac{\mathrm{d}}{\mathrm{d}t} c^{\dagger}_{m}c^{\dagger}_{n} & =
2i \sum_{j} \big( A_{m,j}c^{\dagger}_{j}c^{\dagger}_{n} + B_{m,j}c_{j}c^{\dagger}_{n} \nonumber \\
& + \! A_{n,j} c^{\dagger}_{m}c^{\dagger}_{j} \! + \! B_{n,j} c^{\dagger}_{m}c_{j} \big) \!
- \! \kappa c^{\dagger}_{m} c^{\dagger}_{n} (1 \! + \! \delta_{m,n}) , \! \label{eq:derI} \\
%
\dfrac{\mathrm{d}}{\mathrm{d}t} c_{m}c_{n} & =
2i \sum_{j} \big( -A_{m,j}c_{j}c_{n} - B_{m,j}c^{\dagger}_{j}c_{n} \nonumber \\
& - \! A_{n,j} c_{m}c_{j} \! - \! B_{n,j} c_{m}c^{\dagger}_{j} \big) \!
- \! \kappa c_{m} c_{n} (1 \! + \! \delta_{m,n}) . \! \label{eq:derK}\end{aligned}$$ If we now set $l=n-m, \ l \in [0,L-1]$, for a translational invariant system we can define $\vec {\cal F}$ such that ${\cal F}_{l=n-m} \equiv F_{m,n}$ (and analogously for $\vec {\cal G}$, $\vec {\cal I}$, $\vec {\cal K}$), in such a way that Eq. can be rewritten as $$\begin{aligned}
\dfrac{\mathrm{d}}{\mathrm{d}t} {\cal F}_{l} =
& 2i \sum_{j} \big( A_{m,j} \, {\cal F}_{l+m-j} + B_{m,j} \, {\cal K}_{l+m-j} \nonumber \\
& - A_{m+l,j} \, {\cal F}_{j-m} - B_{l+m,j} \, {\cal I}_{j-m} \big) \! - \! \kappa P_l {\cal F}_l ,
\label{eq:derF2}\end{aligned}$$ where $P_l = 1-\delta_{l,1}$ ($l=1, \ldots L$) are the $L$ components of the vector $\vec P$. The first term on the right-hand side can be manipulated as $$\begin{aligned}
\sum_{j} A_{m,j} {\cal F}_{l+m-j}
&= A_{m,m} {\cal F}_{l} \! + \! A_{m,m+1} {\cal F}_{l-1} \! + \! A_{m,m-1} {\cal F}_{l+1} \nonumber \\
&= A_{l,l} {\cal F}_{l} + A_{l,l-1} {\cal F}_{l-1} + A_{l,l+1} {\cal F}_{l+1} \nonumber \\
&= \sum_{j} A_{l,j} {\cal F}_{j} = \big( A \cdot \vec{\cal F} \big)_{l},\end{aligned}$$ where in the second line we used the fact that $A^{\rm T}=A$; moreover, due to translational invariance of the model, it is possible to shift both indices of $A$ together. Proceeding in an analogous way, for the other terms of we find $$\begin{aligned}
\sum_{j} B_{m,j} {\cal K}_{l+m-j} &= - \big( B \cdot \vec{\cal K} \big)_{l},\\
\sum_{j} A_{l+m,j} {\cal F}_{j-m} &= \big( A \cdot \vec{\cal F} \big)_{l},\\
\sum_{j} B_{l+m,j} {\cal I}_{j-m} &= \big( B \cdot \vec{\cal I} \big)_{l}.\end{aligned}$$
It is possible to follow the same type of calculations for the other two-point correlators, Eqs. -, such that the dynamics of all the two-point correlators defined above can be written in a compact way as a set of time-dependent linear equations $$\dfrac{\mathrm{d}}{\mathrm{d}t}
\begin{bmatrix} \vec{\cal F}\\ \vec{\cal G}\\ \vec{\cal I}\\ \vec{\cal K}\\ \end{bmatrix}
= \left[ M(t) - \kappa \; {\rm diag} \big( \vec P, \vec P, \vec Q, \vec Q \big) \right]
\begin{bmatrix} \vec{\cal F}\\ \vec{\cal G}\\ \vec{\cal I}\\ \vec{\cal K}\\ \end{bmatrix} ,
\label{eq:timedep_homo}$$ where the $4L \times 4L$ matrix $M$ is given by $$M(t) = 2i \begin{bmatrix}
0 & \phantom{-} 0 & -B & -B\\
0 & \phantom{-} 0 & \phantom{-} B & \phantom{-} B\\
B & -B & \phantom{-} 2A(t) & \phantom{-} 0\\
B & -B & \phantom{-} 0 & -2A(t)
\end{bmatrix}$$ and the vector $\vec P$ has been defined above, while the $L$ components of the vector $\vec Q$ are given by $Q_l = 1+\delta_{l,1}$ (with $l = 1, \ldots L$).
Eventually, the instantaneous energy $E(t) \equiv \langle H(t) \rangle$ can be calculated from the quadratic Hamiltonian : $$\begin{aligned}
E &= - \! \sum_{n=1}^{L} (F_{n,n+1} \! - \! G_{n,n+1} \! - \! K_{n,n+1}
\! + \! I_{n,n+1}) \! - \! 2 \Gamma \, F_{n,n} \nonumber \\
&= - L ({\cal F}_{1} - {\cal G}_{1} - {\cal K}_{1} + {\cal I}_{1}) - 2L \, \Gamma \, {\cal F}_{0},
\label{eq:energy_homo}\end{aligned}$$ where we used anti-commutation relations to express all terms such that the resulting index of the translational invariant correlators is non-negative. To match boundary conditions and parity considerations, we use $L$ odd and are therefore in the negative parity sector.
The initial conditions, for a given Hamiltonian $H(t)$ (that is a certain value of $\Gamma$) can be immediately found by a Bogoliubov transformation that generalizes Eq. to non-homogeneous quadratic systems [@Dziarmaga_2006]: $$c_{i} = \sum_{\mu=1}^{L} \big( U_{i,\mu} \gamma_{\mu} + V^{*}_{i,\mu} \gamma^{\dagger}_{\mu} \big) .$$ The transformation satisfies the properties $\langle \gamma_{\mu} \gamma^{\dagger}_{\nu} \rangle_{0} = \delta_{\mu,\nu}$ and $\langle \gamma_{\mu} \gamma_{\nu} \rangle_{0} = \langle \gamma^{\dagger}_{\mu} \gamma^{\dagger}_{\nu} \rangle_{0}
= \langle \gamma^{\dagger}_{\mu} \gamma_{\nu} \rangle_{0} = 0$, where $\langle \dots \rangle_{0}$ indicates the expectation value over the ground state of $H(t_{\rm in})$, that is with $\Gamma = +\infty$. This yields $$\begin{aligned}
F_{m,n} (t_{-\infty}) &= \langle c^{\dagger}_{m} c_{n} \rangle_{0} = \left( V V^{\dagger} \right)_{m,n}, \\
G_{m,n} (t_{-\infty}) &= \langle c_{m} c^{\dagger}_{n}\rangle_{0} = \left( U U^{\dagger} \right)_{m,n},\\
I_{m,n} (t_{-\infty}) &= \langle c^{\dagger}_{m} c^{\dagger}_{n}\rangle_{0} = \left( V U^{\dagger} \right)_{m,n},\\
K_{m,n} (t_{-\infty}) &= \langle c_{m} c_{n}\rangle_{0} = \left( U V^{\dagger} \right)_{m,n}.
\end{aligned}
\label{eq:CondIn}$$ From these equations, exploiting the translational invariance of the system, we can choose the initial conditions of the system by selecting the first column of each of those four matrices.
[100]{}
Nature Physics Insight on [*Quantum Simulators*]{}: Nat. Phys. Vol. [**8**]{}, No.4 (2012).
A. D. Córcoles, E. Magesan, S. J. Srinivasan, A. W. Cross, M. Steffen, J. M. Gambetta, and J. M. Chow, Nat. Commun. [**6**]{}, 6979 (2015).
J. M. Chow, J. M. Gambetta, A. D. Córcoles, S. T. Merkel, J. A. Smolin, C. Rigetti, S. Poletto, G: A. Keefe, M. B. Rothwell, J. R. Rozen, M. B. Ketchen, and M. Steffen, Phys. Rev. Lett. [**109**]{}, 060501 (2012).
M. Nielsen and I. L. Chuang, [*Quantum Computation and Quantum Information*]{} (Cambridge University Press, Cambridge, UK, 2000).
E. Fahri, J. Goldstone, S. Gutmann, J. Lapan, A. Lundgren, and D. Preda, Science [**292**]{}, 472 (2001).
G. E. Santoro, R. Martonak, E. Tosatti, and R. Car, Science [**295**]{}, 2427 (2002).
V. S. Denchev, S. Boixo, S. V. Isakov, N. Ding, R. Babbush, V. Smelyanskiy, J. Martinis, and H Neven, Phys. Rev. X [**6**]{}, 031015 (2016).
T. Lanting et al., Phys. Rev. X [**4**]{}, 021041 (2014).
A D. King, E. Hoskinson, T. Lanting, E. Andriyash, and N. H. Amin, Phys. Rev. A [**93**]{}, 052320 (2016).
D. Venturelli, S. Mandrà, S. Knysh, B. O’Gorman, R. Biswas, and V. Smelyanskiy, Phys. Rev. X [**5**]{}, 031040 (2015).
M. W. Johnson [*et al.*]{}, Nature [**473**]{}, 194 (2011).
S. Boixo, T. Albash, F. M. Spedalieri, N. Chancellor, and D. A. Lidar, Nat. Commun. [**4**]{}, 2067 (2013).
N. G. Dickson [*et al.*]{}, Nat. Commun. [**4**]{}, 1903 (2013).
S. Boixo, T. F. Rønnow, S. V. Isaksov, Z. Wang, D. Wecker, D. A. Lidar, J. M. Martinis, and M. Troyer, Nat. Phys. [**10**]{}, 218 (2014).
T. F. R[ø]{}nnow, Z. Wang, J. Job, S. Boixo, S. V. Isakov, D. Wecker, J. M. Martinis, D A. Lidar, and M. Troyer, Science [**25**]{}, 420 (2014).
S. Boixo, S. V. Isakov, V. N. Smelyanskiy, R. Babbush, N. Ding, Z. Jiang, J. M. Martinis, and H. Neven, arXiv:1608.00263
J. Roland and N. J. Cerf, Phys. Rev. A [**65**]{}, 042308 (2002).
E. Torrentegui, S. Ibáñez, S. Martínez-Garaot, M. Modugno, A. del Campo, D. Guéry-Odelin, A. Ruschhaupt, Xi Chen, and J. G. Muga, Adv. At. Mol. Opt. Phys. [**62**]{}, 117 (2013).
P. Doria, T. Calarco, and S. Montangero, Phys. Rev. Lett. [**106**]{}, 190501 (2011); T. Caneva, T. Calarco, R. Fazio, G. E. Santoro, and S. Montangero, Phys. Rev. A [**84**]{}, 012312 (2011).
E. Farhi, J. Goldstone, and S. Gutmann, arXiv:quant-ph/0208135; M. Stewart Siu, Phys. Rev. A [**75**]{}, 062337 (2007); G. Schaller, Phys. Rev. A [**78**]{}, 032328 (2008); E. Farhi, J. Goldstone, D. Gosset, S. Gutmann, H. B. Meyer, and P. Shor, Quantum Inf. Comput. [**11**]{}, 181 (2011).
D. Patanè, A. Silva, L. Amico, R. Fazio, and G. E. Santoro, Phys. Rev. Lett. [**101**]{}, 175701 (2008).
T. W. B. Kibble, J. Phys. A [**9**]{}, 1387 (1976).
W. H. Zurek, Nature [**317**]{}, 505 (1985).
J. Dziarmaga, Adv. Phys. [**59**]{}, 1063 (2010).
P. Silvi, G. Morigi, T. Calarco, and S. Montangero, Phys. Rev. Lett. [**116**]{}, 225701 (2016).
P. Nalbach, S. Vishveshwara, and A. A. Clerk, Phys. Rev. B [**92**]{}, 014306 (2015).
S. Boixo, V. N. Smelyanskiy, A. Shabani, S. V. Isakov, M. Dykman, V. S. Denchev, M. H. Amin, A. Yu Smirnov, M. Mohseni, and H. Neven, Nat. Commun. [**7**]{}, 10327 (2016).
V. N. Smelyanskiy, D. Venturelli, A. Perdomo-Ortiz, S. Knysh, and M. I. Dykman, Phys. Rev. Lett. [**118**]{}, 066802 (2017).
K. Kechedzhi and V. N. Smelyanskiy, Phys. Rev. X [**6**]{}, 021028 (2016).
Z. Jiang, V. N. Smelyanskiy, S. V. Isakov, S. Boixo, G. Mazzola, M. Troyer, and H. Neven, Phys. Rev. A [**95**]{}, 012322 (2017).
T. Albash and D. A. Lidar, Phys. Rev. A [**91**]{}, 062320 (2015).
T. Albash, S. Boixo, D. A. Lidar, and P. Zanardi, New J. Phys. [**14**]{}, 123016 (2012); G. Vacanti, R. Fazio, S. Montangero, G. M. Palma, M. Paternostro, and V. Vedral, New J. Phys. [**16**]{}, 53017 (2014).
P. Ao and J. Rammer, Phys. Rev. Lett. [**62**]{}, 3004 (1989).
M Wubs, K. Saito, S. Kohler, P. Hänggi, and Y. Kayanuma, Phys. Rev. Lett. [**97**]{}, 200404 (2006).
S. Ashhab, J. R. Johansson, and F. Nori, Phys. Rev. A [**74**]{}, 052330 (2006).
D. Zueco, P. Hänggi, and S. Kohler, New J. Phys. [**10**]{}, 115012 (2008).
S. Javanbakht, P. Nalbach, and M. Thorwart, Phys. Rev. A [**91**]{}, 052103 (2015).
L. Arceci, S. Barbarino, R. Fazio, and G. E. Santoro, Phys. Rev. B [**96**]{}, 054301 (2017).
T. Prosen, New J. Phys. [**10**]{}, 043026 (2008).
V. Eisler, J. Stat. Mech. (2011) P06007.
B. Horstmann, J. I. Cirac, Phys. Rev. A [**87**]{}, 012108 (2013).
F. Verstraete, J. J. García-Ripoll, and J. I. Cirac, Phys. Rev. Lett. [**93**]{}, 207204 (2004).
M. Zwolak and G. Vidal, Phys. Rev. Lett. [**93**]{}, 207205 (2004).
S. Sachdev, [*Quantum Phase Transitions*]{} (Cambridge University Press, Cambridge, UK, 1999).
D. S. Fisher, Phys. Rev. B [**51**]{}, 6411 (1995).
W. H. Zurek, U. Dorner, and P. Zoller, Phys. Rev. Lett. [**95**]{}, 105701 (2005).
J. Dziarmaga, Phys. Rev. Lett. [**95**]{}, 245701 (2005).
J. Dziarmaga, Phys. Rev. B [**74**]{}, 064416 (2006).
T. Caneva, R. Fazio, and G. E. Santoro, Phys. Rev. B [**76**]{}, 144427 (2007).
E. Lieb, T. Schultz, and D. Mattis, Ann. Phys. (N.Y.) [**16**]{}, 407 (1961).
A. Young, Phys. Rev. B [**56**]{}, 11691 (1997).
A. Dutta, A. Rahmani, A. del Campo, Phys. Rev. Lett. [**117**]{}, 080402 (2016).
L. Campos Venuti, T. Albash, D. A. Lidar, and P. Zanardi, Phys. Rev. A [**93**]{}, 032118 (2016); L. Campos Venuti, T. Albash, M. Marvian, D. Lidar, and P. Zanardi, Phys. Rev. A [**95**]{}, 042302 (2017).
E. Barouch, B. M. McCoy, and M. Dresden, Phys. Rev. A [**2**]{}, 1075 (1970).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '[The equivalence between Tsallis Thermodynamics and Hill’s Nanothermodynamics is established. The correct thermodynamic forces in Tsallis thermodynamics are established. Through this connection we also find a general expression for the entropic index $q$ which we illustrate with two physical examples, allowing in both cases to relate $q$ to the underlying dynamics of the Hamiltonian systems.]{}'
address: 'Departament de Termodinàmica, Universitat de València, C/Dr. Moliner 50, E-46100 Burjassot, Spain'
author:
- 'Vladimir García-Morales'
- Javier Cervera
- Julio Pellicer
title: 'Correct thermodynamic forces in Tsallis Thermodynamics: connection with Hill Nanothermodynamics'
---
,
,
Thermodynamics ,Statistical Mechanics ,Critical point phenomena
Introduction
============
During the last decade there has been growing interest in Tsallis (nonextensive) thermostatistics [@Tsallis88; @web]. This formalism has been successfully applied to a wide variety of statistical systems at scales ranging from particle and nuclear physics [@Navarra03] to astrophysics [@Torres97] and situations involving long range interactions [@Long] as well as low dimensional maps and multifractals [@multifractals]. Another important success is the straightforward explanation of the occurrence of Levy distributions in nature [@Levy]. The building block of Tsallis thermostatistics, known as Tsallis entropy, was introduced in $1988$ [@Tsallis88] $$\mathcal{S}^{*}=\frac{\sum_{j}^{\Omega}p_{j}^{q}-1}{1-q} \qquad (q
\in \Re) \label{entropy}$$ Here $p_{j}$ is the probability of the microstate $j$ and $\mathcal{S}^{*}$ is a generalized entropic measure which reduces to the Gibbs-Shannon one $S=-k\sum^{\Omega}_{j}p_{j}\ln p_{j}$ when the entropic parameter $q$ tends to unity. The main feature of this entropic form is its nonadditivity. Given two subsystems $A$ and $B$ of a composite nonextensive system $A+B$ it can be easily checked [@Tsallis88] that $$\mathcal{S}^{*}(A+B)=\mathcal{S}^{*}(A)+\mathcal{S}^{*}(B)+
(1-q)\mathcal{S}^{*}(A)\mathcal{S}^{*}(B) \label{nonadditivity}$$ Despite the wide variety of applications that the formalism has proven [@web] there are some important points concerning Hamiltonian systems that have not yet been elucidated. Among these, the problem of how can $q$ be calculated for these systems has attracted great attention [@qmeaning] and is, arguably, the main opened issue of Tsallis thermodynamics (TT) [@Cho02]. Another point that has not been clarified is what is the correct form of the thermodynamic forces in TT [@Vives02; @AbeToral; @critics]. It has also been stated that Tsallis entropy cannot have a well defined thermodynamics because of its nonadditivity [@Salinas97; @critics].
Quite interestingly also, there exist already another previous important generalization of traditional thermostatistics which is well grounded physically and which was originally conceived for dealing with nonextensive systems [@Chamberlin02]. This formalism is Hill’s Nanothermodynamics [@founds] (NT) and was originated in 1962 generalizing the concept of Gibbs’ chemical potential to complex nonextensive systems that can vary its entropy by fragmentating in smaller subsystems. NT has been applied succesfully since then to ferromagnetism [@Chamberlin00], glassy systems [@Chamberlin99] and liquid-vapor interfaces [@Hill98; @GarciaMorales03]. NT is well grounded physically from its very beginning in contrast with TT which relies in the *ad hoc* (although beautiful) postulate for the entropy inspired in the multifractal formalism [@Tsallis88]. TT has, however, many practical advantages and allows to deal quite straigthforwardly with complex systems (specially those exhibiting power laws statistically since these can be closely described by means of Tsallis q-exponential distributions). NT could be misunderstood as applicable only to small sizes or small numbers of particles. But stars, clusters and multifractals can be considered also as NT systems (the range of the interactions and/or correlations are of the order of the size of the system or longer) and, in general, all systems in which their equilibrium properties depart from the standard description, which relies in the concept of extensivity [@Chamberlin02]. NT is a rigorous statistical foundation of thermodynamic finite-size effects as well as others coming from the complexity of the interactions. It is therefore interesting questioning whether if there is any connection between TT and NT.
In this article we clarify all these points. Specifically: i) We establish the equivalence between TT and NT, ii) this connection allows us then to introduce an adequate form for the thermodynamic forces in TT, overcoming previous difficulties [@critics] iii) the nonadditivity property is then understood for complex systems thermodynamically and, importantly, iv) we provide an expression for $q$ that can be used to *calculate* this quantity in Hamiltonian many particle systems relating it to its internal dynamics.
Formal equations of Tsallis Thermodynamics (TT)
===============================================
We present first previous rigorous results of TT and NT to better clarify our approach. Maximization of the Tsallis entropy imposing the constraints $\langle X_{\alpha} \rangle_q = \sum_{j}^\Omega
p_{j}^{q} X_{\alpha}(j)/\sum_{j}^\Omega p_{j}^{q}$ for the biased average $\langle X_{\alpha} \rangle_q$ of an extensive quantity $X_{\alpha}$ leads to recover a Gibbs-like differential equation for the entropy $$d \mathcal{S}^{*} = \sum_{\alpha} y_{\alpha} d \langle
X_{\alpha} \rangle_q, \label{termo}$$ The rest of the formal equations of TT have been recently established [@Vives02]. The Euler equation of TT is $$\sum_{\alpha} \langle X_{\alpha} \rangle_q y_{\alpha} =
\frac{1}{1-q} \left [ 1+ (1-q) \mathcal{S}^{*} \right ] \ln \left
[ 1+ (1-q) \mathcal{S}^{*} \right ]. \label{Euler}$$ and the Gibbs-Duhem equation has the form $$\sum_{\alpha} \langle X_{\alpha} \rangle_q dy_{\alpha} = \ln
\left [ 1+ (1-q) \mathcal{S}^{*} \right ] d\mathcal{S}^{*}
\label{GibbsDuhem}$$ A reasonable condition (local additivity) that variables $\langle
X_{\alpha} \rangle_q$ must satisfy and which is used in the derivation of Eqs. (\[Euler\]) and (\[GibbsDuhem\]) is $
\left.d \langle X_{\alpha}(\lambda A)
\rangle_q/d\lambda\right|_{\lambda=1}= \langle X_{\alpha}(
A)\rangle_q \label{condition} $ for a system $\lambda$ times bigger than another $A$. Note that this by no means imply that $\langle X_{\alpha}(\lambda A) \rangle_q=\lambda\langle
X_{\alpha}(A) \rangle_q$ (global additivity) [@Vives02], since $\lambda$ is replaced by unity. Global additivity generally holds only when an ensemble of $\lambda$ independent noninteracting systems is considered (see below) but not necessarily upon subdivision of a system into smaller fractions (it is necessary, for example, to provide some energy to separate two droplets $A$ and $B$ from a bigger droplet $A+B$ because of the energetic cost of creating additional interfaces). Tsallis entropy satisfies [@Vives02] $\left.d \mathcal{S}^{*}(\lambda A)
/d\lambda\right|_{\lambda=1}=\left [ 1+ (1-q) \mathcal{S}^{*}(A)
\right ] \ln \left [ 1+ (1-q) \mathcal{S}^{*}(A) \right ]/(1-q).$ It is to be noted that, as occurs with their respective counterparts in standard thermodynamics, Eqs. (\[termo\]) to (\[GibbsDuhem\]) are not independent, i.e. knowing two of them the other one can be trivially deduced. The above general equations and the Legendre transform mechanism allows to determine all thermodynamic quantities of a nonextensive system.
Hill Nanothermodynamics (NT)
============================
Let us now introduce Hill’s formalism of NT. It is based in maximization of the Gibbs-Shannon entropy but contains a new entropic thermodynamic potential, the subdivision entropic potential (SEP) ${
\ensuremath{\mathcal{J}}
}$, which couples nonlinearly the natural thermodynamic variables of the system. Introduction of this new potential is made at the ensemble level, in which a set of $\mathcal{N}$ identical, noninteracting small systems is considered [@founds]. In this case, the set of formal equations are ${
\ensuremath{\mathcal{J}}
}d{
\ensuremath{\mathcal{N}}
}+\sum_{\alpha}y_{\alpha,H}d\left<X_{\alpha}\right>_{H,t}=dS_{t}$ (Gibbs); ${
\ensuremath{\mathcal{J}}
}{
\ensuremath{\mathcal{N}}
}+\sum_{\alpha}y_{\alpha,H}\left<X_{\alpha}\right>_{H,t}=S_{t}$ (Euler) and ${
\ensuremath{\mathcal{N}}
}d{
\ensuremath{\mathcal{J}}
}+\sum_{\alpha}\left<X_{\alpha}\right>_{H,t}dy_{\alpha,H}=0$ (Gibbs-Duhem). Here, subindex $H$ means “Hill’s variables” which are the *physical* (averaged) locally extensive ($\left<X_{\alpha}\right>_{H,t}$) and intensive ($y_{\alpha,H}$) ones. $t$ means “total” and denotes the properties of the whole ensemble. Thus $S_{t}$ is the total entropy for the ensemble of $\mathcal{N}$ systems. We have $S_{t}={
\ensuremath{\mathcal{N}}
}S$ and $\left<X_{\alpha}\right>_{H,t}={
\ensuremath{\mathcal{N}}
}\left<X_{\alpha}\right>_{H}$ in terms of the respective variables for one system (note that here the members of the ensemble do not interact directly although, of course, can exchange heat, volume and particles depending on the ensemble considered, this being itself a formal construction in which total properties are ${
\ensuremath{\mathcal{N}}
}$ times those of one system). The formal equations of NT become then for one system [@founds] $$\begin{aligned}
\sum_{\alpha}y_{\alpha,H}d\left<X_{\alpha}\right>_{H}&=&dS \label{HGi}\\
\sum_{\alpha}y_{\alpha,H}\left<X_{\alpha}\right>_{H}&=&S-{
\ensuremath{\mathcal{J}}
}\label{HEuler}\\
\sum_{\alpha}\left<X_{\alpha}\right>_{H}dy_{\alpha,H}&=&-d{
\ensuremath{\mathcal{J}}
}\label{HGD}\end{aligned}$$
Connection between TT and NT
============================
Eqs. (\[HGi\]) to (\[HGD\]) are to be compared with Eqs. (\[termo\]) to (\[GibbsDuhem\]) respectively. It can be seen that if we make the following connection $S \equiv
\mathcal{S}^{*}$, $y_{\alpha,H} \equiv y_{\alpha}$, $\left<X_{\alpha}\right>_{H} \equiv \left<X_{\alpha}\right>_{q}$ and $${
\ensuremath{\mathcal{J}}
}\equiv \mathcal{S}^{*}-\left[1+(1-q)\mathcal{S}^{*}\right]\frac{\ln \left[1+(1-q)\mathcal{S}^{*}\right] }{1-q}
\label{def}$$ the structure of the formal equations of TT and NT is the same. Here the Tsallis entropy $\mathcal{S}^{*}$ is not only the physical one: its property of nonadditivity, see Eq. (\[nonadditivity\]), is also the basis for the SEP ${
\ensuremath{\mathcal{J}}
}$ which is found to be necessary to explain the thermal behavior of small systems (at least). TT describes, thus, the most general thermal equilibrium, the nanothermodynamic equilibrium [@Chamberlin03], in which the new potential ${
\ensuremath{\mathcal{J}}
}$ plays a decisive role. ${
\ensuremath{\mathcal{J}}
}$ vanishes for a macroscopic system, for which one has also $q=1$ in Eq.(\[def\]), and is a measure of the (thermodynamic) smallness of the system. In making this correspondence it could seem counterintuitive equating an additive entropy $S$ to a nonadditive one $\mathcal{S}^{*}$. However, $S$ is Gibbs-Shannon like because Hill’s entropy is written statistically in terms of thermal averages considering a nanosystem in equilibrium with a macroscopic heat bath at a given average temperature. The nonlinear coupling of the intensive variables is accounted for thermodynamically in Eq.(\[HEuler\]) by means of the generalized potential ${
\ensuremath{\mathcal{J}}
}$. This is totally equivalent to consider the nanosystem at the same average inverse temperature arising this time not from coupling to a macroscopic heat bath but to a neighbouring environment with fluctuating temperature following a chi-square distribution (without need of introducing ${
\ensuremath{\mathcal{J}}
}$ in this case). The latter leads to Tsallis statistics [@qmeaning; @Wilk], the entropy being nonextensive and the probability distributions depending on the parameter $q$ (that comes from the order of the chi-square (Gamma) distribution considered). Some authors have shown that the Tsallis distribution is the Laplace transform of the chi-square distribution [@qmeaning; @Wilk]. This is the key point in passing from Boltzmann-Gibbs distributions to Tsallis q-exponential distributions.
Nonadditivity in the entropy (a statistical approach) and introduction of the potential ${
\ensuremath{\mathcal{J}}
}$ (a thermodynamic approach) are only different pathways that lead to the same thermodynamics. As shown below, thermodynamic excess functions (Hill’s NT) are linked to entropic nonadditivity (Tsallis TT). It is also to be noted that the ensemble approach followed by Hill to derive the thermodynamics of a small system is equivalent to assume that local additivity of parameters and Eq. (\[HGi\]) hold as well as the following expression for the additivity breaking of the entropy $\left.d S(\lambda
A)/d\lambda\right|_{\lambda=1}=S(A)-{
\ensuremath{\mathcal{J}}
}.$
Despite the different statistical behavior both Hill and Tsallis entropies can be made to coincide numerically and the differences of both formalisms concern only the way of averaging. It is worthy insisting on the fact that thermodynamically speaking, however, these differences are irrelevant, since the structure of the formal equations is the same. Both formalisms, TT and NT, preserve the Gibbsian structure of thermodynamics [@Tsallis88; @founds]. Fluctuations are also gaussian-like in both formalisms [@founds; @Vives02]. Furthermore, if the range $0<q<1$ is considered to be the physically meaningful one [@Vives02] fluctuations in TT are larger for $q \ne 1$ than in the extensive case [@Vives02] which is again consistent with the fact that fluctuations in NT for a system with ${
\ensuremath{\mathcal{J}}
}\ne 0$ are also larger than in the ${
\ensuremath{\mathcal{J}}
}=0$ case [@founds]. Although Tsallis probability distributions are more general than Boltzmann-Gibbs ones, it is important to note that, in NT, ${
\ensuremath{\mathcal{J}}
}$ introduces additional degrees of freedom in the evaluation of the relevant partition function [@Chamberlin00] (which in NT is always related to excess quantities besides the traditional thermodynamic ones). Furthermore, and quite importantly, both TT and NT share the property of ensemble non-equivalence for $q \ne 1$ and ${
\ensuremath{\mathcal{J}}
}\ne
0$, respectively.
Correct form for the Lagrange parameters
========================================
It has been questioned the physical meaning of the parameters $y_{\alpha}$, and some authors [@Vives02; @AbeToral] have concluded previously that the physical quantities of interest are ${\hat y}_{\alpha} = y_{\alpha}/[1+(1-q) \mathcal{S}^{*}].$ This conclusion, however, is wrong. Parameters ${\hat y}_{\alpha}$ are naturally related to a thermodynamic formalism in which use of Renyi entropy (and *not* Tsallis entropy) is made. As a function of Tsallis entropy, the Renyi entropy $\mathcal{\widehat{S}}^{*}$ has the form $\mathcal{\widehat{S}}^{*} = \ln \left [1+ (1-q) \mathcal{S}^{*}
\right]/(1-q)$ [@Tsallis88]. However, while Tsallis entropy is nonextensive and stable [@AbeW], Renyi entropy is extensive but thermodynamically unstable [@Lesche82] and cannot be employed to generalize standard thermodynamics [@AbeW]. Parameters ${\hat y}_{\alpha}$ also arise when it is assumed *ad hoc* global additivity of energy and the maximum Tsallis entropy for two systems brought into contact is considered [@AbeToral]. This, however, is also incorrect, since energy cannot be globally additive where entropy does not satisfy this property (otherwise, temperature could not be an intensive variable [@critics]). It is to be noted therefore that neither the correct thermodynamic forces nor the physical meaning of the formal relations were previously established. Through the above correspondence and from Hill’s NT, it is now known that the $y_{\alpha}$ must be equal for two different systems put in contact at equilibrium. *The $y_{\alpha}$ are then the physically meaningful* thermodynamic forces. This allows to establish equilibrium TT, which is now free from recent criticisms [@critics] arising from the use of incorrect thermodynamic forces.
Excess quantities. Meaning of the nonadditivity of Tsallis entropy
==================================================================
Excess thermodynamic quantities $\langle
X_{\alpha}\rangle_q^{(x)}$, as considered, for example, in surface physics [@founds; @Rowlinson82], can be defined as $\langle
X_{\alpha}(A,B) \rangle_q^{(x)}= \langle X_{\alpha}(A+B)
\rangle_q-\langle X_{\alpha}(A) \rangle_q-\langle X_{\alpha}(B)
\rangle_q$ By applying Eq. (\[termo\]) to the composite system $A+B$ and then to each system $A$ and $B$ separately and using these definitions, we obtain $ d
[(1-q)\mathcal{S}^{*}(A)\mathcal{S}^{*}(B)] = \sum_{\alpha}
y_{\alpha} d \langle X_{\alpha} \rangle_q^{(x)}$. This clearly leads to quantify the entropy excess in TT as $\mathcal{S}^{*(x)}=(1-q)\mathcal{S}^{*}(A)\mathcal{S}^{*}(B)$ [@GarciaMorales04b]. We see, thus, that TT provides also a specific microscopic statistical model for the thermodynamic excess entropy and this is the very physical meaning of the nonadditivity property. Each independent system $A$ or $B$ is analogous to a bulk phase. Despite their statistical independency, a thermodynamic coupling exists however between both systems. This can be modelled as an interface separating them and contributing with an excess entropy $\mathcal{S}^{*(x)}$ to the properties of the total system $A+B$. A nonextensive system in TT can hence be understood as a two-phase-like system at the vicinity of the critical point (since the nonadditivity property holds at all scales which means having always significant interfaces separating clusters of $A$ from clusters of $B$). It is interesting that Eq.(\[def\]) can be recasted as ${
\ensuremath{\mathcal{J}}
}=
\mathcal{S}^{*}-\mathcal{\widehat{S}}^{*}-(1-q)\mathcal{S}^{*}\mathcal{\widehat{S}}^{*}$ which is also similar to the Tsallis entropy scheme of nonadditivity, see Eq.(\[nonadditivity\]). The SEP ${
\ensuremath{\mathcal{J}}
}$ represents the net balance between fragmentation or aggregation of the system [@founds; @Chamberlin00; @Chamberlin99]. This SEP is hence to be viewed in TT as a thermodynamic force which can be understood *formally* in terms of an equilibrium of the actual physical nonextensive system (with physical entropy $\mathcal{S}^{*}$) with its completely fragmentated and uncorrelated state in which the system is reduced to a gas of its noninteracting constituents. The gas has entropy $-\mathcal{\widehat{S}}^{*}$. The minus sign comes from the fact that this gas represents disintegration, against the cohesion (the positive Tsallis physical contribution) which preserves the integrity of the nonextensive system. This equilibrium is made through an interface contributing with $-(1-q)\mathcal{S}^{*}\mathcal{\widehat{S}}^{*}$. Note that Renyi entropy $\mathcal{\widehat{S}}^{*}$ is extensive, and hence it describes well this completely fragmentated extensive dilute gas (which is an unstable and supersaturated one) of the physical system.
It can be seen also that ${
\ensuremath{\mathcal{J}}
}$ and $\mathcal{S}^{*}$ are related differentially as $d{
\ensuremath{\mathcal{J}}
}= -(1-q)\mathcal{\widehat{S}}^{*}d
\mathcal{S}^{*}$. This equation is somewhat unusual and has no counterpart in standard thermodynamics. If we define $f \equiv d{
\ensuremath{\mathcal{J}}
}/d\mathcal{S}^{*}$ we see that $f(\lambda A)=\lambda f(A)$. This means that the chance to aggregate depends linearly on the size of the system, which is quite reasonable.
General expression for q and examples
=====================================
By using our definition of excess entropy, we obtain $$1-q=\frac{\mathcal{S}^{*(x)}}{\mathcal{S}^{*}(A)\mathcal{S}^{*}(B)}=
\frac{S^{(x)}}{S(A)S(B)} \label{defq}$$ since $\mathcal{S}^{*}(A)\equiv S(A)$ and $\mathcal{S}^{*}(B)
\equiv S(B)$ through our connection while Eq. (\[HGi\]) holds also in NT for the composite system $A+B$ as well as for each separate system $A$ and $B$. In Eq. (\[defq\]) all the quantities appearing in the r.h.s can be always *calculated* from NT, which means that there is a way to know $q$ from first principles estimations. We now illustrate this assertion by calculating $q$ for nonpolar, nonstructured liquids (as argon) at the vicinity of the critical point. Liquid clusters develop inside the supersaturated vapor phase at all scales. From critical point theory [@Rowlinson82] it is well known that the surface tension $\sigma$ scales at the vicinity of the critical point ($T_{r}\to 0$) with the reduced temperature $T_{r} \equiv
(1-T/T_{c})$, where $T_{c}$ is the critical temperature, as $\sim
\sigma_{0}T_{r}^{\mu}$ (with $\mu \approx 1.26$ in three dimensions, and $\sigma_{0}$ being the surface tension amplitude). The correlation length of phase $i$ ($i=A, B$), $\xi_{i}$, scales as $\sim \xi_{0i}T_{r}^{-\mu/(D-1)}$ [@Rowlinson82] where $D$ is the dimension and $\xi_{0i}$ the amplitude of the correlation length. In this limit, the bulk entropies can be written in terms of the critical bulk entropic density $s_{c}$ as $S_{i}=s_{c}V_{i}
\approx s_{c}\xi^{D}_{i}$, where $V_{i}$ are the volumes occupied by each phase. The excess entropy $S^{(x)}$ is equal to $-ad\sigma/dT$ [@founds] where $a \approx \xi^{D-1}$ is the surface area. Replacing all these quantities (taking $A$ as the liquid phase) in Eq. (\[defq\]) and considering $D=3$ we obtain $$1-q=\frac{\mu}{4\pi\omega_{c}s_{c}^{2}\xi_{0A}^{3}\xi_{0B}^{3}}
T_{r}^{3\mu-1} \label{defq2}$$ where we have introduced the critical wetting parameter $\omega_{c}=kT_{c}/(4\pi\xi_{0A}^{2}\sigma_{0}) \approx 0.78$ which is a universal constant for all fluids [@Fisher92; @GarciaMorales03]. We have shown recently that $\omega_{c}$ *is related directly to the dynamics of the critical clusters* [@GarciaMorales03]. Concretely, a value $\omega_{c}=0.78$ means that the critical clusters behave *universally* as a combination of translation and vortex rotational motion [@GarciaMorales03]. This calculation then relates the entropic parameter $q$ to the dynamical quantity $\omega_{c}$ and the universal exponent $\mu$ that can be obtained separately from renormalization group theory [@Rowlinson82]. It is to be noted that although renormalization group theory is an exact approach given the Hamiltonian, real systems exhibit nonideal features due to the complexity of the interactions that can lead to departures from the translationally invariant hamiltonians usually considered in these cases. This is particularly dramatic when studying ferromagnetism [@Chamberlin00; @Chamberlin03]. In our specific example, classical scaling laws are used only here as an approximation. It is to be noted that far from the critical point (always following the coexistence line) $q$ does not necessarily departs from unity since the bulk entropies do not scale with the correlation lengths. Eq. (\[defq2\]) holds only approximately as the critical point is approached (as assumed from the scaling laws above used). It is to be noted that at the critical point $q=1$. This is consistent with the fact that the two phases in coexistence merge then into a single macroscopic homogeneous one provided that the size of the clusters tend to infinity. This means that, in addressing clusters, the thermodynamic limit is taken at the critical point ($N \to \infty$ and $V \to \infty$) and hence standard thermodynamics should be regained as is the case here. The r.h.s. of this equation is vanishingly small for non-structured simple fluids, as argon, because of the large numerical values for $s_{c}$. For these fluids, then, $q \approx
1$ and this justifies the success of the Gibbs method of the dividing surface within standard thermostatistics in regarding these systems. Departures from this ideal behavior are observed, however, for complex structured liquids as water [@Rowlinson82] in which there exists anomalies due to the presence of hydrogen bonding and dipole orientation which drammatically complicates the statistical description. The presence of structuredness reduce the bulk entropy of the liquid and, furthermore, interfacial effects strongly increase (as it is known from the value of the surface tension of water which surprisingly high when compared to simple fluids [@Rowlinson82]). It is then to be expected $q$ to depart significantly from unity in liquid water.
There are many other systems as Ag nanoparticles for which surface functions can be significantly higher than bulk values [@Nanda03] which would lead, from Eq.(\[defq\]), to values of $q$ significantly different from unity. Ferromagnetic materials have been also proven to significantly depart from standard thermostatistics [@Chamberlin00], and this has been interpreted succesfully in terms of a nanothermodynamic equilibrium [@Chamberlin00; @Chamberlin03].
Another example of a formal two-phase-like system is a correlated ionic liquid in the vicinity of a strongly charged macroion [@GarciaMorales04]. In this situation the behavior of the correlated counterions at the vicinity of the macroion is quite different to those far from the macroion (constituting a Poisson-Boltzmann uncorrelated and disordered liquid) [@Netz01; @GarciaMorales04]. For weak coupling, the correlated liquid is still a 3D fluid-like one but the rescaled interfacial area available to each molecule is affected by correlations and is approximately given by $2\Xi$ (where $\Xi$ is the coupling parameter entering in the Hamiltonian) [@Netz01]. It is clear, then, that in the weak coupling regime ($0<\Xi<1$, $q \lesssim 1$) and from Eq. (\[defq\]), $1-q \approx c\Xi$, where $c$ is a small constant (since $S^{(x)}$ depends linearly in the interfacial area). This is consistent with the previously fitted curve for $q(\Xi)$ since in the weak coupling limit we have $1-q=1-1/(1+0.091\Xi)^{0.68} \approx 0.062\Xi$ [@GarciaMorales04].
Concluding remarks
==================
We have established a connection between TT and NT which provides a sound and unambiguous physical basis for TT. As a bonus, it has allowed us to justify that the $y_{\alpha}$’s are the correct physical thermodynamic forces (difficulties pointed out in [@critics] are now overcomed) and has led us to two expressions that can be used quantitatively to evaluate the degree of nonextensivity in Hamiltonian systems. These are Eq. (\[def\]), which introduces the connection between both formalisms relating thermodynamic smallness to nonextensivity, and Eq. (\[defq2\]), which allows for an straightforward evaluation of $q$ from NT. A nonextensive system with $q \ne 1$ can now be understood in the thermodynamic limit as one composed of two entangled phases in a critical regime in which the (fractal) interface separating them, besides other finite size effects, contributes significantly to the total entropy at all scales.
Acknowledgements {#acknowledgements .unnumbered}
================
We acknowledge financial support from the MCYT (Ministry of Science and Technology of Spain) and the European Funds for Regional Development (FEDER) under project No. MAT2002-00646.
C. Tsallis, J. Stat. Phys., 52 (1988) 479. See http://tsallis.cat.cbpf.br/biblio.htm for an updated bibliography on the subject. F. S. Navarra, O. V. Utyuzh, G. Wilk, Z. Wlodarczyk, Phys. Rev. D, 67 (2003) 114002. D. F. Torres, H. Vucetich, A. Plastino, Phys. Rev. Lett. 79 (1997) 1588. R. Salazar, R. Toral, Phys. Rev. Lett., 83 (1999) 4233; S. A. Cannas, F. A. Tamarit, Phys. Rev. B, 54 (1996) R12661. M. L. Lyra, C. Tsallis, Phys. Rev. Lett., 80 (1998) 53; E. P. Borges, C. Tsallis, G. F. J. Añaños, P. M. C de Oliveira, Phys. Rev. Lett. 89 (2002) 254103; G. F. J. Añaños, C. Tsallis, Phys. Rev. Lett., 93 (2004) 020601. C. Tsallis, S. V. F. Levy, A. M. C. Souza, R. Maynard, Phys. Rev. Lett., 75 (1995) 3589; D. Prato and C. Tsallis, Phys. Rev. E, 60 (1999) 2398. G. Wilk and Z. Wlodarczyk, Phys. Rev. Lett., 84 (2000) 2770 (2000); C. Beck, Phys. Rev. Lett., 87 (2001) 180601. A. Cho, Science, 297 (2002) 1268. M. Nauenberg, Phys. Rev. E 67 (2003) 036114; D. H. E. Gross, Physica A 305 (2002) 99. E. Vives, A. Planes, Phys. Rev. Lett. 88 (2002) 020601. S. Abe, S. Martinez, F. Pennini, A. Plastino, Phys. Lett. A 281 (2001) 126; R. Toral, Physica A 317 (2003) 209. S. R. A. Salinas, Introduçao à Física Estatística (Editora da Universidade de Sao Paulo, Sao Paulo, Brasil, 1997). R. V. Chamberlin, Science, 298 (2002) 1172. T. L. Hill, J. Chem. Phys. 36 (1962) 3182; T. L. Hill, Thermodynamics of Small Systems (Dover, New York, 1994) R. V. Chamberlin, Nature 408 (2000) 337. R. V. Chamberlin, Phys. Rev. Lett. 82 (1999) 2520. T. L. Hill and R. V. Chamberlin, Proc. Natl. Acad. Sci. USA 95 (1998) 12779. V. Garcia-Morales, J. Cervera, J. Pellicer, Phys. Rev. E 67 (2003) 062103. R. V. Chamberlin, Phys. Lett. A 315 (2003) 313. G. Wilk, Z. Wlodarczyk, Chaos Soliton Fract. 13 (2002) 581. S. Abe, Phys. Rev. E 66 (2002) 046134. B. Lesche, J. Stat. Phys. 27 (1982) 419. J. S. Rowlinson and B. Widom, Molecular Theory of Capillarity (Clarendon Press, Oxford, 1982). V. Garcia-Morales, unpublished result. M. E. Fisher, H. Wen, Phys. Rev. Lett. 68 (1992) 3654. K. K. Nanda, A. Maisels, F. E. Kruis, H. Fissan and S. Stappert, Phys. Rev. Lett. 91 (2003) 106102. V. Garcia-Morales, J. Cervera, J. Pellicer, Physica A 339 (2004) 482. R. R. Netz, Eur. Phys. J. E 5 (2001) 557.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Takeshi YOSHIDA[^1] and Mitsusada M. SANO'
title: |
Numerical Simulation of Vortex Crystals and Merging\
in $N$-Point Vortex Systems with Circular Boundary
---
Introduction
============
It is well known that strongly magnetized pure electron plasma is approximately equivalent to 2D inviscid incompressible fluid. In a strong uniform magnetic field $\bm{B}_0=B_0\hat{\bm{z}}$ the drift velocity of the guiding center of plasma is given by $\bm{v}(x,y)=-\nabla \phi(x,y)\times{\hat{\bm{z}}}/B_0$, where $\phi$ is the electrostatic potential and $\hat{\bm{z}}$ is a unit vector normal to the plane of the flow. Thus, this flow of the drift velocity is incompressible, i.e., $\nabla \cdot \bm{v}=0$. The vorticity of this velocity field is given by $\bm{\zeta}(x,y)=\nabla\times\bm{v}=\hat{\bm{z}}{\nabla}^2\phi/B_0$. Using the Poisson equation, the vorticity of this velocity field is shown to be proportional to the electron density. Therefore, the vorticity is governed by the following equations, $$\begin{aligned}
\frac{\partial \zeta}{\partial t}+\bm{v}\cdot\nabla \zeta&=0,
&\bm{v}&=\hat{\bm{z}}\times\nabla\Psi,
&\zeta &= {\nabla}^2\Psi. \end{aligned}$$ Here $\zeta(x,y)=(e/\varepsilon_0 B_0)n(x,y)$ ( $n(x,y)$ is the electron density ) and $\Psi(x,y)=(\varepsilon_0/e)\phi(x,y)$ is the stream function. $\varepsilon_0$ is the dielectric constant in vacuum. Thanks to the above relation, the behavior of vortices in incompressible Euler fluid is experimentally realized on non-neutral plasma. The vortex merging process and the motions of point vortices in vacuum or in the background vorticity distribution were experimentally observed. [@2; @3; @4]
Theoretically, stability of point vortices array had been studied. [@Saffman; @Kurakin] Furthermore equilibrium states of $N$-point vortex system had been also investigated. [@kida; @Ralph; @Chavanis; @Ons] The equilibrium states of this $N$-point vortex system under various initial conditions were obtained by using statistical mechanical method. The theory using statistical mechanics predicted that the vortices of this system merge one after another to be a large vortex in equilibrium. This phenomenon occurs when the “temperature” of point vortices is negative. The positive “temperature” leads to merger between vortices of opposite sign, or with boundary. [@Ons]
However, about ten years ago, a new phenomenon was observed in the experiment by Fine *et al*. [@Fine] Under certain initial conditions, vortex merging is “cooled” and a few intense vortices (which is called “clump” in this paper to distinguish a point vortex) form a regular lattice in a low background vorticity distribution. This phenomenon which is called “vortex crystals” is the highlight of this field. This vortex crystals are investigated from various aspects. For example, a power law decay of the number of intense vortices of this system, namely $N\sim t^{-\xi}, \xi=0.2-0.7$ was studied. [@JinDubin; @Fine] The formation of the vortex crystals was predicted by two theories. One is a kinetic theory using the Langevin equation. [@Kono] The other is a theory by the maximizing entropy principle. [@JinDubin2]
Moreover, this phenomenon, namely vortex crystal, which is experimentally observed was reproduced by numerical simulation. [@Schecter1] This numerical result showed quantitatively good agreement with the experimental one. Time evolution of the number of intense vortices both in experimental flow and in numerical one has similar decay in time.
In the recent experiment, [@Sanpei] it is observed that three clumps arrayed on a straight line form an equilateral triangular lattice under certain conditions of the background. When the background is very dense, merger immediately occurs instead of crystallization. In the very dilute background, chaotic motion of the three clumps which occurs without the background is not cooled by the background. Thus the crystallization does not occur and the three clumps in the dilute background do not exhibit a triangular configuration. Therefore the existence and the density of the background are supposed to be significant in the process of vortex crystals.
In this paper, we report the results of our studies on these phenomena, namely the vortex crystals of three clumps, regarded as a $N$-point vortex system. The vortex crystal was numerically reproduced by Schecter *et al*. [@Schecter1] However their simulation is based on the Euler equation (VIC simulation). [@Leonard] Therefore the details on the dynamics of vortices are not visible, namely how vortices in clumps and in background mix each other. Hence we treat the discrete $N$-point vortex system to pursue the behavior of each point vortex. In the discrete $N$-point vortex system, the detail of the dynamics such as mixing of vortices can be observed. In order to compare our simulation to the experiments on non-neutral plasma, the signs of circulation of vortices are all defined to be positive for our $N$-point vortex system in this paper: in the experiments on non-neutral plasma, the sign of vorticities are all the same.
The $N$-point vortex system is described by the following Hamilton equations of motion for open boundary cases, $$H=-\frac{1}{4\pi}\sum_m\sum_{n\neq m}\Gamma_m\Gamma_n\log | z_m-z_n |,$$ $$\begin{aligned}
\Gamma_m\frac{\text{d}x_m}{\text{d}t}&=\frac{\partial H}{\partial y_m},&
\Gamma_m\frac{\text{d}y_m}{\text{d}t}&=-\frac{\partial H}{\partial x_m}.
\label{eq:canoHami}\end{aligned}$$ Here $z_m=x_m+\text{i}y_m, \text{i}=\sqrt{-1}$. $(x_m,y_m)$ and $\Gamma_m$ are the position and the circulation of the $m$-th vortex, respectively. In this case, constants of motion are the total circulation $\sum_m \Gamma_m$, the angular impulse $I=\sum_m\Gamma_m({x_m}^2+{y_m}^2)$, the total energy $E=-\frac{1}{4\pi}{\sum\sum}_{(m\neq n)}\Gamma_m\Gamma_n\log |z_m-z_n|$, and the center of vorticity $\sum_m \Gamma_m z_m / \sum_m \Gamma_m$.
For the system with circular boundary, the Hamiltonian is given by $$H =-\frac{1}{4\pi}\sum_m\sum_{n\neq m}\Gamma_m\Gamma_n\log| z_m-z_n|
+\frac{1}{4\pi}\sum_{m}\sum_{n}\Gamma_m\Gamma_n\log| R^2-z_m\bar{z}_n|
-\frac{\log{R}}{4\pi}(\sum_m \Gamma_m)^2,
\label{eq:hamikabe}$$ where $R$ is the radius of the circular boundary. The equations of motion are $$\begin{aligned}
\frac{\text{d}x_m}{\text{d}t}
&=-\sum_{n\neq m}\frac{\Gamma_n}{4\pi}\frac{y_m-y_n}{|z_m-z_n|^2}
+\sum_n
\frac{\Gamma_n}{4\pi}\frac{y_m-{y_n}^{\prime}}{|z_m-{z_n}^{\prime}|^2},
\label{eq:dxkabe}\\
%
\frac{\text{d}y_m}{\text{d}t}
&=\sum_{n\neq m}\frac{\Gamma_n}{4\pi}\frac{x_m-x_n}{|z_m-z_n|^2}
-\sum_n
\frac{\Gamma_n}{4\pi}\frac{x_m-{x_n}^{\prime}}{|z_m-{z_n}^{\prime}|^2},
\label{eq:dykabe}\end{aligned}$$ where $$\begin{aligned}
{z_m}^{\prime} &= \frac{R^2}{\bar{z}_m}, &
{x_m}^{\prime} &= \frac{R^2}{|z_m|^2}x_m, &
{y_m}^{\prime} &= \frac{R^2}{|z_m|^2}y_m.\end{aligned}$$ The constants of motion for this system with circular boundary are the total circulation $\sum_m \Gamma_m$, the angular impulse $I$, and the total energy $E$.
The organization of this paper is as follows. In a circular domain, a system of more than or equal to three point vortices exhibits chaotic behavior. Therefore, as the most fundamental system of vortices, we investigate a three-point vortex system in §2 for the purpose of examining how the system depend on the parameters $(I,E)$, namely the angular impulse and the total energy. By plotting the Poincaré map, we examine the dynamical behavior of this system in detail. We consider a parameter space which is spanned by two constants of motion $(I,E)$. For various values of these two parameters, the Poincaré plots are calculated. We confirm that the topology of the Poincaré plot drastically changes when the values of the parameters are varied and that the property of the Poincaré plot is influenced by the sign of “temperature” of point vortices introduced by L. Onsager. [@Ons] In §3, we introduce an expression of the energy spectrum for this $N$-point vortex system with circular boundary. This energy spectrum is used in order to understand the irreversible merging process in which a $N$-point vortex system eventually grows into one large vortex, which is the result of negative “temperature”. In §4, we present the results of simulations. In a system of discrete point vortices, we reproduce vortex merger and a vortex crystal that is observed in the experiments on non-neutral plasma and in the simulations based on the 2D Euler equation. In §5, we summarize the results.
Poincaré Section
================
In this section, we present the results of the study on a three-point vortex system. We investigate the topology of the Poincaré plot for the three-point vortex system with 2D circular boundary under various initial conditions, which are determined by the parameters $(I,E)$, namely the angular impulse and the total energy. It was confirmed experimentally and will be shown by our numerical simulation in the later section that the three clumps system exhibits a vortex crystal and vortex merger in the background vorticity distribution. Studying a three-point vortex system in vacuum is important in order to isolate the most fundamental properties from these phenomena, i.e., the vortex crystal and vortex merger: because it is supposed to be the backbone process of the three clumps system in the background vorticity distribution.
It is known that in a system of point vortices without boundary chaotic behavior appears for more than or equal to four point vortices, and that with circular boundary it appears for more than or equal to three. Aref and Pomphrey [@Aref] obtained the Poincaré plot for a non-boundary system of four vortices and demonstrated that its dynamics is chaotic. Using their procedure, we calculate the Poincaré plots for the dynamics of three point vortices with boundary. For the system with circular boundary, there are two constants of motion, namely the angular impulse $I$ and the total energy $E$. We obtained the Poincaré plots of this system for various values of these two constants.
In the following, according to Aref and Pomphrey [@Aref] we derive the Hamiltonian with two degrees of freedom for our system by carrying out canonical transformations. For simplicity, we take the circulation $\Gamma_m=1$ for $m=1,2$ and $3$. Then the Hamiltonian for this system is (The third term of eq.(\[eq:hamikabe\]) is constant. So we ignore it in the following calculation.) $$\begin{aligned}
H=&-\frac{1}{8\pi}
\sum_{m=1}^3\sum_{n=1(\neq m)}^3\log(z_m-z_n)(\bar z_m-\bar z_n)
+\frac{1}{8\pi}
\sum_{m=1}^3\sum_{n=1}^3\log(R^2-z_m\bar z_n)(R^2-\bar z_mz_n)\label{eq:Hami}\\
=&\frac{1}{4\pi}\log h(z_1,z_2,z_3), \notag \end{aligned}$$ where $$\begin{aligned}
h(z_1,z_2,z_3)&=(R^2-|z_1|^2)(R^2-|z_2|^2)(R^2-|z_3|^2)
\frac{|R^2-z_1\bar z_2|^2|R^2-z_2\bar z_3|^2|R^2-z_3\bar z_1|^2}
{|z_1-z_2|^2 |z_2-z_3|^2 |z_3-z_1|^2}.\end{aligned}$$ The vortex coordinates $\{z_i\}_{i=1,2,3}$ are expanded in Fourier series $$\begin{split}
\begin{cases}
z_1=\frac{1}{\sqrt{3}}
(\sqrt{2J_1}\text{e}^{\text{i}\theta_1}+\sqrt{2J_2}\text{e}^{\text{i}\theta_2}+\sqrt{2J_3}\text{e}^{\text{i}\theta_3}),\\
z_2=\frac{1}{\sqrt{3}}
(\sqrt{2J_1}\text{e}^{\text{i}\theta_1}+\text{e}^{-\text{i}\frac{2\pi}{3}}\sqrt{2J_2}\text{e}^{\text{i}\theta_2}
+\text{e}^{-\text{i}\frac{4\pi}{3}}\sqrt{2J_3}\text{e}^{\text{i}\theta_3}),\\
z_3=\frac{1}{\sqrt{3}}
(\sqrt{2J_1}\text{e}^{\text{i}\theta_1}+\text{e}^{-\text{i}\frac{4\pi}{3}}\sqrt{2J_2}\text{e}^{\text{i}\theta_2}
+\text{e}^{-\text{i}\frac{2\pi}{3}}\sqrt{2J_3}\text{e}^{\text{i}\theta_3}).
\label{eq:z123}
\end{cases}
\end{split}$$ Here $J_n$’s and $\theta_n$’s are the action-angle variables. We further transform the variables in the following way. $$\begin{cases}
\phi_1=\frac{1}{2}(\theta_1-\theta_3),\\
\phi_2=\frac{1}{2}(\theta_1-2\theta_2+\theta_3),\\
\phi_3=\theta_2,
\end{cases}$$ and $$\begin{cases}
I_1=J_1-J_3,\\
I_2=J_1+J_3,\\
I_3=J_1+J_2+J_3.
\end{cases}$$ Substituting them into the Hamiltonian, we get the reduced Hamiltonian $$H_{\text{r}}=H_{\text{r}}(I_1,I_2,\phi_1,\phi_2).\label{eq:cHami}$$ Here, since $2I_3$ is just the angular impulse (i.e., the constant of motion, $I\equiv 2I_3=\sum_{m=1}^3|z_m|^2$), then $\phi_3$ is a cyclic coordinate. Therefore it does not appear in the reduced Hamiltonian.
Furthermore in order to get the Poincaré section, we introduce canonical variables, $$\begin{cases}
R_1=\sqrt{(I_1+I_3)/2I_3}\cos2\phi_1,\\
P_1=\sqrt{(I_1+I_3)/2I_3}\sin2\phi_1,
\end{cases}$$ $$\begin{cases}
{R_2}=\sqrt{(I_3-I_2)/2I_3}\sin2\phi_2,\\
{P_2}=\sqrt{(I_3-I_2)/2I_3}\cos2\phi_2.
\label{eq:RP}
\end{cases}$$ The calculation of the Poincaré plot is done by the following procedure. Since the reduced Hamiltonian eq.(\[eq:cHami\]) is too complicated, time evolution is calculated by using the Hamiltonian eq.(\[eq:Hami\]). We use the 4th order Runge-Kutta method for our numerical calculation. 10000 random initial points are given on the $(R_1,P_1)$-plane satisfying two constants of motion, namely the angular impulse and the total energy, and time evolution is done until these each orbit of initial points crosses the $(R_1,P_1)$-plane 50 times. The Poincaré section is set on the plane with $R_2=0, \dot{R_2}>0$. In this simulation, relative errors of both constants of motion, namely $I$ and $E$, are $10^{-10}$. First, we show the density of states as a function of the angular impulse $I$ and the total energy $E$ as Johnson did (Fig.\[fig:DHI\]). [@Darell] A rigid line represents the values of the parameters at which three point vortices form an equilateral triangle configuration; the center of mass of the equilateral triangle is at the origin. The form of the rigid line is given by $$E_{\text{r}}(I)=\frac{1}{4\pi}
\log \left( \frac{1}{27}\frac{[R^6-(I/3)^3]^3}{(I/3)^3}\right).
\label{eq:regtri}$$ Here $R$ is the radius of the circular boundary. The dotted lines indicate several contour levels of density of states. The innermost level is the highest (from outer to inner, $1.0\times 10^{-5},3.0\times 10^{-5},1.0\times 10^{-4},
3.0\times 10^{-4},1.0\times 10^{-3},3.0\times 10^{-3},
1.0\times 10^{-2},3.0\times 10^{-2},1.0\times 10^{-1}$(%) of all configurations, respectively).
In the region for small $I$ in Fig.\[fig:DHI\], the states with $E<E_{\text{r}}$ do not exist. The reason why the density in this region of parameters is zero is as follows. When vortices are close each other, the energy $E$ increases. If vortices are close to boundary, the energy decreases. In small $I$ range, because of the restriction of the angular impulse $I$, no vortices can be close to the boundary. Therefore, in the region of small $I$, the equilateral triangular configuration (rigid line) gives the minimum energy. This is actually observed in Fig.\[fig:DHI\].
Figure \[Fig:PS\] shows the Poincaré plots for some parameters. The values of these parameters are indicated in Fig.\[Fig:souzu\] and Table \[tbl:vl-PS\]. The density of the states reflects the area of the plots on the Poincaré section. For example, we compare the areas of Fig.\[Fig:PS\](F) with that of Fig.\[Fig:PS\](G). Here we express the region of the plots in Fig.\[Fig:PS\](F) as $\mathcal{D}_{\text{(F)}}$ on the complex plane $Z=P_1+\sqrt{-1} R_1$. Thus the region $\mathcal{D}_{\text{(F)}}$ is expressed as $\mathcal{D}_{\text{(F)}}
={\mathcal{D}_{\text{(F)}}}^{(0)}
\bigcup{\mathcal{D}_{\text{(F)}}}^{(1)}
\bigcup{\mathcal{D}_{\text{(F)}}}^{(2)}
\bigcup{\mathcal{D}_{\text{(F)}}}^{(3)}$, where ${\mathcal{D}_{\text{(F)}}}^{(0)}=\{Z|0<s<|Z|<l\}$, and ${\mathcal{D}_{\text{(F)}}}^{(1)}$, ${\mathcal{D}_{\text{(F)}}}^{(2)}$ and ${\mathcal{D}_{\text{(F)}}}^{(3)}$ represent three protuberances sprouted from ${\mathcal{D}_{\text{(F)}}}^{(0)}$ toward the origin. As the energy increases, three regions of the protuberances become thinner along the azimuthal direction. The region ${\mathcal{D}_{\text{(F)}}}^{(0)}$ also becomes thinner (Fig.\[Fig:PS\](G)). This tendency is observed for general values of $I$.
When $E=E_{\text{r}}$ (hence the parameters $(I,E)$ are on the rigid line in Fig.\[fig:DHI\]), we can see from eqs.(\[eq:z123\]-\[eq:RP\]) that the plots on the Poincaré section are lying on the circle of which radius is $\sqrt{1/2}$ and at the origin. When the value of $E$ changes from $E_{\text{r}}$ slightly for the same value of $I$, the region of the plots becomes concentric rings as Fig.\[Fig:PS\](A): if we write this two regions on the complex plane $Z=P_1 + \sqrt{-1}R_1$ as $\mathcal{D}_1$ and $\mathcal{D}_2$, when $E=E_{\text{r}}$ these two region is expressed as $\mathcal{D}_1=\{Z| |Z|=0\}$ and $\mathcal{D}_2=\{Z| |Z|=\sqrt{1/2}\}$. When $E$ changes from $E_{\text{r}}$, these regions change to $\mathcal{D}_1=\{Z|a_1\leq |Z|\leq a_2\}$ and $\mathcal{D}_2=\{Z|b_1\leq |Z|\leq b_2\}$, where $0<a_1<a_2<b_1<b_2$. Therefore, the orbit in the phase space do not pass the origin on the $(P_1,R_1)$-plane. Actually, in Fig.\[Fig:PS\], there are no points around the origin on the Poincaré section. This means that the topology of the Poincaré plot changes when the value of $E$ crosses $E_{\text{r}}$. As $E$ increases, this two regions of the concentric rings are connected as seen in Fig.\[Fig:PS\](B): these two regions $\mathcal{D}_1$ and $\mathcal{D}_2$ defined above change to $\mathcal{D}_1=\{Z|a_1\leq |Z|\leq a_2\}$ and $\mathcal{D}_2=\{Z|b_1\leq |Z|\leq b_2\}$, where $0<a_1<b_1<a_2<b_2$. As $E$ increases more, the region $\mathcal{D}_1\bigcup\mathcal{D}_2$ is scooped out: $\mathcal{D}_1\bigcup\mathcal{D}_2\setminus\bigcup_{i=1}^m\mathcal{D}^{(i)}$, where $\mathcal{D}^{(i)}$ are the scooped region as seen in Fig.\[Fig:PS\](C) (for m=6) and Fig. \[Fig:PS\](G) (for m=3).
Furthermore, the property of Poincaré plot is apparently changed when the sign of the gradient of the density of states $\partial{W}/\partial{E}$ is changed to opposite sign. This gradient is proportional to the inverse temperature $1/T$ defined by “statistical mechanics”. $$\begin{split}
S&=\log W, \\
\frac{1}{T}
&=\frac{\partial{S}}{\partial{E}}
=\frac{1}{W} \frac{\partial{W}}{\partial{E}},
\end{split}$$ where $S$ is the entropy and $E$ is the energy of the system. See Fig.\[Fig:souzu\]. The values of the energy of points (A) and (B) are lower than the ridge of contour line (positive “temperature”), and (D) is around the ridge (“temperature” is infinite). The others are on the region upper than the ridge (negative “temperature”). When the “temperature” is negative, the region of Poincaré plot is scooped out. This shows that the sign of “temperature” influences the behavior of the Poincaré plot and the phase space structure.
In the following, we take notice of the properties of the motion in the three-point vortex system with the parameter $(I,E)=(100,2.25)$. Figure \[Fig:PS-VC\] shows the Poincaré plot and the locus of three point vortices under different initial conditions on the real $(x,y)$-plane with $(I,E)=(100,2.25)$. The plots of Figs.\[Fig:PS-VC\](P-1) - \[Fig:PS-VC\](P-3) are generated from a single trajectory under different initial conditions. Figs.\[Fig:PS-VC\](O-1) - \[Fig:PS-VC\](O-3) show the motions of rotating three vortices. In the case of Fig.\[Fig:PS-VC\](O-1), two vortices rotate each other around the origin apart from the other vortex which runs near the boundary. We will call the motion of this two vortices rotating each other in Fig.\[Fig:PS-VC\](O-1) the “binary star motion”. As seen in Fig.\[Fig:PS-VC\](P-1), this motion of voritices plots tori on the $(P_1,R_1)$-plane. Tori on the Poincaré section which does not correspond to the “binary star motion” of vortices are shown in Fig.\[Fig:PS-VC\](P-2). The corresponding locus is depicted in Fig.\[Fig:PS-VC\](O-2). Though the loci of these two type of motions in Figs.\[Fig:PS-VC\](O-1) and \[Fig:PS-VC\](O-2) are apparently different, these Poincaré plots in Figs.\[Fig:PS-VC\](P-1) and \[Fig:PS-VC\](P-2) are similar in its regular behavior on the Poincaré section (Figs.\[Fig:PS-VC\](P-1) and \[Fig:PS-VC\](P-2)). The other type of motions of vortices is shown in Figs.\[Fig:PS-VC\](P-3) and \[Fig:PS-VC\](O-3). In this case, “chaotic sea” emerges on the $(P_1,R_1)$-plane in Fig.\[Fig:PS-VC\](P-3). The vortex orbits show cusps as in Fig.\[Fig:PS-VC\](O-3). As mentioned above, the motion of the three-point vortex system can be classified at least three types, namely the types of Figs.\[Fig:PS-VC\](O-1) - \[Fig:PS-VC\](O-3). The behavior of the Poincaré plot for the motion of vortices enable us to distinguish the properties of vortices motion as in Figs.\[Fig:PS-VC\](P-2) and \[Fig:PS-VC\](P-3): this difference is not distinguishable by only seeing the loci of three vortices in Figs.\[Fig:PS-VC\](O-2) and \[Fig:PS-VC\](O-3).
It is difficult to calculate the Poincaré plot for the parameters in the region $E<E_{\text{r}}$ and in the region of large $I$. We suppose that there are two reasons why this calculation is difficult. The first reason is the complexity of the formula of the reduced Hamiltonian eq.(\[eq:cHami\]). This complexity causes numerical errors. The second reason is the following. When $E<E_{\text{r}}$, a vortex runs quite near the boundary (hence it runs very fast). In this situation we have to take the time step to be small. In actual numerical computation we fixed the time step. Therefore in that case numerical error becomes large in the calculation of the Poincaré plot. This difficulty of numerical integration is a subject to be solved. A variable time step scheme would be needed.
![ The Poincaré plots on the $(P_1,R_1)$-plane for various values of parameters. The values of parameters are summarized in Table \[tbl:vl-PS\]. (E-2) is the enlargement of (E). []{data-label="Fig:PS"}](Fig2.eps)
[-\[Fig:PS\](G). (A)-(G) correspond to the sections (A)-(G) in Fig.\[Fig:PS\], respectively. []{data-label="Fig:souzu"}](Fig3.eps "fig:"){width="15.0cm"}]{}
$I$ $E_{\text{r}}$ $E$ index
----- ---------------- ------- -------
50 2.361 2.4 (A)
2.42 (B)
2.8 (C)
100 2.189 2.202 (D)
2.5 (E)
150 2.070 2.25 (F)
2.5 (G)
: The values of the parameters $(I,E)$ for the Poincaré plots shown in Fig.\[Fig:PS\]. $E_{\text{r}}$ indicates the value of the energy for the equilateral triangle configuration. []{data-label="tbl:vl-PS"}
. []{data-label="Fig:PS-VC"}](Fig4.eps)
Energy Spectrum
===============
In a $N$-point vortex system, when the temperature is negative, many like-signed vortices merge each other and tend to become a large vortex. [@Ons] This indicates that the energy is transferred from small scale to large scale in two dimensional case. This phenomenon is known as the inverse cascade [@Ranryuu0; @Ranryuu1; @Ranryuu2] in general two dimensional turbulence.
Novikov gave a formula for the energy spectrum for a system of point vortices without boundary. [@Novikov] In the following, according to his procedure, we derive an explicit expression of the energy spectrum for a $N$-point vortex system with circular boundary.
The energy is defined by $$E=\int_0^{\infty}\tilde{E}(k)\text{d} k
=\int_0^{\infty}\frac{| \tilde{\bm v}(\bm k)|^2}{2} \text{d}k
=\int_0^{\infty}\frac{| \tilde{\omega}(\bm k)|^2}{2k^2} \text{d}k.$$ Here, $\tilde{\bm{v}}(\bm{k})$ is the Fourier transform of velocity field, $${\tilde{\bm{v}}}(\bm{k})
=\frac{1}{2\pi}\int_{-\infty}^{\infty}
\bm{v(r)}\exp(\text{i}\bm{k}\cdot\bm{r})\text{d}\bm{r},$$ and $\tilde{\omega}(\bm{k})$ is the Fourier transform of the vorticity field. In this system, the velocity field is restricted to be within the circular domain. Thus we introduce an induced vortex sheet on the circular boundary whose radius is R. The vorticity field is given by $$\begin{aligned}
\omega(\bm{r})
&=\sum_m\Gamma_m\delta(\bm{r}-\bm{r}_m)
-\left[\nabla\Psi\cdot\bm{n}\right]_R\delta(r-R) \\
&=\sum_m\Gamma_m\delta(\bm{r}-\bm{r}_m)
-v_\theta |_R \delta(r-R) \label{eq:IdcVor} \\
&=\sum_m\Gamma_m\left[\delta(\bm{r}-\bm{r}_m)
-\frac{R^2-{r_m}^2}{2\pi R | \bm{R}-\bm{r}_m|^2}
\delta(r-R) \right].\label{eq:IdcVor2}\end{aligned}$$ The effect of the induced vortex sheet is given by the second term in each line. Here $\bm{r}=(x,y)$, $\bm{r}_m$ is the position of the $m$-th vortex, and $\bm{R}=R\bm{r}/r$. $\Psi(\bm{r})$ is the stream function by vortices which is within the boundary, $\bm{n}$ is a unit vector perpendicular to the boundary, and $\nabla{\Psi}\cdot\bm{n}=-\partial\Psi/\partial r=v_{\theta}$. The derivation of the second term in eq.(\[eq:IdcVor2\]) is given in Appendix A.
Thus the Fourier transform of the vorticity field $\omega(\bm{r})$ is given by, $$\begin{aligned}
\tilde\omega(\bm{k})
&=\frac{1}{2\pi}
\int\omega(\bm{r})\exp(\text{i}\bm{k}\cdot\bm{r})\text{d}\bm{r} \\
&=\frac{1}{2\pi}\int\sum_m\Gamma_m\left[\delta(\bm{r}-\bm{r}_m)
-\frac{R^2-{r_m}^2}{2\pi R | \bm{R-r_m}|^2}\delta(r-R) \right]
\exp(\text{i}\bm{k}\cdot\bm{r})\text{d}\bm{r}.\end{aligned}$$ This can be easily rewritten into $$\begin{aligned}
\tilde\omega(\bm{k})
&=\frac{1}{2\pi}\sum_m \Gamma_m
\left[
\int\delta(\bm{r-r_m})\exp(\text{i}\bm{k}\cdot\bm{r})\text{d}\bm{r}
-\frac{R^2-{r_m}^2}{2\pi R} \oint_{r=R}
\frac{\exp(\text{i}\bm{k}\cdot\bm{R})}{| \bm{R}-\bm{r}_m|^2}\text{d}\bm{r}
\right]\\
%&=\frac{1}{2\pi}\sum_m \Gamma_m\left[
%\text{e}^{\text{i}\bm{k}\cdot\bm{r}_m}
%-\frac{R^2-{r_m}^2}{2\pi R}
%\oint_{C=R}
%\frac{\text{e}^{\text{i}\bm{k}\cdot(\bm{R}-\bm{r}_m)}}
%{| \bm{R-r_m}|^2} d\bm{r} \, \text{e}^{\text{i}\bm{k}\cdot\bm{r}_m} \right] \\
&=\frac{1}{2\pi}\sum_m \Gamma_m\left[
1
-\frac{R^2-{r_m}^2}{2\pi R}
\oint_{r=R}
\frac{\exp[\text{i}\bm{k}\cdot(\bm{R}-\bm{r}_m)]}
{| \bm{R}-\bm{r}_m|^2} \text{d}\bm{r} \right]
\exp(\text{i}\bm{k}\cdot\bm{r}_m) \\
&=\sum_m\tilde{\Gamma}_m(\bm{k})\exp(\text{i}\bm{k}\cdot\bm{r}_m),\end{aligned}$$ where we define $$\tilde{\Gamma}_m(\bm{k}) \equiv \frac{1}{2\pi}\Gamma_m\left[
1
-\frac{R^2-{r_m}^2}{2\pi R}
\oint_{r=R}
\frac{\exp[\text{i}\bm{k}\cdot(\bm{R}-\bm{r}_m)]}
{| \bm{R}-\bm{r}_m|^2} \text{d}\bm{r}
\right].$$ In the limit $k\to 0$, $\tilde{\Gamma}_m(\bm{k})$ is vanished. Actually, the vorticity disappears in the outside of the circular domain.
By definition, $\overline{\tilde{\Gamma}_m(\bm{k})}=\tilde{\Gamma}_m(-\bm{k}) $. Thus we have $$| \tilde{\bm v}(\bm k)|^2 = \frac{| \tilde\omega(\bm{k})|^2}{k^2}
=\frac{1}{k^2}
\sum_m\sum_n\tilde{\Gamma}_m(\bm{k})\tilde{\Gamma}_n(-\bm{k})
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)].$$ After some calculation, we have the formula for the energy spectrum, $$\begin{split}
\tilde{E}(k)=&\frac{1}{4\pi k} \left\{
\sum_n{\Gamma_n}^2 \right.
+2\sum_m\sum_{n<m}\Gamma_m\Gamma_nJ_0(k| \bm{r}_m-\bm{r}_n|) \\
&\qquad -2\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}\varepsilon_l\left(\frac{r_m}{R}\right)^lJ_l(kR)J_l(kr_n)
\cos[l(\varphi_m-\varphi_n)] \\
&\qquad +\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty} \varepsilon_l
\left[J_l(kR)\right]^2
\left. \left(\frac{r_mr_n}{R^2}\right)^l\cos[l(\varphi_m-\varphi_n)] \right\},
\label{eq:ES}
\end{split}$$ where $\varphi_m$ is the angle between $\bm{r}_m$ and the zero axis, and $z_m = r_m \text{e}^{\text{i}\varphi_m}$. $J_l(x)$ is the $l$-th Bessel function. $\varepsilon_l$ is defined as $$\varepsilon_l= \begin{cases}
1 & (l=0) \\
2 & (l\geq1).
\end{cases}$$ Detail on the derivation is shown in Appendix B. Note that the first two terms in eq.(\[eq:ES\]) are just the formula derived by Novikov [@Novikov] and then the rest terms represent the effect of the circular boundary.
Numerical Simulation
====================
In this section, we show the results of our numerical simulations on a vortex crystal and vortex merger. We confirm that the system of a few hundreds discrete point vortices is also able to exhibit a vortex crystal and a vortex merger.
Non-neutral plasma experiments and 2D Euler equation numerical simulations found that by the existence of continuous low background vorticity distribution, in certain initial conditions intense vortices (clumps) are arrayed in a triangular lattice (vortex crystal). When the initial condition changes slightly to other distribution, clumps merge each other (vortex merger). This merger is the phenomenon seen when the “temperature” is negative. In our numerical calculations, the Hamilton equations of the $N$-point vortex system are numerically integrated, not using the Euler equation. Numerical simulation is performed in the following procedure. We numerically integrate the equations of motion of point vortices with circular boundary (whose radius R=10), namely eqs.(\[eq:dxkabe\]) and (\[eq:dykabe\]), by the 4th-order Runge-Kutta method. In our simulation, a relative error of the constants of motion is $10^{-7}-10^{-8}$ for energy $E$ and $10^{-10}-10^{-11}$ for angular impulse $I$. In the simulation using the Euler equation by Schecter *et al*., [@Schecter1] errors of both constants of motion are about $10^{-3}$.
Vortex crystals
---------------
In the experiment by Sanpei *et al*., [@Sanpei] it is found that three clumps are accelerated to form an equilateral triangular configuration by interaction with continuous low-level background vorticity distribution. Since three clumps make a unit cell of vortex crystals, it is regarded that this process is a fundamental process of forming vortex crystals. The existence of continuous low-level background vorticity distribution and the interaction between the background and the clumps are regarded as a key factor for making vortex crystals.
We reproduce qualitatively a vortex crystal of three clumps in a system of point vortices. In Fig.\[Fig:VrCry\], our result is depicted. In our simulation, for the initial condition, each clump (densely distributed regions of point vortices) contains about 60 vortices, and the background distribution consist of about 180 vortices. The circulation of each vortex is 0.2. Three Clumps are arrayed on a straight-line with one clump at equal distances from others (Fig.\[Fig:VrCry\](a)) so as to be the same as the experiment by Sanpei *et al*. [@Sanpei] The middle clump contains more point vortices than others because of the overlap with background point vortices.
![ The vortex crystal of point vortices. Each clump contains 60 vortices. The background consists of 180 vortices. The circulation of each point vortex is 0.2. []{data-label="Fig:VrCry"}](Fig5.eps)
Before seeing the result of vortex crystallization of our simulation, we define the property of clumps configuration, namely “arranging order”. Labeling each clump by $C_i (i=1,2$ and $3$), rotating sequence of clumps in anti-clockwise is only $(C_1C_2C_3)$ or $(C_3C_2C_1)$. We call these rotating sequence the “arranging order”. Vortex crystallization occurs as the following. At early steps, the clumps are at vertices of non-equilateral triangle rotating in the background point vortices (in Figs.\[Fig:VrCry\](a)-\[Fig:VrCry\](c)). Although the arranging order of the clumps do not change, the side length of the triangle of clumps changes. At later steps, the clumps form equilateral triangle and these rotate quasi-stationarily for a long time (we calculated until $T=200$) (in Figs.\[Fig:VrCry\](d)-\[Fig:VrCry\](f)).
In addition to the “arranging order”, we introduce a signed symmetry parameter $S$. A symmetry parameter is defined as $S=12\sqrt{3}A/l^2$ by Sanpei *et al*.[@Sanpei] $A$ is the area and $l$ is the peripheral length of the triangle whose vertices are the clumps. We define $A$ as $\bm{e}_z\cdot(\bm{r}_1-\bm{r}_3)\times(\bm{r}_2-\bm{r}_3)/2$. Here $\bm{e}_z$ is a unit vector perpendicular to $(x,y)$-plane, and $\bm{r}_i$ is the position of the $i$-th clump. An absolute value of $S$ is maximized at 1 when the clumps form an equilateral triangle, and the sign of $S$ is changed when the “arranging order” changes. In the following result of our simulation, the sign of $S$ does not change because the “arranging order” does not change.
We confirmed that the energy of clumps is transferred to the background vortices in the process of vortex crystallization (Fig.\[Fig:EEE\]). The energy of three clumps decreases and the energy of the background increases while total energy of the system is conserved. In this process, the tendency of the energy of the background vortices is similar to that of the signed symmetry parameter(Fig.\[Fig:S-E\]). Therefore, this means that if there is an outlet of energy of clumps, i.e., the background, three clumps tend to form a triangular lattice. However, it is necessary for the clumps not to occur merging.
In this simulation, the arranging order of the clumps (namely, the sign of the signed symmetry parameter $S$) does not change. However, in several parameters (the number of vortices, the circulation of background or clumps, and the ratio of these) this arranging order does change, i.e., the system shows the “binary star motion”, and the system does not exhibit vortex crystals. In ref. , it is noticed that the ratio of the circulation of background to that of clumps influences the relaxation rate toward vortex crystals. According to the simulation by Schecter et. al, [@Schecter1] the relaxation rate toward vortex crystals is the fastest when $\Gamma_{\text{background}}/\Gamma_{\text{total}}$ is $0.2-0.4$. In our case, $\Gamma_{\text{background}}/\Gamma_{\text{total}}=0.5$. We suppose that the background distribution with the proper ratio to total circulation restrains the “binary star motion” to crystallize the clumps.
It is observed that the point vortices of the clumps and the background mix each other (Fig.\[Fig:VrCry\]). However, the number of vortices of each clump is conserved around 60. The vortices of the clumps indicated by $\blacktriangle$, $\blacktriangledown$ and $\blacklozenge$ are oozed out to the background. The vortices of the background indicated by $+$ permeate to the clumps.
![ The energy of three clumps, the energy of the background vortices and the total energy of the system. The energy of the clumps tends to transfer toward the background. In later steps (the time crystallization occur), the energy transfer saturates. []{data-label="Fig:EEE"}](Fig6.eps){width="15.0cm"}
![ The signed symmetry parameter and the energy of the background vorticity. These show similar tendency that when the clumps form triangular lattice these value are growing. []{data-label="Fig:S-E"}](Fig7.eps){width="15.0cm"}
Merging
-------
The background vortices causes not only crystallization but also merging. In this subsection, we consider the $N$-point vortex system as a model of vortex merging. Merging is highly important in a process of decay into vortex crystals, or into equilibrium states. Furthermore the power law behavior of the number of clumps ($N\sim t^{-\xi}, \xi=0.2-0.7$) [@Fine; @JinDubin] in vortex crystals occur through the merging process of clumps. In merging process, we also calculate the energy spectrum of the system. In our simulation, each clump and the background contain $30$ point vortices. The circulation $\Gamma_m$ is set as $\Gamma_m=1$. For the initial configuration, the clumps are set at the vertices of equilateral triangle whose center of mass is at the origin. The energy spectrum is calculated by the formula derived in the previous section. We show the configuration of point vortices and the energy spectrum of this system in Fig.\[Fig:VrMg\]. The circular boundary is out of view (the area shown in Fig.\[Fig:VrMg\] is the region $[-6,6]\times[-6,6]$, the radius of boundary is $R=10$). The behavior of peaks of the energy spectrum of the range $10^0<k<10^1$ drastically changes in merging process. Before merging (in a lattice state, Figs.\[Fig:VrMg\](a) and \[Fig:VrMg\](b)), the peaks of the energy spectrum are oscillated with relatively large period in the range $10^0<k<10^1$. However, after merger (Figs.\[Fig:VrMg\](c) and \[Fig:VrMg\](d)) these become very sharp and wildly oscillates (i.e., plenty of peaks).
The energy spectrum shows the power law $E(k) \sim k^{-2.8}$ in the range $10^{0}<k<10^1$, and $E(k) \sim k^{-1.1}$ in the range $k>10^1$ (Fig.\[Fig:Beki\]). The origin of the power $-1$ region is the self energy (the first term of eq.(\[eq:ES\])). The former power is near to the value of the prediction for the 2D turbulence, [@Ranryuu0; @Ranryuu1; @Ranryuu2; @Ranryuu3] i.e., $E(k)\sim k^{-3}$. However, the other initial condition leads to different values, namely $E(k)\sim k^{-\alpha}, \alpha\approx 2.2-2.8$. Moreover, the range of this power is changed for different distributions of vortices. The range $10^0<k<10^1$ and the shape of spectrum in this range are an effect of the distribution of vortices in the merged clump.
In order to confirm the effect of the background vorticity distribution, we calculate the system without the background of which configuration of vortices in the clumps are the same as the system with the background (Fig.\[Fig:VrNoMg\]). In this case, the clumps rotate preserving an equilateral triangular configuration for long time. Therefore this means that the background vortices merge clumps which are located separately so as not to merge without the background.
![ The configurations of vortices and the energy spectrum in the merging process. Each clump and the background consist of 30 point vortices. Around $T=2.0$, a vortex merger occurs. []{data-label="Fig:VrMg"}](Fig8.eps)
![ The power law of the energy spectrum after merging. $E(k)\sim k^{-2.8}$ in the range $10^0<k<10^1$. The power $-1$ range is originated in the self energy. []{data-label="Fig:Beki"}](Fig9.eps){width="10.0cm"}
![ The clumps without background vorticity distribution. Each clump contains 30 vortices. The initial configuration of vortices of clumps are same as Fig.\[Fig:VrMg\]. []{data-label="Fig:VrNoMg"}](Fig10.eps)
Discussion and Conclusions
==========================
In this paper, the vortex crystal and the vortex merger in the system of a few hundred point vortices are studied. The three-point vortex system without the background is investigated as a fundamental process of a three clumps system in the background vorticity distribution.
In §2, we studied the three-point vortex system with boundary. In order to investigate chaotic behavior of this system, the Poincaré plot is calculated. The size of the plotted region on the section is closely related to the density of states: in the parameter where the density of state is low, the plotted area on the Poincaré section is narrow and vice versa. The topology of the plotted area changes when the energy is varied across the value of the energy in which three point vortices are at the vertices of an equilateral triangular configuration. When point vortices are in an equilateral triangular configuration, the plots on the Poincaré section are at the origin and on the circle of which radius is $\sqrt{1/2}$. When the parameters change from this value, there are no plots at the origin on the section. We confirm that the sign of “temperature” influence the property of Poincaré plot. When the “temperature” is negative, the region of Poincaré plot is scooped out from inner of the region. The positive “temperature” leads to the Poincaré plot in which there is no Tori. The property of the Poincaré plot reflects the motion of point vortices. Tori or “chaotic sea” on this section have the relationship with the type of the motion of point vortices. The orbit that two point vortices rotate each other apart from the other point vortex draws tori on $(P_1,R_1)$-plane. The cusps are seen at the locus of the motion in which “chaotic sea” emerges on the section. The former indicates regular, and the latter shows chaotic behavior of point vortices.
The energy spectrum of this system is introduced in §3. Using this result, we calculate the energy spectrum of the $N$-point vortex system in merging process in §4. The energy spectrum in the range around $10^0<k<10^1$ relaxes to the form of $E(k)\sim k^{-\alpha},(\alpha\approx 2.2-2.8)$. This range and the power are changed as the distribution of vortices is varied. In §4, the vortex crystal is reproduced in the system of point vortices. This shows that continuous distribution of vorticity is not necessary for vortex crystals. We find the energy transfer is occurred in the process of the vortex crystal of our simulation. The energy of clumps is disgorged to the background, and the clumps form triangle lattice. It is observed that the point vortices of the clumps and of the point vortices of the background are mixing each other. However, the number of voritices in each clump is conserved. This mixing effect of point vortices in the crystallization of clumps must be investigated.
In our simulation, in the process of the vortex crystal, the arranging order of the clumps along the direction of rotation (namely, the sign of symmetry parameter $S$ introduced at §4.1) does not change; however the arranging order of the point vortices are changed in a three-point vortex system as Fig.\[Fig:PS-VC\](O-1). When this order of the clumps is permutated, vortex crystals does not occur in our simulations. This suggests a possibility that there are a relation between the crystallization of three clumps and the motion of three point vortices mentioned above. In a three-point vortex system, three vortices change the arranging order at some initial conditions and do not change at the other. We suppose that there are at least three factors for the crystallization of clumps, namely the parameters $(I,E)$, the ratio of circulation $\Gamma_{\text{background}}/\Gamma_{\text{total}}$, and the ratio of the number of vortices $N_{\text{background}}/N_{\text{total}}$. Of course, the location of clumps also influences the crystallization. It must be investigated which parameter restrain the change of the order of rotating vortices.
Acknowledgments {#acknowledgments .unnumbered}
===============
We are indebted to Prof. H. Tomita, Prof. H. Hayakawa and Dr. Y. Yatsuyanagi for valuable discussions about this work. We also acknowledge Prof. Y. Kiwamoto and Dr. A. Sanpei for inspiration of this work.
Appendix A {#appendix-a .unnumbered}
==========
We want to calculate $v_{\theta}=-\partial \Psi/\partial r$ which is the second term of the vorticity field in eq.(\[eq:IdcVor\]); the term of induced vortex sheet on the circular boundary. When the $m$-th vortex is at the position $\bm{r}_m=(r_m,0)$, using the method of images, the stream function of this vortex at the position $\bm{r}=(r,\theta)$ is $$\begin{aligned}
\Psi_m(r,\theta)=&-\frac{\Gamma_m}{2\pi}\log r^{\prime}
+\frac{\Gamma_m}{2\pi}\log r^{\prime\prime}
-\frac{\Gamma_m}{2\pi}\log{\frac{R}{r_m}} \notag \\
=&-\frac{\Gamma_m}{4\pi}\log(r^2+{r_m}^2-2rr_m \cos\theta)
+\frac{\Gamma_m}{4\pi}\log(r^2+{{r_m}^{\prime}}^2
-2r{r_m}^{\prime}\cos\theta)-\frac{\Gamma_m}{2\pi}\log{\frac{R}{r_m}} \notag \\
=&-\frac{\Gamma_m}{4\pi}\log(r^2+{r_m}^2-2rr_m \cos\theta) \notag \\
&+\frac{\Gamma_m}{4\pi}
\log\left[r^2+\left(\frac{R^2}{r_m}\right)^2
-2r\left(\frac{R^2}{r_m}\right)\cos\theta\right]-\frac{\Gamma_m}{2\pi}\log{\frac{R}{r_m}},\end{aligned}$$ where in the last line, the relation ${r_m}^{\prime}=R^2/r_m$ is used. $r^{\prime}$, $r^{\prime\prime}$ and ${r_m}^{\prime}$ are, as given by Fig.\[fig:ImVor\], the distance between $\bm{r}$ and the $m$-th vortex, the distance between $\bm{r}$ and the image vortex of the $m$-th vortex, and the distance between the origin and the image vortex, respectively. $\theta$ is the angle between $\bm{r}$ and $\bm{r}_m$.
![The image vortex[]{data-label="fig:ImVor"}](Fig11.eps){width="10.0cm"}
Therefore $\partial \Psi_m/\partial r$ is written as $$\frac{\partial\Psi_m}{\partial r}
=-\frac{\Gamma_m}{4\pi}
\left[
\frac{2r-2r_m \cos \theta}{r^2+{r_m}^2-2rr_m\cos\theta}
-\frac{2r-\frac{2R^2}{r_m}\cos\theta}
{r^2+\frac{R^4}{{r_m}^2}-2\frac{R^2}{r_m}r\cos\theta} \right].\label{eq:PR}$$ When $r=R$, eq.(\[eq:PR\]) becomes $$\begin{aligned}
\left.\frac{\partial\Psi_m}{\partial r}\right|_{r=R}
&=-\frac{\Gamma_m}{4\pi}
\left[
\frac{2R-2r_m \cos \theta}{R^2+{r_m}^2-2Rr_m\cos\theta}
-\frac{2\frac{{r_m}^2}{R}-2r_m\cos\theta}{R^2+{r_m}^2-2Rr_m\cos\theta}\right] \\
&=-\frac{\Gamma_m}{4\pi}
\frac{2(R^2-{r_m}^2)}{R(R^2+{r_m}^2-2Rr_m\cos\theta)} \\
&=-\Gamma_m \frac{R^2-{r_m}^2}{2\pi R{| \bm{R}-\bm{r}_m |}^2},\end{aligned}$$ where $\bm{R}=R\bm{r}/r$. The sum about $m$ of the above equation is the second term of eq.(\[eq:IdcVor2\]).
Appendix B {#appendix-b .unnumbered}
==========
In this Appendix, the derivation of the energy spectrum $\tilde{E}(k)$ is shown. The starting point is $$\begin{aligned}
\tilde{E}(k)
&=\frac{| \tilde{\bm v}(\bm k)|^2}{2}
=\frac{| \tilde\omega(\bm k)|^2}{2k^2} \notag \\
&=\frac{1}{2k^2}
\sum_m\sum_n\tilde{\Gamma}_m(\bm{k})\tilde{\Gamma}_n(-\bm{k})
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)],\end{aligned}$$ where, $$\begin{aligned}
\tilde{\Gamma}_m(\bm{k}) &= \frac{1}{2\pi}\Gamma_m\left[
1-\frac{R^2-{r_m}^2}{2\pi R}\oint_{r=R} \frac{\exp(\text{i}\bm{k}\cdot\bm{R})}
{| \bm{R}-\bm{r}_m|^2} \text{d}\bm{r} \cdot \exp(-\text{i}\bm{k}\cdot\bm{r}_m)
\right]. \label{eq:tGamma}\end{aligned}$$ First, for the second term in eq.(\[eq:tGamma\]), we use the following relation $$\begin{split}
\exp(\text{i}\bm{k}\cdot\bm{R}) &= \text{e}^{\text{i}kR\cos(\theta-\phi_m)}
= \text{e}^{\text{i}kR\sin(\theta-\phi_m-3\pi/2)} \\
&=\sum_{l=-\infty}^{\infty}J_l(kR)\text{e}^{\text{i}l(\theta-\phi_m-3\pi/2)},
\end{split}$$ where $\theta$ and $\phi_m$ are the angles between $\bm{r}_m$ and $\bm{R}$, and between $\bm{k}$ and $\bm{r}_m$, respectively. Therefore we have $$\begin{aligned}
\oint_{r=R}
\frac{\exp(\text{i}\bm{k}\cdot\bm{R})}
{| \bm{R}-\bm{r}_m|^2} \text{d}\bm{r}
&=\int_0^{2\pi} \text{d}\theta R
\frac{\sum_{l=-\infty}^{\infty}J_l(kR)\text{e}^{\text{i}l(\theta-\phi_m-3\pi/2)}}
{R^2\left[1+\left(\frac{r_m}{R}\right)^2-2\frac{r_m}{R}\cos\theta\right]} \\
&=\frac{1}{R}\sum_{l=-\infty}^{\infty}J_l(kR)(\text{i})^l
\int_0^{2\pi} \text{d}\theta
\frac{\text{e}^{\text{i}l(\theta-\phi_m)}}
{1+\left(\frac{r_m}{R}\right)^2-2\frac{r_m}{R}\cos\theta}.\label{eq.VnsSin}\end{aligned}$$ The imaginary part of the integral in the last line of eq.(\[eq.VnsSin\]) vanishes because it is an odd function. The integral is evaluated as $$\int_0^{2\pi} \text{d}\theta
\frac{\text{e}^{\text{i}l\theta}}
{1+\left(\frac{r_m}{R}\right)^2-2\frac{r_m}{R}\cos\theta}
=\frac{2\pi}{1-\frac{{r_m}^2}{R^2}}\left(\frac{r_m}{R}\right)^{|l|}.$$ Since ${r}_m/R<1$, the following relation can be used ($|a|<1$). $$\int_0^\pi\frac{\cos\alpha x}{1-2a\cos x +a^2 }\,\text{d}x
=\frac{\pi a^{|\alpha|}}{1-a^2}.$$ Thus, we have $$\begin{aligned}
\oint_{r=R}\frac{\exp(\text{i}\bm{k}\cdot\bm{R})}{| \bm{R}-{r_m}|^2}\text{d}\bm{r}
&=\frac{1}{R}\sum_{l=-\infty}^{\infty}J_l(kR)(\text{i})^l
\frac{2\pi}{1-\frac{{r_m}^2}{R^2}}\left(\frac{r_m}{R}\right)^{|l|}
\text{e}^{-\text{i}l\phi_m}\\
&=\frac{2\pi R}{R^2-{r_m}^2}
\left[J_0(kR)
+2\sum_{l=1}^{\infty}J_l(kR)(\text{i})^l\cos(l\phi_m)
\left(\frac{r_m}{R}\right)^l
\right].\end{aligned}$$ Finally eq.(\[eq:tGamma\]) is given by $$\begin{split}
\tilde{\Gamma}_m(\bm{k})
&= \frac{1}{2\pi}\Gamma_m\left[1-\frac{R^2-{r_m}^2}{2\pi R}
\oint_{r=R} \frac{\exp(\text{i}\bm{k}\cdot\bm{R})}
{| \bm{R}-\bm{r}_m|^2} \text{d}\bm{r} \cdot \exp(-\text{i}\bm{k}\cdot\bm{r}_m)
\right] \\
&=\frac{1}{2\pi}\Gamma_m\left[1-
\sum_{l=0}^{\infty}\varepsilon_lJ_l(kR)\cos(l\phi_m)
\left(\frac{\text{i}r_m}{R}\right)^l \exp(-\text{i}\bm{k}\cdot\bm{r}_m)
\right],
\end{split}$$ where $$\varepsilon_l= \begin{cases}
1 & (l=0) \\
2 & (l\geq1).
\end{cases}$$ Substituting this into $|\tilde\omega(\bm k)|^2$, we get $$\begin{aligned}
|\tilde\omega(\bm k)|^2
&=\sum_m\sum_n\tilde{\Gamma}_m(\bm{k})\tilde{\Gamma}_n(-\bm{k})
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)] \\
&=\frac{1}{(2\pi)^2}\sum_m\sum_n\Gamma_m\Gamma_n
\left\{
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)]
-{\sum_l}^{(m)}-{\sum_{l^{\prime}}}^{(n)}
+{\sum_{l,l^\prime}}^{(m,n)}
\right\}.\end{aligned}$$ Here we define $$\begin{split}
{\sum_l}^{(m)}
& \equiv
\sum_{l=0}^{\infty}\varepsilon_lJ_l(kR)\cos(l\phi_m)
\left(\frac{\text{i}r_m}{R}\right)^l \exp(-\text{i}\bm{k}\cdot\bm{r}_n), \\
{\sum_{l^{\prime}}}^{(n)}
&\equiv
\sum_{l=0}^{\infty}
\varepsilon_{l^\prime}J_{l^\prime}(kR)\cos(l^\prime\phi_n)
\left(\frac{-\text{i}r_n}{R}\right)^{l^\prime} \exp(\text{i}\bm{k}\cdot\bm{r}_m), \\
{\sum_{l,l^\prime}}^{(m,n)}
& \equiv
\sum_{l=0}^{\infty}
\sum_{l^\prime=0}^{\infty}
\varepsilon_l\varepsilon_{l^\prime}
J_l(kR)\cos(l\phi_m) J_{l^\prime}(kR)\cos(l^\prime\phi_n)
\left(\frac{\text{i}r_m}{R}\right)^l
\left(\frac{-\text{i}r_n}{R}\right)^{l^\prime}.
\end{split}$$ Integrating over angles $\varphi$ of $\bm{k}$, the energy spectrum can be written as $$\begin{split}
\tilde{E}(k)&=\int_0^{2\pi}\text{d}\varphi\,k\frac{|\tilde{\bm v}(\bm k)|^2}{2}
=\frac{1}{2k}\int_0^{2\pi}\text{d}\varphi|\tilde\omega(\bm k)|^2 \\
&=\frac{1}{2(2\pi)^2k}\int_0^{2\pi}\text{d}\varphi
\sum_m\sum_n\Gamma_m\Gamma_n
\left\{
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)]
-{\sum_l}^{(m)}-{\sum_{l^{\prime}}}^{(n)}
+{\sum_{l,l^\prime}}^{(m,n)}
\right\}\\
&\equiv (A)+(B)+(C)+(D),\label{eq:Esp}
\end{split}$$ where, $$\begin{aligned}
(A)=&\frac{1}{2(2\pi)^2k}\int_0^{2\pi}\text{d}\varphi
\sum_m\sum_n\Gamma_m\Gamma_n
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)]
,\\
(B)=&-\frac{1}{2(2\pi)^2k}\int_0^{2\pi}\text{d}\varphi
\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}\varepsilon_lJ_l(kR)\cos(l\phi_m)
\left(\frac{\text{i}r_m}{R}\right)^l \exp(-\text{i}\bm{k}\cdot\bm{r}_n)
,\\
(C)=&-\frac{1}{2(2\pi)^2k}\int_0^{2\pi}\text{d}\varphi
\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}
\varepsilon_{l^\prime}J_{l^\prime}(kR)\cos(l^\prime\phi_n)
\left(\frac{-\text{i}r_n}{R}\right)^{l^\prime} \exp(\text{i}\bm{k}\cdot\bm{r}_m)
,\\
(D)=&\frac{1}{2(2\pi)^2k}\int_0^{2\pi}\text{d}\varphi
\sum_m\sum_n\Gamma_m\Gamma_n \notag \\
&\times
\left\{
\sum_{l=0}^{\infty}
\sum_{l^\prime=0}^{\infty}
\varepsilon_l\varepsilon_{l^\prime}
J_l(kR)\cos(l\phi_m) J_{l^\prime}(kR)\cos(l^\prime\phi_n)
\left(\frac{\text{i}r_m}{R}\right)^l\left(\frac{-\text{i}r_n}{R}\right)^{l^\prime}
\right\}.\end{aligned}$$ We calculate each term as follows.
In order to calculate the first term $(A)$, suppose that the angle between $\bm{k}$ and $(\bm{r}_m-\bm{r}_n)$ is $\psi$. We can expand $\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)]$ in the series of the Bessel functions, $$\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)]=\exp[\text{i}k|\bm{r}_m-\bm{r}_n|\cos\psi]
=\sum_{l=-\infty}^{\infty}J_l(k|\bm{r}_m-\bm{r}_n|)\text{e}^{\text{i}l(\psi+\pi/2)}.$$ Thus we have $$\begin{aligned}
\int_0^{2\pi}\text{d}\varphi
\exp[\text{i}\bm{k}\cdot(\bm{r}_m-\bm{r}_n)]
&=\int_0^{2\pi}\text{d}\varphi
\sum_{l=-\infty}^{\infty}J_l(k| \bm{r}_m-\bm{r}_n|)\text{e}^{\text{i}l(\psi+\pi/2)}\notag \\
&=\sum_{l=-\infty}^{\infty}J_l(k| \bm{r}_m-\bm{r}_n|)\int_0^{2\pi}\text{d}\varphi
\,\text{e}^{\text{i}l(\psi+\pi/2)} \notag\\
&=2\pi J_0(k| \bm{r}_m-\bm{r}_n|),\end{aligned}$$ where $\psi=\chi-\varphi$ and $\chi$ is the angle of $\bm{r}_m-\bm{r}_n$. Then, the first term $(A)$ is written as $$\begin{split}
(A) &= \frac{1}{4\pi k}
\sum_m\sum_n\Gamma_m\Gamma_nJ_0(k| \bm{r}_m-\bm{r}_n|)\\
&=\frac{1}{4\pi k}\left[\sum_n{\Gamma_n}^2
+2\sum_m\sum_{m<n}\Gamma_m\Gamma_nJ_0(k| \bm{r}_m-\bm{r}_n|) \right].
\end{split}$$
The second term $(B)$ of eq.(\[eq:Esp\]) is $$(B)
=-\frac{1}{2(2\pi)^2 k}\int_0^{2\pi}\text{d}\varphi\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}\varepsilon_lJ_l(kR)\cos(l\phi_m)
\left(\frac{\text{i}r_m}{R}\right)^l \exp(-\text{i}\bm{k}\cdot\bm{r}_n) \label{eq:2term}.$$ Suppose that $\varphi$ and $\varphi_n$ are the angle components of $\bm{k}$ and $\bm{r}_n$ respectively. We have $$\exp(-\text{i}\bm{k}\cdot\bm{r}_n)
=\sum_{l=-\infty}^{\infty}J_l(kr_n)
\text{e}^{-\text{i}l\left(\varphi-\varphi_n+\frac{\pi}{2}\right)}$$ and $$\cos(l\phi_n)
=\frac{1}{2}\left\{\text{e}^{\text{i}l(\varphi-\varphi_n)}+\text{e}^{-\text{i}l(\varphi-\varphi_n)}
\right\},$$ where $\phi_n=\varphi-\varphi_n$. Substituting these into the sum over $l$ in eq.(\[eq:2term\]), we obtain $$\begin{aligned}
&\sum_{l=0}^{\infty} \varepsilon_lJ_l(kR)\cos(l\phi_m)
\left(\frac{\text{i}r_m}{R}\right)^l \exp(-\text{i}\bm{k}\cdot\bm{r}_n) \notag \\
&=\frac{1}{2}
\sum_{l=0}^{\infty}\varepsilon_l\left(\frac{\text{i}r_m}{R}\right)^lJ_l(kR)
\sum_{l^\prime=-\infty}^{\infty}J_{l^\prime}(kr_n)
\text{e}^{-\text{i}l^\prime\left(\varphi-\varphi_n+\frac{\pi}{2}\right)}
\left\{\text{e}^{\text{i}l(\varphi-\varphi_m)}+\text{e}^{-\text{i}l(\varphi-\varphi_m)} \right\}.
\label{eq:lpart}\end{aligned}$$ Integrating over the angle $\varphi$, the sum over $l^{\prime}$ in eq.(\[eq:lpart\]) is written as $$\begin{aligned}
&\sum_{l^\prime=-\infty}^{\infty}
\int_0^{2\pi}\text{d}\varphi \, J_{l^\prime}(kr_n)
\text{e}^{-\text{i}l^\prime\left(\varphi-\varphi_n+\frac{\pi}{2}\right)}
\left\{\text{e}^{\text{i}l(\varphi-\varphi_m)}+\text{e}^{-\text{i}l(\varphi-\varphi_m)} \right\} \notag\\
&=J_l(kr_n)
(-\text{i})^l\cdot 2\cos[l(\varphi_m-\varphi_n)] \cdot 2\pi,\end{aligned}$$ where the integral over $\varphi$ of $\text{e}^{\text{i}\varphi(l-l^{\prime})}$ is zero when $l\neq l^{\prime}$ and is $2\pi$ when $l=l^{\prime}$. Then we can write the second term $(B)$ as $$\begin{aligned}
(B)
&=-\frac{1}{4\pi k}\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}\varepsilon_l\left(\frac{r_m}{R}\right)^lJ_l(kR)J_l(kr_n)
\cdot \cos[l(\varphi_m-\varphi_n)] .\end{aligned}$$
For the third term $(C)$, a similar calculation to the second term $(B)$ can be applied. Then we have $$(C)=(B)
=-\frac{1}{4\pi k}\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}\varepsilon_l\left(\frac{r_m}{R}\right)^lJ_l(kR)J_l(kr_n)
\cdot \cos[l(\varphi_m-\varphi_n)]$$
The last term $(D)$ is $$\begin{aligned}
(D)&=\frac{1}{2(2\pi)^2 k}\sum_m\sum_n\Gamma_m\Gamma_n \notag \\
&\times\sum_{l=0}^{\infty}\sum_{l^\prime=0}^{\infty} \int_0^{2\pi}\text{d}\varphi
\varepsilon_l\varepsilon_{l^\prime}
J_l(kR)J_{l^\prime}(kR)\cos(l\phi_m)\cos(l^\prime\phi_n)
\left(\frac{\text{i}r_m}{R}\right)^l\left(\frac{-\text{i}r_n}{R}\right)^{l^\prime}.
$$ Since $l$ and $l^\prime$ are both integers ( $l,l^\prime \geq 0$ ), then we have $$\begin{aligned}
&\int_0^{2\pi}\text{d}\varphi
\cos(l\phi_m)\cos(l^\prime\phi_n)
=\int_0^{2\pi}\text{d}\varphi
\cos[l(\varphi-\varphi_m)]\cos[l^\prime(\varphi-\varphi_n)] \notag \\
&= \begin{cases}
2\pi & (l=l^\prime=0) \\
\pi\cos[l(\varphi_m-\varphi_n)] & (l=l^\prime\neq 0)\\
0 & (\text{otherwise}).
\end{cases}\end{aligned}$$ Thus the fourth term $(D)$ is given by $$\begin{aligned}
(D)&=\frac{1}{4\pi k}\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty} \varepsilon_l
\left[J_l(kR)\right]^2
\left(\frac{r_mr_n}{R^2}\right)^l\cos[l(\varphi_m-\varphi_n)] .\end{aligned}$$ Substituting these four terms into eq.(\[eq:Esp\]), the formula for the energy spectrum is given by $$\begin{split}
\tilde{E}(k)=&\frac{1}{4\pi k} \left\{ \sum_n{\Gamma_n}^2 \right.
+2\sum_m\sum_{n<m}\Gamma_m\Gamma_nJ_0(k| \bm{r}_m-\bm{r}_n|) \\
&-2\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty}\varepsilon_l\left(\frac{r_m}{R}\right)^lJ_l(kR)J_l(kr_n)
\cdot \cos[l(\varphi_m-\varphi_n)] \\
&+\sum_m\sum_n\Gamma_m\Gamma_n
\sum_{l=0}^{\infty} \varepsilon_l
\left[J_l(kR)\right]^2
\left . \left(\frac{r_mr_n}{R^2}\right)^l\cos[l(\varphi_m-\varphi_n)] \right\}.
\end{split}$$
[99]{} K. S. Fine, C. F. Driscoll, J. H. Malmberg and T. B. Mitchell: Phys. Rev. Lett. **67** (1991) No.5, 588.
T. B. Mitchell, C. F. Driscoll and K. S. Fine: Phys. Rev. Lett. **71** (1993) No.9, 1371.
Y. Kiwamoto, K. Ito, A. Sanpei, A. Mohri, T. Yuyama and T. Michishita: J. Phys. Soc. Jpn. **68** (1999) No.12, 3766.
P. G. Saffman: *Vortex Dynamics*, (Cambridge univ. press, Cambridge, 1992), Chap.7 p.116.
L. G. Kurakin and V. I. Yudovich: Chaos **12** (2002) No.3, 574.
S.Kida: J. Phys. Soc. Jpn. **39** (1975) No.5, 1395.
R. A. Smith and Thomas M. O’Neil: Phys. Fluids B **2** (1990) No.2, 2961.
L. Onsager: Nuovo Cimento **6** (1949) 279.
P. H. Chavanis: preprint, arXiv : cond-mat/0212223 v1 10 Dec. 2002.
K. S. Fine, A. C. Cass, W. G. Flynn, and C. F. Driscoll: Phys. Rev. Lett **75** (1995) No.18, 3277.
D. Z. Jin and D. H. E. Dubin: Phys. Rev. Lett. **84** (2000) No.7, 1443.
M. Kono, H. L. Pécseli and J. Trulsen: Pysica Scripta **61** (2000) 489.
D. Z. Jin and D. H. E. Dubin: Phys. Rev. Lett. **80** (1998) No.20, 4434.
D. A. Schecter, D. H. E. Dubin, K. S. Fine, and C. F. Driscoll: Phys. Fluids **11** (1999) No.4, 905.
A. Sanpei, Y. Kiwamoto, K Ito, and Y. Soga: Phys. Rev. E **68** (2003) 016404.
A. Leonard: J. Comp. Phys. **37** (1980) 289.
H. Aref and N. Pomphrey: Proc. R. Soc. London, Ser. A **380** (1982) 359.
D. J. Johnson: Phys. Fluids **31** (1988) No.7, 1856.
R. H. Kraichnan: Phys. Fluid. **10** (1967) No.7, 1417.
R. H. Kraichnan and D. Montgomery: Rep. Prg. Phys. **43** (1980) 547.
P. Tabeling: Phys. Rep. **362** (2002) 1.
G. K. Batchelor: Phys. Fluids **12** (1969) Suppl. II, II 233.
E. A. Novikov: Sov. Phys.-JETP **41** (1975) No.5, 937.
[^1]: E-mail address: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Zhenning Cai[^1], Yuwei Fan[^2], Ruo Li[^3]'
bibliography:
- '../article.bib'
title: 'Globally Hyperbolic Regularization of Grad’s Moment System in One Dimensional Space'
---
[^1]: School of Mathematical Sciences, Peking University, Beijing, China, email: [[email protected]]{}.
[^2]: School of Mathematical Sciences, Peking University, Beijing, China, email: [[email protected]]{}.
[^3]: CAPT, LMAM & School of Mathematical Sciences, Peking University, Beijing, China, email: [[email protected]]{}.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Klaus Aehlig$^1$, Helge Dietert$^2$, Thomas Fischbacher$^1$, and Jochen Gerhard$^3$\
${}^1$ University of Southampton, Faculty of Engineering and the Environment,\
Southampton, UK\
${}^2$ University of Cambridge, UK\
${}^3$ Frankfurt Institute for Advanced Studies,\
Institut für Informatik,\
Goethe Universität Frankfurt, DE
title: 'Casimir Forces via Worldline Numerics: Method Improvements and Potential Engineering Applications'
---
Introduction
============
Casimir forces, i.e., electromagnetic forces arising due to quantum effects between uncharged conductors at distances much larger than characteristic atomic radii, have first been predicted via theoretical considerations in 1948 [@Casimir48], and have since been verified experimentally to 1% accuracy [@PhysRevLett.81.4549]. While initially a fringe subject – albeit unfortunately one that drew considerable attention from pseudoscientific authors – research interest in Casimir forces strongly increased in particular in the last decade, and a number of novel theoretical tools were developed [@Scardicchio:2004fy; @2003hep.th...11168M; @Rahi:2010zr; @PhysRevA.76.032106] that allow the study of the Casimir effect in considerably more involved situations (geometries, material properties, temperature) than earlier investigations. At present, one major obstacle to research is that Casimir force calculations often still are computationally very demanding. Nevertheless, the development of theoretical tools and methods must go hand in hand with progress in nanoscale manufacturing, for it is clear that a sound understanding of the role of Casimir forces in nano machines will become increasingly important as we learn to manufacture on shorter length scales.
One approach to the calculation of Casimir forces is based on the “Worldline approach” developed by Gies, Klingmüller, Langfeld and Moyaerts [@2002IJMPA..17..966G; @2003hep.th...11168M; @2006JPhA...39.6415G]. While this mostly has been used to study a simplified field theoretic model with massles scalars instead of vector gauge bosons (photons), and nowadays alternative methods are available to directly calculate electrodynamic effects even with frequency-dependent optical properties of materials [@2009arXiv0904.0267R; @2010PhRvA..81a2119M], the worldline approach is interesting for a number of reasons:
- Due to the probabilistic nature of the method, it is sometimes computationally comparatively cheap (depending on the geometry) to obtain a rough estimate of Casimir forces.
- With very little effort, the calculation can be modified in such a way that it simultaneously gives all the forces on a number of bodies, making it potentially attractive for problems requiring a geometric shape optimization approach.
- Finally, it is formulated in a way that is suggestive of a remarkably intuitive interpretation. This may manage to make this conceptually subtle quantum effect somewhat accessible to wider audiences not necessarily deeply familiar with quantum field theory. (One should compare this with the didactic problems related to the “mesomeric effect” in chemistry.)
In this article, we will focus mostly on the second item in the list above, which is elucidated in detail in Section \[sec:autodiff\]. The key insight is that a deformable object, such as a beam bending under the influence of Casimir forces as shown in figure \[fig:beam\], when discretized into $N$ elements (blocks) that are partially restricted in their relative motion, may require a certain computational effort, $T$ processor-seconds, for a reasonable estimate of the Casimir energy of a given configuration, but will then always allow the simultaneous computation of Casimir forces (and moments) on *all* $N$ elements in a way that requires at most $5T$ processor-seconds, *irrespective of the number of bodies $N$*.
\[fig:beam\]
{width="8cm"}
[Beam bending under the influence of external forces and moments, a standard problem in Engineering, may be regarded as an energy minimization problem. Geometries such as the one shown above that involve bending beams at tens-of-nanometers length scales are presently being discussed as nanoelectromechanical memory devices [@4160125]. Here, the beam also serves as a floating gate that allows electronic read-out of its mechanical state (upward bent or downward bent). Discretizing the beam into elements with one translational degree of motion, the nature of the Worldline numerics algorithm makes it easy to *simultaneously* calculate the Casimir forces on all elements with a computational effort less than one order of magnitude above the effort required to get a reliable estimate for the Casimir energy of the given configuration – *independent of the number of elements*. This is not easily achieved with a number of alternative methods to calculate Casimir forces.]{}
Computational Methods
=====================
In contradistinction to other approaches, Worldline Numerics is remarkably simple to implement in a computer program, requiring almost no advanced software library infrastructure to deal with issues such as parallel sparse matrix linear algebra, multipole moments, or large matrix eigenvalues. Still, there are a number of useful improvements of the basic method that help to ensure making effective use of computational resources. As these are often linked to one another, it makes sense to address them in a coherent fashion.
Monte Carlo integration over Loops
----------------------------------
The Casimir energy for a static geometry that can be modeled by a position-dependent potential $V(x)$ is given as the quantum effective action per unit time: $$E_{\rm Casimir}=\frac{\Gamma[V]}{\int_{\tau=\tau_-}^{\tau_+}\,d\tau}$$
The “Worldline Numerics” approach by Gies, Langfeld, and Moyaerts [@Gies:2003] is based on re-writing the logarithm in the effective action as an integral a la $\ln
(p/q) \rightarrow \int_0^{\infty} \frac{dT}{T}\left(\exp(-px)-\exp(-qx)\right)$ and re-expressing the operator trace as a Feynman path integral. This leads, for a real scalar field of mass $m$ interacting only with the external potential $V\!$, to an expression for the effective action that is numerically tractable via Monte Carlo methods. The key expression from [@Gies:2003], which we repeat here for the convenience of the readers, is: $$\label{eq:effective-action}
\Gamma_\Lambda[V] = - \frac 12 \frac 1{(4\pi)^{2}}\int_{1/\Lambda^2}^\infty
\frac{dT}{T^{3}} e^{-m^2T}\int d^4x
\left[\left\langle W_V[y;x,T]\right\rangle_y - 1\right]$$ where an UV cut-off regulator $\Lambda$ has been introduced. Here, the expectation value $\langle\cdot\rangle_y$ is the ensemble average over all closed loop (c.l.) gaussian random walks $y:[0;1]\mapsto\mathbb R, y(0)=y(1)$ of “Wilson loops” re-scaled to proper time $T$. Let the statistical weight of the loop $y$ be $$p[y]=\exp\left(-\int_{t=0}^{t=1}dt\,\dot y(t)^2/4\right)$$ then: $$\left\langle W_V[y;x,T]\right\rangle_y=
\frac{\int_{y\;\rm c.l.}\mathcal{D}y\,W_V[y;x,T]\,p[y]}
{\int_{y\;\rm c.l.}\mathcal{D}y\,p[y]}$$ where $W_V$ depends on the *path* $y$ and on position shift $x$ and propertime $T$: $$W_V[y;x,T]=\exp\left(-T\int_{t=0}^{t=1}dt\,V(x+\sqrt{T}y(t))\right).$$
From this expression for the effective action of a free scalar interacting with a potential, Casimir forces can be obtained by using the position dependency of the potential $V(x)$ to model the geometry, and calculating energy changes associated with changes to the geometry.
While the applicability of this model for the calculation of real Casimir forces is questionable (even for perfect conductors) as the physics of photons is quite different from that of a scalar field, the remarkable conceptual simplicity of the above expressions certainly warrants a deeper investigation of its properties and potential utility, for it might actually allow a (yet undiscovered) generalization to the photon case. For the electromagnetic case, one would naturally want to start with investigations of perfect conductor surfaces, and the (obvious) scalar pendant of this is a potential $V(x)$ that suppresses all quantum fluctuations *inside* the given bodies. It is not difficult to see that one may alternatively restrict the potential to have non-zero values only close to the surfaces of bodies, taking $$V(x) = \lambda \int_\Sigma d^2\sigma \delta^3(x-x_\sigma)$$ and considering the limit $\lambda\to\infty$. Then, $W_V[y;x,T]$ reduces to: $$\exp\left[-T\int_0^1 dt V(x+\sqrt T y(t))\right]
=
\begin{cases}
1 & \text{Loop pierces a surface}\\
0 & \text{Loop does not pierce a surface.}
\end{cases}$$ Substituting $s=\sqrt{T}$ to eliminate the square root, and using translation invariance of the integral to ensure all loops have center of gravity at the origin, the expression for the geometry-dependent regularized Casimir energy then is: $$\Gamma_\Lambda = -\frac{1}{(4\pi)^2} \int_{1/\Lambda}^\infty \frac{ds}{s^5} \langle\Theta_V(sy,x)\rangle_y$$ where the mean value is over all unit loops, $sy$ is the loop $y(t)$ scaled pointwise around its center of gravity $\bar y$ by the factor $s$, and $$\Theta_V(sy,x) =
\begin{cases}
0 & \text{Loop does not pierce a surface}\\
1 & \text{Loop pierces a surface.}
\end{cases}$$
The problem with this approach is that the Casimir energy attributed to the surface of any single body goes to infinity as we send the energy regulator $\Lambda$ to infinity, due to the contribution from very short loops close to the surface. (This is, of course, a non-physical artefact related to the “geometry much larger than atomic scales” approximation.) In order to predict Casimir forces between different objects we are only interested in the dependency of the energy on the relative position of these objects. Therefore, it makes sense to modify this scheme in a way that (i) the contribution of each loop is taken into account relative to a configuration in which all objects are at infinite separation from one another, and (ii) the Casimir energy contribution attributed to loops piercing only one object surface (hence, “belonging” to that object) is taken as zero.
Consider $n$ objects with potentials $V_1,\dots, V_n$. Then the total potential is $V=V_1+V_2+\dots +V_n$ and we use the freedom to shift the absolute energy level to define the “interaction Casimir energy” $E$ as in [@Gies:2003] as the energy difference relative to a configuration in which every body is at an effectively infinite distance from every other body: $$\label{eq:subtraction}
E = \left(\Gamma[V] - \Gamma[V_1] - \Gamma[V_2] - \dots - \Gamma[V_n]\right)/\Delta \tau$$
Using this, we get $$E = -\frac{1}{(4\pi)^2}\int_{1/\Lambda}^\infty \frac{ds}{s^5} \langle\Theta(sy,x)\rangle_y
\label{eq:loopenergy.final}$$ with $\Theta$ given by $$\Theta(sy,x) =
\begin{cases}
0 & \text{Re-scaled loop does not pierce any surface} \\
1-n & \text{Re-scaled loop pierces the surfaces of $n\ge 1$ objects}
\end{cases}$$
If $n$ objects come close to one another, every loop that pierces all of them can be regarded as the image of $n$ loops, each to be considered as being attached to (and moving with) that body. Hence, when objects are in proximity, we count a loop *once* that would have been counted $n$ times instead for separated objects. (Note that the counting weight of both a loop that pierces no surface, and a loop that pierces only one surface, is zero.) If the objects are now spatially separated the integral $\int_{1/\Lambda}^\infty
\frac{ds}{s^5} \langle\Theta(sy,x)\rangle_y$ is finite and well behaved for $\Lambda\to0$, so we can safely set $\Lambda=0$. One is easily convinced that this is indeed the correct expression by considering a simple geometry (such as two parallel flat slabs) and requesting that the Casimir force does not change if one object is instead thought of as being made of two adjacent bodies. The counting weights are dictated by the convention for the “zero energy” configuration.
For each loop $y$, the weight $\Theta$, as a function of the re-scaling factor $s$, is piece-wise constant. The $s$-integral hence can easily be performed analytically. Rather than being only a convenient simplification that saves computing time, this property plays a crucial role for the efficient simultaneous computation of multibody forces, cf. Section \[sec:autodiff\].
Loop Generation {#subsubsec:generation}
---------------
When trying to evaluate Equation \[eq:loopenergy.final\], one naturally would try to discretize the loop as consisting of a finite number of straight sections. Taking the procedure literally, the presence of complicated curved geometries would mandate computationally fairly expensive ray-surface intersection checks. In many cases, a better investment of the computational effort may be to instead make the number of discretization points on the loop sufficiently large to ensure that simple inside/outside checks applied to each point give a reasonably close approximation. Still, generic ray/surface intersection checks can become useful, especially if the complicated multiple integral in Equation \[eq:loopenergy.final\] (over loop shapes, loop sizes, and loop centers of gravity) can partially be evaluated by analytic, or rather semi-analytic means that involve numerical approximation of integral boundaries. This is relevant for the discussion in Section \[sec:scaling\], and may make major computational improvements of the method possible. In this context, we want to point out the existence of advanced algorithms useful for ray/surface intersection problems, such as Comba and Stolfi’s Affine Arithmetic [@CombaStolfiSiggraph93; @Knoll09]. In order to generate a properly distributed random sample of loops, we first generalize the problem to finding a process that produces piecewise straight paths with gaussian length distribution (of given standard deviation) for a given starting and end point (not necessarily coincident). This problem can be re-phrased as finding pairs of such paths, each with its own starting point but at first without any constraint on their endpoints, and imposing the condition that they meet at their endpoints. Concatenating the first path to the reverse of the second solves the problem of finding a path with the correct distribution between two given points. One easily sees that the distribution of the midpoint is still gaussian (being the product of two gaussian distributions). Hence, we can sample a loop by recursively sampling an intermediate point in the interval between a given start and end point.
This method, known as the “d (‘doubling’) loop algorithm” [@Gies:2005sb], manages to generate closed loops with the desired distribution with very little effort. A slight problem of this approach is that it only generates loops for which the number of vertices is a power of two, but as one usually is only interested in getting below a given resolution, and this method will at worst require an effort too large by a factor 2, this is practically irrelevant. Nevertheless, the reasoning presented above is readily generalized to non-equal subdivisions (e.g. 1/3 plus 2/3), and from there to also allow recursive subdivision into other numbers of parts, but this is typically not needed. In order to generate a loop containing 10 points using the extended d-loop method, one would first determine the opposite point (point \#6) of the starting point (point \#1), then for both arcs use a $2:3$ split of the associated variance budget and work out the gaussian distribution of the corresponding intermediate point’s position. Sub-division of the arc of length 2 is straightforward, while the length 3 arc would first be split $1:2$ using the same approach.
Rather than choosing starting points randomly in space and then determining the location of the halfway-round-the-loop point, it makes sense to perform stratified spatial sampling on a lattice. To do so, we choose the first point to be the lattice point and take the halfway point to be gaussian distributed with mean the first point and standard deviation a characteristic length. We then continue sampling a loop of that length and later scale it around the midpoint between the first point and the halfway point appropriately to obtain a unit loop. In that way, the midpoint is gaussian distributed around the lattice point with a given length scale.
If we want unbiased integration by taking loops for each lattice point, the lattice points have to be representative for loops sampled in their vicinity, with a characteristic length being that of the grid. This is certainly true if the grid is fine compared to any characteristic length of the geometry. However, the same can be achieved for arbitrary grids, if we take the characteristic length in the just described stratified sampling to be that of the grid.
A different kind of lattice effects has to be taken into account for methods computing a force as a difference in energy for two given geometries. Such methods would typically put a fixed set of loops on each lattice point and add up their energy contributions. Then they would do the same for the same geometry with one object moved in a particular direction. The difference in energy is then proportional to the force component on the moved object in the given direction.
To focus on the net effect, as we do with symmetries (see \[subsubsec:cancellation\]), one typically would use the same set of loops for both geometries. Also, to have for each loop a corresponding shifted loop, the amount the object is moved has to be a multiple of the grid length (in this direction). While doing otherwise would not necessarily yield a bias, doing so significantly improves the convergence speed of this method.
For our purposes, we typically only ask simple questions about each loop, such as “which objects does it hit?”, or (at most) “For what scaling intervals does this loop, centered at $x_{cm}$ but re-scaled in size, hit object $O_n$?”. In order to answer these, only very little information needs to be stored when visiting the loop point by point. So it is possible to implement the relevant algorithms in such a way that the loop is generated on the fly, and we never have to store the entire loop in memory – the number of points we have to remember is about the binary logarithm on the loop length. This yields an algorithm with a very small memory footprint and attractive characteristics for computing architectures that emphasize a high degree of parallelism between very simple cores.
Numerical Integration over the Scaling Factor {#subsec:directintegration}
---------------------------------------------
One approach to obtain energies—and so ultimately, by comparing energies for different geometric configurations, forces—is to directly evaluate the integral in Equation \[eq:loopenergy.final\] numerically. Naively, one would have to, for various values of $s$, estimate $\langle\Theta(sy,x)\rangle_y$ and summing up. Since the order of summing up does not matter, we can as well compute the expectation value of the following process.
> Choose $s$ uniformly at random from the interval $[a,b]$. Randomly generate a loop $sy$ of size $s$ and count $(1-n)/s^5$ if the loop hits $n\geq 2$ objects, and $0$ otherwise.
Here $[a,b]$ is an interval big enough so that integrating over that interval does not differ noticeably from integrating over all positive reals.
Looking at that random process more closely, one notes that the information about the random loop we use is the number of objects it hits. We have to pay particular attention to short loops that are just long enough to barely touch multiple objects, for they give the largest contribution to the sum. One should note that it is not possible to attribute a useful physical meaning to absolute differences in loop scaling factors $s$: for a loop that hits (at least) two objects, the effect of changing $s$ to $s+0.1$ very much depends on what the magnitude of $s$ is. As relative changes of the scaling factor hence are more important than absolute changes, we much prefer a distribution, when sampling loops, that handles all orders of magnitude equally. In other words, we prefer a distribution where the logarithm of $s$ is uniformly distributed on $[\ln(a),\ln(b)]$.
When changing the distribution of $s$, we also have to transform the weight attributed to each sample accordingly. Taking the logarithm of $s$ to be uniformly distributed, rather than $s$ itself, each value $s$ will be $1/s$ times as likely as before. To still get the same expectation, we have to multiply each value by $s$. Hence, we are finally left with estimating the expectation of the following process:
> Chose $\sigma=\ln s$ uniformly at random in the interval $[\ln(a),\ln(b)]$. Randomly generate a loop of size $e^\sigma$ and count $(1-n)e^{-4\sigma}=(1-n)s^{-4}$ if loop hits $n\geq 2$ elements, and $0$ otherwise.
Symbolic Integration over the Scaling Factor {#sec:scaling}
--------------------------------------------
It makes sense to try to perform at least part of the integration needed to evaluate eq. \[eq:effective-action\] symbolically, for two independent reasons. While this may on the one hand help to simplify the problem, it also gives us a much more useful handle on problems that involve changing geometries. As we are much more interested in Casimir forces (and moments) than just energies, this is obviously desirable.
In particular, we can, as in [@Gies:2006cq], typically perform the integration over the loop scaling factor $\int_{s=0}^\infty \frac{ds}{s^5} \Theta(sy,x)$ symbolically.
If we have sampled a loop $y$, we can compute for each sampling point the values of $s$ for which this sampling point is inside a given object. Often, this is just an interval, or at worst the union of a few intervals. By merging these intervals for each sampling point, we can compute the set of $s$ values for which the loop hits the given object. Now, as $\Theta$ counts the number of objects hit by the loop, it is piecewise constant on the partitioning so obtained; if $\Theta=n$ for $T\in[a,b]$, we have $\int_a^b \frac{ds}{s^5}\, \Theta(sy,x) = n(a^{-4}-b^{-4})/4$.
Note that this means that we also do not need to specify the region of $s$ which we want to sample, i.e. our method does not need to know a geometric length-scale.
Forces on Multiple Bodies via Sensitivity Backpropagation {#sec:autodiff}
---------------------------------------------------------
In order to calculate forces, we have to determine by how much Casimir energies change when changing the geometry. Taking the limit $\Lambda\to\infty$, the subtraction scheme eq.\[eq:subtraction\] is only compatible with geometry changes that change the position and orientation, but not shape, of individual bodies: evidently, the Casimir energy attributed to a single body via this scheme is zero, regardless of its shape. As we naturally would expect the Casimir force between two parallel flat plates to not disappear when connecting them with a thin wire (so that they become a single object), the regularization prescription that amounts to attributing forbidden loops to specific objects cannot be compatible with (unconstrained) shape changes.
Even if we limit ourselves to shifts and rotations of bodies, retaining eq. \[eq:subtraction\], it makes sense to describe geometry changes in a more general way. We hence consider the potential $V$ to be a function of multiple geometry parameters (positions and angles), e.g. $V=V(r^a_1,r^a_2,r^a_3,r^b_1,r^b_2,r^b_3,\alpha^a_1,\beta^a_2,\gamma^a_3;\ldots)$.
If we perform integration over the scaling parameter $s$ analytically for every loop and consider the intervals over which the subset of objects pierced by that loop does not change, then the interval endpoints, will become analytic functions of the geometry parameters – and so will the loop’s contribution to the total Casimir energy.
For any given loop, we have to perform a fairly simple computation, which in the end gives a single number. Furthermore, it is perfectly feasible to design the algorithm in such a way that at the end of the computation of the loop’s contribution to the Casimir energy, we still remember all the intermediate values that entered that calculation. In such a situation, there is a standard method (in the sense of an algorithmic transformation on the program that calculates the loop energy) that allows a fast evaluation of the gradient with respect to all the geometry parameters. Irrespective of the number $N$ of such parameters, and the complexity of the intermediate expressions, it is possible to obtain the gradient to full numerical accuracy with at most $5$ times the computational effort needed to calculate the scalar function (often even much less). In comparison, the naive direct method of calculating the gradient by comparing function values would require at least $N+1$ full evaluations of the function (and then only give a result with reduced numerical accuracy).
The generic approach that makes this possible, which has become known under the names “automatic/algorithmic differentiation”, “sensitivity backpropagation”, or “adjoint code” [@Speelpenning80; @Griewank89onautomatic] is essentially based on this idea:
- Every intermediate result used in the calculation gets stored away for later use (i.e. none may be dropped or over-written).
- To each such intermediate quantity $I_k$ stored, we associate a buffer that can store another number $\bar I_k$, initialized to zero at the beginning of the program. Ultimately, these will each end up holding the answer to the question: “If, at the point when $I_k$ became first known during the calculation of the function value, we interrupted the computation, changed that value from $I_k$ to $I_k'=I_k+\epsilon$, and then allowed the computation to proceed without further modifications now using $I_k'$ instead of $I_k$, by how much would the final result then change, relative to $\epsilon$ (in the limit of small $\epsilon$)?”. Hence, the $\bar I_k$ will eventually become *sensitivities* that describe the dependence of the result on the given intermediate quantity.
- Treating input parameters in the same way as intermediate values, the sensitivities on all the input values give the function’s gradient.
- Sensitivities are calculated in a two-step process: first, the function is evaluated once to obtain all the intermediate quantities for the given choice of input parameters. Then, starting from the result, which has been obtained from an arithmetic operation involving intermediate values, sensitivities for the last intermediate get updated. As these again have been the result of some arithmetic operation, the sensitivities for the intermediate values they have been obtained from can be determined, etc. As one intermediate quantity may be used multiple times throughout the calculation, it is important to collect incremental contributions to sensitivities when going backwards through the computation.
The theoretical maximum effort factor of $5$ can be traced back to the effort required to handle multiplication/division of intermediate values. Evidently, if the sensitivity of the result on the intermediate quantity $I_k$ is known in $\bar I_k$, and $I_k$ was obtained as the sum of $I_i+I_j$, then the sensitivities of $I_i$ and $I_i$ must be increased by $\bar I_k$ (for if e.g. $I_j$ gets used multiple times, then $\bar I_j$ receives multiple increments). If, however, $I_k$ is the *product* of $I_i$ and $I_j$, then $\bar
I_i$ must be increased by $I_j\cdot \bar I_k$ and vice versa – the combined read/add/store operations give rise to a bounded multiplicative factor for the total effort.
In practice, the corresponding calculations are even considerably simpler than what the general theory of algorithmic differentiation suggests, as the partial derivatives $\partial E/\partial g_j$ with respect to the geometry parameters $g_j$ are, at least for simple surfaces, obtainable directly as a by-product of the forward calculation. Comparing the Worldline method with the remarkable approach discovered by Rodriguez, Ibannescu, Iannuzzi, Capasso, Joannopoulos, and Johnson [@2007PhRvL..99h0401R; @2007PhRvA..76c2106R] or for that matter any method that involves numerically solving discretized sparse linear operator equation systems, one would naturally not expect sensitivity backpropagation to be as readily applicable with these approaches as with Worldline Numerics. On the one hand, sensitivity backpropagating an iterative linear solver would require remembering the intermediate values from all iteration steps (otherwise thrown away), and also, the calculation may have happened to produce a solution before having explored some of the dependencies sufficiently well. While research has been done on backpropagating linear solvers, the incorporation of this strategy into Worldline Numerics undeniably is much easier to accomplish.
Adaptive Sampling {#subsec:adaptive}
-----------------
In a typical geometry, essentially the whole energy or force is contributed by few, comparably small regions. These are typically the regions where two objects come closely together.
While we still have to sample loops in such a way that we integrate over all of the relevant region of space, it is worthwhile to focus effort mainly on these highly contributing areas, as the absolute uncertainty of our Monte-Carlo estimation is much higher there. We achieve this in the following way: We first specify an absolute accuracy to which we want the density estimated to at every point. When later sampling the density at a given point, we first take a specified minimum of samples. From that we estimate the (unbiased) variance of our sampling at this point. We continue sampling until a pre-defined (95%) confidence interval for the sampling mean is smaller than the prespecified accuracy.
Living with Cancellation {#subsubsec:cancellation}
------------------------
Some geometries, like the “cylinders with sidewalls” geometry studied in [@2008PhRvA..77c0101R], show a high degree of symmetry. While perfect symmetry helps to reduce the computational effort as the calculation can be restricted to a fundamental domain, slightly non-symmetric configurations often are a problem if we want to compute the force on an object that gets pulled in different (perhaps opposing) directions: most of the contributions cancel, giving rise to a small residual force.
A naive approach would compute the contribution at both sides of the object separately and then add up. This, however, would yield a huge variance for a comparably small resulting value. Fortunately, the force contribution of a loop and its mirrored image are highly correlated in these situations. Often, one is the negative value of the other. So we have a better way of estimating the contribution by estimating the expectation of the following process:
> Randomly pick a loop and also consider its mirror image under the symmetry; then add up the force contributions of both these loops.
In that way, we do not change the expectation value of the sum, but, due to the correlation, the variance is much smaller. In that way, it is possible to compute force contributions where a naive approach would require excessive effort due the huge variation, as in the system discussed in Section \[sec:cylinders\].
Parallelization
---------------
In the Worldline formalism the calculation of contributions to the total energy (or force) can be performed independently for each grid point. Also, for each grid point, the contribution of each loop to the energy (or force) does not depend on the contribution of the other loops. Thus the problem is easily seen as being embarrassingly parallel, and furthermore the basic component – processing a loop – does not require overy complex calculations (in the sense of memory requirements and algorithmic effort). As the worldline method is a probabilistic approach, the accuracy of the calculation can be increased (within reasonable limits dictated by computational effort) by increasing the number of grid points, number of loops, and the number of points per loop in an appropriate way.
As the computation for each point and loop follows the same algorithm, this approach fits the (single instruction, multiple data) processing approach very well. Powerful massively parallel hardware is now available at a highly competitive price in the form of specialized Graphics Processing Units (s), where development has – to a large extent – been driven by the video game industry.
When implementing Worldline Numerics on hardware, one has to bear in mind certain constraints of programming. The memory hierarchy of most s discerns between global memory and a special form of fast local memory called “shared memory”. It is often favourable to perform most of the computations in shared memory. This fast shared memory typically is quite small and has to be divided between simultaneously running work items, thus limiting the number of loop points that can be computed concurrently. One also has to keep in mind that s perform rather poorly on complex branching patterns and the hardware optimizations on s mainly target a high throughput of floating point operations, often neglecting the performance on integer calculations needed for most Pseudo Random Number Generators (s).
On account of shared memory size limitations, the authors developed a version of the d-loop algorithm that generates, for each loop, the loop points on the fly – without ever storing the entire loop in memory. As s do not support recursion directly, it is advantageous to instead manage a stack of arcs yet to be split in half, as sketched in figure \[fig:gpustack\].
while (stacksize>0) {
pop(StartPos, EndPos, &level);
if (level > 0){
MidPos = (StartPos + EndPos) / 2 + gauss_normal(0, sigma(level));
push(MidPos, EndPos, level-1);
push(StartPos, MidPos, level-1);
}
else calc_contribution(StartPos);
}
This manages to reduce the memory footprint of loop generation and processing from $\mathcal{O}(N)$ to $\mathcal{O}(\log(N))$, $N$ being the number of loop points. This makes it possible in principle to shift loop generation to cores even for loops too large to fit into the shared memory. In most cases, however, it seems more appropriate to instead pre-generate loop shapes on the and subsequently upload them into read-only memory visible to each work item. The threads then perform intersection checks for loops of pre-determined shape shifted to different grid points. Using the same set of loops at all grid points also helps in terms of statistics as there then can be direct cancellation between opposing forces arising from similar geometric structures. (See also the discussion in Section \[subsubsec:cancellation\].)
The authors so far used s mainly with the direct numerical integration method described in section \[subsec:directintegration\] and have at the time of this writing not yet generalized the (algorithmically much less uniform) handling of scale factor intervals to the for different geometries. First computations in the plate-plate geometry show a much better convergence than a full Monte-Carlo integration.
As language for the implementation of the algorithm the authors have used in combination with [@pyopencl]. This helps code re-use across a broad spectrum of multi- or manycore architectures.
Generalization to Electrodynamics?
----------------------------------
While Worldline Numerics has drawbacks in comparison to other computational approaches, such as e.g. its inability to easily incorporate frequency-dependent optical properties of immersion media, it also has, from an applications perspective, some potentially very attractive characteristics (as we have reasoned out in this article). The biggest present obstacle to the utilization of Worldline Numerics for engineering applications is that it has not yet been generalized from (massless and massive) scalar fields to photons interacting with conductors. As we demonstrate in section \[sec:cylinders\], trying to use scalars to approximate the behaviour of photons can easily give qualitatively wrong results – so, the need to be able to handle photons is quite pressing from an application perspective.
While worldline methods can in principle be adopted to dealing with vector bosons, as has been explored e.g. in [@2001PhR...355...73S] for gluon loop radiative corrections, the challenge is in properly modeling the boundary conditions for photon-conductor interaction. Considering the structure of the theory, one would at least expect that it should be achievable to couple the photon to a charged scalar undergoing spontaneous symmetry breaking, giving it an effective mass inside each object. This would then amount to modeling electromagnetic Casimir forces in the presence of superconductors.
Rather than trying to construct a generalization of Worldline Numerics for photons and conductors starting from quantum electrodynamics principles, we for now approach this problem in a more adventurous way, if only to generate ideas: here, our guiding question is what the simplest conceivable generalization of Worldline Numerics might be that could possibly stand a chance of modeling electrodynamics. Naturally, the subsequent discussion in this section will be of highly speculative nature.
Obviously, such a generalization will involve having photons propagate in loops. As we are not particularly interested in manifest Lorentz invariance here, we may just as well straightaway choose to gauge away the non-dynamic timelike component of the electromagnetic vector potential $A_0$, leaving us with a three-dimensional vector describing the photon polarization state. This is, of course, still redundant, as the photon only has two (transversal) physical states rather than three, but it might be conceivable to ultimately end up with a formulation in which, after evaluating the integrals over loop shapes and sizes, it turns out that the boundaries do not interact with (i.e. never see) the longitudinal photons. In such a case, they would drop out from all Casimir forces.
It makes sense to assume that, as in the scalar case, bodies can be modeled as hollow thin shells; if we accept the perfect conductor approximation (which in particular states that characteristic length scales from the geometry are large in comparison to characteristic atomic distances), we would not expect to be able to probe the inner structure of conductors by looking at Casimir forces between them. So, we need to only concern ourselves with what happens when a photon loop pierces a conductor surface. The boundary conditions of a perfect conductor ($\vec n\times\vec E=0$ – no electric field parallel to surface, as this would immediately be compensated by charges shifting accordingly, and $\vec n \cdot\vec B=0$ – no magnetic field perpendicular to surface, as this would be compensated by eddy currents, at least for frequencies $\omega$ sufficiently large for the perfect conductor approximation to hold) have to be implemented in some way. We want to consider all possible (including un-physical longitudinal) photon polarizations simultaneously, and hence should associate to every loop edge a $3\times 3$ matrix acting on the polarization state. In the end, the contribution to the Casimir energy from the loop under consideration should be taken to be the trace of the product of all these matrices; free propagation will amount to the identity, and the matrix corresponding to an edge that pierces a conductor surface would involve a projection eliminating some polarizations. What form may such a projection matrix have? As we already accounted for photon polarization directions, the only directions available are the surface normal $\vec n$, as well as the local velocity $\dot{\vec{ y}}$. The $\vec E=\partial_t A\sim\omega \vec A$ condition only involves perpendicularity to $\vec n$, and while the $\vec B$-condition would be expected to pick up spatial components of $\dot{\vec y}$, re-scaling these in order to form a projector seems to also bring us to a $\vec A\perp \vec n$ condition, which means that one should take as projector that restricts to forbidden states the matrix $I-\vec n\otimes\vec n$. Evidently, in the case of parallel plates, the trace would just introduce a factor two relative to the scalar field, as desired.
![Casimir forces in the piston geometry as a function of barrier height, calculated for scalars and photons using the analytic expressions given in [@2007PhRvD..76d5016H].[]{data-label="fig:exact"}](exact.pdf){width="80.00000%"}
![Comparison of predicted forces in the Casimir Piston geometry: The graph shows the ratio of the Casimir forces computed with our method to the exact, analytically computed, values [@2007PhRvD..76d5016H], see Figure \[fig:exact\]. This ratio is plotted for computations based on scalar fields (upper curve) and photons (lower curve). A constant ratio would indicate perfect coincidence of the methods; note that constant scaling factors (e.g., due to fundamental constants of physics) have not been taken into account in our computations, as they were only carried out to test the functional dependency.[]{data-label="fig:EM-Casimir"}](em-casimir-graph.pdf){width="80.00000%"}
The prescription described above seems, at the superficial level, quite obviously wrong – this is most easily seen by considering a loop close to a box corner made up of different objects: if it pierces three mutually perpendicular faces, it would contribute three forbidden polarizations, where the physical photon only has two. Still, it is valid to ask whether this loop processing prescription is related to some idealization of boundaries in a quantum field theory of vectors – and ask what it might look like.
One would, for the reason given above, naturally expect such a prescription to give energies that do not have much in common with photon Casimir energies. An attractive nontrivial test geometry for which scalar and photon Casimir forces are known exactly, where discrepancies could reasonably be expected to manifest, and in which this scheme is easily implemented, is the piston geometry studied in [@2007PhRvD..76d5016H]. If we plot the quotient of the (not yet properly normalized) energies predicted by Worldline Numerics and the analytic result for the scalar field, we get curve 1 in figure \[fig:EM-Casimir\]. If we compute the same ratio for our aventurous “Vector Worldline” generalization and the known result for photons, we get curve 2.
We consider a coordinate aligned closed box with square cross section of side length $b=c=1$ and height height $h\gg b$ – here, we choose $h=100$. This box is sub-divided by a coordinate-aligned box-shaped movable barrier of identical cross section and infinitely small thickness; the distance between barrier and base of the box is taken to lie in the interval $[0;1]$.
The spatial sampling of loop positions was done on a Cartesian grid with lattice spacing $0.05$, covering a coordinate-parallel cube of edge length $3$ centered at the point 0.5 above the center of the box base plate. Forces have been calculated for barrier heights from $0.3$ to $0.8$ in steps of $0.1$, using a central difference quotient with $\delta h=0.1$.
Since the barrier is taken as being as wide as the box, there is no positive minimum to the contributing loop size. However, for loops much smaller than the distance between barrier and base of the box the local geometry on either side of the barrier symmetric. So the contribution of sufficiently small loops cancels out. In order to avoid infinities, the loop scaling factor $s$ was constrained to the range $[0.02;6]$ and sampled as described in Section \[subsec:directintegration\].
In this system, one must keep in mind that, for the distances involved, Casimir energies, both in the scalar and photon case, vary by more than three orders of magnitude – as shown in Figure \[fig:exact\]. While the ratio clearly shows some drift for the photon case in this experimental calculation, it cannot yet be excluded that this may be a discretization related artefact. Given that the simplistic ansatz proposed here would be expected to quite clearly show up as being deeply wrong (considering the ratios involved), this outcome is somewhat remarkable. At the very least, it seems to make sense to try and explore a small number of other geometries to learn whether this is a curious coincidence for this particular geometry (as one would expect), whether some not yet understood crazy cancellations may actually make this method work as a model for photons (unlikely), or whether it may turn out as being somewhat useful pragmatically in the sense of a heuristic engineering method that is known to be mathematically flawed (as many such engineering heuristics are), but manages to give a reasonably good estimate for some applications.
Parallel Cylinders between Plates, revisited {#sec:cylinders}
============================================
![The “cylinders with sidewalls” geometry and the dependency of the attractive scalar Casimir force between the cylinders on the Position of the plates. At position 1.0, the plates would touch the cylinders.[]{data-label="fig:cylindersplates"}](cylinders.pdf "fig:"){width="30.00000%"} ![The “cylinders with sidewalls” geometry and the dependency of the attractive scalar Casimir force between the cylinders on the Position of the plates. At position 1.0, the plates would touch the cylinders.[]{data-label="fig:cylindersplates"}](cylinders-graph.pdf "fig:"){width="60.00000%"}
The “cylinders with sidewalls” geometry studied in [@2008PhRvA..77c0101R] (Figure \[fig:cylindersplates\], on the left) has been shown to nicely demonstrate that Casimir forces are essentially multi-body forces. It is given by two parallel, infinitely long cylinders between two parallel infinite plates. Focusing on the attractive force between cylinders, one finds that this depends in a fairly subtle way on the distance between the plates.
For the calculation whose results are shown in figure \[fig:cylindersplates\], the cylinders used have radius 1.0 and centres at ($x$, $y$)-coordinates (-2.0, 0.0) and (2.0, 0.0), respectively. The plates are given by the equation $z=p$, where $p$ is varied in the range 1.02…2.50. The calculation of the force on the left cylinder used the scaling method, sampling around a grid with spacing 0.05 and exploiting the mirror method (Section \[subsubsec:cancellation\]) by taking a reflection-symmetric loop ensemble w.r.t. the plane $x$=-2.0 to reduce the variance due to cancellation. This sampling was done in an adaptive manner (Section \[subsec:adaptive\]). In total, between $1.9\cdot 10^7$ and $3.9\cdot 10^7$ loops of $2^{13}$ points were sampled for each geometry.
The results are shown at the right-hand side of Figure \[fig:cylindersplates\]. As opposed to methods such as the proximity force approximation, we do see a dependency of the forces on the cylinders on the plate distance. We however could not find the non-monotonic behaviour reported in the literature [@2008PhRvA..77c0101R]. So, once again, we have an example where scalars behave in a qualitatively different way than photons [@2007PhRvD..76d5016H].
Conclusions
===========
From a microsystems engineering perspective, the Worldline Numerics / Loop Cloud Method has a number of attractive properties, such as the ability to quickly give crude estimates, considerable potential to solve geometric optimization related problems, and of course its conceptual simplicity and intuitiveness that make it a useful educational tool with the potential to give a simple yet quantitatively correct mental model of the origin of Casimir forces. Quite remarkably, the operational procedure can be explained using very simple concepts only – in fact, even without having to use much linear algebra.
At present, the biggest obstacle to its utilization for engineering applications is the method’s inability to handle photons in the presence of conducting boundaries instead of scalar particles in the presence of Dirichlet boundary conditions. As we have demonstrated in section \[sec:cylinders\] through an example calculation, the problem is that any attempt to use scalars in order to approximate photon Casimir forces is questionable as this can easily give predictions that are wrong already at the qualitative level.
In addition to this calculation, and to discussing some methodological improvements of the Loop Cloud Method, this article provided some early speculation on whether a simple extension of the method may be able to give useful estimates for photon Casimir forces; despite obvious conceptual issues, the accompanying computation turned out to produce numbers that match the known exact result remarkably well, in particular considering that Casimir forces range over a few orders of magnitude for the problem studied. Still, one naturally would next try to refute the viability of this simplistic approach by checking its predictions for another not too trivial geometry, before starting any attempt to formally prove its validity.
#### Acknowledgments
It is a pleasure to thank Holger Gies and Roman Zwicky for useful discussions and comments on this subject. calculations have been performed on the University of Southampton’s Iridis3 cluster and the cluster of the Goethe Universität Frankfurt. calculations have been performed on the cluster of the Goethe Universität Frankfurt. J. Gerhard received funding for this project by the HGS-HIRe Abroad programm of the *Helmholtz Graduate School for Hadron and Ion Research*. Financial support by the Engineering and Physical Sciences Research Council under Grant is gratefully acknowledged.
[10]{}
Hendrik Brugt Gerhard Casimir. On the attraction between two perfectly conducting plates. , 51(7):793–795, 1948.
U. Mohideen and Anushree Roy. Precision measurement of the casimir force from 0.1 to $0.9\mathit{\mu}m$. , 81:4549–4552, Nov 1998.
A. Scardicchio and R. L. Jaffe. . , B704:552–582, 2005, quant-ph/0406041.
L. [Moyaerts]{}, K. [Langfeld]{}, and H. [Gies]{}. . , November 2003, arXiv:hep-th/0311168.
Sahand Jamal Rahi, Thorsten Emig, and Robert L. Jaffe. . 2010, 1007.4355.
Alejandro Rodriguez, Mihai Ibanescu, Davide Iannuzzi, J. D. Joannopoulos, and Steven G. Johnson. Virtual photons in imaginary time: Computing exact casimir forces via standard numerical electromagnetism techniques. , 76:032106, Sep 2007.
H. [Gies]{} and K. [Langfeld]{}. . , 17:966–976, 2002, arXiv:hep-ph/0112198.
H. [Gies]{} and K. [Klingm[ü]{}ller]{}. . , 39:6415–6421, May 2006, arXiv:hep-th/0511092.
A. W. [Rodriguez]{}, A. P. [McCauley]{}, J. D. [Joannopoulos]{}, and S. G. [Johnson]{}. . , April 2009, 0904.0267.
A. P. [McCauley]{}, A. W. [Rodriguez]{}, J. D. [Joannopoulos]{}, and S. G. [Johnson]{}. . , 81(1):012119–+, January 2010, 0906.5170.
T. Nagami, H. Mizuta, N. Momo, Y. Tsuchiya, S. Saito, T. Arai, T. Shimada, and S. Oda. Three-dimensional numerical analysis of switching properties of high-speed and nonvolatile nanoelectromechanical memory. , 54(5):1132 –1139, may 2007.
H. [Gies]{}, K. [Langfeld]{}, and L. [Moyaerts]{}. . , 6:18–+, June 2003, arXiv:hep-th/0303264.
Joao Luiz Dibl Comba and Jorge Stolfi. Affine arithmetic and its applications to computer graphics. , 1993.
A. Knoll, Y. Hijazi, A. Knesler, M. Schott, Hansen C., and Hagen H. Fast ray tracing of arbitrary implicit surfaces with interval and affine arithmetic. , 28(1):26–40, 2009.
H. Gies, J. Sánchez-Guillén, and R. A. Vázquez. Quantum effective actions from nonperturbative worldline dynamics. , 0508, 2005, hep-th/0505275.
Holger Gies and Klaus Klingmuller. . , D74:045002, 2006, quant-ph/0605141.
B. Speelpenning. Compiling fast partial derivatives of functions given by algorithms. Technical Report COO-2383-0063; UILU-ENG-80-1702; UIUCDCS-R-80-1002, Illinois Univ., Urbana (USA). Dept. of Computer Science, 1980.
Andreas Griewank. On automatic differentiation. In [*IN MATHEMATICAL PROGRAMMING: RECENT DEVELOPMENTS AND APPLICATIONS*]{}, pages 83–108. Kluwer Academic Publishers, 1989.
A. [Rodriguez]{}, M. [Ibanescu]{}, D. [Iannuzzi]{}, F. [Capasso]{}, J. D. [Joannopoulos]{}, and S. G. [Johnson]{}. . , 99(8):080401–+, August 2007, 0704.1890.
A. [Rodriguez]{}, M. [Ibanescu]{}, D. [Iannuzzi]{}, J. D. [Joannopoulos]{}, and S. G. [Johnson]{}. . , 76(3):032106–+, September 2007, 0705.3661.
S. J. [Rahi]{}, A. W. [Rodriguez]{}, T. [Emig]{}, R. L. [Jaffe]{}, S. G. [Johnson]{}, and M. [Kardar]{}. . , 77(3):030101–+, March 2008, 0711.1987.
Andreas [Kl[ö]{}ckner]{}, Nicolas [Pinto]{}, Yunsup [Lee]{}, B. [Catanzaro]{}, Paul [Ivanov]{}, and Ahmed [Fasih]{}. . Technical Report 2009-40, Scientific Computing Group, Brown University, Providence, RI, USA, November 2009.
C. [Schubert]{}. . , 355:73–234, December 2001, arXiv:hep-th/0101036.
M. P. [Hertzberg]{}, R. L. [Jaffe]{}, M. [Kardar]{}, and A. [Scardicchio]{}. . 76(4):045016–+, August 2007, 0705.0139.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A core-collapse supernova will produce an enormous burst of neutrinos of all flavors in the few-tens-of-MeV range. Measurement of the flavor, time and energy structure of a nearby core-collapse neutrino burst will yield answers to many physics and astrophysics questions. The neutrinos left over from past cosmic supernovae are also observable, and their detection will improve knowledge of core collapse rates and average neutrino emission. This review describes experimental techniques for detection of core-collapse neutrinos, as well as the sensitivities of current and future detectors.'
author:
- |
Kate Scholberg\
Department of Physics, Duke University, Durham NC 27708, USA;\
email: [email protected]
bibliography:
- 'refs.bib'
title: Supernova Neutrino Detection
---
epsf.def psfig.sty
INTRODUCTION
============
When a massive star has exhausted its nuclear fuel, it collapses to form a compact object such as a neutron star or a black hole. A prominent feature of the collapse is that $\sim$99% of the gravitational binding energy of the resulting remnant is converted to neutrinos with energies of a few tens of MeV over a timescale of a few tens of seconds. This highly efficient energy loss via neutrinos occurs because the neutrinos interact only via the weak interaction and can escape easily, whereas photons are trapped.
Neutrinos were observed for the celebrated 1987A core-collapse supernova (SN1987A) in the Large Magellanic Cloud (LMC), 50 kpc away from Earth. Two water Cherenkov detectors, Kamiokande-II [@Hirata:1987hu] and the Irvine-Michigan-Brookhaven (IMB) experiment [@Bionta:1987qt], observed 19 neutrino interaction events between them over a 13-s interval at a time consistent with the estimated time of core collapse. Two scintillator detectors, Baksan [@Alekseev:1987ej] and LSD [@Aglietta:1987it], also reported observations; the latter report was controversial because the events were recorded several hours early. The SN1987A events were certainly nearly all of $\bar{\nu}_e$ flavor. Although these events were a meager sample, the SN1987A neutrino events were sufficient to confirm the baseline model of core collapse. Beyond that, they have provided a very wide range of constraints on astrophysics and physics (e.g., References [@Schramm:1990pf; @Koshiba:1992yb; @Raffelt:1999tx]), resulting in the publication of hundreds of papers, which continues to this day.
Worldwide capabilities for supernova neutrino detection have increased by orders of magnitude since 1987. The next observation of a nearby core-collapse supernova will provide a great deal of information for both physics and astrophysics. The rate of core-collapse supernovae is estimated to be a few per century [@Tammann:1994ev; @Cappellaro:2000ez] in a galaxy such as the Milky Way, so the chance of observing one in the next few decades is not negligible. The most likely distance of the next core-collapse supernova from Earth is between 12 abd 15 kpc, according to the distribution of possible supernova progenitors in the Milky Way [@Mirizzi:2006xx].
This review concentrates on experimental aspects of the detection of neutrinos from core-collapse supernovae. The physics of neutrino interactions in the tens-of-MeV regime imposes limitations on the quality and type of information that can be obtained from the neutrinos.
Section \[signal\] briefly describes the main features of the signal. Section \[interactions\] describes the neutrino interactions that are relevant for current detectors. Section \[detectors\] describes generic detection issues and backgrounds and covers different detector technology types as well as specific detector examples. Section \[pointing\] addresses techniques for pointing to the supernova. Sections \[extragalactic\] and \[diffuse\] consider detection of neutrinos originating from beyond the Galaxy. The final section summarizes future prospects.
THE SUPERNOVA NEUTRINO SIGNAL {#signal}
=============================
Despite enormous recent progress, much about the physics of core collapse is not well understood. The neutrino messengers from deep inside the supernova will help us understand many aspects of the supernova mechanism and associated phenomena. The neutrinos are probably intimately involved with the explosion mechanism; imprinted on the flux will be signatures of shock waves, accretion, cooling, possible formation of exotic matter, and further collapse to a black hole, and an improved understanding of supernova nucleosynthesis will result from a detection. A detailed discussion of the physics of core collapse and associated neutrino signatures is beyond the scope of this review; some example reviews can be found in References [@Mezzacappa:2005ju; @Janka:2006fh; @Raffelt:2007nv; @Dighe:2008dq].
Some general features of the neutrino signal are described briefly as follows. At the beginning of collapse, one expects a short (tens-of-milliseconds), bright neutronization or breakout burst dominated by $\nu_e$ from electron capture: $p+e^-\rightarrow n + \nu_e$. This burst is followed by an accretion phase, tens to hundreds of milliseconds long, over which electron flavors dominate; at this stage, there may be a complex energy and time structure that represents effects of the standing accretion shock instability phenomenon. The next phase is cooling, which lasts a few tens of seconds, during which the core sheds most of its gravitational binding energy. During this phase, $\nu\bar{\nu}$ pairs dominate neutrino production, and luminosities and temperatures gradually decrease. An overall feature of the neutrino flux is that luminosity is roughly equally divided among flavors. Furthermore, the flavors have an expected energy hierarchy (where the robustness varies by model) according to $\langle E_{\nu_e}\rangle < \langle E_{\bar{\nu}_e}\rangle < \langle E_{\nu_x}\rangle$, where $\nu_x$ represents $\nu_{\mu},\nu_{\tau},\bar{\nu}_\mu$, or $\bar{\nu}_\tau$ (the $\mu$ and $\tau$ flavors have identical production at the energies involved and can be considered collectively). These energies are in decreasing order of strength of interaction with matter: $\nu_e$ have more interactions than $\bar{\nu}_e$, because of the excess of neutrons in the core; in turn, $\bar{\nu}_e$ have more interactions than $\nu_x$, which are restricted to neutral currents (NCs). The weaker the interactions are, the denser and deeper the decoupling of the neutrinos from the star and, therefore, the hotter the temperature of the neutrinos at the surface of last scattering (the neutrinosphere). Some examples of neutrino flux predictions in the literature (given in approximate order of modernity) can be found in References [@Burrows:1991kf; @Totani:1997vj; @Raffelt:2003en; @Fischer:2008rh; @Gava:2009pj; @Huedepohl:2009wh].
Figure \[figure1\] shows an example of a flux prediction. Modeling has steadily improved over the past few decades, with inclusion of more and more effects as techniques become more sophisticated and as available computing power increases. Some of the more recent models tend to produce somewhat cooler spectra and therefore fewer interaction events than older ones. There may be significant variations in the expected flux from supernova to supernova due to differences in the mass and composition of the progenitor, and possibly asymmetries, rotational effects, or magnetic field effects.
Fluxes are expected to have the approximate spectral form parameterized by (e.g., References [@Huedepohl:2009wh] and [@Vaananen:2011bf]) $$\label{eq:PLdistr}
\phi(E_{\nu}) =N
\left(\frac{E_{\nu}}{\langle E_{\nu} \rangle}\right)^{\alpha} \exp\left[-\left(\alpha + 1\right)\frac{E_{\nu}}{\langle E_{\nu} \rangle}\right] \ ,$$ where $\alpha$ is often referred to as the “pinching parameter”, because it controls the high-energy tail of the distribution; $E_\nu$ is the neutrino energy; $\langle E_\nu \rangle$ is the mean neutrino energy; and $N = \frac{(\alpha+1)^{\alpha+1}}{\langle E_{\nu}\rangle\Gamma(\alpha+1)}$is the normalization constant, where $\Gamma$ is the Gamma function. The different $\nu_e$, $\bar{\nu}_e$ and $\nu_x$ flavor components are generally expected to have different $\langle E_\nu \rangle$ and $\alpha$ parameters. The spectra are expected to evolve in time (e.g., Reference [@Huedepohl:2009wh]).
An enormous pulse of neutrinos from a core collapse will help us understand the physics of neutrinos as well as the astrophysics of core collapse. Interesting signatures will manifest themselves as modulations of the observed spectra, potentially with significant time dependence. We now know that neutrinos have mass and that the three flavors of neutrinos mix with one another; the supernova neutrino fluxes will also depend on neutrino oscillation parameters. Matter oscillation (Mikheyev-Smirnov-Wolfenstein) effects modify the spectra as the neutrinos traverse dense matter, so the signals will depend both on neutrino oscillation parameters and on supernova matter profiles. In fact, conditions in the supernova will be so extreme that neutrino-neutrino interactions will become important, which will lead to a complex phenomenology of collective effects. These effects include spectral swap or spectral split features in the spectra, for which flavors are effectively swapped above or below a particular energy threshold. The recent literature exploring supernova collective oscillation physics is extensive: A non-comprehensive sample can be found in References [@Duan:2005cp; @Fogli:2007bk; @Raffelt:2007cb; @Raffelt:2007xt; @EstebanPretel:2008ni; @Duan:2009cd; @Dasgupta:2009mg; @Duan:2010bg; @Duan:2010bf].
The nature of the neutrino spectra and their time evolution depend on mass and oscillation parameters, such as $\theta_{13}$ and the mass hierarchy. For the normal hierarchy, $m_3 >> m_2,m_1$, there are two light and one heavy neutrino mass states, and for the inverted hierarchy, $m_1,m_2 >> m_3$, there are two heavy and one light neutrino mass states. In particular, the spectra of $\nu_e$ and $\bar{\nu}_e$ will differ according to the mass hierarchy, so supernova neutrino measurements could tell us what the hierarchy is (e.g., References [@Dighe:1999bi; @Dighe:2003be; @Lunardini:2003eh; @Dighe:2008dq; @Chakraborty:2011ir; @Choubey:2010up]). Note that although reactor and beam experiments have now measured the value of $\theta_{13}$, the mass hierarchy may not be determined by laboratory experiments for a decade or more [@Akiri:2011dv]. Furthermore, the chance that the supernova neutrinos will traverse Earth matter on their way to a detector is not negligible [@Mirizzi:2006xx] and oscillations in the Earth modulate the supernova neutrino spectrum for either $\nu_e$ or $\bar{\nu}_e$ [@Lunardini:2001pb; @Takahashi:2001dc; @Takahashi:2002cm] depending on the hierarchy. Therefore, information on oscillation parameters can be determined in a supernova-model-independent way if measurements at different locations on the Earth can be compared. Even in a single detector, an Earth-matter-induced spectral modulation may give information about oscillations [@Dighe:2003be; @Dighe:2003jg; @Dighe:2003vm]. Furthermore the flux may carry signatures of oscillations involving sterile neutrino states (e.g., [@Fetter:2002xx]).
Although they probably will not be competitive with near-future laboratory measurements (or cosmological measurements), constraints on neutrino absolute mass may also become possible via measurements of an energy-dependent time delay, especially if some sharp feature in the spectrum is available to set a time reference (e.g., References [@Beacom:1998ya; @Beacom:2000qy]). Other neutrino properties, such as lifetime, magnetic moments, nonstandard interactions, and the speed from time of flight over a known distance, can also be constrained [@Schramm:1990pf; @Raffelt:1999tx]. The physics to be learned is not confined to the neutrino sector. There will be opportunities to shed light on other particle physics as well [@Raffelt:1999tx], given that the neutrino signal will provide a measure of the total energy of the collapse and constrains energy loss via new particles and exotic mechanisms, including axions [@Raffelt:1999tx] and extra dimensions [@Hannestad:2001jv].
A generic potential difficulty is that both core-collapse physics and neutrino physics affect the nature of the neutrino burst, and it may not be trivial to disentangle the two. Relatively robust and model-independent signatures do exist, however. Clearly, the more experimental data we can gather about the flavor, energy and time structure of the burst, in as many detectors around the world as possible, the better our chances will be of disentangling the various effects.
NEUTRINO INTERACTIONS IN THE TENS-OF-MEV RANGE {#interactions}
==============================================
Neutrinos are detected via electromagnetically or strongly interacting products of weak charged-current (CC) and NC interactions with electrons and nuclei. This section describes what is known about neutrino interactions in the tens-of-MeV range relevant for current detectors, along with their observables. Unfortunately, relatively few interactions have precisely known cross sections. Except for elastic scattering and inverse $\beta$ decay (IBD) interactions, both theoretical and experimental knowledge is limited.
Inverse Beta Decay {#ibd}
------------------
Relatively cheap detector materials such as water and hydrocarbon-based scintillator have many free protons. The most significant interaction in these materials is IBD, which is the interaction between $\bar{\nu}_e$ and free protons, $\bar{\nu}_e + p \rightarrow n + e^+$. The IBD kinematic threshold is $E_{\nu_{\rm thr}}=1.8$ MeV. The positron’s energy loss can typically be observed. In the supernova energy regime, to a good approximation $E_{e^+}=E_\nu-1.3$ MeV. In scintillation detectors, the 0.511-MeV positron annihilation $\gamma$s may also be observed. The neutron may be captured on free protons, with an approximately 200-$\mu$s thermalization and capture time, producing a deuteron and a 2.2 MeV $\gamma$. The neutron may also be captured on another nucleus; in particular, the detector may be doped with some material with a high neutron capture cross section, such as gadolinium (Gd). Gd of natural isotopic composition has average neutron capture cross section some $1.6\times 10^5$ times that of free protons, and a thermalization and capture time of a few tens of microseconds when dissolved in water or scintillator. Neutron capture on a Gd nucleus is followed by a deexcitation cascade of $\gamma$s summing to approximately 8 MeV of energy. In scintillator, most of the energy is visible via Compton scattering of the $\gamma$s. Reference [@Beacom:2003nk] proposes doping of water with a Gd compound, for which approximately $4$ MeV of energy is visible per neutron capture [@Watanabe:2008ru].
The calculated IBD cross section can be found in [@Strumia:2003zx]. The interaction has only a slight energy-dependent anisotropy [@Vogel:1999zy], so in general it is not very useful for pointing to a supernova.
Elastic Scattering on Electrons {#es}
-------------------------------
Although the cross section is relatively small compared with other interactions, neutrino-electron elastic scattering, $\nu_x+e^-\rightarrow \nu_x+e^-$, is important because of its directionality. The electron is scattered in the direction of the incoming neutrino; thus, for detectors that can reconstruct electron tracks, such as water Cherenkov and liquid argon time-projection chambers (TPCs), the elastic scattering sample can be selected against background. Furthermore, elastic scattering events can be used to point to the supernova (see Section \[pointing\]). Elastic scattering proceeds via both CC and NC interactions for $\nu_e$ and $\bar{\nu}_e$ (the latter are helicity suppressed), and via NC interaction for $\nu_x$. The cross sections, and the angular and recoil energy distributions of the scattered electrons, are based on very well understood weak interaction physics [@Marciano:2003eq].
Charged- and Neutral-Current Interactions with Nuclei
-----------------------------------------------------
Neutrinos also interact with nucleons in nuclei via CC and NC interactions, although cross sections are typically somewhat smaller for bound than for free nucleons. CC interactions proceed via interaction of $\nu_e$ and $\bar{\nu}_e$ with neutrons and protons, respectively, in nuclei, $\nu_e+(N,Z)\rightarrow (N-1,Z+1) +e^-$, and $\bar{\nu}_e+(N,Z)\rightarrow (N+1,Z-1) +e^+$. The kinematic threshold is $E_{\nu_{\rm thr}}=\frac{M_f^2+m_e^2+2 M_f m_e - M_i^2}{2M_i}\sim M_f-M_i+m_e$, where $M_f$ and $M_i$ are the initial and final state nuclear masses and $m_e$ is the electron mass. Note that at supernova energies, $\nu_\mu$ and $\nu_\tau$ are below the CC interaction threshold and thus are kinematically unable to produce their partner leptons. In nuclei, the $\bar{\nu}_e$ interaction is typically suppressed at a given energy with respect to the $\nu_e$ interaction due to Pauli blocking. Typically the energy loss of the charged lepton is observable. In the $\bar{\nu}_e$ case, the $\gamma$s produced by annihilation of the positron may be observable. Ejecta (nucleons and $\gamma$s) produced by the final nucleus as it deexcites may also be observable and may help to tag the interaction. The neutrino interactions may furthermore leave radioactive nuclear products, for which decays correlated in space and time with the primary lepton could be observed [@Kolbe:2002gk].
NC interactions on nucleons in nuclei may also produce observable signals via ejected nucleons or deexcitation $\gamma$s. Of particular note are the 15.1-MeV $\gamma$ from $^{12}$C deexcitation and the ejection of neutrons from lead (see Sections \[scint\] and \[lead\]).
In CC cases, produced leptons retain memory of the incoming neutrino energy, as the heavy recoil nucleus tends to take away little energy. In both CC and NC cases, specific interactions have different thresholds and energy-dependent cross sections, so even if there is no product to measure that remembers the neutrino energy on an event-by-event basis, there is neutrino spectral information to be had by measuring statistical distributions of products.
Neutrino interactions on nuclei in the tens-of-MeV range are relatively poorly understood theoretically, in terms of both the interaction rate and the angular and energy distributions of the resulting interaction products. These interactions and their products tend to be sensitive to the details of the nuclear physics involved. Nevertheless, several calculations exist. Cross sections on carbon are considered in References [@Kolbe:1995af; @Kolbe:1999au; @Volpe:2000zn]. Cross sections and signatures for oxygen that are relevant for water detectors are considered in References [@Haxton:1987kc; @Haxton:1988mw; @Haxton:1990ks; @Langanke:1995he; @Kolbe:2002gk]. Calculations relevant for the neutrino response of argon detectors can be found in References [@Raghavan:1986hv; @Ormand:1994js; @Bhattacharya:1998hc; @GilBotella:2003sz; @Bueno:2003ei; @GilBotella:2004bv; @SajjadAthar:2004yf], and studies relevant for lead and iron appear in References [@Fuller:1998kb; @Kolbe:2000np; @Volpe:2001gy; @Toivanen:2001re; @Engel:2002hg; @Samana:2008pt].
Uncertainties in theoretical calculations are no better than the $\sim$10-20% level, and unfortunately experimental measurements of neutrino interactions in this energy range are poor or nonexistent. The only measurements with $\sim$10-20% uncertainties are for $^{12}$C by the Liquid Scintillator Neutrino Detector (LSND) at Los Alamos National Laboratory and KARMEN at Rutherford Appleton Laboratory [@Armbruster:1998gk; @Auerbach:2001hz]. Future programs to measure cross sections will be highly desirable for an effective interpretation of the next observed supernova burst signal, as well as for the understanding of processes in the supernova itself. A stopped-pion source (as used for the LSND and KARMEN measurements), such as the Spallation Neutron Source [@Efremenko:2005nf; @SajjadAthar:2005ke] or a cyclotron-based source [@Alonso:2010fs] would be nearly ideal: The well-understood stopped-pion spectra of $\nu_\mu$, $\bar{\nu}_\mu$ and $\nu_e$, with neutrino energies around $\sim$20-50 MeV, are a rather close match to the expected supernova spectrum. Low-energy $\beta$ beams, neutrinos originating from the decay of radioactive ion beams, with tunable spectra, are also promising sources of well-understood flux for measurements of supernova-relevant cross sections [@Serreau:2004kx; @Jachowicz:2008kx].
Coherent Elastic Neutrino-Nucleus Scattering
--------------------------------------------
Neutrinos at MeV energies will also coherently scatter off protons or entire nuclei via NC weak interactions. The interaction rate is relatively high; however recoil energies tend to be rather low, so detection thresholds need to be very low. For coherent scattering on protons, recoil energies are less than a few MeV but within the reach of low-background large scintillation detectors [@Beacom:2002hs]. When entire nuclei are the target, recoil energies tend to be in the few-keV to few-tens-of-keV range [@Freedman:1977xn; @Drukier:1983gj], out of reach of conventional detectors. However, detectors designed to search for weakly interacting massive particle (WIMP) dark matter recoils will be sensitive [@Horowitz:2003cz].
Summary of Interactions
-----------------------
Table \[tab:interactions\] summarizes interactions on different targets relevant for existing detector types and lists the primary detection modes. Estimates of the number of interactions per kt (before considering the detector response) are given in the table for two different models from the literature, the Livermore model [@Totani:1997vj] and the GKVM (Gava-Kneller-Volpe-McLaughlin) model [@Gava:2009pj]. Note that there can be considerable variation in total neutrino interaction yield from model to model, depending on temperature and pinching, because of the energy dependence of the cross sections. Figure \[figure2\] shows cross sections for relevant processes as a function of neutrino energy. See `http://www.phy.duke.edu/\simschol/snowglobes` for complete references and assumptions.
Channel Observable(s) Interactions
----------------------------------------------------------------------- --------------- --------------
$\nu_x + e^- \rightarrow \nu_x + e^-$ C 17/10
$\bar{\nu}_e+ p \rightarrow e^+ + n$ C, N, A 278/165
$\nu_x+ p \rightarrow \nu_x + p$ C 682/351
$\nu_e + {}^{12}{\rm C} \rightarrow e^- + {}^{12}{\rm N^{(*)}}$ C, N, G 3/9
$\bar{\nu}_e + {}^{12}{\rm C} \rightarrow e^+ + {}^{12}{\rm B^{(*)}}$ C, N, G, A 6/8
$\nu_x + {}^{12}{\rm C} \rightarrow \nu_x+ {}^{12}{\rm C}^*$ G 68/25
$\nu_e + {}^{16}{\rm O} \rightarrow e^- + {}^{16}{\rm F^{(*)}}$ C, N, G 1/4
$\bar{\nu}_e + {}^{16}{\rm O} \rightarrow e^+ + {}^{16}{\rm N^{(*)}}$ C, N, G 7/5
$\nu_x + {}^{16}{\rm O} \rightarrow \nu_x+ {}^{16}{\rm O}^*$ G 50/12
$\nu_e + {}^{40}{\rm Ar} \rightarrow e^- + {}^{40}{\rm K}^*$ C, G 67/83
$\bar{\nu}_e + {}^{40}{\rm Ar} \rightarrow e^+ + {}^{40}{\rm Cl}^*$ C, A, G 5/4
$\nu_e + {}^{208}{\rm Pb} \rightarrow e^- + {}^{208}{\rm Bi}^*$ N 144/228
$\nu_x + {}^{208}{\rm Pb} \rightarrow \nu_x+ {}^{208}{\rm Pb}^*$ N 150/55
$\nu_x + A \rightarrow \nu_x + A$ C 9,408/4,974
: Summary of relevant interactions for current and near-future detectors. The observables column lists primary observable products relevant for interactions in current detectors. Abbreviations: C, energy loss of a charged particle; N, produced neutrons; G, deexcitation $\gamma$s; A, positron annihilation $\gamma$s. Note there may in principle be other signatures for future detector technologies or detector upgrades. The interactions column gives interactions per kt at 10 kpc for two different neutrino flux models (Livermore/GKVM), for neutrino energies greater than 5 MeV, computed according to `http://www.phy.duke.edu/\simschol/snowglobes`. No detector response is taken into account here, and actual detected events may be significantly fewer. For elastic scattering and inverse $\beta$ decay, the numbers per kt refer to water; for other detector materials, the numbers need to be scaled by the relative fraction of electrons or protons, respectively. For neutrino-proton elastic scattering, the numbers per kt refer to scintillators. \[tab:interactions\]
TECHNOLOGIES FOR SUPERNOVA NEUTRINO DETECTION {#detectors}
=============================================
The particles produced by supernova neutrinos are detected via numerous standard particle-detection techniques, namely by collecting photons or charge proportional to ionization energy loss, collecting Cherenkov photons, or detecting neutrons. A general review of relevant detection physics can be found in Reference [@Nakamura:2010zzi]. The specific experimental techniques employed depend on the characteristics of the target; in many cases, the target material itself is the detection medium.
A neutrino experimentalist’s basic strategy is to collect as many neutrino events from a supernova burst as possible, of as many flavors as possible. For current detectors, typical event yields are a few hundred events per kt of detector material for a core-collapse event just beyond the center of the Milky Way, 10 kpc away. Desirable for measurement are event-by-event timing resolution, the ability to measure neutrino energies with good resolution, and the ability to use the neutrino information to point back to the supernova. Sensitivity to all flavors of the burst is especially desirable: $\nu_\mu$ and $\nu_\tau$ flavors constitute two thirds of the burst’s luminosity, but NC sensitivity is required to detect them. Note that the interaction rate is not the only thing that matters: It is especially valuable for detectors to have the ability to tag interactions as $\nu_e$, $\bar{\nu}_e$, and $\nu_{x}$, in addition to simply collecting them. An ideal detector would measure the flavor, energy, time, and direction of the neutrinos on an event-by-event basis (with no background), but in reality, one must settle for imperfectly reconstructed events and inferred statistical information.
The expected number of events from a supernova should scale simply with distance as 1/$D^2$, where $D$ is the distance to the supernova. To a good approximation for most technologies (to the extent that detection efficiency is independent of detector size), event rates scale linearly with detector mass.
Expected differential event rates of observed particles for a given neutrino interaction process for a realistic detector can be computed by folding a given supernova neutrino flux with the cross section and detector response according to:
$$\frac{dn}{dE'} = \int_0^\infty \int_0^\infty dE d\hat{E}\Phi(E) \sigma(E) k(E-\hat{E}) T(\hat{E}) V(\hat{E}-E')$$
where $E$ is the neutrino energy; $\hat{E}$ is the produced particle energy; $E'$ is the measured product particle energy, $\sigma(E)$ is the total cross section of the process, $k(E-\hat{E})$ gives the energy distribution of the produced particle; and $T(\hat{E})$ and $V(\hat{E}-E')$ model the detector threshold and energy response, respectively. Here we employ (`http://www.phy.duke.edu/\simschol/snowglobes`), which makes use of (`http://www.mpi-hd.mpg.de/lin/globes`) to compute the event rates.
For a successful supernova burst observation, the detector background rate must not greatly exceed the signal rate in a $\sim$10-s burst. Backgrounds for supernova neutrino detection vary by detector type and location. Ambient radioactivity, from the environment or detector materials (*e.g.* uranium/thorium chain isotopes and radon), is nearly always of concern, although typically such background events do not have energies higher than $\sim$7-10 MeV, and a relatively high fraction of burst signal events exceed that energy threshold. Nevertheless, radioactivity can be troublesome for measurements of the low-energy end of the signal, possibly at late times, and for low-threshold detectors (see Section \[lowthresh\]).
Cosmic-ray-related backgrounds can be suppressed by siting detectors deep underground. However, cosmic-ray muons can still penetrate to deep sites and produce nuclear fragments via spallation or capture processes with atoms in a detector or surrounding materials. These spallation fragments include radioactive nuclei and neutrons or other hadrons, which later interact with nuclei; the various products can decay on timescales ranging from less than milliseconds to hours, days or longer. The distribution of products is not well understood for all target nuclei. Muon spallation events can produce fake bursts over timescales of tens of seconds. In principle, if one can identify the parent muon of a spallation event or require spatial uniformity (because spallation products are correlated with the linear track of a muon), one can reject spallation background, at some cost in dead time; however in no detector can this be done with perfect efficiency. Although cosmic-ray muons get rarer the deeper one goes, their energy spectrum gets harder and the likelihood per penetrating muon of producing spallation increases.
Other backgrounds include reactor $\bar{\nu}_e$, solar $\nu_e$, and low-energy atmospheric neutrinos and antineutrinos– however, all of these backgrounds should be very small for Galactic bursts. In addition, detectors are subject to various kinds of instrumental noise or misreconstruction, the nature of which is highly specific to the particular detector setup (and in practice can vary in time significantly). However for current underground detectors, background rates should be very low for the duration of a Galactic supernova burst. Supernova neutrino detection is conceivable even for some surface detectors [@Sharp:2002as; @Ayres:2004js]. Backgrounds become a dominant concern for more distant supernovae (see Sections \[extragalactic\] and \[diffuse\]), even for deep detectors.
Scintillation Detectors {#scint}
-----------------------
Scintillator detectors are composed of hydrocarbons, which have the approximate chemical formula C$_n$H$_{2n}$. The energy loss of charged particles is observed via light emitted from deexcitation of molecular energy levels, and a very large number of photons may be released. Large-mass scintillator detectors typically take the form of large homogeneous volumes of liquid viewed by photomultiplier tubes (PMTs); they may also be segmented into smaller volumes. The energy loss is proportional to the number of photons collected, and the interaction vertices may be reconstructed through the use of the time-of-arrival information of the photons. Large numbers of photoelectrons may be collected for typical PMT densities, leading to excellent energy resolution and low thresholds.
Because of the presence of free protons in scintillator, IBD is hugely dominant for a supernova burst signal. In scintillator, a neutron produced in an IBD interaction is thermalized and captured on a free proton, $n+p \rightarrow d+ \gamma$, with a $\sim$200-$\mu$s delay; the energy loss from Compton scattering of the produced 2.2-MeV $\gamma$ is also observable. Furthermore, the two 0.511-MeV $\gamma$s from the annihilation of the positron also contribute to the observed energy loss (although fine segmentation is usually required to distinguish them). Elements with a high neutron capture cross section may be dissolved in the scintillator. In particular, Gd is highly effective for enhancing neutron capture-detection efficiency: see Section \[ibd\]. The time sequence of a prompt positron signal followed by a delayed capture signal provides a clean IBD tag.
Examples of large homogeneous scintillation detectors are KamLAND [@Eguchi:2002dm] in Japan, and Borexino [@Cadonati:2000kq; @Monzani:2006jg] in Italy. Segmented detectors with supernova sensitivity include the past MACRO detector [@Ambrosio:1997hh] and the current detectors LVD [@Aglietta:1992dy; @Agafonova:2007hn] in Italy and Baksan [@Alekseev:1993dy] in Russia. Future detectors include SNO+ [@Kraus:2010zzb], currently under construction at SNOLAB in Canada, and the proposed experiments LENA [@Wurm:2011zn] and HanoHano [@Learned:2008zj]. Some surface detectors may also have sensitivity; these include MiniBooNE [@Sharp:2002as; @AguilarArevalo:2009ju], a kt mineral-oil detector at Fermi National Accelerator Laboratory that makes use primarily of Cherenkov light for particle detection, and NO$\nu$A [@Ayres:2004js], a segmented liquid scintillator detector. For such detectors, despite relatively large cosmic-ray rates, a burst from a nearby supernova would stand out above the background. Smaller surface experiments with dissolved Gd, designed for reactor oscillation studies– Double Chooz [@Ardellier:2006mn], RENO [@Ahn:2010vy] and Daya Bay [@OchoaRicoux:2011zz]– will be able to gather some tens of IBD events at 10 kpc with a good signal-to-background ratio, thanks to excellent tagging.
Elastic scattering will contribute a few percent to the total supernova burst event rate; however, because light emission is isotropic, little supernova direction information will be obtained. Nevertheless, for very good vertex resolution, some directional information can be observed via separation of neutron capture and positron signal sites for IBD (see Section \[pointing\]).
Neutrino interactions on carbon also occur. In particular, a 15.1-MeV deexcitation $\gamma$ from NC excitation of $^{12}$C may be observable, and good energy resolution in scintillator would allow determination of the total neutrino flux.
Finally, in scintillator, it may be possible to observe NC neutrino-proton elastic scattering. This signal produces very low recoil energy protons; in scintillator, quenching effects suppress light from this signal. Nevertheless, significant numbers of events could be observed [@Beacom:2002hs], and measurements of the recoil energy distribution is one of the few ways to obtain information about the $\nu_x$ spectra [@Dasgupta:2011wg].
Figure \[figure3\] shows an example of expected event rates in a scintillator detector, computed according to `http://www.phy.duke.edu/\simschol/snowglobes` and assuming an estimated energy resolution from Reference [@Eguchi:2002dm]. The assumed flux is from Reference [@Gava:2009pj].
Water Cherenkov Detectors {#wc}
-------------------------
Water is another substance with an abundance of free protons; also, it has the advantage of being inexpensive, so large detectors are possible. Like scintillator detectors, water Cherenkov detectors consist of homogeneous volumes of liquid instrumented with PMTs. Charged particles are detected via their Cherenkov light emission, which forms a 42$^\circ$ cone for relativistic particles. The energy loss is proportional to the number of photons detected, and one may reconstruct the charged particle’s interaction vertex and direction via the Cherenkov ring pattern. Cherenkov detection produces relatively little light compared with that produced by scintillator (by approximately a factor of 50) and has the disadvantage that particles must have speeds exceeding $c/n$ to be visible (here, $n$ is the index of refraction of the material). Water’s index of refraction is $n\sim 1.34$; for this value, the (total-energy) Cherenkov threshold for electrons is 0.8 MeV, and for protons it is 1,400 MeV. As a result, protons induced by supernova neutrinos are always invisible, as are sub-MeV electrons; low energy $\gamma$s, which interact by Compton scattering and for which scattered electrons are often below the Cherenkov threshold, often have rather low light yields and poor detection efficiency. However, Cherenkov detection has the advantage that the observed photons carry directional information. Therefore, for anisotropic interactions such as elastic scattering, angular distributions of products can be measured and used both for supernova pointing and for the disentanglement of flavor components [@Scholberg:2011zzb; @Akiri:2011dv].
As for scintillator, interaction rates in water are dominated by IBD. Due to the Cherenkov threshold, the detection of the 2.2-MeV $\gamma$ from neutron capture on free protons is very difficult. It may be possible to dissolve Gd compounds in the water to enhance neutron tagging [@Beacom:2003nk]. Approximately 4 MeV of equivalent energy can be detected per neutron capture; in Super-Kamiokande (Super-K), the neutron tagging efficiency is estimated to be approximately 67% [@Watanabe:2008ru]. A possible Gd enhancement is being studied for Super-K [@Kibayashi:2009ih]. Neutron tagging will be helpful for determining the total flavor components of the supernova burst signal [@Akiri:2011dv].
Other interactions in oxygen nuclei of water will contribute to the total burst signal in a water detector. Deexcitation $\gamma$s from NC interactions [@Langanke:1995he] may be visible, although due to Cherenkov threshold effects, detection efficiency is poor and requires relatively high PMT coverage [@Scholberg:2011zzb; @Akiri:2011dv]. Some interactions have ejected neutrons [@Kolbe:2002gk]; these may be captured and detected with Gd loading (which may lead to confusion with IBD events). Angular information obtained from the predicted CC $^{16}$O backwards anisotropy may also be usable [@Haxton:1988mw].
Past examples of water Cherenkov detectors include IMB [@BeckerSzendy:1992hr] and Kamiokande [@Hirata:1991ub], both of which are famous for the SN1987A observation. The only current example of a large water Cherenkov detector is Super-K [@Ikeda:2007sa] in Japan. Future concepts include an initially proposed water Cherenkov detector design for the Long-Baseline Neutrino Experiment (LBNE) [@Akiri:2011dv], Hyper-Kamiokande [@Abe:2011ts], and MEMPHYS [@Borne:2011zz].
Although there are no current or planned future instances of heavy-water detectors, the Sudbury Neutrino Observatory (SNO) detector [@Boger:1999bb] in Canada, which took data between 1999 and 2006, was sensitive to supernova neutrinos via NC and CC breakup interactions ($\nu_e+d\rightarrow p+p+e^-$, $\bar{\nu}_e+d\rightarrow n+ n +e^+$), in its 1 kt of heavy water. These interactions were observable via Cherenkov light and multiple methods of neutron detection, in addition to interactions in SNO’s 1.7 kt of light water [@Fleurot:2007zz].
Figure \[figure4\] shows an example of expected event rates from a water Cherenkov detector, computed according to `http://www.phy.duke.edu/\simschol/snowglobes` and assuming detector response similar to that of Super-K I [@Akiri:2011dv].
Long-String Water Cherenkov Detectors
-------------------------------------
Long-string water Cherenkov detectors are arrays of long vertical strings of PMTs in water or ice. Although such detectors are nominally designed for the study of very high energy astrophysical neutrinos (up to the TeV range or higher), such detectors may also be capable of supernova neutrino detection, assuming that the background rates are sufficiently low [@Halzen:1994xe; @Halzen:1995ex]. The idea is that neutrino interactions in the water or ice surrounding the PMTs (mostly IBD) will create enough photons to produce a coincident increase in the single-count rate in the PMTs, observable over background counts. The IceCube detector, which consists of 86 strings with 60 PMTs each and is embedded 1-2 km below the surface of the Antarctic ice, has had a supernova trigger installed [@Abbasi:2011ss] and has demonstrated sensitivity to supernovae within the Milky Way. Future infill upgrades of higher-PMT-density strings could improve supernova sensitivity by allowing for the detection of coincidences between PMTs from individual neutrino interactions.
Currently, IceCube cannot reconstruct individual neutrino interaction events, so it is insensitive to spectral and directional information. However, thanks to the huge photon statistics, it has very good timing, which will render it sensitive to the time structure of the supernova burst (*e.g.* References [@Halzen:2009sm; @Lund:2010kh]). Some other long-string detectors, such as ANTARES [@Ageron:2011pe], are so noisy that they may not be able to trigger themselves but they may be able to receive external triggers to retain extra data. The future KM3NET [@Leisos:2012dk] detector may be able to suppress background by use of PMT coincidences.
Liquid Argon Time Projection Chambers
-------------------------------------
Liquid argon detectors will have excellent sensitivity to $\nu_e$ via the CC interaction on $^{40}$Ar, $\nu_e+{}^{40}{\rm Ar}\rightarrow e^{-}+{}^{40}{\rm K}^{*}$. In principle, this is a taggable interaction, for which the deexcitation $\gamma$s from $^{40}{\rm K}^{*}$ can be observed. The $\bar{\nu}_e$ interaction, $\bar{\nu}_e+{}^{40}{\rm Ar}\rightarrow e^{-}+{}^{40}{\rm Cl}^{*}$ , will also occur and can be tagged via the pattern of $\gamma$s. NC excitations are also possible, although little information is currently available in the literature about cross sections and observables. Finally, there will be elastic scattering of neutrinos on electrons.
Large liquid argon detectors suitable for supernova neutrino detection are primarily large TPCs, in which ionization charge is drifted by an electric field and signals are collected on wire planes. Using the time of arrival of charge at the readout planes, one can reconstruct three-dimensional track; particles can be identified by their rate of energy loss along a track. Argon also scintillates, and scintillation photons collected by PMTs enable fast timing of signals and enhance event localization within the detector. Liquid argon TPC detection technology offers good energy resolution and full particle reconstruction, unaffected by Cherenkov threshold. For sufficiently fine wire spacing, millimeter position resolution and very high quality tracking can be achieved. Energy thresholds as low as a few MeV may be possible. The direction of the scattered electron for elastic scattering can be determined [@Bueno:2003ei]. Cosmogenic backgrounds should be a function of depth and location and are relatively unknown, although some preliminary studies exist [@Barker:2012nb].
A current example is ICARUS in Italy [@Bueno:2003ei], which will record events as long as suitable triggering can be implemented. Future ideas include the LBNE liquid argon detector [@Akiri:2011dv], GLACIER [@Angus:2010sz] in Europe, and possibilities in Japan [@Hasegawa:2011zz]. Smaller surface detectors such as MicroBooNE at Fermilab [@Soderberg:2009rz] may also record some events.
Figure \[figure5\] shows an example of expected event rates from an argon detector, computed according to `http://www.phy.duke.edu/\simschol/snowglobes` and assuming the detector resolution from Reference [@Amoruso:2003sw].
Detectors Employing Heavy Nuclei {#lead}
--------------------------------
Interactions with heavier nuclei such as iron and lead may yield quite high rates [@Hargrove:1996zv; @Fuller:1998kb; @Engel:2002hg; @Boyd:2002cq; @Zach:2002is] of both CC and NC interactions. Observables include leptons and ejected nucleons. Single- and multi-neutron ejections are possible. Because the cross sections for single- and double-neutron ejections strongly depend on neutrino energy, a measurement of the relative rates of these processes provides information about the incoming neutrino spectra. Lead is especially promising for neutron detection because of its stability and ease of handling in large quantities. Furthermore, its most abundant isotopes have low neutron capture cross sections, making it nearly transparent to supernova-neutrino-induced neutrons, allowing the neutrons to reach a region of the detector system that is sensitive to neutron capture.
Although detectors employing iron and/or lead in different configurations have been proposed (e.g., OMNIS [@Boyd:2002cq; @Elliott:2000su]), the only current example of such a detector is HALO in Canada [@Duba:2008zz]. HALO detects neutrons by using recycled SNO $^3$He counters in combination with 79 tons of lead. The relevant interactions are $\nu_e+{}^{A}{\rm Pb}\rightarrow e^{-}+{}^{A}{\rm Bi}^{*}$ and $\nu_x+{}^{A}{\rm Pb}\rightarrow \nu_x+{}^{A}{\rm Pb}^{*}$. For both CC and NC cases, the resulting nuclei deexcite via nucleon emission. Antineutrino CC interactions are strongly suppressed by Pauli blocking because of lead’s neutron excess. Although natural lead contains isotopes other than $^{208}$Pb, the neutrino cross section for $A=208$ should be similar for other components [@Kolbe:2000np; @Engel:2002hg]. In HALO, neutrons are moderated in polypropylene before being captured, and the energy loss of the particles resulting from neutron capture on $^3$He is recorded in proportional counters. The detection efficiency of the capture of a neutron generated in the lead is on the order of tens of percent. There is no event-by-event energy information, but numbers of single- and double-neutron-producing interactions can be inferred and these relative numbers are very sensitive to the neutrino spectrum. HALO2 is a proposed upgrade to approximately the kt scale.
Reference [@Vaananen:2011bf] explores the physics sensitivity for HALO. Figure \[figure6\] shows an example of expected interactions in lead, computed according to `http://www.phy.duke.edu/\simschol/snowglobes` .
Low Threshold Detectors {#lowthresh}
-----------------------
Because of its very low energy threshold, only very clean, radioactively quiet detectors can detect nuclear recoils from coherent elastic neutrino-nucleus scattering; they must also have very good rejection of electronic recoils. Detectors developed for WIMP detection (or for neutrinoless double-$\beta$-decay searches or low-energy solar neutrino detection) fit the bill. Examples are described in References [@Gaitskell:2004gd] and [@Akimov:2011zz]. Possibilities include solid-state, noble liquid and bubble technologies, as well as new spherical TPC proposals [@Vergados:2006jg]. Noble liquid detectors with xenon, argon or neon targets are particularly promising for scale-up to multiton scales; they have recoil energy thresholds of a few tens of keV. These detectors are sensitive to all flavor components of the flux via the NC coherent interaction, and some spectral information can be obtained via the measured recoil energy spectrum. Expected supernova neutrino event rates are a few events per ton at 10 kpc, so current-generation detectors will have only modest sensitivity. However, future experiments of 10-ton-or-greater scale are very promising, in that they may provide spectral information about all flavors.
Summary of Detectors
--------------------
Detector Type Mass (kt) Location Events Live period
---------------------- --------------- ----------- ------------ --------- --------------
Baksan C$_n$H$_{2n}$ 0.33 Caucasus 50 1980-present
LVD C$_n$H$_{2n}$ 1 Italy 300 1992-present
Super-Kamiokande H$_2$O 32 Japan 7,000 1996-present
KamLAND C$_n$H$_{2n}$ 1 Japan 300 2002-present
MiniBooNE$^*$ C$_n$H$_{2n}$ 0.7 USA 200 2002-present
Borexino C$_n$H$_{2n}$ 0.3 Italy 100 2005-present
IceCube Long string 0.6/PMT South Pole N/A 2007-present
Icarus Ar 0.6 Italy 60 Near future
HALO Pb 0.08 Canada 30 Near future
SNO+ C$_n$H$_{2n}$ 0.8 Canada 300 Near future
MicroBooNE$^*$ Ar 0.17 USA 17 Near future
NO$\nu$A$^*$ C$_n$H$_{2n}$ 15 USA 4,000 Near future
LBNE liquid argon Ar 34 USA 3,000 Future
LBNE water Cherenkov H$_2$O 200 USA 44,000 Proposed
MEMPHYS H$_2$O 440 Europe 88,000 Future
Hyper-Kamiokande H$_2$O 540 Japan 110,000 Future
LENA C$_n$H$_{2n}$ 50 Europe 15,000 Future
GLACIER Ar 100 Europe 9,000 Future
: Summary of neutrino detectors with supernova sensitivity. Neutrino event estimates are approximate for 10 kpc; note that there is significant variation by model. Not included are smaller detectors (e.g., reactor neutrino scintillator experiments) and detectors sensitive primarily to coherent elastic neutrino-nucleus scattering (e.g., WIMP dark matter search detectors). The entries marked with an asterisk are surface or near-surface detectors and will have larger backgrounds. []{data-label="tab:detectors"}
Table \[tab:detectors\] summarizes existing and future detectors. Note that the live detector mass used for supernova neutrino detection may be greater than the restricted “fiducial” mass often employed for physics analyses of typical beam, atmospheric, or solar neutrino experiments, because the background rates during a supernova burst are low compared with the signal rates. For example, in Super-K, supernova burst events could be analyzed in the full 32-kt inner-detector volume, whereas the typical mass used for beam, atmospheric and solar neutrino analyses is 22.5 kt or smaller.
POINTING TO THE SUPERNOVA WITH NEUTRINOS {#pointing}
========================================
It will be tremendously valuable to determine the direction of the supernova from the neutrino signal itself. First, this information will be useful for an early alert. For obvious reasons, astronomers want to know where to point their telescopes. Second, a possible scenario is that the supernova will have no signal in electromagnetic radiation, or only a very faint signal, and pointing information will be useful for locating a remnant (or a progenitor from catalogs). Doing so could, for example, allow one to learn the distance of travel by the neutrinos through the Earth for matter effect evaluation.
The most promising method for neutrino pointing is via elastic scattering (see Section \[es\]), in which the electron gets kicked in the direction of the neutrino. In a Cherenkov detector, the directionality of the electron can be determined from the Cherenkov ring. Because elastic scattering represents only a few percent of the total signal, the problem becomes one of finding a small bump on a nearly isotropic background. In the absence of background, pointing quality goes roughly as $\sim 25^\circ/\sqrt{N}$, where $N$ is the observed number of elastic scattering events. Reduction of the nearly isotropic background (mostly IBD) can be achieved through the use of neutron tagging with Gd (see Section \[wc\]). When background is taken into account, the expected pointing quality for Super-K at 10 kpc [@Beacom:1998fj; @Tomas:2003xn] is about 8$^\circ$, which improves to approximately 3$^\circ$ with good IBD tagging. For a megaton detector, better-than-1$^\circ$ pointing is achievable.
IBD positrons have only a mild anisotropy [@Vogel:1999zy], which nevertheless could be exploited in a very high statistics measurement. In principle it is possible to recover some directional information in scintillator by reconstructing the relative positions of neutron capture and positron energy loss. This IBD directionality has been measured in the CHOOZ scintillator detector for reactor $\bar{\nu}_e$ interactions [@Apollonio:1999jg], and future scintillator detectors may be able to exploit the effect. Other possible methods include detection of TeV neutrinos from the supernova [@Tomas:2003xn] (which point well, but may not accompany all core-collapse supernovae and may have significant delay), and use of the matter oscillation pattern in a detector with good energy resolution (which requires very large statistics) [@Scholberg:2009jr].
In principle, determination of the supernova direction by use of triangulation can be achieved through relative timing of signals from multiple detectors around the Earth. By using a time signal in two detectors, one can locate the supernova within a ring on the sky; with three time measurements, two intersecting rings narrow down the location to two spots; and with timing from four detectors, one can point to a single spot. However triangulation is likely difficult in practice, at least on a short timescale. Times of flight through the Earth are on the order of milliseconds; because the neutrino signal is spread over $\sim$10 s, it requires large statistics [@Beacom:1998fj] or else a relatively high flux short-timescale component of the signal. Nevertheless, the triangulation technique may be feasible for future megadetectors. At this time and in the near future, the best bet for supernova neutrino pointing is elastic scattering in a water Cherenkov detector.
AN EARLY ALERT: SNEWS {#earlyalert}
=====================
Neutrinos emerge promptly from a collapsing star, on a $\sim$10-s timescale, and the first electromagnetic radiation may not appear for hours or even days, depending on the nature of the stellar envelope. Therefore, neutrinos can be used to provide an early warning of an imminent visible supernova.
Very early light from extragalactic supernovae is rarely observed, so a prompt alert would give astronomers valuable time to find the supernova. The environment immediately surrounding the progenitor star would be probed by the initial stages of the supernova. Any effects of a close binary companion upon the blast would occur very soon after shock breakout. Advance warning could enable observation of ultraviolet and soft X-ray flashes, which are predicted at very early times. There could also be entirely unexpected effects at early times. A Galactic supernova is rare enough that it will be critical to save all available information.
The SN1987A neutrino events were recorded approximately 2.5 h before the inferred time of the supernova’s first light; in real time, the experimentalists found the events in their data only after the fact [@Koshiba:1992yb]. The situation will be different for the next nearby supernova. The SuperNova Early Warning System (SNEWS) is an international network of detectors that aims to provide an early alert to astronomers of a supernova’s occurrence [@Antonioli:2004zb]. At the time of this writing, Super-K, LVD, IceCube, and Borexino are active participants in the network. Each detector involved sends a prompt datagram to a central computer if a signal consistent with a burst of supernova neutrinos is observed, and if there is coincidence within 10 s an alert message is automatically sent. The idea is to require a coincidence for a very high confidence early alert, so as to suppress false alarm background. The criteria for the alert are fairly conservative; they require less than one accidental coincidence per century. The details of the coincidence conditions are described in References [@Antonioli:2004zb] and [@Scholberg:2008fa]. Any pointing information sent to SNEWS will be forwarded to astronomers (although depending on the detectors involved, no pointing information may be initially available).
Note that some detectors that can provide useful information are not necessarily capable of triggering themselves on a supernova burst and may not be continuously archiving information. They may be noisy and/or may not know what kind of signal to look for from a supernova. Some examples of detectors in this category include some of the long string detectors (ANTARES), gravitational-wave detectors (if not all data are archived), and surface neutrino-sensitive detectors with a high rate of cosmic-ray background. A SNEWS neutrino coincidence will be a high confidence indication that a supernova has occurred. Noisy supernova detectors would therefore arrange to use the SNEWS coincidence as an input – they would set up a buffering system to record data (for hours or days, depending on the resources available) that would routinely be overwritten but that could be saved to permanent storage in the case of a SNEWS coincidence. This approach would greatly enrich the world’s supernova data sample.
An interesting possibility for an early alert for the collapse itself is the detection of neutrinos from a presupernova star [@Odrzywolek:2003vn]. Very large ($\sim$20 $M_\odot$) stars would produce pair-annihilation neutrinos during the final silicon-burning phase before core collapse. The flux is small and the neutrinos are cool enough ($\sim$MeV average energies) to be difficult to detect, but in the few days before collapse there would be a distinctive hardening and an event rate increase. Tagged IBDs from the high-energy tail of the spectrum are the most promising prospect, so scintillator or Gd-loaded water detectors would be required. When realistic backgrounds are accounted for, for a $\sim$100 kt water detector the reach for this kind of early warning would probably be less than 1 kpc. This represents only a small fraction of stars in the Milky Way, but it does include Betelgeuse, a nearby core-collapse candidate at 200 pc.
EXTRAGALACTIC SUPERNOVA NEUTRINO SEARCHES {#extragalactic}
=========================================
Table \[tab:detectors\] shows event rates at 10 kpc, which is just beyond the center of the Galaxy. According to $1/D^2$ scaling, most detectors will observe robust bursts of events for any collapse that occurs within the Galaxy. Beyond the edge of the Galaxy, there are few core-collapse candidate stars. For a collapse in the LMC at 50 kpc, a few hundred events would be expected in Super-K, and approximately one event for Andromeda at 770 kpc. A next-generation few-hundred-kt detector would detect dozens of events from Andromeda. However, beyond that, the inverse square law strongly suppresses the signal and neutrino interactions become scarce.
Reference [@Ando:2005ka] points out that with megaton-scale detectors, one might detect of the order of one neutrino per year from extragalactic sources. The number of potential sources goes as $\sim D^3$. There is evidence that the historical rate of supernovae within 10 Mpc or so is slightly higher than one would predict from simple scaling; one expects approximately one core collapse per year within that radius. Furthermore, some black hole-forming collapses may produce hotter-than-usual spectra and therefore higher neutrino event rates, and they may occur as often as once per decade within a few megaparsecs [@Yang:2011xd].
However, for a search for extragalactic supernova neutrinos, background suppression becomes critical. The time window for an optical trigger would be several hours, and typical detectors would need to have a very low rate of background in this time window. For example, at a 17-MeV threshold, Super-K’s background rate is $\sim$1 per day [@Ikeda:2007sa]; scaling to 1 Mt, this rate would be on the order of $\sim$10 background events in a several-hour time window, with only a $\sim$10% chance of detecting a single signal event at 10 Mpc. Requiring time clusters of two or more events reduces efficiency. Backgrounds are typically highly site and detector dependent, however. In particular, spallation backgrounds depend on the cosmic-ray rate and spectrum; therefore the simple scaling from the rates in Reference [@Ikeda:2007sa] may not necessarily be appropriate. There are also plausible strategies to reduce backgrounds (see Section \[diffuse\]), and over megaton-decades, one might collect an observable signal over background. Reference [@Kistler:2008us] explores what one might learn from miniburst measurements of distant supernovae with a multimegaton detector.
Searches looking for coincidences with nonelectromagnetic collapse signatures, namely gravitational waves, may have more promise [@Pagliaroli:2009qy; @Leonor:2010yp] because the signal will also be prompt after core collapse. However, a disadvantage of such searches is that the mechanism for gravitational-wave production and the nature of the signal are more uncertain; the distance range of gravitational-wave observability is poorly understood.
THE DIFFUSE SUPERNOVA NEUTRINO BACKGROUND {#diffuse}
=========================================
Looking even farther out for sources of neutrinos, one can imagine measuring the flux of neutrinos from all the supernovae in cosmic history. This so-called diffuse supernova neutrino background (DSNB) is sometimes referred to as the relic supernova neutrino flux. The physics of the DSNB is reviewed in References [@Beacom:2010kk] and [@Lunardini:2010ab].
The DSNB flux depends on the historical rate of core collapse, average neutrino production, cosmological redshift effects and neutrino oscillation effects. For neutrino energies above $\sim$19 MeV, estimates of the $\bar{\nu}_e$ component of the DSNB range from $\sim$0.1 to 1 cm$^{-2}$s$^{-1}$.
The detection interactions remain the same as for burst neutrinos; however, the experimental issues become entirely background dominated, as there would be no external trigger at all and events would be measured singly. Overall, one would expect $\sim$0.1 IBD per kt per year in water or scintillator detectors from the $\bar{\nu}_e$ component of the DSNB. At lower energies, solar neutrinos dominate the neutrino background, although reactor $\bar{\nu}_e$ fluxes also contribute. At higher energies, atmospheric neutrino backgrounds dominate. The atmospheric and reactor backgrounds vary by detector location. There is an energy window between $\sim$20 and 40 MeV in which the diffuse background dominates the neutrino flux incident on the Earth (Figure \[figure7\]) [@Lunardini:2010ab]. Detector-specific backgrounds for DSNB are the main issue in this energy window. In a water detector such as Super-K, toward the lower end of this range, cosmic ray-induced spallation backgrounds leak into the signal window. In the 20-50 MeV range, atmospheric neutrino-induced muons dominate the background: In water, low-energy atmospheric CC $\nu_\mu$ can produce “invisible” muons below Cherenkov threshold, that then decay; the resulting Michel electrons mimic positrons from IBD. So far, the best limits on the DSNB are from Super-K [@Malek:2002ns; @Bays:2011si]; the newer result has an updated limit of $< 2.8-3.0$ $\bar{\nu}_e$ cm$^{-2}$s$^{-1}$ for neutrino energies greater than $17.3~$MeV. The new limit is slightly less stringent than the earlier result but is the outcome of improved analysis.
The KamLAND scintillator detector also has rather less stringent limits on the DSNB $\bar{\nu}_e$ flux, but they extend to somewhat lower energies [@Gando:2011jza]. The background for DSNB $\bar{\nu}_e$ in scintillator differs from that in water: The muons are not invisible because there is no Cherenkov threshold, and IBD interactions can be effectively tagged with neutrons. However, NC interactions of atmospheric neutrinos on carbon (ejected neutrons in coincidence with deexcitation $\gamma$s) dominate the background in this energy range in scintillator. The SNO heavy-water experiment [@Aharmim:2006wq] has published a limit on the $\nu_e$ component of the DSNB flux, 70 cm$^{-2}$s$^{-1}$ at 90% CL between 22.9 MeV and 36.9 MeV, via a search for high-energy CC interactions of $\nu_e$ on deuterium.
Because tagging of $\bar{\nu}_e$ could reduce the Michel background (for atmospheric neutrino interactions unaccompanied by neutrons), the Gd proposal for Super-K [@Beacom:2003nk; @Kibayashi:2009ih] is promising for future measurements of the DSNB. Future large water detectors are very promising, especially if Gd loaded [@Akiri:2011dv], as are future large scintillator detectors [@Wurm:2011zn] as long as their NC background can be reduced. Liquid argon detectors are sensitive to the $\nu_e$ component of the DSNB [@Cocco:2004ac; @Autiero:2007zj]. For argon, background is the key issue but is generally unknown at this time, although it could plausibly be small with good $\gamma$ tagging of $\nu_e$ absorption events and deep detectors.
The rate of DSNB neutrino events is steady, sure, and low: a constant rate of 0.1 events per kt per year can be expected. In contrast, nearby collapse supernovae will provide copious bursts of neutrinos, but only sporadically: One expects thousands of neutrinos, but only every few decades, so building a detector with a short lifetime is a risky proposition due to potentially long Poissonian gaps between supernovae. However, if one is prepared to wait, our own galaxy will provide, in the very long term, a much higher average rate of neutrino events: $\sim$10 events per kt per year. The best experimental strategy is clearly a diversified portfolio: A large, long-running, low-background detector will win on both types of investment.
PROSPECTS {#summary}
=========
A Galactic supernova will offer unprecedented opportunities for diverse neutrino detectors around the world to gather critical information about astrophysics and particle physics. Flavor sensitivity –not only the interaction rate but also the ability to tag different interaction channels– will be critical for maximizing the science harvest from a burst observation. Current-generation detectors are sensitive primarily to $\bar{\nu}_e$; however, next-generation detectors will expand worldwide flavor sensitivity. Potential upgrades and new detectors will also enhance the prospects for detection of the so-far-unobserved DSNB. If future megaton-scale detectors are built, prospects for a vast yield of information from a nearby burst are excellent, and we hope for a reach extending well beyond the Milky Way.
ACKNOWLEDGMENTS {#acknowledgments .unnumbered}
===============
The author’s research activities are supported by the U.S. Department of Energy and the National Science Foundation. The author thanks her Super-K, LBNE, SNEWS and HALO collaborators (special thanks to Tarek Akiri and Cecilia Lunardini), as well as the contributors Farzan Beroz, Rachel Carr, Huaiyu Duan, Alex Friedland, Nicolas Kaiser, Jim Kneller, Alexander Moss, Diane Reitzner, David Webber, and Roger Wendell.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We compute the effect of collapse models on neutrino oscillations. The effect of the collapse is to modify the evolution of the [*spatial*]{} part of the wave function and we will show that this indirectly amounts to a change on the flavor components. For the analysis we use the mass proportional CSL model, and perform the calculation to second order perturbation theory. As we will show, the CSL effect is very small—mainly due to the very small mass of neutrinos—and practically undetectable.'
author:
- Sandro Donadi
- Angelo Bassi
- Catalina Curceanu
- Luca Ferialdi
title: The effect of spontaneous collapses on neutrino oscillations
---
Introduction
============
The general validity of the superposition principle of quantum mechanics is questioned by an increasing number of scientists [@Bel; @Ad; @We; @Le], and is subject to an intense experimental verification [@Ar1; @Ar2; @Ar3; @Ar4; @Bo; @Mo]. The possibility that the superposition principle might have only a limited range of validity is foreseen by collapse models [@Grw; @Cslmass; @Fu; @Csl; @adlerphoto; @Pr; @Ad2; @Pe1; @Pe2; @Pe3; @Di1; @Di2], which predict small deviations from standard quantum mechanics, for all those cases where quantum linearity plays a fundamental role. Neutrino oscillations [@Beuthe; @Neu1; @Neu2; @Neu3] are one such a case, and the goal of this article is to present the theoretical analysis and estimate of the effect of spontaneous collapses on the oscillatory behavior of neutrinos.
A previous analysis of this kind was proposed in [@Ch], based on the Penrose model of gravity induced collapse [@Pd1; @Pd2]. This model however lacks a fully consistent dynamical equation, and previous attempts to fill this gap [@Dg] have been criticized [@Gpd]. Moreover, as shown in [@Gpd], the model fails when applied to single constituents (protons, electrons, ...), since in this case its predictions are in conflict with known experimental data. Therefore, the application of gravity induced collapse models to neutrino oscillations is rather delicate.
Here we will compute the spontaneous collapse effect on neutrino oscillations using the mass proportional version [@Cslmass; @Fu] of the CLS model [@Csl], which is widely used in physical applications, together with the GRW model [@Grw]. Its dynamics is described by the following stochastic differential equation: $$\label{eq:csl-massa}
d|\phi_{t}\rangle=\left[-\frac{i}{\hbar}Hdt+\frac{\sqrt{\gamma}}{m_{0}}\int d\mathbf{x}\,\left(M(\mathbf{x})-\left\langle M(\mathbf{x})\right\rangle \right)dW_{t}(\mathbf{x})-\frac{\gamma}{2m_{0}^{2}}\int d\mathbf{x}\,\left(M(\mathbf{x})-\left\langle M(\mathbf{x})\right\rangle \right)^{2}dt\right]|\phi_{t}\rangle,$$ where the operator $H$ is the standard quantum Hamiltonian of the system and the other two terms induce the collapse of the wave function in space. The mass $m_0$ is a reference mass, which is taken equal to that of a nucleon. The parameter $\gamma$ is a positive coupling constant which sets the strength of the collapse process and $\left\langle M(\mathbf{x})\right\rangle =\left\langle \phi_{t}\left|M(\mathbf{x})\right|\phi_{t}\right\rangle $, where $M({\bf x})$ is a smeared mass density operator: $$M\left(\mathbf{x}\right)=\underset{j}{\sum}m_{j}\underset{s}{\sum}\int
d\mathbf{y}g\left(\mathbf{x-y}\right)
\psi_{j}^{\dagger}\left(\mathbf{y},s\right)\psi_{j}\left(\mathbf{y},s\right),$$ $\psi_{j}^{\dagger}\left(\mathbf{y},s\right)$, $\psi_{j}\left(\mathbf{y},s\right)$ being, respectively, the creator and annihilation operators of a particle of type $j$, having mass $m_j$ and spin $s$, in the space point $\mathbf{y}$. The smearing function $g({\bf x})$ is taken equal to $$\label{eq:nnbnm}
g(\mathbf{x}) \; = \; \frac{1}{\left(\sqrt{2\pi}r_{C}\right)^{3}}\;
e^{-\mathbf{x}^{2}/2r_{C}^{2}},$$ where $r_C$ is the other new phenomenological constant of the model. Finally, $W_{t}\left(\mathbf{x}\right)$ is an ensemble of independent standard Wiener processes, one for each point in space. The standard numerical value of the correlation length $r_C$ is [@Csl]: $$r_C \; \simeq \; 10^{-5}\text{cm},$$ while, in the literature, two different values for the collapse strength $\gamma$ have been proposed. The first value has been originally proposed by Ghirardi, Pearle and Rimini [@Csl]: $$\gamma \; \simeq \; 10^{-30}\text{cm}^{3}\text{s}^{-1}$$ in analogy with the GRW model [@Grw]. The second value has been proposed by Adler, inspired by the analysis of the process of latent image formation according to collapse models, and amounts to [@adlerphoto]: $$\gamma \; \simeq \; 10^{-22}\text{cm}^{3}\text{s}^{-1}.$$
Aim of this work is to understand if, as claimed in [@Ch], neutrino oscillations can be used to test collapse models, and in particular to improve the upper bounds on the collapse strength $\gamma$. The idea is the following: it is well known that, since flavour eigenstates are linear superposition of mass eigenstates, standard quantum mechanics predicts neutrino oscillations. In the CSL model the dynamics is driven by Eq. , which differs from the Schrödinger equation for the two terms accounting for the collapse in space of the wave function. As a consequence, the CSL model predicts a different time evolution of mass eigenstates with respect to that of standard quantum mechanics. This implies, as an indirect consequence, that also flavour eigenstates evolve differently, and therefore that neutrinos are expected to oscillate in a different manner. In some sense, it is as if neutrinos were traveling through a random medium, instead of free space. It is well known that neutrino oscillations are affected by a random medium [@Sm1; @Sm2]. However, we stress that this is more a mathematical analogy as in our case the origin of the randomness is different, and is due to the spontaneous collapse of the wave function. The fact that the collapse mechanism acts on the spatial part of the wave function implies that we have to consider the whole Hilbert space of the system, not just the part related to the flavour degrees of freedom. Technical details about this issue are given in Sec.V.
The paper is organized as follows: in Section II we state the main result of the computation and quantify the damping of neutrino oscillation, as predicted by the mass proportional CSL model. In Section III we compare our result with that of [@Ch]. In Section IV we discuss decoherence effects on neutrino oscillation and compare them with the collapse effects. The remaining sections of the paper are devoted to computing the formulas, which are used in Section II.
The CSL prediction for neutrino oscillation
===========================================
According to the mass proportional CSL model, the transition probability of finding a neutrino in a flavour eigenstate $\beta$, when it was initially in the flavour state $\alpha$, is: $$\label{eq:dfggkdse}
P_{\alpha\rightarrow\beta}
=
\sum_{k=1}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha k}\text{U}_{\beta k}+\sum_{{k \neq j}}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j} \; e^{-\xi_{jk}t} \; \cos\left[\frac{1}{\hbar}(E_{i}^{\left(k\right)}-E_{i}^{\left(j\right)})t\right].$$ $\text{U}$ is the $n\times n$ mixing matrix, which relates the flavour and mass bases, $E_i^{(k)}$ is the initial energy of the neutrino in the mass eigenstate with mass $m_k$, and: $$\label{eq:xi}
\xi_{jk} \; \equiv \; \frac{\gamma}{16\pi^{3/2}r_{C}^{3}m_{0}^{2}c^{4}}\left(\frac{m_{j}^{2}c^{4}}{E_{i}^{\left(j\right)}}-\frac{m_{k}^{2}c^{4}}{E_{i}^{\left(k\right)}}\right)^{2}$$ is the decay-rate of neutrino oscillations, as predicted by collapse models. Note that the frequency of the oscillations is the same as the one predicted by quantum mechanics[^1]. The prediction of the CSL model differs from the standard formula only for a damping factor in front of the oscillating term, with a decay rate given by Eq. . Eqs. and , which will be derived in Sections V-VIII are significant because they allow to precisely quantify the collapse effect on neutrino oscillations. Since collapse master equations have the same structure as decoherence master equations for open quantum systems, it does not come as a surprise that Eq. is in agreement with general arguments, which fix the form that the damping terms coming from decoherence effect should take [@floreanini; @deco1; @deco2].
Having the above equations at hand, we can give a quantitative estimate the CSL effect on neutrino oscillations, by computing the damping factor in Eq. . We consider the stronger value $\gamma\simeq \; 10^{-22}\text{cm}^{3}\text{s}^{-1}$ for the collapse parameter, suggested in [@adlerphoto]. Substituting the numeric values of the constants in Eq. , one finds that: $$\label{eq:xi2}
\xi_{ij}t\simeq (7.33 \times 10^{-36} \text{s$^{-1}$eV$^2$}) \frac{t}{E^2} \,.$$ Here we have taken the largest possible squared mass difference $m_1^2c^4-m_2^2c^4 = 7.59 \times 10^{-5} \text{eV$^2$}$ [@deltamass], where $m_1$ and $m_2$ are respectively the first and the second mass eigenstate and we have considered the ultra-relativistic approximation $E_{i}^{\left(j\right)}=\sqrt{\mathbf{p}_i^{2}c^{2}+m_j^{2}c^{4}}\simeq p_ic\equiv E$, and the same approximation for $E_{i}^{\left(k\right)}$. The energy $E$ and the time $t$ are free and depend on the nature of the neutrinos under study. Neutrinos detected in laboratories have mainly three origins: cosmogenic, solar and those produced in labs. Table I displays the typical values for the energy and time of flight (for simplicity, we assume that neutrinos travel at the speed of light) for these three types of neutrinos. The magnitude of the damping of the oscillations, as predicted by the mass proportional CSL model, has been evaluated using Eq. . As we can see, in all three cases the CSL damping effect on neutrino oscillations is very small. The main reason is that the masses here involved—those of neutrinos—are very small, thus hampering the collapse mechanism.
To conclude we compare the effect of the CSL model for the oscillation formula with the error embodied in the ultra-relativistic approximation. This approximation is usually done in the literature since the error introduced is very small. In order to estimate the error due to the ultra-relativistic approximation, we expand in series the energies $E_i$ of Eq. in the ultra-relativistic regime ($pc>>mc^{2}$). The energy difference, at the second order, becomes: $$E_{i}^{\left(k\right)}-E_{i}^{\left(j\right)}\simeq\frac{m_{k}^{2}c^{4}-m_{j}^{2}c^{4}}{2p_{i}c}-\frac{1}{8}\left(\frac{m_{k}^{4}c^{8}-m_{j}^{4}c^{8}}{p_{i}^{3}c^{3}}\right)\simeq\frac{\triangle m^{2}}{2E}-\frac{\triangle m^{2}}{8E^{3}}\left(m_{k}^{2}c^{4}+m_{j}^{2}c^{4}\right)$$ where we introduced $\triangle m^{2}=m_{k}^{2}c^{4}-m_{j}^{2}c^{4}$ and we approximated, only for the denominator, $E_{i}^{\left(k\right)}\simeq E_{i}^{\left(j\right)}\simeq p_{i}c:=E$. The first term is the oscillation frequency usually considered in the literature [@deco1]. The other one is the most relevant correction. Even taking the upper value for neutrinos masses of order of 2.2 eV [@upper] and considering the case of solar neutrinos, those having a lower energy ($E=10^{6}$ eV), the second term on the right hand side of the above equation is 12 order of magnitude smaller than the first term. This shows that the ultra-relativistic approximation is very good in general.
The error in the oscillation formula due to the ultra-relativistic approximation is: $$\frac{\triangle m^{2}}{8E^{3}}\left(m_{k}^{2}c^{4}+m_{j}^{2}c^{4}\right)\frac{t}{\hbar}=\left(1.01\times10^{10}\;\textrm{s}^{-1}\textrm{eV}^{3}\right)\frac{t}{E^{3}}.$$ Using the data in table I, the effect is $\sim10^{-29}$ for cosmogenic neutrinos, $\sim10^{-6}$ for solar neutrinos and $\sim10^{-22}$ for laboratory neutrinos. Compared with the effect due to the collapse, reported in the last line of table I, the error due to the ultra-relativistic approximation is bigger. This is the reason why we did not make such an approximation.
cosmogenic solar laboratory
-------------- ---------------------- ---------------------- ----------------------
E(eV) $10^{19}$ $10^{6}$ $10^{10}$
t(s) $3.15\times10^{18}$ $5\times10^2$ $2,13\times10^{-2}$
$\xi_{ij} t$ $2.31\times10^{-55}$ $3.66\times10^{-45}$ $1.56\times10^{-57}$
: We consider three types of neutrinos: cosmogenic, solar and laboratory neutrinos. For each type, the table shows: the typical order of magnitude of the energies (first line), the time of flight (second line) and the damping factor as predicted by the mass proportional CSL model (third line).
Comparison with the Diosi-Penrose gravity-induced collapse model
================================================================
As mentioned in the introduction, the damping of neutrino oscillations due to gravitational collapse, as described by the Diosi-Penrose model, was first studied in [@Ch]. In this work, the author argued that the decaying factor (the analog of $\xi_{jk} t$ for the CSL model) has the following form: $$\Lambda_{G}^{j,k}\equiv \int_{D}^{L}\triangle E_{G}^{j,k}\left(L'\right)dL'\,,$$ where $D$ is the distance such that $\triangle E_{G}^{j,k}\left(D\right)=0$, $L$ is the distance traveled by the neutrino, and $$\triangle E_{G}^{j,k}=4\pi\bar{\xi}\int\int\frac{\left[\rho_{j}\left(\mathbf{r}\right)-\rho_{k}\left(\mathbf{r}\right)\right]\left[\rho_{j}\left(\mathbf{r}'\right)-\rho_{k}\left(\mathbf{r}'\right)\right]}{\left|\mathbf{r}-\mathbf{r}'\right|}d\mathbf{r}d\mathbf{r}'\,,$$ where $\bar{\xi}$ is a parameter that we will set equal to $-G$, with $G$ the gravitational constant, like in the original paper by Penrose [@Pd1]. Moreover, $\rho_{1}\left(\mathbf{r}\right)$ and $\rho_{2}\left(\mathbf{r}\right)$ are the two mass distributions, one for each different mass eigenstate. Since these two distributions travel at different velocities, $\triangle E_{G}^{j,k}$ has a dependence on the traveled distance $L'$ (see Eq. (14) of [@Ch]). Following the computation done in [@Ch] and keeping all constants explicit, one finds: $$\Lambda_{G}^{j,k}\simeq8\pi \frac{G}{\hbar c}\left[\frac{3\left(m_{j}+m_{k}\right)\hbar^{2}}{5G_{F}}-\frac{m_{j}m_{k}E}{2\pi\hbar c}\ln\left(\frac{6\left(m_{j}+m_{k}\right)\pi\hbar^{3}c}{5m_{j}m_{k}G_{F}E}\right)\right]L$$ where $G_{F}$ is the Fermi constant and $m_{j}, m_k$ are the neutrino masses. In the case of cosmogenic neutrinos which have an energy of about $E=10^{19}$eV and travel a distance $L\simeq10^{25}$m, the magnitude of the damping factor $\Lambda_G^{j,k}$ lies between 1 and $10^{-2}$, depending on the mass of the neutrino. Therefore, for the Diosi-Penrose model damping effect is by far stronger than that computed with the CSL model. This does not come as a surprise, because—as we stated in the introduction—it is well known that the model predicts, for single elementary constituents, a too-strong collapse of the wave function, which is incompatible with known experimental data [@Gpd]. The reason is the following. It is clear that the model gives rise to divergences in the point-like limit, therefore one has to introduce a cutoff [@Dg2]. One way of doing it, is to consider elementary particles as spherical mass-distributions with a finite radius $R$. In [@Dg] it was proposed to take $R \sim 10^{-15}$m, i.e. the nuclear size. However in [@Gpd] it was shown that also in this case the model is consistent with known facts (the energy increase of isolated systems, due to the collapse, is too large), and proposed a much larger radius, namely $R \sim 10^{-5}$cm, in order to restore compatibility. On the contrary, in [@Ch] the radius ($a_j$, according to the paper’s notation) $R \sim G_F m / \hbar^2 \sim 10^{-31\pm1}$m (where the uncertainty depends on the chosen value for the neutrino’s mass) was considered. This cutoff is too small, therefore the result cannot be trusted.
Decoherence effects
===================
While traveling through the Universe and in particular through the atmosphere, neutrinos interact with the surrounding environment and scatter with other particles, mainly protons, electrons and other neutrinos. These interactions give rise to decoherence effects which also modify neutrino oscillations. Since protons interact with neutrinos only via the neutral weak current, both neutrino families[^2] are affected in the same way by this kind of interaction and neutrino oscillations are not modified. Unlike protons, electrons and neutrinos interact with the incoming neutrino both via neutral and charged weak currents: since these scatterings have different cross sections depending on the neutrino flavor, they contribute to decoherence [@Stol2].
A natural phenomenological estimate for the order of the decoherence rate is: $\Lambda_{\text{\tiny DEC}}\sim n\,v\,\sigma$ with $v$ the relative velocity of the incoming neutrinos, $n$ the density of the environmental leptons, and $\sigma$ the relevant scattering cross section whose values are known in the literature [@sigmanue; @sigmanunu]: $$\begin{aligned}
\label{eq:s1}\sigma_{\nu_{e},e}&\simeq&7\times10^{-42}(\mathrm{E}_{\nu}/\mathrm{GeV})\mathrm{cm}^2\,,\\
\sigma_{\nu_{\mu},e}&\simeq&10^{-42}(\mathrm{E}_{\nu}/\mathrm{GeV})\mathrm{cm}^2\,,\\
\sigma_{\nu_e,\nu_e}&\simeq&2,8\times10^{-47}(\mathrm{E}_{\nu}/\mathrm{GeV})\mathrm{cm}^2\,,\\
\label{eq:s2}\sigma_{\nu_e,\nu_{\mu}}&\simeq&4\times10^{-48}(\mathrm{E}_{\nu}/\mathrm{GeV})\mathrm{cm}^2\,.\end{aligned}$$ The average density of electrons in the outer space and in the atmosphere are respectively $n_e^{\text{\tiny OUT}}\sim1/\mathrm{m}^3$ and $n_e^{\text{\tiny ATM}}\sim2\times10^{26}/\mathrm{m}^3$, while the average density of neutrinos is about $n_{\nu}\sim10^8/\mathrm{m}^3$ everywhere. Assuming $v$ to be the velocity of light in vacuum, then one finds: $$\begin{aligned}
\Lambda_{\text{\tiny DEC}}^{\text{\tiny OUT}} \sim 10^{-43}
(E/\text{eV})\,\text{Hz}\,,\qquad
\Lambda_{\text{\tiny DEC}}^{\text{\tiny ATM}}\sim10^{-20}(E/\text{eV})
\,\text{Hz}.\end{aligned}$$ with $\Lambda_{\text{\tiny DEC}}^{\text{\tiny OUT[ATM]}}$ the decoherence rate in the out-space \[atmosphere\]. Neutrinos travel through the atmosphere within $\sim10^{-4}\,$s. Using this data with the time-of-flights and energies in Table I, for the decoherence damping factor of the cosmogenic neutrinos, one finds: $\sim10^{-5}$. For solar neutrinos instead, one gets: $\sim10^{-18}$, thus the damping of solar neutrino oscillations is hardly detectable, in agreement with experimental results [@decoexp; @vogel].
This estimate shows that, since environmental decoherence on neutrino oscillations is much stronger than the CSL collapse effect, these spontaneous collapse effects cannot be observed experimentally, even if the technology were sophisticated enough to reach such sensitivities. They would anyhow be masked by decoherence effects. Moreover, decoherence effects are not far away from the collapse effect predicted in [@Ch], which—as we argued—is anyhow overestimated. Therefore, also gravity-induced collapse effects cannot be detected via neutrino oscillations.
The rest of the paper is devoted to deriving Eqs. and . The calculation is lengthy but instructive because it shows, as already stressed in the introduction, how the collapse, which acts on the spatial part of the wave function, as a byproduct also affects the flavour degrees of freedom. Moreover, a precise calculation clears any possible misunderstanding about the effect of collapse models on neutrino oscillations.
Mathematical setup
==================
Working with non-linear equations such as the CSL equation is notoriously difficult. As shown e.g. in [@Im], the experimentally testable predictions of the model—when averaged over the noise—do not change if the real noise $W_{t}\left(\mathbf{x}\right)$ is replaced by an imaginary noise $i W_{t}\left(\mathbf{x}\right)$. In this way, one loses the collapse properties of the equation. However, the advantage of having an imaginary noise is that the evolution is described by a standard linear Schrödinger equation with a random Hamiltonian: $$\label{eq:htot}
H_{\text{\tiny TOT}} = H - \hbar \sqrt{\gamma} \sum_j \frac{m_{j}}{m_{0}}
\sum_s\int d\mathbf{y}\, w(\mathbf{y},t)
\psi_{j}^{\dagger}(\mathbf{y},s) \psi_{j}(\mathbf{y},s),$$ where $$\label{eq:sfsoi}
w(\mathbf{y},t) = \int d\mathbf{x}\, g(\mathbf{x-y})\xi_{t}(\mathbf{x}),$$ and $\xi_{t}(\mathbf{x}) = dW_{t}(\mathbf{x})/dt$ is a white noise field, with mean equal to zero and correlation function $\mathbb{E}[ \xi_{t}(\mathbf{x}) \xi_{s}(\mathbf{y})] = \delta(t-\nobreak s) \delta({\bf x-y})$. As such, $w(\mathbf{x},t)$ is a Gaussian noises field, with zero mean and correlation function: $$\label{eq:sdfddas}
{\mathbb E}[w(\mathbf{x},t) w(\mathbf{y},s)] \; = \;
\delta(t-s)F({\bf x} - {\bf y}), \qquad F({\bf x}) \; = \;
\frac{1}{(\sqrt{4 \pi} r_C)^3} e^{-{\bf x}^2/4 r_C^2}.$$
We are interested in the relativistic generalization of the Hamiltonian (\[eq:htot\]). The most natural choice is: $H_{\text{\tiny TOT}} = H_{\text{\tiny D}} + N(t)$ where (in the case of just one type of particle): $$H_{\text{\tiny D}}=\int d\mathbf{x}\mathcal{H}_{\text{\tiny D}}\left(x\right)=\int d\mathbf{x}\, \psi^{\dagger}\left(x\right)\left[-i\hbar c\,{\vec{\alpha}}\cdot\vec{\nabla}+mc^{2}{\beta}\right]\psi\left(x\right)$$ is the standard Dirac Hamiltonian. Here we have introduced the four-vector notation $x \equiv (ct, {\bf x})$, $\psi\left(x\right)$ is the Dirac spinor field, $c$ the speed of light, $m$ the mass of the particle associated to this field, $\vec{\alpha} \equiv (\alpha^1, \alpha^2, \alpha^3)$ with $\alpha^{i}=\gamma^{0}\gamma^{i}$ and $\beta=\gamma^{0}$ where the $\gamma^{\mu}$ are the Dirac matrices, which we take in their standard representation: $$\gamma^{0}=\left(\begin{array}{cc}
1 & 0\\
0 & -1\end{array}\right),\qquad\gamma^{i}=\left(\begin{array}{cc}
0 & \sigma^{i}\\
-\sigma^{i} & 0\end{array}\right)\qquad\textrm{with}\;\;\; i=1,2,3.$$ where $1$ is the identity matrix in two dimensions and $\sigma^{i}$ are the Pauli matrices. The noise term instead is given by: $$\label{eq:dfgdf}
N\left(t\right)=\int d\mathbf{x}\,\mathcal{N}\left(x\right)=-\hbar\sqrt{\gamma_{m}}\int d\mathbf{x}w\left(x\right)\overline{\psi}\left(x\right)\psi\left(x\right), \qquad \gamma_{m}=\gamma\left(\frac{m}{m_{0}}\right)^{2}$$ It can be shown[^3] that the Hamiltonian $H_{\text{\tiny TOT}} = H_{\text{\tiny D}} + N(t)$ defined in this way, in the non relativistic limit, reduces to the Hamiltonian of Eq. (\[eq:htot\]).
We chose to treat neutrinos as Dirac particles, despite the fact that is not yet known if they are Dirac or Majorana particles. We expect that the size of the collapse effect does not change significantly if Majorana fields are used in place of Dirac fields. As a matter of fact, in [@Kaons] the effect of the CSL model on kaon oscillations formula was studied. Despite the fact that kaons are different from neutrinos and that in [@Kaons] they were studied in the non-relativistic regime, the result is the same as that of this paper: a damping factor in front of the oscillating term, with a decay rate equivalent of to Eq. when the non-relativistic limit is taken.
Working with plane waves gives rise to unphysical infinities, since they are not normalizable. To avoid potential problems, we use the box normalization, i.e. we confine our fields in a box of length $L$, and we impose periodic boundary conditions: $\psi\left(t,\mathbf{x}\right)=\psi\left(t,\mathbf{x}+\mathbf{L}\right)$, where $\mathbf{L}$ is a vector with all the components equal to $L$. In turn, the momentum is discretized: ${\bf p}=\frac{2\pi\hbar}{L}{\bf k}$ with $k^{i}\in\mathbb{Z}$ and $i=1,2,3$ labels the spatial components. Then the Dirac field, in the interaction picture where we choose $H_{\text{\tiny D}}$ as the unperturbed Hamiltonian and $N\left(t\right)$ as the perturbation (the noise coupling $\sqrt{\gamma_{m}}$ is very small), takes the usual expression[^4]: $$\label{eq:wave}
\psi_{I}\left(x\right)=\sum_{s=1}^{2}\sum_{\mathbf{k}=-\infty}^{+\infty}\frac{1}{\sqrt{L^{3}}}\sqrt{\frac{mc^{2}}{E_{p}}}\left[b\left(p,s\right)u\left(p,s\right)e^{-\frac{i}{\hbar}E_{p}t+i\frac{2\pi}{L}\mathbf{k}\cdot\mathbf{x}}+d^{\dagger}\left(p,s\right)v\left(p,s\right)e^{\frac{i}{\hbar}E_{p}t-i\frac{2\pi}{L}\mathbf{k}\cdot\mathbf{x}}\right],$$ where $u$ and $v$ are the usual Dirac spinors, $E_{p}=\sqrt{\mathbf{p}^{2}c^{2}+m^{2}c^{4}}$ is the energy and $b$ and $d$ are operators satisfying the standard anti-commutation relations. We also recall the relation between the evolution operator $U(t)$ in the Schrödinger picture and $U_{I}\left(t\right)$ in the interaction picture [@Sakurai]: $$\label{evolution}
U\left(t\right)=e^{-\frac{i}{\hbar}H_{\text{\tiny D}}t}U_{I}\left(t\right),$$ and we set the initial time to 0.
One can question why we use a relativistic Hamiltonian in a model which is not relativistic, since the correlation function of the noise is not Lorentz-invariant. Our approach to this issue, is that collapse models are phenomenological models emerging from a pre-quantum theory yet to be discovered. The noise field is a real cosmological field (whose nature is yet to be investigated) which naturally defines a privileged frame, most likely corresponding to the co-moving frame of the universe. Hence we see no contradiction in analyzing relativistic phenomena with the CSL model. For attempts towards a fully relativistic formulation of collapse models, one can refer to [@Tu; @Bedingham].
The transition amplitude
========================
In the more general approach to the problem of neutrino oscillations, we consider $n$ flavour eigenstates, which will be labeled by greek subscripts $\left|\nu_{\alpha}\right\rangle$ and each of them is a linear combination of $n$ mass eigenstates that will be labeled by latin subscripts $\left|\nu_{j}\right\rangle$: $$\left|\nu_{\alpha}\right\rangle =\sum_{j=1}^{n}\text{U}_{\alpha j}\left|\nu_{j}\right\rangle.$$ Here, $\text{U}$ is the $n\times n$ mixing matrix, which relates the two different bases; since the flavour eigenstates are supposed to be orthonormal, as well as the mass eigenstates, $\text{U}$ must be unitary.
We take a neutrino in an initial flavour eigenstate, and compute the probability of finding it in another flavour eigenstate, after some time $t$, assuming that the dynamics is governed by the Hamiltonian $H_{\text{\tiny TOT}}$. We assume that the neutrino has definite initial and final momenta. This means that its initial and final states are plane waves and that both mass eigenstates have the same momenta. As discussed in [@Beuthe], in order to have a more consistent description, one should use wave packets instead of plane waves. However, a wave packet analysis goes beyond the scope of this paper: it would make the calculations much more difficult, and the expected result should not be much different from the one here derived. Moreover, in the standard treatment of neutrino oscillations, the plane wave analysis already gives a satisfactory description, to some degree, both in vacuum and in matter. Mathematically, we will compute the following quantity: $$\label{eq:Tab}
T_{\alpha\rightarrow\beta}\equiv\left\langle \nu_{\beta};\mathbf{p}_{f},s_{f}\left|U\left(t\right)\right|\nu_{\alpha};\mathbf{p}_{i},s_{i}\right\rangle =\sum_{i,j=1}^{n}\text{U}_{\alpha j}\text{U}_{\beta i}^{*}\left\langle \nu_{i};\mathbf{p}_{f},s_{f}\left|U\left(t\right)\right|\nu_{j};\mathbf{p}_{i},s_{i}\right\rangle$$ here $U\left(t\right)$ is the time-evolution operator of Eq. while $\left|\nu_{\alpha};\mathbf{p}_{i},s_{i}\right\rangle$ is the flavour eigenstate $\alpha$ with initial momentum $\mathbf{p}_{i}$ and spin $s_{i}$ and $\left|\nu_{\beta};\mathbf{p}_{f},s_{f}\right\rangle$ is the final state with momentum $\mathbf{p}_{f}$ and spin $s_{f}$. Since the Hamiltonian is the sum of Hamiltonians associated to different mass eigenstates ($H=\sum_{j=1}^{n}H_{j}$), it is convenient to expand the flavour eigenstates into the mass eigenstates, as we did in Eq. (\[eq:Tab\]). The form of the Hamiltonian also implies that $U(t)$ factorizes: $U\left(t\right)=\prod_{k=1}^{n}U_{k}\left(t\right)$. Here $U_{k}\left(t\right)$ is the time evolution operator related to the Fock space of the neutrino having a definite mass $m_{k}$. This is an important property, because it implies that if $i\neq j$: $$\begin{aligned}
\langle \nu_{i};\mathbf{p}_{f},s_{f}|U(t)|\nu_{j};\mathbf{p}_{i},s_{i}\rangle & = & \left\langle \Omega_{1}\left|U_{1}\left(t\right)\right|\Omega_{1}\right\rangle ...\left\langle \nu_{i};\mathbf{p}_{f},s_{f}\left|U_{i}\left(t\right)\right|\Omega_{i}\right\rangle ...\left\langle \Omega_{j}\left|U_{j}\left(t\right)\right|\nu_{j};\mathbf{p}_{i},s_{i}\right\rangle... \nonumber \\
& & ...\left\langle \Omega_{n}\left|U_{n}\left(t\right)\right|\Omega_{n}\right\rangle =0\end{aligned}$$ since $\left\langle \nu_{i};\mathbf{p}_{f},s_{f}\left|U_{i}\left(t\right)\right|\Omega_{i}\right\rangle =0$, as one can check with a direct calculation. Therefore we can write: $$\label{eq:cfhgfd}
T_{\alpha\rightarrow\beta} =\sum_{j=1}^{n}\text{U}_{\alpha j}\text{U}_{\beta j}^{*}\left[\left\langle \Omega_{1}\left|U_{1}\left(t\right)\right|\Omega_{1}\right\rangle ...\left\langle \nu_{j};\mathbf{p}_{f},s_{f}\left|U_{j}\left(t\right)\right|\nu_{j};\mathbf{p}_{i},s_{i}\right\rangle ....\left\langle \Omega_{n}\left|U_{n}\left(t\right)\right|\Omega_{n}\right\rangle \right],$$ which reduces the entire calculation to a 1-particle computation. In the next section we will focus our attention on the matrix element $\left\langle \nu_{j};\mathbf{p}_{f},s_{f}\left|U_{j}\left(t\right)\right|\nu_{j};\mathbf{p}_{i},s_{i}\right\rangle$ since, as we will show, the remaining terms contribute with an unimportant global phase factor.
The matrix elements
===================
We now focus on the main part of this work. What we need to compute the 1-particle matrix element: $$T\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)
\equiv
\left\langle \mathbf{p}_{f},s_{f}\left|U\left(t\right)\right|\mathbf{p}_{i},s_{i}\right\rangle = e^{-\frac{i}{\hbar}E_{f}t}\left\langle \mathbf{p}_{f},s_{f}\left|U_{I}\left(t\right)\right|\mathbf{p}_{i},s_{i}\right\rangle.$$ Since this part of the computation is the same for every mass eigenstate, we have dropped the label $j$. We expand the evolution operator by means of the Dyson series up to the second order: $$U_{I}\left(t,0\right)\simeq1-\frac{i}{\hbar}\int_{0}^{t}dt_{1}:N_{I}\left(t_{1}\right):-\frac{1}{\hbar^{2}}\int_{0}^{t}dt_{1}\int_{0}^{t_{1}}dt_{2}:N_{I}\left(t_{1}\right)::N_{I}\left(t_{2}\right):,$$ where $N_{I}\left(t\right)$ is the interaction picture representation of Eq. (\[eq:dfgdf\]) and $: ... :$ denotes the normal ordering[^5]. Accordingly, the transition probability is the sum of three terms: $$\label{eq:uioi}
T\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)=e^{-\frac{i}{\hbar}E_{f}t}\left[T^{\left(0\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)+T^{\left(1\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)+T^{\left(2\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)\right],$$ corresponding to the first three terms of the Dyson series. We now give a pictorial representation of each term by means of Feynman diagram and we compute each of them. The first term corresponds to the free propagation: $$\begin{array}{cclcclccl}
\displaystyle T^{\left(0\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right) & = & \Diagram{\vertexlabel^{i}fsfAfs\vertexlabel^{f}}
\end{array}$$ where the solid line represent the particle. This term is trivial: $$\label{eq:cvhjcg}
T^{\left(0\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)
\equiv
\left\langle \mathbf{p}_{f},s_{f}|\mathbf{p}_{i},s_{i}\right\rangle =\delta_{s_{f}s_{i}}\delta_{\mathbf{p}_{f},\mathbf{p}_{i}}.$$ The second term correspond to the diagram: $$\begin{array}{cclcclccl}
\displaystyle T^{\left(1\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right) & = & \Diagram{\vertexlabel^ifA\vertexlabel_1 hu \\ f0 fdA\vertexlabel^{\;\;f}}
\end{array}$$ where the dotted line represents the noise field. This term is: $$\label{eq:T1}
T^{\left(1\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)
\equiv
i\sqrt{\gamma_{m}}\int_{0}^{t}dt_{1}\int d\mathbf{x}_{1}w\left(x_{1}\right)\langle \mathbf{p}_{f},s_{f}|:\bar{\psi}_{I}\left(x_{1}\right)\psi_{I}\left(x_{1}\right):|\mathbf{p}_{i},s_{i}\rangle.$$ In order to compute the matrix element in Eq. (\[eq:T1\]), we use the series expansion of the fields as given in Eq. (\[eq:wave\]). The non-null terms are those containing two $b$ and two $b^{\dagger}$ operators. After some calculations, one finds that: $$\begin{aligned}
\label{eq:T1me}
\lefteqn{\langle \Omega|b\left(p_{f},s_{f}\right):\bar{\psi}_{I}\left(x_{1}\right)\psi_{I}\left(x_{1}\right):b^{\dagger}\left(p_{i},s_{i}\right)|\Omega\rangle =} \nonumber \\
& = & \sum_{s,s'=1}^{2}\sum_{\mathbf{p},\mathbf{p'}=-\infty}^{+\infty}\frac{1}{L^{3}}\frac{mc^{2}}{\sqrt{E_{p}E_{p'}}}e^{\frac{i}{\hbar}\left(p'^{\mu}-p^{\mu}\right)x_{1\mu}}\overline{u}\left(p',s'\right)u\left(p,s\right)\langle \Omega|b\left(p_{f},s_{f}\right)b^{\dagger}\left(p',s'\right)b\left(p,s\right)b^{\dagger}\left(p_{i},s_{i}\right)|\Omega\rangle \nonumber \\
& = & \frac{1}{L^{3}}\frac{mc^{2}}{\sqrt{E_{i}E_{f}}}e^{\frac{i}{\hbar}\left(p_{f}^{\mu}-p_{i}^{\mu}\right)x_{1\mu}}\overline{u}\left(p_{f},s_{f}\right)u\left(p_{i},s_{i}\right). \end{aligned}$$ Here we introduced the four momentum $p^{\mu}=\left(E_{p}/c,\mathbf{p}\right)$. If we substitute Eq. (\[eq:T1me\]) in the definition of $T^{\left(1\right)}$, we get: $$\label{eq:T1again}
T^{\left(1\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)=i\sqrt{\gamma_{m}}\frac{mc^{2}}{\sqrt{E_{i}E_{f}}}\overline{u}\left(p_{f},s_{f}\right)u\left(p_{i},s_{i}\right)\frac{1}{L^{3}}\int_{0}^{t}dt_{1}\int d\mathbf{x}_{1}w\left(x_{1}\right)e^{\frac{i}{\hbar}\left(p_{f}^{\mu}-p_{i}^{\mu}\right)x_{1\mu}}.$$
The last term in Eq. (\[eq:uioi\]) is the more complicated to compute, since it involves the product of four fields. It gives the following contribution: $$\begin{aligned}
\lefteqn{T^{\left(2\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)
\equiv
-\frac{1}{\hbar^{2}}\langle \mathbf{p}_{f},s_{f}|\left[\int_{0}^{t}dt_{1}\int_{0}^{t_{1}}dt_{2}:N_{I}\left(t_{1}\right)::N_{I}\left(t_{2}\right):\right]|\mathbf{p}_{i},s_{i}\rangle =} \\
& = & -\frac{\gamma_{m}}{2}\int_{0}^{t}\!dt_{1}dt_{2}\int\! d\mathbf{x}_{1} d\mathbf{x}_{2}\,w\left(x_{1}\right)w\left(x_{2}\right) \langle \mathbf{p}_{f},s_{f}|T\left[:\bar{\psi}_{I}\left(x_{1}\right)\psi_{I}\left(x_{1}\right)::\bar{\psi}_{I}\left(x_{2}\right)\psi_{I}\left(x_{2}\right):\right]|\mathbf{p}_{i},s_{i}\rangle, \nonumber\end{aligned}$$ here “$T$” is the time-ordering product. Using Wick’s theorem, and discarding all tadpole terms, which involve a contraction between two fields at the same spacetime point, we have: $$\begin{aligned}
\label{eq:dsfdfbv}
T\left[\bar{\psi}_{1a}\psi_{1a}\bar{\psi}_{2b}\psi_{2b}\right]
& = & :\bar{\psi}_{1a}\psi_{1a}\bar{\psi}_{2b}\psi_{2b}:-S_{ab}\left(x_{1}-x_{2}\right)S_{ba}\left(x_{2}-x_{1}\right) \nonumber \\
& & +iS_{ab}\left(x_{1}-x_{2}\right):\bar{\psi}_{1a}\psi_{2b}:-iS_{ba}\left(x_{2}-x_{1}\right):\psi_{1a}\bar{\psi}_{2b}:,\end{aligned}$$ where $a$ and $b$ label the spinor components and the Dirac propagator is: $ iS_{ab}\left(x_{1}-x_{2}\right)\equiv\langle \Omega|T\left[\psi_{1a}\bar{\psi}_{2b}\right]|\Omega\rangle$. Here, Einstein’s summation convention is used for the spinor indices. We momentarily drop the pedex $I$ related to the interaction picture and we write the dependence on $x_{1}$ and $x_{2}$ simply as $_{1}$ and $ _{2}$. The diagramatic representation of the different terms in Eq. (\[eq:dsfdfbv\]) is: $$\begin{array}{cclcclccl}
\Diagram{\vertexlabel^ifA\vertexlabel_1 hu \\ f0 fdA\vertexlabel^{\;\;f}}\;\;\;\;\Diagram{\vertexlabel^ifA\vertexlabel_2 hu \\ f0 fdA\vertexlabel^{\;\;f}} & = & \displaystyle :\bar{\psi}_{1a}\psi_{1a}\bar{\psi}_{2b}\psi_{2b}:\;, \qquad &
\feyn{h\vertexlabel_1 f0 {flA}{}
{fluV}{} f0 \vertexlabel_2h} & = & \displaystyle -S_{ab}\left(x_{1}-x_{2}\right)S_{ba}\left(x_{2}-x_{1}\right)\;,
\end{array}$$ $$\begin{array}{cclcclccl}
\Diagram{\vertexlabel_i hd\vertexlabel_{2} fA \vertexlabel_{1}
hu\\
fuA f0 fdA \vertexlabel^{\;\;f}} & = & \displaystyle iS_{ab}\left(x_{1}-x_{2}\right):\bar{\psi}_{1a}\psi_{2b}:\;,\qquad\qquad\qquad
\Diagram{\vertexlabel_i hd\vertexlabel_{1} fA \vertexlabel_{2}
hu\\
fuA f0 fdA \vertexlabel^{\;\;f}} & = & \displaystyle -iS_{ba}\left(x_{2}-x_{1}\right):\psi_{1a}\bar{\psi}_{2b}:\;.
\end{array}$$ We can easily see that the first term is zero, since we are studying the case with only one particle in the initial and final states. Regarding the second diagram, an important issue here arises. This diagam represents a vacuum fluctuation term, which is divergent. As well known [@GreinerFIELD], all vacuum fluctuations diagrams of any order sum up to a phase factor $\left\langle \Omega\left|U_{I}\left(t,0\right)\right|\Omega\right\rangle$, and all divergences cancel each other. Therefore we can write: $$\left\langle \mathbf{p}_{f},s_{f}\left|U_{I}\left(t,0\right)\right|\mathbf{p}_{i},s_{i}\right\rangle =\left\langle \Omega\left|U_{I}\left(t,0\right)\right|\Omega\right\rangle \cdot \left\langle \mathbf{p}_{f},s_{f}\left|U_{I}\left(t,0\right)\right|\mathbf{p}_{i},s_{i}\right\rangle _{\textrm{ext}},$$ where $\left\langle \mathbf{p}_{f},s_{f}\left|U_{I}\left(t,0\right)\right|\mathbf{p}_{i},s_{i}\right\rangle_{\textrm{ext}}$ denotes the contribution from diagrams with external fermionic legs. This vacuum fluctuation term is important because, together with those of Eq. (\[eq:cfhgfd\]), it gives a global phase $\prod_{k=1}^{n}\left\langle \Omega_{k}\left|U_{kI}\left(t,0\right)\right|\Omega_{k}\right\rangle$ independent of $j$, which factorizes out of the sum. Therefore, such terms are physically unimportant, the only relevant part being $\left\langle \mathbf{p}_{f},s_{f}\left|U_{I}\left(t,0\right)\right|\mathbf{p}_{i},s_{i}\right\rangle_{\textrm{ext}}$. From now on, we will work only with diagrams with external fermionic legs, and we drop the pedex “ext”.
Coming back to Eq. (\[eq:dsfdfbv\]), we can now focus our attention on the third and the fourth term, that correspond to the last two diagrams. Since $:\psi_{2b}\bar{\psi}_{1a}:\; = - :\bar{\psi}_{1a}\psi_{2b}:$ for fermions, these two terms give the same contribution. The Dirac propagator reads: $$\begin{aligned}
iS_{ab}\left(x_{1}-x_{2}\right)
& =&
\sum_{s=1}^{2}\sum_{\mathbf{p}=-\infty}^{+\infty}\frac{1}{L^{3}}\frac{mc^{2}}{E_{p}}\left\{ \theta\left(t_{1}-t_{2}\right)e^{-\frac{i}{\hbar}p^{\mu}\left(x_{1\mu}-x_{2\mu}\right)}u_{a}\left(p,s\right)\overline{u}_{b}\left(p,s\right) \right. \nonumber \\
& & \left.\qquad\qquad\qquad\quad\;\; -\theta\left(t_{2}-t_{1}\right)e^{-\frac{i}{\hbar}p^{\mu}\left(x_{2\mu}-x_{1\mu}\right)}\overline{v}_{b}\left(p,s\right)v_{a}\left(p,s\right)\right\}, \end{aligned}$$ while the matrix element gives: $$\langle \mathbf{p}_{f},s_{f}|:\bar{\psi}_{1a}\psi_{2b}:|\mathbf{p}_{i},s_{i}\rangle
=
\frac{1}{L^{3}}\frac{mc^{2}}{\sqrt{E_{f}E_{i}}}e^{\frac{i}{\hbar}p_{f}^{\mu}x_{1\mu}}e^{-\frac{i}{\hbar}p_{i}^{\mu}x_{2\mu}}\overline{u}_{a}\left(p_{f},s_{f}\right)u_{b}\left(p_{i},s_{i}\right).$$ In the following we will need only the case $\mathbf{p}_{f}=\mathbf{p}_{i}$ and $s_{f}=s_{i}$. In this case, we have: $$\begin{aligned}
T^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};\mathbf{p}_{i},s_{i};t\right)
& = &
-\gamma_{m}\frac{1}{L^{3}}\frac{m c^{2}}{E_{i}}\sum_{s=1}^{2}\sum_{\mathbf{p}=-\infty}^{+\infty}\frac{1}{L^{3}}\frac{m c^{2}}{E_{p}} \nonumber \\
& &
\cdot\left\{ \int_{0}^{t}dt_{1}\int_{0}^{t_{1}}dt_{2}\int d\mathbf{x}_{1}\int d\mathbf{x}_{2}\, w(x_{1})w(x_{2})e^{\frac{i}{\hbar}(p_{i}^{\mu}-p^{\mu})\left(x_{1\mu}-x_{2\mu}\right)}\right. \nonumber \\
& & \cdot \; \overline{u}_{a}(p_{i},s_{i})u_{a}(p^,s)\overline{u}_{b}(p^,s) u_{b}(p_{i},s_{i}) \nonumber \\
& &
-\int_{0}^{t}dt_{2}\int_{0}^{t_{2}}dt_{1}\int d\mathbf{x}_{1}\int d\mathbf{x}_{2}w\left(x_{1}\right)w\left(x_{2}\right)e^{\frac{i}{\hbar}(p_{i}^{\mu}-p^{\mu})\left(x_{1\mu}-x_{2\mu}\right)} \nonumber \\
& & \cdot \left.
\overline{u}_{a}(p_{i},s_{i})v_{a}(p,s)\overline{v}_{b}(p,s)u_{b}(p_{i},s_{i})\phantom{\frac{1}{2}}\!\!\!\!\right\}.\end{aligned}$$ Using the standard relations [@Greiner]: $$\sum_{s=1}^{2}u_{a}(p,s)\overline{u}_{b}(p,s)=\left(\frac{p^{\mu}\gamma_{\mu}+mc}{2mc}\right)_{ab},\;\;\;\;\;\;\sum_{s=1}^{2}v_{a}(p,s)\overline{v}_{b}(p,s)=\left(\frac{p^{\mu}\gamma_{\mu}-mc}{2mc}\right)_{ab},$$ we can see that the terms containing $p^{\mu}\gamma_{\mu}$ cancel each other, while those containing the mass give a $\delta_{ab}$. Thus, if we also use: $\overline{u}(p_{i},s_{f})u(p_{i},s_{i})=\delta_{s_{f},s_{i}}$, we obtain: $$\begin{aligned}
\label{eq:fdhgfg}
T^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};\mathbf{p}_{i},s_{i};t\right)
& = &
-\gamma_{m}\frac{1}{L^{3}}\frac{mc^{2}}{E_{i}}\sum_{\mathbf{p}=-\infty}^{+\infty}\frac{1}{L^{3}}\frac{mc^{2}}{E_{p}}\int_{0}^{t}dt_{1}\int_{0}^{t_{1}}dt_{2}\int d\mathbf{x}_{1}\int d\mathbf{x}_{2}w\left(x_{1}\right)w\left(x_{2}\right) \nonumber \\
& & \times e^{\frac{i}{\hbar}(p_{i}^{\mu}-p^{\mu})\left(x_{1\mu}-x_{2\mu}\right)}.\end{aligned}$$ Now we have all the elements we need, in order to compute the transition probability. We will do this in the next section.
The transition probability
==========================
The physical quantity we are interested in, is the transition probability, which corresponds to $\left|T_{\alpha\rightarrow\beta}\right|^{2}$, averaged over the noise, and integrated over the final momentum and polarization states: $$\label{eq:sfdso}
P_{\alpha\rightarrow\beta}
\equiv
\sum_{s_{f}}\sum_{\mathbf{p}_{f}=-\infty}^{+\infty}\mathbb{E}\left|T_{\alpha\rightarrow\beta}\right|^{2}
=
\sum_{k=1}^{n}\sum_{j=1}^{n}\text{U}_{\alpha k}^{*}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j}^{*}P_{kj}\left(\mathbf{p}_{i},s_{i};t\right),$$ where: $$P_{kj}\left(\mathbf{p}_{i},s_{i};t\right)\equiv\sum_{s_{f}}\sum_{\mathbf{p}_{f}=-\infty}^{+\infty}e^{\frac{i}{\hbar}(E_{f}^{\left(k\right)}-E_{f}^{\left(j\right)})t}\,\mathbb{E}\left[T_{k}^{*}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right) T_{j}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)\right],$$ and $T_{j}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)$ is given by Eq. (\[eq:uioi\]), where now we have explicitly indicated the label $j$ associated to the mass eigenstate $m_{j}$ and $E_{f}^{\left(j\right)}=\sqrt{\mathbf{p}_f^{2}c^{2}+m_j^{2}c^{4}}$. When averaging, one has to remember that only terms containing an even number of noises survive (in the Feynman representation, all products of diagrams containing an even number of dotted legs). Using this fact, and exploiting the Kronecher deltas of $T^{\left(0\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)$ (see Eq. (\[eq:cvhjcg\])), we can write: $$\label{eq:sdgrst}
P_{kj}\left(\mathbf{p}_{i},s_{i};t\right)
=
e^{\frac{i}{\hbar}(E_{i}^{\left(k\right)}-E_{i}^{\left(j\right)})t}\left[1+ I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right) + I_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};t\right)+I_{k}^{\left(2\right)*}\left(\mathbf{p}_{i},s_{i};t\right)\right],$$ where we have defined: $$\begin{aligned}
\label{eq:rtrtsfds}
I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right)
& \equiv &
\sum_{s_{f}}\sum_{\mathbf{p}_{f}=-\infty}^{+\infty}e^{\frac{i}{\hbar}(E_{f}^{\left(k\right)}-E_{i}^{\left(k\right)}-E_{f}^{\left(j\right)}+E_{i}^{\left(j\right)})t}\mathbb{E}\left[T_{k}^{\left(1\right)*}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)T_{j}^{\left(1\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)\right], \nonumber \\
I_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};t\right)
& \equiv &
\mathbb{E}\left[T_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};\mathbf{p}_{i},s_{i};t\right)\right].\end{aligned}$$ We focus our attention on $I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right)$. Using Eq. (\[eq:T1again\]), keeping in mind the spinor relation $\left(\overline{u}_{f}u_{i}\right)^{*}=\overline{u}_{i}u_{f}$, and performing the average over the noise, which brings in a Dirac delta in time which cancels one of the two time-integrals, one obtains: $$\begin{aligned}
\label{eq:sdfgdr}
\lefteqn{\mathbb{E}\left[T_{k}^{\left(1\right)*}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)T_{j}^{\left(1\right)}\left(\mathbf{p}_{f},s_{f};\mathbf{p}_{i},s_{i};t\right)\right]=} \qquad\qquad\nonumber \\
& = &
\sqrt{\gamma_{m_{j}}\gamma_{m_{k}}}\frac{m_{j}m_{k}c^{4}}{\sqrt{E_{i}^{\left(j\right)}E_{f}^{\left(j\right)}E_{i}^{\left(k\right)}E_{f}^{\left(k\right)}}}\overline{u}(p_{f}^{(j)},s_{f})u(p_{i}^{(j)},s_{i})\overline{u}(p_{i}^{(k)},s_{i})u(p_{f}^{(k)},s_{f}) \nonumber \\
& &
\cdot \int_{0}^{t}dt_{1}e^{\frac{i}{\hbar}(E_{f}^{\left(j\right)}-E_{i}^{\left(j\right)}-E_{f}^{\left(k\right)}+E_{i}^{\left(k\right)})t_{1}}S\left(\mathbf{p}_{i},\mathbf{p}_{f}\right)\end{aligned}$$ where $p_{f}^{(j)} \equiv (E_{f}^{(j)}/c, {\bf p}_{f})$ and similarly for $p_{i}^{(j)}$. Moreover: $$\label{eq:zxgf}
S\left(\mathbf{p}_{i},\mathbf{p}_{f}\right)
\equiv
\frac{1}{L^{6}}\int_{-\frac{L}{2}}^{+\frac{L}{2}}d\mathbf{x}_{1}\int_{-\frac{L}{2}}^{+\frac{L}{2}}d\mathbf{x}_{2}\frac{e^{-\left(\mathbf{x}_{1}-\mathbf{x}_{2}\right)^{2}/4r_{C}^{2}}}{\left(\sqrt{4\pi}r_{C}\right)^{3}}e^{-\frac{i}{\hbar}\left[\left(\mathbf{p}_{f}-\mathbf{p}_{i}\right)\cdot\left(\mathbf{x}_{1}-\mathbf{x}_{2}\right)\right]}$$ (now we have explicitly indicated the integration volume). In order to compute $S\left(\mathbf{p}_{i},\mathbf{p}_{f}\right)$, we change integration variables as follows: $$\mathbf{y}=\left(\mathbf{x}_{1}+\mathbf{x}_{2}\right)\qquad\textrm{and}\qquad\mathbf{x}=\left(\mathbf{x}_{1}-\mathbf{x}_{2}\right),$$ and use the relation: $$\int_{-\frac{L}{2}}^{+\frac{L}{2}}dx_{1}\int_{-\frac{L}{2}}^{+\frac{L}{2}}dx_{2}f\left(x_{1},x_{2}\right)=\frac{1}{2}\int_{0}^{+L}dx\int_{-\left(L-x\right)}^{+\left(L-x\right)}dy\left[f\left(x,y\right)+f\left(-x,y\right)\right].$$ Accordingly, we have: $$S\left(\mathbf{p}_{i},\mathbf{p}_{f}\right)=\frac{1}{L^{3}}\int_{0}^{L}d\mathbf{x}\frac{e^{-\mathbf{x}^{2}/4r_{C}^{2}}}{\left(\sqrt{4\pi}r_{C}\right)^{3}}2\cos\left[\frac{1}{\hbar}\left(\mathbf{p}_{f}-\mathbf{p}_{i}\right)\cdot\mathbf{x}\right]\frac{1}{2^{3}}\prod_{i=1}^{3}2\left(1-\frac{x_{i}}{L}\right)$$ Let us now take the limit $L \rightarrow \infty$, which amounts to making the replacement: $$\sum_{\mathbf{p}_{f}=-\infty}^{+\infty}\longrightarrow\int d\mathbf{p}_{f}\qquad\textrm{and}\qquad\frac{1}{L^{3}}\longrightarrow\frac{1}{\left(2\pi\hbar\right)^{3}}.$$ In this limit, the term $x_i/L$ gives a vanishingly small contribution. Therefore we can write: $$S\left(\mathbf{p}_{i},\mathbf{p}_{f}\right)
=
\frac{1}{\left(2\pi\hbar\right)^{3}} \int_{-\infty}^{+\infty}d\mathbf{x}\frac{e^{-\mathbf{x}^{2}/4r_{C}^{2}}}{\left(\sqrt{4\pi}r_{C}\right)^{3}}e^{\frac{i}{\hbar}(\mathbf{p}_{f}-\mathbf{p}_{i})\cdot\mathbf{x}}
=
\frac{1}{\left(2\pi\hbar\right)^{3}} e^{-\frac{(\mathbf{p}_{f}-\mathbf{p}_{i})^{2}r_{C}^{2}}{\hbar^{2}}}.$$ The time integral in Eq. (\[eq:sdfgdr\]) is trivial, and one arrives easily at the formula: $$\begin{aligned}
\label{eq:dasdasd}
I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right)
& = &
\sum_{s_{f}}\int d\mathbf{p}_{f}\sqrt{\gamma_{m_{j}}\gamma_{m_{k}}}\frac{m_{j}m_{k}c^{4}}{\sqrt{E_{i}^{\left(j\right)}E_{f}^{\left(j\right)}E_{i}^{\left(k\right)}E_{f}^{\left(k\right)}}}\overline{u}(p_{f}^{\left(j\right)},s_{f})u(p_{i}^{\left(j\right)},s_{i})\overline{u}(p_{i}^{\left(k\right)},s_{i})u(p_{f}^{\left(k\right)},s_{f}) \nonumber \\
& & \cdot
\frac{1-e^{\frac{i}{\hbar}(E_{f}^{\left(k\right)}-E_{i}^{\left(k\right)}-E_{f}^{\left(j\right)}+E_{i}^{\left(j\right)})t}}{\frac{i}{\hbar}(E_{f}^{\left(j\right)}-E_{i}^{\left(j\right)}-E_{f}^{\left(k\right)}+E_{i}^{\left(k\right)})}\frac{1}{\left(2\pi\hbar\right)^{3}}e^{-\frac{(\mathbf{p}_{f}-\mathbf{p}_{i})^{2}r_{C}^{2}}{\hbar^{2}}}.\end{aligned}$$ As it is shown in the Appendix C, the integrating function (except for the Gaussian term) changes slowly within the region where the Gaussian term is appreciably different from zero. Therefore we can approximate it with the value it takes in the center of the Gaussian (where $\mathbf{p}_{f}=\mathbf{p}_{i}$) and bring it out of the integral. Taking into account that $\overline{u}\left(p,s_{i}\right)u\left(p,s_{f}\right)=\delta_{s_{i}s_{f}}$ and performing the integration of the Gaussian part, Eq. (\[eq:dasdasd\]) takes the very simple expression: $$\label{eq:syksdgar}
I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right)
=
\sqrt{\gamma_{m_{j}}\gamma_{m_{k}}}\frac{m_{j}m_{k}c^{4}}{E_{i}^{\left(j\right)}E_{i}^{\left(k\right)}}\frac{t}{\left(2\pi\right)^{3}}\frac{\pi^{3/2}}{r_{C}^{3}}.$$
We now turn our attention to the term $I_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};t\right)$ in Eq. (\[eq:rtrtsfds\]). Substituting Eq. (\[eq:fdhgfg\]) in Eq. (\[eq:rtrtsfds\]), we have: $$I_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};t\right)
=
-\gamma_{m_{j}}\frac{m_{j}c^{2}}{E_{i}^{\left(j\right)}}\sum_{\mathbf{p}=-\infty}^{+\infty}\frac{m_{j}c^{2}}{E_{p}^{\left(j\right)}}\frac{t}{2}\frac{1}{L^{6}}\int d\mathbf{x}_{1}\int d\mathbf{x}_{2}\frac{e^{-\left(\mathbf{x}_{1}-\mathbf{x}_{2}\right)^{2}/4r_{C}^{2}}}{\left(\sqrt{4\pi}r_{C}\right)^{3}}e^{-\frac{i}{\hbar}\left(\mathbf{p}_{i}-\mathbf{p}\right)\left(\mathbf{x}_{1}-\mathbf{x}_{2}\right)}.$$ The spatial integrals are equal to $S\left(\mathbf{p},\mathbf{p}_{i}\right)$ defined in Eq. (\[eq:zxgf\]). Performing the same type of calculation as before, and taking the limit $L \rightarrow \infty$, one arrives at the result: $$\label{eq:vcvxvxcy}
I_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};t\right)
=
-\gamma_{m_{j}}\frac{m_{j}c^{2}}{E_{i}^{\left(j\right)}}\int d\mathbf{p}\frac{m_{j}c^{2}}{E_{p}^{\left(j\right)}}\frac{1}{2}t\frac{1}{\left(2\pi\hbar\right)^{3}}e^{-\frac{\left(\mathbf{p}-\mathbf{p}_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}.$$ Once again, one can show that the integrating function (Gaussian term apart) varies slowly within the region where the Gaussian function is appreciably different from zero. Therefore one can bring this function out of the integral, fixing its value at the center of the Gaussian ($\mathbf{p}=\mathbf{p}_{i}$), and perform the Gaussian integration. The final expression is: $$\label{eq:izgdj}
I_{j}^{\left(2\right)}\left(\mathbf{p}_{i},s_{i};t\right)
=
-\frac{\gamma_{m_{j}}}{2}\frac{m_{j}^{2}c^{4}}{E_{i}^{2\left(j\right)}}\frac{t}{\left(2\pi\right)^{3}}\frac{\pi^{3/2}}{r_{C}^{3}}.$$
Having computed explicitly all the terms of Eq. (\[eq:sdgrst\]), we can turn our attention to Eq. (\[eq:sfdso\]). It is convenient to split the sum of Eq. (\[eq:sfdso\]) in one part with $k=j$ and the another part with $k\neq j$: $$\begin{aligned}
\label{eq:ennesimaP}
P_{\alpha\rightarrow\beta}
& = &
\sum_{k=1}^{n}\text{U}_{\alpha k}^{*}\text{U}_{\beta k}\text{U}_{\alpha k}\text{U}_{\beta k}^{*}P_{kk}\left(\mathbf{p}_{i},s_{i};t\right) \nonumber \\
& + &
\sum_{{k=2\atop j<k}}^{n}\left[\text{U}_{\alpha k}^{*}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j}^{*}P_{kj}\left(\mathbf{p}_{i},s_{i};t\right)+\text{U}_{\alpha j}^{*}\text{U}_{\beta j}\text{U}_{\alpha k}\text{U}_{\beta k}^{*}P_{jk}\left(\mathbf{p}_{i},s_{i};t\right)\right]\,.\end{aligned}$$ Using the symmetry relation $I_{jk}^{\left(1\right)}=I_{kj}^{\left(1\right)*}$, which implies that $P_{kj}\left(\mathbf{p}_{i},s_{i};t\right)=P_{jk}^{*}\left(\mathbf{p}_{i},s_{i};t\right)$, one can rewrite Eq. as follows : $$\label{eq:dgglkjgh}
P_{\alpha\rightarrow\beta}=\sum_{k=1}^{n}\text{U}_{\alpha k}^{*}\text{U}_{\beta k}\text{U}_{\alpha k}\text{U}_{\beta k}^{*}+\sum_{{k=2\atop j<k}}^{n}2\textrm{Re}\left[\text{U}_{\alpha k}^{*}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j}^{*}P_{kj}\left(\mathbf{p}_{i},s_{i};t\right)\right]\,,$$ where we have exploited the identity: $
P_{kk}\left(\mathbf{p}_{i},s_{i};t\right)=1.
$ In the physically interesting case where the mixing elements $U_{\alpha k}$ are real, Eq. (\[eq:dgglkjgh\]) takes a very simple expression: $$\label{eq:dfggkds}
P_{\alpha\rightarrow\beta}
=
\sum_{k=1}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha k}\text{U}_{\beta k}+\sum_{{k \neq j}}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j} \; \left[1- \xi_{jk} t\right] \; \cos\left[\frac{1}{\hbar}(E_{i}^{\left(k\right)}-E_{i}^{\left(j\right)})t\right],$$ with: $$\label{eq:xi1}
\xi_{jk} \; \equiv \; \frac{1}{16\pi^{3/2}r_{C}^{3}}\left(\sqrt{\gamma_{m_{j}}}\frac{m_{j}c^{2}}{E_{i}^{\left(j\right)}}-\sqrt{\gamma_{m_{k}}}\frac{m_{k}c^{2}}{E_{i}^{\left(k\right)}}\right)^{2}= \frac{\gamma}{16\pi^{3/2}r_{C}^{3}m_{0}^{2}c^{4}}\left(\frac{m_{j}^{2}c^{4}}{E_{i}^{\left(j\right)}}-\frac{m_{k}^{2}c^{4}}{E_{i}^{\left(k\right)}}\right)^{2}\,.$$ As a check, one can easily see that the probability is conserved, i.e.: $$\sum_{\beta}P_{\alpha\rightarrow\beta}=1.$$ This is the result we wanted to arrive at. It shows that, also in collapse models, the number of particles is conserved, but the oscillations are damped according to Eq. (\[eq:dfggkds\]), by a factor equal to $\left[1- \xi_{jk} t\right] $. This is in perfect agreement[^6] with well established results concerning the effect of decoherence on oscillatory systems like those here considered [@Be; @Adlerdeco]. Our calculation gives an analytical expression for the damping rate $\xi_{jk}$, as predicted by the mass-proportional CSL model. Note that the calculation has been carried out to second perturbative order, which means $\xi_{jk} t \ll 1$. Therefore, the fact that the probability in Eq. (\[eq:dfggkds\]) becomes negative for $\xi_{jk} t > 1$ is of no concern, because this range of times goes beyond the limits of validity of the present result. Actually, one can try to stretch the above result beyond the second perturbative order, and guess the following expression for the transition probability: $$\label{eq:dfggkdse1}
P_{\alpha\rightarrow\beta}
=
\sum_{k=1}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha k}\text{U}_{\beta k}+\sum_{{k \neq j}}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j} \; e^{-\xi_{jk}t} \; \cos\left[\frac{1}{\hbar}(E_{i}^{\left(k\right)}-E_{i}^{\left(j\right)})t\right].$$
The above results can be easily generalized to oscillatory systems, which decay in time. On the phenomenological level, one takes the decay into account by adding an imaginary term to the Hamiltonian: $$H\longrightarrow H-\frac{i}{2}\Gamma.$$ The calculation remains unaltered, and one arrives at the final result: $$\begin{aligned}
\label{eq:dfggkdsss}
P_{\alpha\rightarrow\beta}
& = &
\sum_{k=1}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha k}\text{U}_{\beta k} \; e^{-\frac{\Gamma^{\left(k\right)}}{\hbar}t} \nonumber \\
& + & \sum_{{k=2\atop j<k}}^{n}\text{U}_{\alpha k}\text{U}_{\beta k}\text{U}_{\alpha j}\text{U}_{\beta j} \; \left[1- \xi_{jk} t\right] \; e^{-\frac{\Gamma^{\left(k\right)}+\Gamma^{\left(j\right)}}{2\hbar}t} \; 2\cos\left[\frac{1}{\hbar}(E_{i}^{\left(k\right)}-E_{i}^{\left(j\right)})t\right],\end{aligned}$$ which generalizes Eq. (\[eq:dfggkds\]) to decaying particles. This concludes our analysis.
Acknowledgements
================
The authors wish to thank S.L. Adler, M. Bahrami, A. Di Domenico and B. Hiesmayr for many useful and enjoyable conversations on this topic. They also acknowledge partial financial support from NANOQUESTFIT, INFN, the COST Action MP1006 “Fundamental problems in Quantum Physics" and the John Templeton Foundation project “Experimental and theoretical exploration of fundamental limits of quantum mechanics”.
APPENDIX A: Dimensional estimate of the decay rate {#appendix-a-dimensional-estimate-of-the-decay-rate .unnumbered}
==================================================
Here we wish to discuss about the possibility of estimate the predictions of the CSL model in neutrino oscillations by using dimensional analysis. Collapse models, as discussed in section IV, are described by the same type of master equations as open quantum systems (which experience decoherence due to interactions with an external environment). In such a case it is well known that the effect of decoherence is to suppress exponentially flavour oscillations. So it does not come as a surprise that for collapse models the effect is the same. Then one could try to guess the decay rate with dimensional analysis, using the relevant constants and parameters of the model. First of all is reasonable to suppose that the effect is proportional to the strength of the noise $\gamma$. Moreover, since we are using the mass proportional CSL model, for which $\gamma$ is replaced by $\gamma_{m_{j}}\equiv\gamma\left(\frac{m_{j}}{m_{0}}\right)^{2}$, one expects also a factor $m_{0}^{2}$ in the denominator. Because $\left[\gamma\right]=\textrm{cm}^{3}\textrm{s}^{-1}$, and the decay rate must have dimension $\textrm{s}^{-1}$, we need to introduce terms with dimension $\textrm{cm}^{-3}$. Since the parameter $r_{C}$ has the dimension of a length, then it is natural to introduce an $r_{C}^{3}$ in the denominator. Finally, we need to introduce terms with the dimension of a squared mass. The simplest choices are: $$\xi_{jk}^{(1)}\sim\frac{\gamma}{r_{C}^{3}m_{0}^{2}}\left(m_{j}-m_{k}\right)^{2}\;\;\;\;\;\textrm{or}\;\;\;\;\;\xi_{jk}^{(2)}\sim\frac{\gamma}{r_{C}^{3}m_{0}^{2}}\left(m_{j}^{2}-m_{k}^{2}\right).$$ Both these formulas are different compared to the correct one given by Eq. (\[eq:xi\]). If we substitute the values of the constants and the parameters and we consider, for example, the case of the cosmological neutrinos, we get: $$\begin{aligned}
\xi_{jk}^{(1)}t_{cosm}&\sim&\frac{\gamma}{r_{C}^{3}m_{0}^{2}}\left(m_{j}-m_{k}\right)^{2}t_{cosm}\sim10^{-17}\\
&& \nonumber\\
\xi_{jk}^{(2)}t_{cosm}&\sim&\frac{\gamma}{r_{C}^{3}m_{0}^{2}}\left(m_{j}^{2}-m_{k}^{2}\right)t_{cosm}\sim10^{-11}\end{aligned}$$ The formula derived with dimensional analysis shows that the CSL effect on neutrino oscillations is very small, practically undetectable. However, it differs by many orders of magnitude from the exact (perturbative) result. We performed the lengthy calculation present here in order to arrive at a fully trustable result. As we have seen here above, dimensional analysis does not allow to reach a firm conclusion.
APPENDIX B: Approximation in the calculation of $I_{j}^{\left(2\right)}$ {#appendix-b-approximation-in-the-calculation-of-i_jleft2right .unnumbered}
========================================================================
In this appendix we justify the approximation we used in order to derive Eq. (\[eq:izgdj\]) from Eq. (\[eq:vcvxvxcy\]). This amounts to proving that: $$\frac{1}{\hbar^{3}}\int d\mathbf{p}\frac{1}{E_{p}^{\left(j\right)}}e^{-\frac{\left(\mathbf{p}-\mathbf{p}_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}\simeq\frac{1}{E_{i}^{\left(j\right)}}\frac{\pi^{3/2}}{r_{C}^{3}}$$ To see this, we can rewrite the integral in polar coordinates and integrate over the angular variables: $$\frac{1}{\hbar^{3}}\int d\mathbf{p}\frac{1}{E_{p}^{\left(j\right)}}e^{-\frac{\left(\mathbf{p}-\mathbf{p}_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}=\frac{2\pi}{\hbar^{3}}\frac{\hbar^{2}}{2p_{i}r_{C}^{2}}\int_{-\infty}^{+\infty}dp\frac{p}{\sqrt{p^{2}c^{2}+m_{j}^{2}c^{4}}}e^{-\frac{\left(p-p_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}$$ Let us introduce the dimensionless variable $s=\frac{\left(p-p_{i}\right)r_{C}}{\hbar}$: $$\frac{1}{\hbar^{3}}\int d\mathbf{p}\frac{1}{E_{p}^{\left(j\right)}}e^{-\frac{\left(\mathbf{p}-\mathbf{p}_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}=\frac{\pi}{p_{i}r_{C}r_{c}^{3}}\int_{-\infty}^{+\infty}ds\frac{\left(s+p_{i}\frac{r_{C}}{\hbar}\right)}{\sqrt{\left(s+\frac{r_{C}}{\hbar}p_{i}\right)^{2}c^{2}+\frac{r_{C}^{2}}{\hbar^{2}}m_{j}^{2}c^{4}}}e^{-s^{2}}$$ If $p_{i}c \gg \hbar c/r_{C} \sim 10\textrm{ eV}$ (the typical range of momenta of neutrinos is between $10^3\textrm{ eV}$ and $10^{19}\textrm{ eV}$) we can disregard $s$ both in the numerator and denominator, obtaining:
$$\frac{1}{\hbar^{3}}\int d\mathbf{p}\frac{1}{E_{p}^{\left(j\right)}}e^{-\frac{\left(\mathbf{p}-\mathbf{p}_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}
\; \simeq \;
\frac{\pi}{r_{C}^{3}E_{i}^{\left(j\right)}}\int_{-\infty}^{+\infty}dse^{-s^{2}}
\; = \;
\frac{\pi^{3/2}}{r_{C}^{3}E_{i}^{\left(j\right)}},$$
which is the desired result.
APPENDIX C: Approximation in the calculation of $I_{jk}^{\left(1\right)}$ {#appendix-c-approximation-in-the-calculation-of-i_jkleft1right .unnumbered}
=========================================================================
Here we justify the approximation we used to pass from Eq. (\[eq:dasdasd\]) to Eq. (\[eq:syksdgar\]). We start with Eq. (\[eq:dasdasd\]): $$\begin{aligned}
\label{eq:sdgyoudgs}
I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right)
& = &
\sum_{s_{f}}\int d\mathbf{p}_{f}\sqrt{\gamma_{m_{j}}\gamma_{m_{k}}}\frac{m_{j}m_{k}c^{4}}{\sqrt{E_{i}^{\left(j\right)}E_{f}^{\left(j\right)}E_{i}^{\left(k\right)}E_{f}^{\left(k\right)}}}\overline{u}(p_{f}^{\left(j\right)},s_{f})u(p_{i}^{\left(j\right)},s_{i})\overline{u}(p_{i}^{\left(k\right)},s_{i})u(p_{f}^{\left(k\right)},s_{f}) \nonumber \\
& &
\times\frac{1-e^{\frac{i}{\hbar}\left(E_{f}^{\left(k\right)}-E_{i}^{\left(k\right)}-E_{f}^{\left(j\right)}+E_{i}^{\left(j\right)}\right)t}}{\frac{i}{\hbar}(E_{f}^{\left(j\right)}-E_{i}^{\left(j\right)}-E_{f}^{\left(k\right)}+E_{i}^{\left(k\right)})}\frac{1}{\left(2\pi\hbar\right)^{3}}e^{-\frac{\left(\mathbf{p}_{f}-\mathbf{p}_{i}\right)^{2}r_{C}^{2}}{\hbar^{2}}}\end{aligned}$$ where: $$E_{f}^{\left(j\right)}=\sqrt{\mathbf{p}_{f}^{2}c^{2}+m_{j}^{2}c^{4}}\;\;\;\textrm{,}\;\;\; u\left(p,s\right)=\frac{p^{\mu}\gamma_{\mu}c+mc^{2}}{\sqrt{2mc^{2}\left(E_{p}+mc^{2}\right)}}u\left(0,s\right)\;\;\;\textrm{and}\;\;\; \sqrt{\gamma_{m_{j}}}=\sqrt{\gamma}\frac{m_{j}}{m_{0}}$$ The first part of the integrand: $$\frac{m_{j}m_{k}c^{4}}{\sqrt{E_{i}^{\left(j\right)}E_{f}^{\left(j\right)}E_{i}^{\left(k\right)}E_{f}^{\left(k\right)}}}\sum_{s_{f}}\overline{u}(p_{f}^{\left(j\right)},s_{f})u(p_{i}^{\left(j\right)},s_{i})\overline{u}(p_{i}^{\left(k\right)},s_{i})u(p_{f}^{\left(k\right)},s_{f})$$ is a composition of polynomial functions of ${\bf p}_f$, and we can safely assume that it does not change too much, within the range where the Gaussian function is appreciably different from zero. Therefore we can then take ${\bf p}_f = {\bf p}_i$; by using also the relation: $$\overline{u}(p_{i}^{\left(1\right)},s_{f})u(p_{i}^{\left(1\right)},s_{i})=\delta_{s_{f},s_{i}},$$ Eq. (\[eq:sdgyoudgs\]) becomes: $$I_{jk}^{\left(1\right)}\left(\mathbf{p}_{i},s_{i};t\right)
\simeq
\sqrt{\gamma_{m_{j}}\gamma_{m_{k}}}\frac{m_{j}m_{k}c^{4}}{E_{i}^{\left(j\right)}E_{i}^{\left(k\right)}}\frac{1}{\left(2\pi\hbar\right)^{3}}\underset{\equiv I}{\underbrace{\int d\mathbf{p}_{f}\frac{1-e^{\frac{i}{\hbar}(E_{f}^{\left(k\right)}-E_{i}^{\left(k\right)}-E_{f}^{\left(j\right)}+E_{i}^{\left(j\right)})t}}{\frac{i}{\hbar}(E_{f}^{\left(j\right)}-E_{i}^{\left(j\right)}-E_{f}^{\left(k\right)}+E_{i}^{\left(k\right)})}e^{-\frac{(\mathbf{p}_{f}-\mathbf{p}_{i})^{2}r_{C}^{2}}{\hbar^{2}}}}}$$ Now we have to focus our attention on the integral $I$, which contains an oscillating term that needs special care. As before, we write the integral in polar coordinates and perform the integration over the angular variables; moreover, we introduce once again the a-dimensional variable $s=\frac{\left(p_{f}-p_{i}\right)r_{C}}{\hbar}$. We have: $$\label{eq:iufsw}
I=\pi\frac{\hbar^{3}}{p_{i}r_{C}^{3}}t\int_{-\infty}^{+\infty}ds\left(\frac{\hbar}{r_{C}}s+p_{i}\right)\frac{e^{ig\left(s\right)}-1}{ig\left(s\right)}e^{-s^{2}},$$ where we have defined: $$\begin{aligned}
\label{eq:sdsdtukz}
g\left(s\right)
& \equiv &
\frac{1}{\hbar}(E_{f}^{\left(k\right)}-E_{i}^{\left(k\right)}-E_{f}^{\left(j\right)}+E_{i}^{\left(j\right)})t \nonumber \\
& = &
\frac{ct}{r_{C}}\left(\sqrt{\left(s+y\right)^{2}+a_{k}}-\sqrt{\left(s+y\right)^{2}+a_{j}}-\sqrt{y^{2}+a_{k}}+\sqrt{y^{2}+a_{j}}\right),\end{aligned}$$ with $$a_{j}\equiv\frac{r_{C}^{2}}{\hbar^{2}}m_{j}^{2}c^{2}=\left(10^{-2}\textrm{eV}^{-2}\right)m_{j}^{2}c^{4}\;\;\;\textrm{e}\;\;\; y\equiv\frac{r_{C}}{\hbar}p_{i}=\left(10^{-1}\textrm{eV}^{-1}\right)p_{i}c$$ Our goal is to show that $g\left(s\right)$ does not vary appreciably, within the range where the Gaussian term is significantly different from zero, and can be approximated with $g\left(0\right)=0$; in this way, the integral can be computed exactly. This kind of approximation is not obvious because the factor $ct/r_{C}$ in front of Eq. (\[eq:sdsdtukz\]) can be very big.
In the ultra-relativistic limit, we approximate the particle’s velocity with the speed of light. In this limit, $y^{2} \gg a_{j},a_{k}$, and so we can expand the square roots in $g(s)$ using the Taylor series $\sqrt{x+\epsilon}=\sqrt{x}+\frac{\epsilon}{2\sqrt{x}}$ and we get: $$g\left(s\right)\simeq\frac{ct}{r_{C}}\left(a_{j}-a_{k}\right)\frac{s}{\left(s+y\right)y}=\frac{ct}{r_{C}}\frac{\left(a_{j}-a_{k}\right)}{\left(y+\frac{y^{2}}{s}\right)}.$$ In order for $g(s)$ to remain small within the interval where the Gaussian term of Eq. (\[eq:iufsw\]) is appreciably different from zero, we need: $$\label{eq:condizione}
\left(y+\frac{y^{2}}{s}\right) \; \gg \; \frac{ct}{r_{C}}\left(a_{j}-a_{k}\right)$$ In all physically interesting situations, the term on the right hand side of Eq. (\[eq:condizione\]) is bigger than 1; moreover $s$ is of the order of unity, because of the Gaussian in Eq. (\[eq:iufsw\]). Inequality is verified if the following condition is true: $$y \; \gg \; \sqrt{\frac{ct}{r_{C}}\left(a_{j}-a_{k}\right)}.$$ Typically, cosmogenic neutrinos have energies bigger than $10^{18}\textrm{eV}$ and travel distances of at most $10^{9}$ light-years [@Ch]. This means that, in the worst case, $ct/r_{C} \sim 10^{32}$ while $(a_{j}-a_{k})$ $ =(10^{-2}\textrm{eV}^{-2})$ $(m_{j}^{2}c^{4}-m_{k}^{2}c^{4})\simeq(10^{-2}\textrm{eV}^{-2})(2\times10^{-3}\textrm{eV}^{2})=10^{-5}$. So we must have $y \gg 10^{14}$, that means $p_{i}c = y/(10^{-1}\textrm{eV}^{-1}) \gg 10^{15}\textrm{eV}$, which is satisfied.
For atmospheric neutrinos, $ct/r_{C}$ is in the range $10^{11}-10^{14}$ while the range of energies is between $10^{-1}\textrm{GeV}$ and $10^{4}\textrm{GeV}$ [@Neu1]. This means that, even in the worse case, the condition to check is $y \gg 10^{5}$, which means $p_{i}c = y/(10^{-1}\textrm{eV}^{-1}) \gg 10^{6}\textrm{eV}$. This is also satisfied.
[99]{}
J.S. Bell, [*Speakable and Unspeakable in Quantum Mechanics*]{}, Cambridge University Press (1986).
S.L. Adler and A. Bassi, Science [**325**]{}, 275 (2009).
S. Weinberg, arXiv:1109.6462v3 \[quant-ph\] (2011).
A. J. Leggett, J. Phys. A [**40**]{}, 3141 (2007).
M. Arndt, O. Nairz, J. Vos-Andreae, C. Keller, G. van der Zouw and A. Zeilinger, Nature [**401**]{}, 680 (1999).
L. Hackermüller, K. Hornberger, B. Brezger, A. Zeilinger and M. Arndt, Nature [**427**]{}, 711 (2004).
S. Gerlich, L. Hackermüller, K. Hornberger, A. Stibor, H. Ulbricht, M. Gring, F. Goldfarb, T. Savas, M. Müri, M. Mayor and M. Arndt, Nature Physics [**3**]{}, 711 (2007).
S. Gerlich, S. Eibenberger, M. Tomandl, S. Nimmrichter, K. Hornberger, P. J. Fagan, J. Tüxen, M. Mayor and M. Arndt, Nature Comm. [**2**]{}, 263 (2011).
W. Marshall, C. Simon, R. Penrose and D. Bouwmeester, Phys. Rev. Lett. [**91**]{}, 130401 (2003).
O. Romero-Isart, Phys. Rev. A [**84**]{}, 052121 (2011).
G.C. Ghirardi, R. Grassi and F. Benatti, Found. Phys. [**25**]{}, 5 (1995).
Fu Q. Phys. Rev. A [**56**]{}, 1806 (1997).
G.C. Ghirardi, P. Pearle and A. Rimini, Phys. Rev. A [**42**]{}, 78 (1990).
G.C. Ghirardi, A. Rimini and T. Weber, Phys. Rev. D [**34**]{}, 470 (1986).
S.L. Adler, J. Phys. A [**40**]{}, 2935 (2007).
A. Bassi and G.C. Ghirardi, Phys. Rept. [**379**]{}, 257 (2003).
S.L. Adler, [*Quantum Theory as an Emergent Phenomenon*]{}, Cambridge University Press (2004).
P. Pearle, Phys. Rev. D [**13**]{}, 857 (1976).
P. Pearle, Phys. Rev. A [**39**]{}, 2277 (1989).
P. Pearle, in Lecture Notes in Physics [**526**]{}, ed. H.-P. Breuer and F. Petruccione (Berlin: Springer-Verlag) (1999).
L. Diósi, J. Phys. A: Math. Gen. [**21**]{}, 2885 (1988).
L. Diósi, Phys. Lett. A [**129**]{}, 419 (1988).
M. Beuthe, Physics Reports [**375**]{}, 105 (2003).
M. C. Gonzalez-Garcia and Y. Nir, Rev. Mod. Phys. [**75**]{}, 345 (2003).
C. Giunti and C. W. Kim, [*Fundamentals of neutrino physics and astrophysics*]{}, Oxford University Press (2007).
M. C. Gonzalez-Garcia and M. Maltoni, Physics Reports [**460**]{}, 1 (2008).
J. Christian, Phys. Rev. Lett. [**95**]{}, 160403 (2005).
R. Penrose, Gen. Rel. Grav. [**28**]{}, 581 (1996).
R. Penrose, in Mathematical Physics 2000, ed. by A. Fokas et al., Imperial College , London, (2000).
L. Diósi, Phys. Rev. A [**40**]{}, 1165 (1989).
G.C. Ghirardi, R. Grassi, A. Rimini, Phys. Rev. A [**42**]{}, 1057 (1990).
L. Wolfenstein, Phys. Rev. D [**17**]{}, 2369 (1978).
S. P. Mikheyev and A. Yu. Smirnov, Sov. J. Nucl. Phys. [**42**]{}, 913 (1985).
F. Benatti, R. Floreanini, Phys. Rev. D [**71**]{}, 013003 (2005).
E. Lisi, A. Marrone, D. Montanino, Phys. Rev. Lett. [**85**]{}, 1166 (2000).
G.L. Fogli, E. Lisi, A. Marrone, D. Montanino, Phys. Rev. D [**67**]{}, 093006 (2003).
V. N. Aseev, A. I. Belesev, A. I. Berlev, E. V. Geraskin, A. A. Golubev, N. A. Likhovid, V. M. Lobashev, A. A. Nozik, V. S. Pantuev, V. I. Parfenov, A. K. Skasyrskaya, F. V. Tkachov, and S. V. Zadorozhny, Phys. Rev. D [**84**]{}, 112003 (2011).
K. Nakamura et al, J. Phys. G: Nucl. Part. Phys. [**37**]{}, 075021 (2010).
L. Diósi, Braz. Journ. Phys. [**35**]{}, 260 (2005).
L. Stodolsky, Phys. Rev. D [**36**]{}, 2273 (1987).
W. J. Marciano, Z. Parsa, J. Phys. G [**29**]{}, 2629 (2003).
E. G. Flowers, P. G. Sutherland, Astrophys. J. [**208**]{}, L19 (1976).
G. L. Fogli [*et al.*]{}, Phys. Rev. D [**76**]{}, 033006 (2007).
see e.g. R.D. McKeown, P. Vogel, Phys. Rept. [**394**]{}, 315-356 (2004) and references therein.
S.L. Adler, A. Bassi, J. Phys. A [**40**]{}, 15083 (2007).
S. Donadi, A. Bassi, C. Curceanu, A. Di Domenico and B. C. Hiesmayr, Found. Phys. [**43**]{}, 813 (2013).
W. Greiner, [*Relativistic quantum mechanics - wave equations*]{}, Springer (2000).
J.J. Sakurai, S.F. Tuan, [*Modern Quantum Mechanics*]{}, Addison-Wesley Publishing Company (1994).
R. Tumulka, J. Stat. Phys. [**125**]{}, 821 (2006).
D. J. Bedingham, Found. Phys. [**41**]{}, 686 (2011).
W. Greiner, [*Quantum Electrodynamics*]{} 3Ed, Springer (2003).
W. Greiner, J. Reinhardt, [*Field Quantization*]{}, Springer (1996).
R.A. Bertlmann, K. Durstberger, B. C. Hiesmayr, Phys. Rev. A [**68**]{}, 012111 (2003).
S.L. Adler, Phys. Rev. D [**62**]{}, 117901 (2000).
[^1]: Eq. differs from those one typically finds in the literature only because the latter are written in the ultra-relativistic approximation $E=\sqrt{\mathbf{p}^{2}c^{2}+m^{2}c^{4}}\simeq pc\left(1+\frac{m^{2}c^{4}}{2p^{2}c^{2}}\right)$.
[^2]: In the following, we will consider decoherence effects only on electronic and muonic neutrinos.
[^3]: A proof of this can be found in section 2.3 of [@Greiner]. There, the proof is worked out for the Dirac equation coupled with an electromagnetic field. This is the same case as ours, if one sets $\mathbf{A}\left(x\right)=0$ and $eA_{0}\left(x\right)=-\hbar\sqrt{\gamma_{m}}w\left(x\right)$.
[^4]: Here and in the following we use the notation of [@Greiner].
[^5]: As well know in Quantum Field Theory, the reason why we used $:N_{I}\left(t\right):$ in place of $N_{I}\left(t\right)$ is that, with this prescription, we can remove all divergent contributions coming from tadpole diagrams. This type of divergences can be absorbed through a renormalization procedure, without giving any physically observable consequence [@GreinerQED; @GreinerFIELD].
[^6]: In general, spontaneous collapses and decoherence are described by similar master equations.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Wind power forecasting (WPF) is significant to guide the dispatching of grid and the production planning of wind farm effectively. The intermittency and volatility of wind leading to the diversity of the training samples have a major impact on the forecasting accuracy. In this paper, to deal with the training samples dynamics and improve the forecasting accuracy, a data mining approach consisting of K-means clustering and bagging neural network is proposed for short-term WPF. Based on the similarity among historical days, K-means clustering is used to classify the samples into several categories, which contain the information of meteorological conditions and historical power data. In order to overcome the over fitting and instability problems of conventional networks, a bagging-based ensemble approach is integrated into the back propagation neural network. To confirm the effectiveness, the proposed data mining approach is examined on real wind generation data traces. The simulation results show that it can obtain better forecasting accuracy than other baseline and existed short-term WPF approaches.'
author:
- '[^1]'
title: 'A Data Mining Approach Combining K-Means Clustering with Bagging Neural Network for Short-term Wind Power Forecasting'
---
Wind power forecasting, k-means clustering, ensemble learning, bagging neural network
Introduction
============
The development and utilization of renewable energy has been one of the hottest spots around the world. Wind power generation is rapidly expanding into a large-scale industry due to the cleanness and wide availability, and has been characterized as an fluctuating and intermittent power. Unfortunately, it is difficult to ensure the security and stability while accessing to the electricity grid, especially for large-scale application. An accurate and reliable wind power forecasting approach is essential for power quality, reliability management while reducing the cost of supplying spinning reserve.
A large number of wind power forecasting approaches have been proposed in many literatures, which can be roughly classified into three categories: 1) physical forecasting approach; 2) statistical forecasting approach; 3) combination approach [@bib:1]. The physical approach uses detailed topological and meteorological descriptions to model the conditions at the location of the wind farm [@bib:2]. Then the wind speed is predicted by numerical weather forecasting approach and converted into wind power via the power curves generated by wind turbines. The accuracy of this approach largely depends on the amount of physical information. The statistical approach aims at establishing the relationship between wind power and a set of variables including historical data and online measured data such as wind speed and wind direction [@bib:3]. This approach is applicable for most scenarios without considering geographical conditions. The core idea of the combination approach is to take advantage of the physical and statistical methods and improve the forecast accuracy [@bib:4].
The physical approach can establish a specialized wind power forecasting model for a wind farm without large amount of historical data. In particular, this approach needs detailed physical characteristics of wind turbines and wind farm to achieve an accurate model [@bib:5]. However, it is difficult to collect these physical characteristics in a short period of time. Besides, the model has poor versatility due to the factors of specific geographical conditions and operation state of wind turbines.
Compared with the physical approach, the statistical approach needs to collect a lot of historical data to build the forecasting model. In conventional statistical approaches, the time series model is usually applied to predict the wind power depending on the principle of continuity. There are several kinds of time series models, including autoregressive model (AR), moving average model (MA), autoregressive moving average model (ARMA), and autoregressive integrated moving average model (ARIMA) [@bib:9]-[@bib:12]. Reference [@bib:13] concentrates on multi-time series and multi-time scale modeling in wind speed and wind power forecasting. Then AR, MA, ARMA and ARIMA approaches are utilized to model multi-time series. The Kalman filter approach [@bib:14] is applied to optimize the time series model, which can modify the weights dynamically on the basis of ARMA model. Much work have shown that the time series models can obtain good forecasting accuracy only in the very short-term scale scenarios. In addition, this approach is applicable for the continuous sequence without considering the similarity among historical data.
Data mining approaches are catching researcher’s attention [@bib:6]-[@bib:8]. It has been reported that neural network (NN) outperformed other methods in short-term forecasting problems [@bib:19]. NN is able to model the complex nonlinear relationship between the historical data and forecasting power [@bib:15]. Besides, NN needs to use the historical data which have a major impact on the wind power as the input variables, such as wind speed and temperature. Back propagation (BP) NN is easy to formulate and often used as a reference model. But there are some potential improvements for the BPNN. Firstly, the result of the BPNN is sensitive to the initialization of weights and biases, and the network is easy to fall into local optimum. Secondly, the overtraining problem may arise if the BPNN has too many parameters to be estimated in the training samples [@bib:16]. Finally, several improvements for the BPNN are achieved by 1): optimizing parameters 2): tuning the structure of NN 3): changing the kernel of NN. Particle swarm optimization (PSO) [@bib:20], the Elman NN [@bib:17] and Radial Kernel Function (RBF) as examples respectively correspond to the above approaches. However, these improvements don’t consider that the selection of training samples has an important impact on the forecasting accuracy.
In order to improve the forecasting accuracy, clustering approaches are also applied to select proper training set. Reference [@bib:17] proposes an approach which selects the similar days as the training samples. The K-means clustering is utilized to classify wind speed into several categories. In reference [@bib:21], the ARMA model is presented and the K-means clustering is used to classify wind direction. However, only the weather conditions are clustered and the actual operation state of wind turbines is not considered.
In this paper, a data mining approach consisting of the K-means clustering and bagging neural network is proposed to predict the short-term wind power of individual wind turbine. The main contributions of this paper are summarized as follows.
1. To mitigate the impact of the diversity of training samples, the K-means clustering is utilized to classify the original data sets into several categories according to meteorological conditions and historical power. Person correlation coefficient is used to calculate the similarity between each category and the forecasting day. The most similar category is selected as the training sample.
2. To overcome the instability and over fitting problems of the BPNN, a bagging-based approach is integrated into the BPNN as the forecasting engine. The random sampling is conducted on the training sample by the Bootstrap sampling technique to form ${N}$ subsets. The BP algorithm is applied to each subset and trained on the individual network. The final result is determined by calculating the average value of ${N}$ networks.
3. The performance of the proposed approach is evaluated. Simulation results show that the root mean squared error (RMSE) and mean absolute error (MAE) of the proposed approach are less than other baseline approaches, which demonstrates the proposed approach can improve the forecasting accuracy.
Besides, many researchers have focued on the platform of collecting and processing data in wind turbines because they are important for managing and forecasting the wind energy. In order to make use of the manifold potential, the available data of wind turbines needs to be analyzed and the advanced measuring technologies and communication networks should be implemented. The Cyber Physical System (CPS) is a powerful approach to develop the wind energy sector, which has been widely used in CMS or SCADA systems [@bib:24].
The rest of the paper is organized as follow. Section presents the process of establishing the prediction model. Section carries out the results of the simulation and Section concludes the paper and summarizes the future work.
The establishment of the wind power forecasting model
=====================================================
Data mining approaches have been widely used for classification and prediction problems. The proposed approach is based on data mining, which consists of the K-means clustering and bagging neural network. Fig. 1 shows the wind power forecasting model. Firstly, data preprocessing is conducted on the vector space to clean unreasonable data, normalize the training samples and select the most related variables as the inputs of the neural network. Secondly, data after preprocessing are clustered by the K-means clustering to select the training set which is most similar to the forecasting day. Finally, the wind power is forecasted by the bagging neural network, which is able to alleviate the instability and over fitting problems of the BPNN.
![**The system architecture for the proposed WPF approach**[]{data-label="HRAN"}](model.eps)
Data preprocessing
------------------
A number of wind turbine parameters are collected as the training samples via the sensor unit. However, these samples may contain unreasonable data. Besides, using too many parameters as the training features would increase the computing complexity and obtain undesired results for the reason that some variables are irrelevant or redundant in this model. Selecting features which are most related to the wind power is able to improve the accuracy. Finally, data normalization has an effect on the convergence rate and accuracy of the training algorithm. Thus, in order to obtain accurate forecasting results, data preprocessing is necessary.
- *Data cleaning*: The original samples may contain data whose values of some characteristics are unreasonable. For example, the values of wind speed and wind power are less than zero. It is obvious that these data are not available and need to be removed. Then, it is necessary to fill the vacancy of the deleted data. The mean value method is applied, which makes use of the mean value ahead and back of the deleted data.
- *Feature selection*: In theory, more input variables can carry more discriminating power. But in practice, excessive variables are prone to cause many problems. Therefore, selecting a suitable set of input variables from the raw data has a great impact on the forecasting performance. Relief algorithm is a kind of feature weighting algorithms. The core idea is that the different weights are assigned to the corresponding features according to the correlation, and the feature whose weight is less than the threshold would be removed. The formulation of Relief algorithm is given by [@bib:18]. The running time of Relief algorithm increases linearly with the sampling times and original features so that this method has high operating efficiency. In addition, Relief algorithm can achieve the purpose of physical dimensions reduction compared with the principal component analysis (PCA).
- *Data normalization*: The aim of data normalization is to transform the raw data to the same orders of magnitude so that the convergence rate and forecasting accuracy can be improved. The min-max method is applied for normalization, which can be expressed as ${\overline{x}=(x-min)/(max-min)}$. In this equation, x is the original data, and max and min represent the maximum and minimum value of the training set. The result ${\overline{x}}$ is mapped to \[0,1\].
The three steps above are significant to more accurate forecasting results. After data preprocessing, the proposed approach can be implemented and compared with other WPF approaches.
Similar day clustering based on meteorological conditions and historical power
------------------------------------------------------------------------------
Large amount of data are generated by wind turbines when operating normally. The training samples can be regarded as the time series and auto regressive method is often utilized to build the forecasting model. However, the random behavior of wind can lead to the inconsistencies of training samples. Then, it is difficult to obtain desire results when using this kind of training samples. Therefore, selecting similar days and classifying them into the same category as the training samples can be considered as a good method to improve the forecasting accuracy.
1. ***The clustering sample***\
Wind speed and temperature have a major impact on wind power generation. Selecting the samples which have the most similar wind speed and temperature with the forecasting day from the historical data is significant. A sample consisting of wind speed and temperature is used as the clustering benchmark, which is constructed as $$S_{1} = [{W{S_{\max }},W{S_{\min }},W{S_{{\mathop{\rm mean}\nolimits} }},{T_{\max }},{T_{\min }},{T_{{\mathop{\rm mean}\nolimits} }}}],$$ where ${WS_{max}}$, ${WS_{min}}$, ${WS_{mean}}$, ${T_{max}}$, ${T_{min}}$ and ${T_{mean}}$ represents the maximum, minimum and average values of wind speed and temperature respectively.
According to the investigation results, wind power generation is largely determined by the meteorological conditions. However, the similar meteorological conditions can not represent that the wind power of the training samples is similar to the same of the forecasting day because of the different geographical conditions and the operation state of wind turbines. Therefore, the historical data which have the similar wind power with the forecasting day is also used for clustering. Then the sample ${S_{2}}$ could be expressed as $$S_{2} = [WP_{max},WP_{min},WP_{mean}],$$ where ${WP_{max}}$, ${WP_{min}}$, ${WP_{mean}}$ represents the maximum, minimum and average values of daily wind power generation. Although the ${S_{2}}$ of the forecasting day is unknown, the day which is before the forecasting day can be assumed to have the same ${S_{2}}$. According to the principle of wind power generation, the daily wind power generation curve has the same trend for next few days in the same season. Then the samples ${S_{1}}$ and ${S_{2}}$ are clustered respectively. It is assumed that cluster 1 from ${S_{1}}$ and cluster 2 from ${S_{2}}$ are most similar to the forecasting day. Finally, the same days from cluster 1 and cluster 2 are collected as the training samples of the forecasting model.
2. ***The clustering algorithm***\
Clustering belongs to unsupervised learning, which has no label on the training data. In this paper, it is needed to use the clustering algorithm to mine the similar samples and category them into one class. Typical clustering algorithms contain the K-means clustering and expectation maximization (EM).
In this paper, the K-means clustering is used as the clustering method for the reason that K-means algorithm can handle large amount of data sets effectively. K-means algorithm is a kind of clustering method based on partitioning, which usually judges the similarity by computing the distance. The core of K-means algorithm is to choose k center points randomly and partition the data according to the distance between the data and k center points. By means to the Euclidean distance, the algorithm assigns each data to its closest center point ${P_{k}}$, which is calculated by $$P_{k} = \frac{1}{{{N_{k}}}} \cdot \sum\limits_{i = 1}^{{N_{k}}} {x_{i}^k},$$ where ${x_{i}^k}$ is the i-th data in the cluster k, and ${N_{k}}$ is the number of data points in the respective cluster. The clustering center needs to be updated by computing the average value of every cluster until the value doesn’t change any more. Then the initial data set is divided into k clusters in which the data have high similarity. The process of clustering is as follows:
- *Step 1*: K objects are selected as the initial clustering centers from the data set which obtains N objects.
- *Step 2*: The distance between the data objects and clustering center is calculated and classifying them to the nearest cluster.
- *Step 3*: The mean value of the cluster is computed and used to update the clustering center.
- *Step 4*: The iterative method is applied in Step 2 and 3 until the value of the clustering center changes no more. If not, the process continues.
The K-means clustering is conducted on the data sets of ${S_{1}}$ and ${S_{2}}$. The data series are cut up into M components, and the training set can be expressed as ${D={d_{1},d_{2},...,d_{M}}}$, where D represents the training days. The K-means algorithm can select similar days from D as one class. Then several classes are produced. According to Pearson correlation coefficient, the most related class is chose as the final input of the neural network.
Bagging neural network
----------------------
Neural network (NN) has been one of the most effective data mining approaches for prediction. NN can deal with nonlinear problems well without establishing complex mathematical model. Back propagation neural network (BPNN) as one of the most common NNs is usually used as the forecasting algorithm. The basic BPNN consists of three layers: an input layer, a hidden layer and an output layer.
![**The model of three-layers BPNN**[]{data-label="HRAN"}](neuralnetwork.eps)
Fig. 2 shows the principle of the BPNN. The BPNN consists of two processes: forward propagation of data stream and back propagation of the error signal. In the process of forward propagation, the state of neurons in each layer only affect ones in the next layer. If the expected output could’t be obtained in the output layer, the algorithm turns to the process of back propagation of the error signal. The gradient descent method is conducted on the weights vector space. It is needed to dynamically search for a set of weights vector and minimize the error function. As for the hidden layer, the neural numbers of this layer is usually ${2M+1}$ according to the experience, where ${M}$ represents the neural numbers of the input layer. However, different neural numbers have an effect on the results of the output layer so that the network is tested when ${M = [2,3,...,10]}$ to find the best forecasting result.
In Fig. 2, ${[{x_{1},x_{2},...,x_{n}}]}$ are the input variables, ${[{w_{1},w_{2},...,w_{n}}]}$ are weights between the input layer and the hidden layer and ${[{v_{1},v_{2},...,v_{n}}]}$ are weights between the hidden layer and the output layer. ${Y(t)}$ is the expected output. In addition, the general Sigmoid function is used as the transfer function to handle the nonlinear problem, which can be expressed as $$y = f\left( x \right) = \frac{1}{{1 + {e^{ - x}}}}.$$
The output of the hidden layer ${z}$ can be formulated as $$z = {f_{1}}\left( {\sum {{w_{i}}{x_{i}}} } \right).$$
Similarly, the output of the out layer can be formulated as $$u = {f_{2}}\left( {\sum {{v_{k}}{z_{k}}} } \right).$$
Thus, the process of forward propagation is completed. Then, the error signal ${e}$ is produced by the function of ${u}$ and ${Y(t)}$, which is calculated as $$e = \frac{1}{2}{\sum {\left( {{y_{i}}\left( t \right) - {u_{i}}} \right)} ^2}.$$
The gradient descent method is utilized to modify ${w_{i}}$ and minimize the error function ${e}$. This method needs to calculate the partial derivatives to update the weighs of ${w_{i}}$ and ${v_{i}}$. It is incessantly required to be repeated until the value of ${e}$ equals to zero, which means that the values of ${u}$ and ${Y(t)}$ are identical. In this process, several weights need to be modified, which consumes a lot of time.
The BPNN can obtain good results compared with other algorithms such as linear aggression and support vector machine (SVM) when dealing with large amount of nonlinear samples. However, it is easy to trap into the local optimization and the forecasting result would be instable when the training samples change little. Besides, the over fitting problem would arise for the BPNN.
In order to improve the problems above, ensemble learning is applied to optimize the BPNN. Ensemble learning can combine several learning modules to enhance the stability and forecasting accuracy of the model. Bootstrap aggregating (bagging) algorithm as a kind of ensemble learning methods is used to improve the performance of the BPNN. It is verified that the bagging-based approach is effective for instable learning algorithms.
The bagging algorithm can be effective on the premise that a weak learner is applied to the training samples. Because of the poor accuracy of the weak learner, it is needed to use the learner several times on the training samples. Then a sequence of forecasting functions are produced and the most accurate function is determined by voting.
![**The process of sampling and training of the bagging algorithm**[]{data-label="HRAN"}](Bagging.eps)
Fig. 3 shows the process of the bagging algorithm. It can be seen that the original training data set is sampled again by Bootstrap sampling technique and the learning algorithm is applied for the subsets. Suppose ${T(x)}$ is a classifier, such as a tree, producing a predicted class label at input point ${x}$. To bag ${T}$, we draw bootstrap samples ${{\left\{ {\left( {x_{i}^*,y_{i}^*} \right)} \right\}^1},...,{\left\{ {\left( {x_{i}^*,y_{i}^*} \right)} \right\}^B}}$ each of size ${n}$ with replacement from the training data. Then the results can be expressed as follows: $${C_{bag}}\left( x \right)=Majority Vote\left\{ {{T^{*b}}\left( x \right)} \right\}_{b = 1}^B$$ The output ${C_{bag}}\left( x \right)$ is the optimal classifier, which is determined by voting. In this paper, the bagging-based method is used for prediction. The key factor that bagging algorithm can improve the forecasting accuracy is the stability of the learning algorithm. Firstly, bagging algorithm can select K sets with n samples randomly from the initial training set. Secondly, BPNN is utilized to train the subsets for several times. Then a sequence consisting of the forecasting models ${{h_{1},h_{2},...,h_{K}}}$ is produced. Finally, the chosen forecasting model ${H}$ is determined by calculating the average value of ${K}$ models. In particular, all the forecasting models have the same importance.
Applying the BPNN based on the bagging algorithm can not only improve the forecasting accuracy but also reduce the time costs.
Simulation results
==================
In this section, the proposed forecasting approach is simulated and compared to other approaches. The 10-min data obtained from the Supervisory Control And Data Acquisition system (SCADA) of a wind farm is adopted as research data. The six numbers for an hour are sum to obtain hourly data, which could avoid that 10-min data fluctuated greatly so that the forecasting accuracy could be improved.
In order to evaluate the performance, two kinds of error calculation methods: 1) root mean squared error (RMSE) and 2) mean absolute error (MAE) are proposed:
$$RMSE = \sqrt {\frac{1}{N}\sum\limits_{i = 1}^N {\mathop {\left( {{y_{i}} - y^{'}_{i}} \right)}\nolimits^2 } }$$
$$MAE = \frac{1}{N}\sum\limits_{i = 1}^N {\left| {{y_{i}} - y^{'}_{i}} \right|}$$
In the above equations, ${y_{i}}$ is the actual power, ${y^{'}_{i}}$ is the predictive power, N is the number of the test data.
Data preparing
--------------
In this paper, the measured data from September 29th to September 30th for a total of 24 hours are selected as test data. Data in three months before this period of time are used as training samples for clustering. Fig. 4 shows the wind power generation of all the year. It can be observed that the power generation is much different before and after July. Even in the same season, the power curve varies dramatically, which means that it is necessary to apply the clustering method.
![**The wind power generation curve from Dec. 2013 to Nov. 2014**[]{data-label="HRAN"}](power.eps)
There are 39 variables collected by the SCADA system. Relief algorithm is conducted on the vector space to select proper variables as the inputs of neural networks. The results of Relief algorithm have been shown in Table , in which the variables whose weight is large than 0.01 are listed. According to the results, the first four variables have better weights. Besides, the ambient temperature is also significant. Then, the whole input variables are selected.
**Variable** **Weight**
-------------- ------------
0.052
0.022
0.016
0.013
0.003
0.003
0.002
0.002
: The results of Relief algorithm[]{data-label="survey"}
Table illustrates that wind speed is the most important variable in wind power forecasting. Although temperature from the wind turbine units plays an important role according to Relief algorithm, it can not be utilized as the forecasting variable for the reason that it is unable to obtain the temperature of machines in the forecasting day. The generator bearing temperature can not be used for the same reason. Therefore, we select average wind speed per second, the angle of blade 1 and the ambient temperature as the input variables, where the angle of blade represents the wind direction. Besides, the value of the angle of blade is decomposed into sine and cosine values.
Fig. 5 shows the relationship between wind speed and power. It can be seen that when actual wind speed is less than cut-in wind speed, and the wind turbines do not operate and the wind power is zero. When the actual wind speed is larger than cut-in wind speed, the value of wind power is increasing as the wind speed growing until the wind speed is less than cut-out wind speed. When the actual wind speed is larger than cut-out wind speed, wind turbines would operate in the state of rated power. It is obvious that the relationship between them is nonlinear so that applying the neural network is necessary.
![**Strong correlation between wind power generation and wind speed**[]{data-label="HRAN"}](relation.eps)
According to the described clustering approach above, it is needed to divide all the training samples into pieces and regard one day as the training unit, which consists of 144 items. The average, maximum and minimum values of wind speed and temperature are calculated as the clustering features. Then the sum, maximum and minimum of the power generation are also collected for clustering. For the forecast period, the power generation before this period is regarded as having the same features of power generation with the forecasting period. Then, the data are clustered by the K-means clustering.
Fig. 6 shows the clustering result, and three lines mean that the initial samples are classified into 3 categories. It can be seen that the three categories are much different in wind speed, temperature and wind power, which are the main factors of prediction. In this process, the blue line is represented by $k = 0$, the green line is for the case of $k = 1$, and the red line is for $k = 2$. The initial samples have different features, and the K-means method can classify the samples which have similar features into one category.
![**The K-means clustering results with meteorological conditions and historical power**[]{data-label="HRAN"}](cluster.eps)
The distance between the forecasting day and the 3 categories is calculated according to the Pearson correlation coefficient: $$P = \frac{{E\left( {XY} \right) - E\left( X \right)E\left( Y \right)}}{{\sqrt {E\left( {{X^2}} \right) - {E^2}\left( X \right)} \sqrt {E\left( {{Y^2}} \right) - {E^2}\left( Y \right)} }}$$ where X-axis and Y-axis represent the forecasting sample and the 3 categories, respectively. The most related category is chosen and there are 29 days totally adopted as the training data. In order to compare the performance of clustering with non-clustering, a data set containing 29 days from September 1th to September 29th is used as the training set.
After the feature selection based on Relief algorithm, the training data consisting of the wind speed, the blade angle, and the temperature, are used as the input variables to construct a BPNN model based on the bagging algorithm.
Forecasting results
-------------------
To better verify the effectiveness of the proposed approach, this paper compares three methods of predicting power generation: the BPNN without clustering pretreatment, the BPNN with clustering pretreatment and the proposed approach.
![**The 24-hours forecasting results of BPNN, BPNN with clustering and the proposed approach**[]{data-label="HRAN"}](matlabResult.eps)
Fig. 7 shows the comparison of the three approaches. It is obvious that the purple line is most similar to the actual wind power value. In particular, the peak values marked on Fig. 7 shows that the proposed approach has better performance compared with other approaches.
**Forecasting approach** **RMSE(kW)** **MAE(kW)**
-------------------------- -------------- -------------
558.098 365.091
487.718 323.596
342.548 255.156
: The performance of baseline approaches and the proposed approach[]{data-label="survey"}
From the prediction results in Table , it can be seen that the BPNN with clustering can reduce the forecasting error compared with the BPNN, in which RMSE and MAE are decreased by 12.7% and 11.5%, respectively. Furthermore, the proposal can obtain the best performance among these three approaches, which can decrease 38.7% and 29.8% for RMSE compared with the BPNN, and 30.1% and 21.1% for MAE compared with the BPNN with clustering. The simulation results verify that the bagging neural network with the similar meteorological conditions and historical power clustering approach can obtain better forecasting accuracy.
Besides, the Approach 1 in [@bib:17], the Approach 2 in [@bib:22], and the Approach 3 in [@bib:23] have been simulated and the results are listed in Table .
**Forecasting approach** **RMSE (kW)** **MAE (kW)**
-------------------------- --------------- --------------
1053.736 956.211
427.912 307.379
389.472 279.528
342.548 255.156
: The performance of three methods and the proposed method[]{data-label="survey"}
It can be seen that the proposed approach has the best forecasting accuracy among these approaches. The Approach 1 has the worst results because the different data has different characteristics, and it may be not applicable to this case. The Approach 2 uses the ANN ensemble based on PCA to predict the wind power. The proposal decreases 19.9% and 17.1% for RMSE and MAE compared with the Approach 2, respectively. The Approach 3 uses the artificial neural network as the forecasting approach and it has better performances than the Approach 1 and 2, but is worse than the proposal. As a result, the proposal can obtain best forecasting accuracy than the other baselines.
Conclusion
==========
In this paper, a data mining approach for wind power forecasting has been proposed, which consists of the K-means clustering method and bagging neural network. The historical data are clustered according to the meteorological conditions and historical power. Pearson correlation coefficient is used to calculate the distance between the forecasting day and the clusters. The input variables of the neural network are selected by Relief algorithm to reduce the complexity and the Bagging algorithm is applied to optimize the stability and accuracy of the BPNN. To demonstrate the effectiveness, the proposed approach has been tested according to the actual data in the practical wind farm. The RMSE and MAE results show that the proposals have significant gains.
Although the proposals are not specially designed for the individual wind turbine, the idea of clustering is still important and effective when a large-scale wind farm is built. Particularly, in a wind farm, the location of wind turbines may lie in one direction, then the wind speed of these wind turbines can be classified into one category. With this way, the proposal can be extended and widely used in all real wind farm, which not only increases the forecasting accuracy, but also reduces the computational complexity.
To improve the forecasting accuracy, the effective meteorological forecasting should be researched, and the corresponding optimal method for the BPNN should be designed. Besides, the multi-dimensional clustering problem should be formed and the wind power forecasting model for wind farms should be researched.
[99]{}
J. Jung and R. P. Broadwater, “Current status and future advances for wind speed and power forecasting," *Renew. Sust. Energ. Rev.*, vol. 31, no. 2, pp. 762-777, Jan. 2014.
D. Carvalho, A. Rocha, M. Gomez-Gesteira, and C. Santos, “A sensitivity study of the WRF model in wind simulation for an area of high wind energy," *Environ. Model. Softw.*, vol. 33, no. 7, pp. 23-34, Mar. 2012.
A. M. Foley, P. G. Leahy, A. Marvuglia, and E. J. McKeogh, “Current methods and advances in forecasting of wind power generation," *Renew. Energy*, vol. 37, no. 1, pp. 1-8, Jan. 2012.
Z. Yin, Z. Chen, D. Yin, and Q. Li, “A new combination model for short-term wind power prediction," *in Proc. IEEE DRPT*, Changsha, China, Nov. 2015, pp. 1869-1873.
Q. Xu, D. He, N. Zhang, C. K, Q.Xia, J. Bai, and J. Huang, “A short-term wind power forecasting approach with adjustment of numberrical weather prediction input by data mining," *IEEE Trans. Sustain. Energy*, vol. 6, no. 4, pp. 1283-1291, Oct. 2015.
M. Khalid and A. Savkin, “Closure to discussion on ’a method for shortterm wind power prediction with multiple observation points’," *IEEE Trans. Power Syst.*, vol. 28, no. 2, pp. 1898-1899, May. 2013.
J. Tastu, P. Pinson, P. J. Trombe, and H. Madsen, “Probabilistic forecasts of wind power generation accounting for geographically dispersed information," *IEEE Trans. Smart Grid*, vol. 5, no. 1, pp. 480-489, Jan. 2014.
C. Wan, Z. Xu, P. Pinson, Z. Y. Dong, and K. P. Wong, “Probabilistic forecasting of wind power generation using extreme learning machine," *IEEE Trans. Power Syst.*, vol. 29, no. 3, pp. 1033-1044, May. 2014.
X. Zhang, J. Zhang, Y. Li, and R. Zhang, “Short-term forecasting of wind speed based on recursive least squares," *in Proc. IEEE ICECE*, Wuhan, China, Oct. 2011, pp. 367-370.
M. S. Miranda and R. W. Dunn, “One-hour-ahead wind speed prediction using a bayesian methodology," *in Proc. IEEE Power Engineering Society General Metting*, Montreal, Canada, Oct. 2006, pp. 1-6.
J. L. Torres, A. Garcia, M. D. Blas, and A. D. Francisco, “Forecast of hourly average wind speed with ARMA models in Spain," *Solar Energy*, vol. 79, no. 1, pp. 65-77, Jul. 2005.
A. Kusiak, H. Zheng, and Z. Song, “Short-term prediction of wind farm power: A data mining approach," *IEEE Trans. Energ. Conver.*, vol. 24, no. 1, pp. 125-136, Mar. 2009.
I. Colak, S. Sagiroglu, M. Yesibudak, E. Kabalci and H. I. Bulbul, “Multi-time series and -time scale modeling for wind speed and wind power forecasting part I: Statistical methods, very short-term and short-term applications," *in Proc. IEEE ICRERA*, Paris, France, Nov. 2015, pp. 209-214.
Y. Tian, Q. Liu, Z. Hu, and Y. Liao, “Wind speed forecasting based on time series - adaptive kalman filtering algorithm," *in Proc. IEEE FENDT*, Chengdu, China, Jun. 2014, pp. 315-319.
K. Bhaskar and S. N. Singh, “AWNN-assisted wind power forecasting using feed-forward neural network," *IEEE Trans. Sustain. Energy*, vol. 3, no. 2, pp. 306-315, Apr. 2012.
S. Li, P. Wang, and L. Goel, “Wind power forecasting using neural network ensembles with feature selection," *IEEE Trans. Sustain. Energy*, vol. 6, no. 4, pp. 1447-1456, Jul. 2015.
X. Zhang, R. Wang, T. Liao, T. Zhang, and Y. Zha, “Short-term forecasting of wind power generation based on the similar day and elman neural network," *in Proc. IEEE SSCI*, Cape Town, South Africa, Dec. 2015, pp. 647-650.
C. Hyuk-Gyu, H. Park, and H. Kwon, “Similarity measurement among sectors using extended Relief-F algorithm for disk recovery," *in Proc. IEEE ICCIT*, Busan, South Korea, Nov. 2008, pp. 790-795.
L. Ma, S. Y. Luan, C. W. Jiang, H. L. Liu, and Y. Zhang, “A review on the forecasting of wind speed and generated power," *Renew. Sust. Energy Rev.*, vol. 13, no. 4, pp. 915-920, May. 2009.
J. P. S. Catalao, H. M. I. Pousinho, and V. M. F. Mendes, “Hybrid wavelet-PSO-ANFIS approach for short-term wind power forecasting in Portugal," *IEEE Trans. Sustain. Energy*, vol. 2, no. 1, pp. 50-59, Jan. 2011.
E. Erdem, J. Shi, and Y. Peng, “Short-term forecasting of wind speed and power - a clustering approach," *in Proc. ISERC*, Toronto, Canada, May. 2014, pp. 1-8.
D. He and R. Liu, “Ultra-short-term wind power prediction using ANN ensemble based on PCA," *in Proc. IEEE IPEMC*, Harbin, China, Aug. 2012, pp. 2108-2112.
A. S. Kumar, T. Cermak, and S. Misak, “Short -term wind power plant predicting with Artificial Neural Network," *in Proc. IEEE EPE*, Kouty nad Desnou, Czech Republic, Jul. 2015, pp. 584-588.
P. Kunzemann, G. Jacobs, and R. Schelenz, “Application of CPS Within Wind Energy-Current Implementation and Future Potential," *in Industrial Internet of Things*, Switzerland: Springer, 2017, pp.647-670.
[Wenbin Wu]{} received his bachelor degree in Telecommunications Engineering with Management from Beijing University of Posts and Telecommunications (BUPT), Beijing, China, in 2014. He is currently a Ph.D. student in the Key Laboratory of Universal Wireless Communications (Ministry of Education) at BUPT. His main research interests include wind power forecasting and non-orthogonal multiple access for cloud radio access network.
[Mugen Peng]{} (M’05-SM’11) received the B.E. degree in electronics engineering from the Nanjing University of Posts and Telecommunications, Nanjing, China, in 2000, and the Ph.D. degree in communication and information systems from the Beijing University of Posts and Telecommunications (BUPT), Beijing, China, in 2005. Afterward, he joined BUPT, where he has been a Full Professor with the School of Information and Communication Engineering since 2012. In 2014, he was an Academic Visiting Fellow with Princeton University, Princeton, NJ, USA. He leads a Research Group focusing on wireless transmission and networking technologies with the Key Laboratory of Universal Wireless Communications (Ministry of Education), BUPT. His main research areas include wireless communication theory, radio signal processing, and convex optimizations, with a particular interests in cooperative communication, selforganization networking, heterogeneous networking, cloud communication, and internet of things. He has authored/coauthored over 60 refereed IEEE journal papers and over 200 conference proceeding papers.
Dr. Peng was a recipient of the 2014 IEEE ComSoc AP Outstanding Young Researcher Award, and the best paper award in IEEE WCNC 2015, WASA 2015, GameNets 2014, IEEE CIT 2014, ICCTA 2011, IC-BNMT 2010, and IET CCWMC 2009. He received the First Grade Award of the Technological Invention Award in the Ministry of Education of China, and the First Grade Award of Technological Invention Award from the China Institute of Communications. He is on the Editorial/Associate Editorial Board of the *IEEE Communications Magazine*, *IEEE Access*, *IEEE Internet of Things Journal*, *IET Communications*, and *China Communications*. He has been the guest leading editor for the special issues in the *IEEE Wireless Communications*. He is the Fellow of IET.
[^1]: Wenbin Wu (e-mail: [email protected]) and Mugen Peng (e-mail: [email protected]) are with the Key Laboratory of Universal Wireless Communications (Ministry of Education), Beijing University of Posts and Telecommunications, Beijing, China.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Two synthetic gene circuits – the genetic toggle switch and the repressilator – are discussed in the context of an educational module on gene circuits and feedback that constitutes the final topic of a year-long introductory physics sequence, aimed at biology and premedical undergraduate students. The genetic toggle switch consists of two genes, each of whose protein product represses the other’s expression, while the repressilator consists of three genes, each of whose protein product represses the next gene’s expression. Analytic, numerical, and electronic treatments of the genetic toggle switch shows that this gene circuit realizes bistability. A simplified treatment of the repressilator reveals that this circuit can realize sustained oscillations. In both cases, a “phase diagram” is obtained, that specifies the region of parameter space in which bistability or oscillatory behavior, respectively, occurs.'
author:
- 'S. B. Cahn'
- 'S. G. J. Mochrie'
bibliography:
- './simon.bib'
title: 'Biologic: Gene circuits and feedback in an introductory physics sequence for biology and premedical students '
---
Å[[A]{}]{}
Background and Introduction
===========================
Two recent reports – the NRC’s “BIO2010: Transforming Undergraduate Education for Future Research Biologists”, [@BIO2010] and the AAMC/HHMI’s “Scientific Foundations for Future Physicians” [@AAMC] – have highlighted the increasing importance of quantitative skills for students who are planning biomedical careers. Since the 2010-2011 academic year, the Yale physics department has offered a new introductory physics sequence – PHYS 170/171 – aimed at biology and premedical students, that seeks to implement a number of the recommendations of these reports. The PHYS 170/171 enrollment was about 100 in the 2010-2011 academic year, but increased to nearly 150 in 2012-2013. The majority of the PHYS 170/171 class (70%) are biology majors, and 80% identify themselves as premedical students. There are roughly equal numbers of sophomores and juniors, with significantly fewer seniors, and two or three freshmen. They are 70% female. 50% self-identify as white, 50% do not. Most come to PHYS 170/171 possessing considerable biological sophistication, because of prior biology and chemistry classes at Yale. Almost all have previously taken a first course in calculus. In considering a new introductory physics syllabus for biology and premedical students, there is a tension between what topics seem likely to be interesting and engaging versus what has traditionally been covered in introductory physics courses. Our starting point for resolving the PHYS 170/171 syllabus is the observation that the majority of these students will not take another physics course after PHYS 170/171. Therefore, we reasoned, there is no rationale to prepare students for more advanced classes or the physics major. Instead, our selection of topics and our approach is informed by the desire to tackle interesting topics, that demonstrate that physics has much to contribute to the life sciences and medicine, and that reflect that biological physics is now a major sub-field of physics, well-represented in physics departments across the country. The principle that we should endeavor to align physics teaching with how we practice physics also resonates with us.[@Handelsman2007] Thus, we have been led to include modules on chemical rate equations, probability, Brownian motion and diffusion, laminar fluid flow, statistical mechanics and Brownian ratchets,[@Mochrie2011] electromagnetic waves, and quantum mechanics – all topics which, between the two of us, we engage with in our own research.
In this paper, we present our PHYS 170/171 module on gene circuits in the hope that it will prove useful to others also thinking about new physics curricula for biology and premedical undergraduates. We decided to include a gene circuits module – humorously called “Biologic” – because of the importance of the concept of feedback to clinicians, and to provide an introduction to biological control, decision making, and time-keeping, which constitute the subject matter of “Systems Biology”, which has emerged as a major subfield of biology over the last decade, and to which physicists and engineers have made key contributions. Our educational goals are to introduce and explore the concept of feedback, to show that feedback can lead to switches and oscillators, both in electronic circuits and in gene circuits, and to alert students to the place of quantitative approaches in Systems Biology. To this end, we present simplified treatments of two [*de novo*]{} designed gene circuits, namely the “genetic toggle switch” [@GTS] and the “repressilator”,[@REPRESS] each of which has been realized experimentally in [*E. coli*]{}. Nature relies on multiple, interconnected gene circuits, that are considerably more refined than these Frankensteinian examples. Nevertheless, analogous natural gene networks are ubiquitous. For example, in the case of the genetic toggle switch, there is an analogy with the well-studied lytic-lysogenic switch in the bacteriophage lambda life cycle [@Ptashne], and a genetic oscillator governs the development of vertebrate segmentation that eventually leads to vertebrae.[@HolleyReview] A version of the repressilator has recently been recognized in the gene circuit of [*Bacillus subtilis*]{}, a common soil bacterium.[@Schultz2009]
”Biologic” is the final module in the year-long PHYS 170/171 introductory physics sequence. By this time, students’ mathematical skills have been practiced by nearly two semesters of physics. From earlier modules, they are familiar with coupled harmonic oscillators, eigenvalues and eigenvectors, and rate equations for chemical reactions, and they have had considerable experience with Wolfram Alpha and Wolfram Demonstrations. Throughout the year, we emphasize using Wolfram Alpha to facilitate mathematical manipulations, including the solution of systems of algebraic equations, the evaluation of derivatives and integrals, and the numerical solution of differential equations, which we believe empowers the students. Because computational approaches constitute an essential aspect of how research is now carried out, both in the physical and life sciences, we also include a number of simulations and visualizations,[@Dori2003; @Chabay2008] implemented as Wolfram Demonstrations,[@DEMOSITE] which run in students’ favorite web browsers. Students’ positive responses to Mathematica Demonstrations assuaged initial doubts amongst faculty colleagues, concerning the students’ ability and willingness to use such software.
In class, we segue from the previous module – electromagnetic waves – to genetic circuits by invoking what is arguably the twentieth century’s greatest invention, namely the transistor. We point out that the proliferation of transistors, which rely on Maxwell’s equations for their function, continues to transform the way we live, and assert that transistors are electronic switches. We further point out that, just as electronic switches and circuits implement electronic “decisions” depending on certain inputs, analogously biology uses biological switches and circuits to implement biological decisions.
To emphasize the intellectual connection between gene circuits and electronic circuits, in the associated laboratory course, PHYS 165/166, we implemented electronic versions of the genetic toggle switch and the repressilator. The electronic toggle switch was built using two of the logical inverters of a 7404 Hex Inverter. The electronic repressilator was built using three NAND gates of a 7400 Quad NAND chip, together with appropriate resistors and capacitors to select the oscillation period. The PHYS 165/166 laboratory handout is included in the Supplementary Information. To provide further opportunities for exploration, we exploit that Mathematica can numerically solve the relevant equations both for the genetic toggle switch and the repressilator. These solutions are presented in Wolfram Demonstrations.[@gtsdemo; @repressdemo]
Phage Lambda: Lysogeny or lysis, that is the question
=====================================================
From their biology classes, many PHYS 170/171 students are familiar with the life cycle of bateriophage lambda, which realizes one of the most studied and best understood biological switches, between the so-called lytic and lysogenic states.[@Ptashne] For infected bacteria in the lysogenic state, the bacteriophage’s genetic material is incorporated in the bacterial chromosome and is replicated along with the host’s genetic material at cell division, but phage capsid proteins, [*etc.*]{} are not expressed. However, in the lytic state, the proteins required to form new phage are expressed, many copies of the phage assemble, and the the host is caused to disintegrate (lyse), releasing many new bacteriophage particles, free now to infect a new host. The write-up for the laboratory module describes that in the lytic state, the phage make lots of copies of themselves and their [*spacesuits*]{}; then they blow up their host bacterium and disperse, presumably to find the next victim. These two different possible outcomes are visualized in the YouTube movie, [http://www.youtube.com/v/sLkZ9FPHJGM](http://www.youtube.com/watch?v=sLkZ9FPHJGM), in which phage infection is signaled by green fluorescence.[@Pierre2008] In the top right of the field of view, early on in the movie, a replicating [*E. coli*]{} becomes infected and undergoes lysis, providing an example of the behavior in the lytic state. By contrast, in the lower left of the field of view, another [*E. coli*]{} becomes infected, but, instead of undergoing lysis, this bacterium undergoes two rounds of cell division before the movie ends, resulting in four infected, fluorescent, daughter [*E. coli*]{} in the lysogenic state.
The existence of two possible outcomes in this system, [*i.e.*]{} bistability, depends on the interaction of two genes, [*cI*]{} and [*cro*]{} and their protein products, cI and cro, each of which represses the other. In the lysogenic state, the concentration of cI is high, expression of cro is repressed, and expression of cI continues at a high level, which continues to repress cro, and so on. Alternatively, in the lytic state, cro is expressed, which represses cI, which therefore is unable to repress cro, therefore cro remains high and the lytic state persists, or would persist, except that this state initiates a pathway to host cell lysis. We note that here feedback is exemplified by the protein product of a certain gene then going on to affect in some way its own expression. In addition to repressing cI, cro is also an activator for cro, realizing a second positive feedback loop, which represents the “suspenders” in a “belt-and-suspenders” approach to maintaining the lysogenic state.
Genetic toggle switch
=====================
The design of the genetic toggle switch is shown in Fig. \[GTSplasmid\]. The genetic toggle switch is an even simpler bistable gene circuit than the lambda switch, consisting of two genes each of which encodes for a protein that represses the other’s gene expression. Because of the even number of components in the circuit, there is a net positive feedback, which can give rise to bistable behavior for certain parameter values. The genetic toggle switch was described and implemented in [*E. coli*]{} in Ref. .
[![ Schematic of the genetic toggle switch plasmid. The solid orange region is the promoter for repressor 1. The solid green region is the promoter for repressor 2. The light orange region codes for repressor 1, which, as indicated, represses expression of repressor 2. The light green region codes for repressor 2, which, as indicated, represses promoter 1. []{data-label="GTSplasmid"}](./GTSplasmid.pdf "fig:"){width="35.00000%"}]{}
Electronic realization of a toggle switch
-----------------------------------------
The concept of bistability and the role of feedback is emphasized in the laboratory component of the course by first introducing and explaining the operation of a logical inverter, shown at the top of Fig. \[inverter\]. We explain that the logical inverter realizes the operation that changes [True]{} to [False]{}, yes to no, 1 to 0, high to low and [False]{} to [True]{}, no to yes, 0 to 1, low to high. We then consider two inverters in series (center of Fig. \[inverter\]). In this case, if the Input (I) is [True]{}, the intermediate result (M) is [False]{}, and the Output (O) is [True]{}. Likewise, if I is [False]{}, M is [True]{}, and O is [False]{}. Finally, we ask students to consider what happens when we introduce feedback, that is, where we take the Output and route it back to the Input, as shown at the bottom of Fig. \[inverter\]. Here, if I is [False]{}, M is [True]{}, O is [False]{}, which feeds back to I, which was already [False]{}. Thus, we see that our “circular” logic is self-consistent in this case. Even if we were to remove the input I, the logic speedway would remain stable. But what if we force M to be [False]{}, which makes O [True]{}, which renders I [True]{} and M [False]{}. Our logic is consistent in this case too. Both conditions are stable, so that two logical inverters with feedback realize a bistable situation. The Output can be set to either value, and it will stay in that state indefinitely, [*i.e.*]{}, it is a Toggle Switch. We point out that such a structure can also be considered a memory element, with a value at the Output of [True]{} or [False]{}, 0 or 1, a binary digit or bit.
![Top: Depiction of a logical inverter. The circle represents inversion. Center: Two cascaded investors. Bottom: Two cascaded inverters with feedback. []{data-label="inverter"}](Inverter "fig:"){width="3in"} ![Top: Depiction of a logical inverter. The circle represents inversion. Center: Two cascaded investors. Bottom: Two cascaded inverters with feedback. []{data-label="inverter"}](TwoInverters "fig:"){width="3in"} ![Top: Depiction of a logical inverter. The circle represents inversion. Center: Two cascaded investors. Bottom: Two cascaded inverters with feedback. []{data-label="inverter"}](TwoInvertersFB "fig:"){width="3in"}
![Top: 7404 Hex Inverter Chip. Bottom: Realization of electronic toggle switch with 7404[]{data-label="7404"}](7404 "fig:"){width="3in"} 0.5 in ![Top: 7404 Hex Inverter Chip. Bottom: Realization of electronic toggle switch with 7404[]{data-label="7404"}](2InverterSchematic "fig:"){width="3in"}
To actually implement an electronic toggle switch in the laboratory module, we use two of the logical inverters of a 7404 hex inverter chip (Fig. \[7404\]). Students are provided with the chip already appropriately wired up to use the first two logical inverters only: The chip is mounted on a breadboard, is powered with $V_{CC}=5$ V at pin 14 and is grounded at pin 7. To construct the feedback circuit, pin 2 is connected to pin 3 and pin 4 is connected to pin 1. In addition, the chip is connected to two “Morse Code” switches, which can bring pin 1 to ground (False) or pin 3 to ground. (The molecular biological analogues of these switches are so-called inducers.) The outputs, pins 2 and 4, are also connected to LED indicators, which are illuminated when the voltage at pin 2 or 4 is high, to provide a readout of the state of the circuit. Students are asked to sketch the circuit in their laboratory notebooks. They are asked what happens when you push one switch, and then push it again, what happens when they push the other switch, and whether this circuit follows the bistable “logic” described above. For most of the students, this exercise represents the first contact they have had with Boolean logic, and with digital electronics..... beyond their role as consumers, of course.
Chemical rate equations for the genetic toggle switch
-----------------------------------------------------
An important theme throughout PHYS 170/171 is that physics is concerned with providing mathematical descriptions of the natural world. Therefore, in the lecture portion of the class, building on earlier modules on chemical rate equations, and on coupled harmonic oscillators, we analyze the genetic toggle switch as follows: First, we write down chemical rate equations for the concentrations of the two repressors. Next, we look for and find steady-state solutions to these equations. Then, we determine which of the steady-state solutions are [*unstable*]{} and which are [*stable*]{}. Finally, we interpret the stable solutions that we find, and examine how these solutions depend on the model’s parameters.
The chemical rate equations that we write to describe the concentrations of repressor 1 ($c_1$) and repressor 2 ($c_2$) are: $$\frac{{\rm d\,}c_1}{{\rm d\,}t} = -K_1 c_1 + \gamma_1 [1-P_1(c_2)].
\label{RATE1}$$ and $$\frac{{\rm d\,}c_2}{{\rm d\,}t} = -K_2 c_2 + \gamma_2 [1-P_2(c_1)],
\label{RATE2}$$ where $c_1$ is the concentration of repressor 1, $c_2$ is the concentration of repressor 2, $K_1$ is the degradation rate of repressor 1, $K_2$ is the degradation rate of repressor 2, $\gamma_1$ is the production rate of repressor 1 in the absence of any repression by repressor 2, $\gamma_2$ is the production rate of repressor 2 in the absence of any repression by repressor 1, $P_1(c_2)$ is the probability that repressor 2 binds promoter 1, and $P_2(c_1)$ is the probability that repressor 1 binds promoter 2, thus repressing expression of repressor 2. Only that fraction of promoter 2 sites that are not occupied by repressor 1 can bind RNA polymerase and give rise to repressor 2 expression. Therefore, the production rate of repressor 2 in the presence of a concentration, $c_1$, of repressor 1 is actually $\gamma_2[1-P_2(c_1)]$. Similarly, the production rate of repressor 1 in the presence of a concentration, $c_2$, of repressor 2 is actually $\gamma_1[1-P_1(c_2)]$. Of course, a number of biological processes – transcription, RNA processing, RNA export, translation [*etc.*]{} – are subsumed into these equations and their parameters, but they do articulate the quote that “Everything should be made as simple as possible, but not simpler”, attributed to Albert Einstein.
For analytic simplicity, we will take $K_1=K_2$, $\gamma_1=\gamma_2$ and $P_1=P_2=P$ with $$P(c_1)=\frac{(c_1/c_0)^n}{1+(c_1/c_0)^n}.
\label{HILL}$$ EQ. \[HILL\] is a so-called Hill function and is an approximate representation of the concentration dependence of the binding probability and where $c_0$ is the repressor concentration for one-half occupancy of the promoter sites. A value of $n$ greater than 1 generally represents cooperativity. Roughly speaking, the larger the value of $n$, the greater is the degree of cooperativity. An example of cooperativity, well-known to the students, occurs for oxygen binding by hemoglobin, which permits oxygen uptake in the lungs where the concentration of oxygen in blood is high and oxygen release where the concentration of oxygen is low and oxygen is needed in the body. In that case, a Hill function (EQ. \[HILL\]) with $n=4$ is often used to describe the probability that hemoglobin binds four oxygen molecules. Importantly, in order to realize bistable behavior in the genetic toggle switch some cooperativity is required, [*i.e.*]{} $n$ must be greater than 1.
With the simplifications described above, EQ. \[RATE1\] and EQ. \[RATE2\] become $$\frac{1}{K} \frac{{\rm d\,}x}{{\rm d\,}t} = - x + \frac{a}{1+y^n}
\label{RATE1B}$$ and $$\frac{1}{K} \frac{{\rm d\,}y}{{\rm d\,}t} = -y + \frac{a}{1+x^n} ,
\label{RATE2B}$$ where $x = c_1/c_0$, $y = c_2/c_0$, and $a = {\gamma}/{(K c_0)}$.
Mathematica solutions for the genetic toggle switch
---------------------------------------------------
Mathematica can numerically solve these equations (EQ. \[RATE1B\] and EQ. \[RATE2B\]) versus time for specified parameters and initial conditions. The solution is presented to the class as a Mathematica Demonstration.[@gtsdemo]
![Output of the Mathematica Demonstration [@gtsdemo] that solves EQ. \[RATE1B\] and EQ. \[RATE2B\] for the genetic toggle switch. In this case, $\gamma/(K c_0)=3.0$, the initial concentrations are $x(0)=0.6$ and $y(0)=0.4$, and the time axis is actually $Kt$. []{data-label="GeneticToggleDemo1"}](GeneticToggleDemo1.pdf)
By varying the sliders in this demonstration, it is possible to vary the parameters of the model and the initial conditions. This exercise reveals that for some parameters there is a steady-state bistable solution, where $x$ (green) is large and $y$ (orange) is small, as shown in Fig. \[GeneticToggleDemo1\], or [*vice versa*]{}, and that it is possible to switch between these two solutions by changing only the initial conditions. For other parameters, there is a steady-state solution with $x=y$, irrespective of the initial conditions.
Steady-state solutions for the genetic toggle switch
----------------------------------------------------
For any parameters and initial conditions, it is evident from the Mathematica demonstration that the concentrations, $x$ and $y$, approach constant values at long times, $x^*$ and $y^*$, respectively. These values are the steady-state solutions of EQ. \[RATE1B\] and EQ. \[RATE2B\] , defined via $$- x^* + \frac{a}{1+(y^*)^n} =0
\label{SS1A}$$ and $$-y^* + \frac{a}{1+(x^*)^n} =0.
\label{SS1B}$$ In order to proceed analytically,[@PBC] we specialize to the case $n=2$, in which case WolframAlpha can solve EQ. \[SS1A\] and EQ. \[SS1B\]. First, we consider non-bistable solutions for which $x=y$. In this case, both EQ. \[SS1A\] and EQ. \[SS1B\] for $n=2$ become equivalent to: $$-x^* + \frac{a}{1+(x^*)^2} =0.
\label{SS1C}$$ To solve EQ. \[SS1C\], we navigate to the Wolfram Alpha website. <http://www.wolframalpha.com>,[@WolframAlpha] and enter:
solve -x+a/(1+x^2)=0 for x
to find: $$x^*=\frac{(\sqrt{3}\sqrt{27a^2+4}+9a)^{\frac{1}{3}}}{2^\frac{1}{3}3^\frac{2}{3}}
-
\frac{(2/3)^\frac{1}{3}}{(\sqrt{3}\sqrt{27a^2+4}+9a)^{\frac{1}{3}}}.
\label{solution1}$$ To solve EQ. \[SS1A\] and EQ. \[SS1B\] in the case that $x \neq y$, we enter
solve -x+a/(1+y^2)=0,-y+a/(1+x^2)=0 for x,y
into Wolfram Alpha. In this case, there are three real solutions. One of them is just that given in EQ. \[solution1\] and is real for all values of $a$. The other two solutions are: $$x=\frac{1}{2} \left ( a+\sqrt{a^2-4} \right ),~
y=\frac{1}{2} \left ( a-\sqrt{a^2-4} \right )
\label{no1}$$ or $$x=\frac{1}{2} \left ( a-\sqrt{a^2-4} \right ),~
y=\frac{1}{2} \left ( a+\sqrt{a^2-4} \right ),
\label{no2}$$ both of which are real only for $a>2$. These two solutions are the bistable solutions: either $x$ is large and $y$ is small (EQ. \[no1\]) or $y$ is large and $x$ is small (EQ. \[no2\]). All three real solutions are plotted together in Fig. \[ROOTS\] in magenta, red and blue. For the magenta solution, $x=y$. For the second solution, $x$ is the blue line and $y$ is the red line. For the third solution, $x$ is the red line and $y$ is the blue line.
![The three real solutions of EQ. \[SS1A\] and EQ. \[SS1B\] plotted versus the (normalized) repressor production rate divided by the repressor degradation rate \[$a = {\gamma}/{(K c_0)}$\]. The two bistable solutions are plotted in blue and red. The other solution is plotted in magenta. []{data-label="ROOTS"}](ROOTS.pdf)
Stable or unstable?
-------------------
To determine whether or not a particular solution is stable, we consider concentrations that are slightly different from the steady-state solutions that we just found, [*i.e.*]{}, we set $$x = x^* +\delta
\label{D1}$$ and $$y = y^* +\epsilon,
\label{D2}$$ where $x^*$ and $y^*$ are the steady state solutions that we just found and $\delta$ and $\epsilon$ are small deviations. For a stable solution, $\delta$ and $\epsilon$ will evolve to smaller values in time. For an unstable solution, $\delta$ and $\epsilon$ will evolve to larger values in time.
To determine the time evolution of $\delta$ and $\epsilon$, we substitute EQ. \[D1\] and EQ. \[D2\] into EQ. \[RATE1B\] and EQ. \[RATE2B\] (with $n=2$) to obtain at linear order in $\delta$ and $\epsilon$. The results are: $$\frac{1}{K} \frac{{\rm d\,} \delta}{{\rm d\,}t} =
-\delta - Y \epsilon,
\label{RATE1d}$$ where $$Y= \frac{2ay^*}{\left[1+(y^*)^2\right]^2},$$ and, similarly $$\frac{1}{K} \frac{{\rm d\,} \epsilon}{{\rm d\,}t} = -\epsilon - X \delta,
\label{RATE2d}$$ where $$X= \frac{2ax^*}{\left[1+(x^*)^2\right]^2}.$$
EQ. \[RATE1d\] and EQ \[RATE2d\] are similar to equations that describe coupled harmonic oscillators, that the students have seen earlier in the year. Consequently, they are comfortable that these equations realize normal modes, characterized by eigenvalues and eigenvectors. Since we are interested in whether $\delta$ and $\epsilon$ shrink or grow versus time, the key quantities that we need to find are the eigenvalues, which report upon the decay (or growth) rates for particular linear combinations of $\delta$ and $\epsilon$. To determine the eigenvalues, we assume that $\delta$ and $\epsilon$ decay exponentially decay in time, [*i.e.*]{}, we assume that $\delta = D e^{-\Gamma t}$ and $\epsilon= E e^{-\Gamma t}$. We then substitute these guesses into EQ. \[RATE1d\] and EQ. \[RATE2d\], and solve for $\Gamma$ in terms of the parameters of the problem. Finally, we decide whether $\Gamma$ is positive, corresponding to $\delta$ and $\epsilon$ that decrease in time and therefore a stable solution, or whether $\Gamma$ is negative, corresponding to an unstable solution.
Following the first step of this procedure, we find $$-\frac{\Gamma}{K} {\delta}= -\delta - Y \epsilon.
\label{RATE2e}$$ $$-\frac{\Gamma}{K} {\epsilon} = -\epsilon - X \delta,$$ whence $$\Gamma = K(1 \pm \sqrt{Y X}).
\label{GAMMA}$$ EQ. \[GAMMA\] is applicable to all of the solutions we have found previously. It is simply necessary to use the appropriate values of $Y$ and $X$.
First, we examine the bistable solutions. Using the expressions given in EQ. \[no1\] and EQ. \[no2\], we find $$\Gamma = K\left(1\pm \frac{2}{a}\right).
\label{GAMMABISTABLE}$$ Recalling that the bistable solutions are real only for $a>2$, EQ. \[GAMMABISTABLE\] informs us that the values of $\Gamma$ corresponding to the two eigenmodes of the bistable solutions are both invariably positive. Consequently, they both correspond to decaying exponentials, and we see that the bistable solutions are, in fact, stable.
For the other real solution, corresponding to the magenta curve in Fig. \[ROOTS\], we have $x^*=y^*$, so that $X=Y$, and $$\Gamma = K(1\pm Y).\label{GAMMANONBISTABLE}$$ It is straightforward to show that $Y$ is less than unity for $a<2$ and greater than unity for $a>2$. It follows that for $a<2$, $\Gamma$ is positive corresponding to a stable solution. However, for $a>2$, $\Gamma$ is negative. We only need one of the eigenvalues to be negative to send us away from the steady-state solution. Therefore, this is indeed an unstable solution, and it is not realized, because any small fluctuation grows away from it. Such a fluctuation away from the unstable solution will eventually approach one of the stable solutions, as is apparent from the Mathematica demonstration. Finally, then, we can report the “equation of state” of the genetic toggle switch in Fig. \[PHASEDIAGRAM\]. For $a<2$, there is a single solution with $c_1=c_2$, [*i.e.*]{}, there is no bistability. By contrast, for $a>2$, there are two bistable solutions with $x>y$ or $y > x$. Importantly, whether or not bistable behavior is realized depends on the parameters of the model. In class, we discuss that the model’s prediction of bistable behavior is just what is observed in the experiments of Ref. .
![Steady-state concentrations for the genetic toggle switch plotted versus the ratio of the production rate divided by the degradation rate. []{data-label="PHASEDIAGRAM"}](GeneticTogglePhaseDiagram.pdf)
The Repressilator
=================
Fig. \[REPRESSILATOR\] shows the genetic architecture of the repressilator. This gene circuit is composed of three genes and their gene products, each of which represses expression of the following gene’s gene product. This circuit uses the same basic element as the gene toggle switch, but uses three of them, rather than two. The odd number of elements going around the complete circuit gives rise to negative feedback. We will see that negative feedback around a circuit can lead to oscillations.
![ Schematic of the repressilator plasmid. The solid orange region is the promoter for repressor 1. The solid green region is the promoter for repressor 2. The solid red region is the promoter for repressor 3. The light orange region codes for repressor 1, which, as indicated, represses expression of repressor 2. The light green region codes for repressor 2, which, as indicated, represses expression of repressor 3. The light red region codes for repressor 3, which, as indicated, represses expression of repressor 1. []{data-label="REPRESSILATOR"}](RepressilatorPlasmid.pdf){width="35.00000%"}
![Top: Three inverters connected so that the output of the third inverter is routed to the input of the first inverter. Middle: Quad NAND gate chip used to produce three inverters. Bottom: Three inverters with resistors and capacitors incorporated in the circuit to achieve a convenient oscillation period. []{data-label="ThreeInverters"}](ThreeInvertersFB "fig:"){width="3in"} 0.5 in ![Top: Three inverters connected so that the output of the third inverter is routed to the input of the first inverter. Middle: Quad NAND gate chip used to produce three inverters. Bottom: Three inverters with resistors and capacitors incorporated in the circuit to achieve a convenient oscillation period. []{data-label="ThreeInverters"}](7400 "fig:"){width="2in"} 0.5 in ![Top: Three inverters connected so that the output of the third inverter is routed to the input of the first inverter. Middle: Quad NAND gate chip used to produce three inverters. Bottom: Three inverters with resistors and capacitors incorporated in the circuit to achieve a convenient oscillation period. []{data-label="ThreeInverters"}](3InverterRepressilatorSchematic "fig:"){width="3.5in"}
An electronic repressilator
---------------------------
In the PHYS 165/166 laboratory class, students are invited to analyze the Boolean logic when three inverters are connected as shown at the top of Fig. \[ThreeInverters\]. In this case, if input A is [True]{}, then B is [False]{}, then C is [True]{}, then A is [False]{}. But this configuration is inconsistent, because our original assumption was that input A was [True]{}. Do things work out better if we start off with input A [False]{}? Then B is [True]{}, then C is [False]{}, then A is [True]{}. The logic seems to fail in both cases. The resolution of this contradiction is to admit that in any actual circuit, whether it is genetic, electronic, or mathematical, it will take a non-zero period of time to switch from [True]{} to [False]{} and [*vice versa*]{}. Then, it turns out, the “logic” succeeds: Each inverter will switch states, then the next, then the next, then the first again, [*ad infinitum*]{}. In this way, we may realize an oscillator.
We realize an electronic version of the repressilator using three NAND gates of a 7400 quad NAND chip (Fig. \[ThreeInverters\]). To give the circuit a convenient oscillation period, we introduced capacitors and resistors as shown at the bottom of Fig. \[ThreeInverters\]. Because the capacitor takes a finite amount of time to charge or discharge through the resistor after the output of each NAND gate has switched, a delay is incorporated between the time at which the output of each NAND gate switches and the time at which the input to the next NAND gate reaches the threshold voltage for switching the next NAND gate. The specific values used are $R=5$ k$\Omega$ and $C=100$ $\mu$F, which yield an oscillation period of approximately 1 s. To provide a vivid readout of the state of the repressilator circuit, red, yellow and green LEDs are introduced (see Fig. \[ThreeInverters\]). As shown in Supplementary Movie 1 on YouTube, [http://www.youtube.com/v/sKtb5SFgv4](http://www.youtube.com/watch?v=S-Ktb5SFgv4), they oscillate happily with the advertised period. In the laboratory class, students are asked to examine the signals on an oscilloscope attached to the inputs of the NAND chip, pins 1,4, and 9, and determine more precisely the period of an entire cycle of flashing lights, as shown in Fig. \[BOTH\], which also illustrates the breadboard implementation of our electronic repressilator.
![ Photograph of our implementation of an electronic repressilator (top) together with the voltage outputs observed on an oscilloscope (bottom). The shape or voltage outputs is strikingly similar to the Mathematica traces in Fig. \[RepressScreenShot\] []{data-label="BOTH"}](RepressE1a.pdf "fig:") ![ Photograph of our implementation of an electronic repressilator (top) together with the voltage outputs observed on an oscilloscope (bottom). The shape or voltage outputs is strikingly similar to the Mathematica traces in Fig. \[RepressScreenShot\] []{data-label="BOTH"}](RepressE2.pdf "fig:")
Chemical rate equations for the repressilator
---------------------------------------------
Our analysis of the repressilator builds directly on the previous analysis of the genetic toggle switch. Consequently, the analysis presented here is simpler and, we believe, more accessible to our students than that given in Ref. . For the repressilator, we simply add a third gene, so that the relevant chemical rate equations become: $$\frac{1}{K} \frac{{\rm d\,}x_1}{{\rm d\,}t} = - x_1 + \frac{a}{1+x_2^n},
\label{RATE99A}$$ $$\frac{1}{K} \frac{{\rm d\,}x_2}{{\rm d\,}t} = -x_2 + \frac{a}{1+x_3^n} ,
\label{RATE99B}$$ and $$\frac{1}{K} \frac{{\rm d\,}x_3}{{\rm d\,}t} = -x_3 + \frac{a}{1+x_1^n},
\label{RATE99C}$$ where now $x_1=c_1/c_0$, [*etc.*]{} In this case, we will leave $n$ as is. The reason is that it will turn out that this system does not realize sustained oscillations for $n \leq 2$ for any value of $a$. We will determine an approximate “phase diagram” of the repressilator as as function of $a$ and $n$.
[![ Screenshot, showing our Mathematica Demonstration [@repressdemo] of the behavior of the repressilator for $a=12$, $n=6$, $x_1(0)=2$, $x_2(0)=2$, and $x_3(0)=4$. To investigate the solutions on their own, students are able to adjust the parameters of the simulation, using the sliders and toggles, directly from their web browsers. []{data-label="RepressScreenShot"}](./RepressilatorShot.pdf "fig:"){width="48.00000%"}]{}
Mathematica solutions for the repressilator
-------------------------------------------
This system of equations (EQ. \[RATE99A\], EQ. \[RATE99B\], and EQ. \[RATE99C\]) can be solved numerically by Mathematica, and we have written and published a [Wolfram Demonstration](http://demonstrations.wolfram.com/Repressilator/) that does just this.[@repressdemo] A screen shot showing $x_1$, $x_2$, and $x_3$ as a function of time is presented in Fig. \[RepressScreenShot\]. For certain parameter values, the numerical solution approaches a steady-state value for the concentrations after decaying oscillatory behavior at early times. For other parameter values, including those corresponding to Fig. \[RepressScreenShot\], the numerical solution shows sustained, constant-amplitude oscillations. In this latter case, this is a biologic clock, which might be a model for the clock that operates in cell division, or establishes a circadian rhythm. These sustained oscillations are not a steady-state solution, but they are the interesting solution in this case. Therefore, in this case, we want to find the range of parameter values for which such unstable, oscillatory solutions are observed. Nevertheless, our procedure in this case will mirror the procedure that we followed in the case of the genetic toggle switch: We will find the steady-state solutions. Next, we will derive the equations that describe how small deviations ($\delta_1$, $\delta_2$, and $\delta_3$) in concentration from the steady state values evolve in time. Then, we will assume an exponential time dependence for $\delta_1$, $\delta_2$, and $\delta_3$ and solve for the corresponding values of $\Gamma$ in terms of the parameters of the problem. Where $\Gamma$ is positive corresponds to a stable solution. Where $\Gamma$ is negative corresponds to an unstable solution, which in this case, is the more interesting solution.
Steady-state solutions for the repressilator
--------------------------------------------
In a steady-state, EQ. \[RATE99A\], EQ. \[RATE99B\] and EQ. \[RATE99C\] become $$x^*_1 = \frac{a}{1+(x^*_2)^n},
\label{RATE99D}$$ $$x^*_2 = \frac{a}{1+(x^*_3)^n} ,
\label{RATE99E}$$ and $$x^*_3 = \frac{a}{1+(x^*_1)^n},
\label{RATE99F}$$ where the $^*$ indicates the steady-state value. These equations initially seem daunting to solve, but exploration of the numerical solution, given in the Mathematica demonstration, does not reveal any steady-state solutions for which the concentrations ($x_1$, $x_2$, and $x_3$) are different from each other. Therefore, in fact, all of the real solutions to EQ. \[RATE99A\], EQ. \[RATE99B\] and EQ. \[RATE99C\] correspond to $x^*_1=x^*_2=x^*_3= x^*$, say. (Since concentrations must be real, we are interested solely in real solutions.) In this case, each of EQ. \[RATE99D\], EQ. \[RATE99E\], and EQ. \[RATE99F\] reduces to $$x^* = \frac{a}{1+(x^*)^n}.
\label{RATE100}$$ It is straightforward to see graphically that EQ. \[RATE100\] has only one real solution: we plot $y = x^*$ and $y={a}/[{1+(x^*)^n]}$, and where these two curves cross corresponds to the real solutions for $x^*$ that we seek. It is clear from this plot (not shown) that there is always one and only one intersection point, and therefore one and only one real solution, irrespective of the value of $a$.
Stability analysis for the repressilator
----------------------------------------
To examine the stability of this solution, we proceed similarly to above. That is, we assume $x_1 = x^*+\delta_1$, $x_2=x^*+\delta_2$, $x_3=x^*+\delta_3$, and carry out a linear expansion of EQ. \[RATE99A\], \[RATE99B\], and \[RATE99C\] leading to $$\frac{1}{K} \frac{{\rm d\,} \delta_1}{{\rm d\,}t} = -\delta_1 - X_n \delta_2,
\label{RATE3d}$$ $$\frac{1}{K} \frac{{\rm d\,} \delta_2}{{\rm d\,}t} = -\delta_2 - X_n \delta_3,
\label{RATE3e}$$ and $$\frac{1}{K} \frac{{\rm d\,} \delta_3}{{\rm d\,}t} = -\delta_3 - X_n \delta_1,
\label{RATE3f}$$ where $$X_n = \frac{n\left(x^*\right)^{n-1}a}{\left(1+\left(x^*\right)^n\right)^2}.
\label{YYY}$$ Next, we assume that $\delta_1 = D_1 e^{-\Gamma t}$, $\delta_2 = D_2 e^{-\Gamma t}$, and $\delta_3 = D_3 e^{-\Gamma t}$, with the result that $${\Gamma} =K\left[1+X_n\right],
\label{Y4a}$$ or $${\Gamma} =K\left[1+\frac{X_n}{2} + i\frac{\sqrt{3}X_n}{2}\right],
\label{Y4b}$$ or $${\Gamma} =K\left[1-\frac{X_n}{2} + i\frac{\sqrt{3}X_n}{2}\right].
\label{Y4c}$$ Since $X_n$ is positive, EQ. \[Y4a\] corresponds to an exponential decay. Since $1+X_n/2$ is also positive, EQ. \[Y4b\] corresponds to an exponentially decaying oscillation. Therefore, both of these eigenvalues correspond to stable solutions. EQ. \[Y4c\] is stable for $X_n<2$. However, for $X_n > 2$, $1 - X_n/2$ is negative, and this solution grows exponentially in an oscillating fashion. Therefore, the condition to realize an unstable solution is $X_n >2$, which in fact corresponds to sustained oscillations.
“Phase diagram” for the repressilator
-------------------------------------
![Approximate repressilator phase diagram. The region above the red line corresponds to sustained oscillations – a limit cycle. The region below the red line corresponds to stable solutions for which $x_1^*=x_2^*=x_3^*$. []{data-label="RPD"}](RepressilatorPhaseDiagram.pdf){width="3.2in"}
What is the value of $X_n$? So far, we have not calculated an explicit value for $X_n$, because we have not calculated an explicit value for $x^*$. This is because it is not possible to write down the solution of EQ. \[RATE100\] analytically for arbitrary $n$. However, we can find a useful, approximate solution as follows. First we note that using EQ. \[RATE100\], EQ. \[YYY\] may be rewritten: $$X_n = \frac{n\left(x^*\right)^{n+1}}{a}.
\label{YYY2000}$$ We may also re-write EQ. \[RATE100\] as $$(x^*)^{n+1} + x^* = a.
\label{RATE10000}$$ If we assume that $(x^*)^n \gg 1$, it follows from EQ. \[RATE10000\] that $$(x^*)^{n+1} \simeq a-a ^{\frac{1}{n+1}}.
\label{ABAB}$$ Combining EQ. \[YYY2000\] and EQ. \[ABAB\], we find $$X_n \simeq n(1-a^{-n/(n+1)}),
\label{YOY}$$ which represents a useful approximate expression for $X_n$ that we can use to determine the stability of the steady state solution. First, notice that for $n=2$, $X_2$ is inevitably less than 2. This result survives an exact calculation, which is possible for $n=2$. Therefore, there are no sustained oscillations for $n=2$. It’s a different story for $n>2$. In this case, according to EQ. \[Y4c\] the steady state solution is unstable for $$n(1-a^{-n/(n+1)}) > 2.$$ After some algebra, this condition becomes that the steady state solution is unstable for $$a>\left(1-\frac{2}{n}\right)^{-\left(1+\frac{1}{n}\right)}.
\label{REPRESS_PHASE_DIA}$$ EQ. \[REPRESS\_PHASE\_DIA\] represents a “phase diagram” for the repressilator, specifying the region of $n$-$a$ “phase space” in which the steady-state solution is unstable and which therefore realizes sustained oscillations. This (approximate) repressilator phase diagram is show in Fig. \[RPD\]. The region above the line corresponds to sustained oscillations – a limit cycle. The region below the line corresponds to a stable fixed point. Fig. \[RPD\] is approximate because EQ. \[YOY\] is approximate. Nevertheless, when the students run the repressilator Mathematica demonstration, they are soon able to convince themselves that our analytic phase diagram is qualitatively correct. Specifically, they find that to realize sustained oscillations requires larger values of $a$ at smaller values of $n$ and relatively smaller values of $a$ at larger values of $n$. Again, we see that the behavior realized depends on biochemical parameters of the model. And again, in class, we discuss that the model’s prediction of oscillatory behavior is what is observed in the experiments of Ref. .
Summary
=======
We have discussed and analyzed two prototypical gene circuits with feedback, namely the genetic toggle switch and the repressilator, which together constitute the final topic in a year-long introductory physics sequence for biology and pre-medical students at Yale. Our analytic, numerical, and electronic treatments of the genetic toggle switch, which consists of two genes, whose protein products each represses the other’s gene expression, reveals that this circuit realizes bistability. Our new, simplified treatment of the repressilator, which consists of three genes, each of whose protein product represses the next gene’s expression, reveals that this circuit realizes sustained oscillations for certain parameter values. In both cases, we obtained a “phase diagram” that specifies the region of parameter space in which bistability or oscillatory behavior, respectively, are realized.
We thank the PHYS 170/171 and PHYS 165/166 classes for their participation, and Sean Barrett, Ross Boltyanskiy, Diego Caballero, Rick Casten, Betsy Cowell, Jane Cummings, Stefan Elrington, Merideth Frey, Eric Holland, Syed Hussaini, Sohrab Ismail-Beigi, Anna Kashkanova, Peter Koo, Andrew Mack, Wambui Muturi, Rona Ramos, Raphael Sarfati, William Segraves, Gennady Voronov, Christine Willinger, and Yao Zhao for valuable discussions. SGJM acknowledges support from the NSF via PHY 1019147.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this work we find necessary and sufficient conditions for a free nilpotent or a free metabelian nilpotent Lie algebra to be endowed with an ad-invariant metric. For such nilpotent Lie algebras admitting an ad-invariant metric the corresponding automorphisms groups are studied.'
address: 'CONICET and ECEN-FCEIA, Universidad Nacional de Rosario, Pellegrini 250, 2000 Rosario, Argentina. '
author:
- 'Viviana J. del Barco'
- 'Gabriela P. Ovando'
title: 'Free nilpotent Lie algebras admitting ad-invariant metrics'
---
[^1] [^2]
Introduction
============
An ad-invariant metric on a Lie algebra ${\mathfrak g }$ is a nondegenerate symmetric bilinear form $\la \,,\, \ra$ which satisfies $$\label{adme}
\la [x, y], z\ra + \la y, [x, z]\ra = 0 \qquad \mbox{ for all }x, y, z \in {\mathfrak g }.$$
Lie algebras endowed with ad-invariant metrics (also called “metric” or “quadratic”) became relevant some years ago when they were useful in the formulation of some physical problems such as the known Adler-Kostant-Symes scheme. They also constitute the basis for the construction of bialgebras and they give rise to interesting pseudo-Riemannian geometry [@Co]. For instance in [@Ov1] a result originally due to Kostant [@Kos] was revalidated for pseudo-Riemannian metrics: it states that the Lie algebra of the isometry group of a naturally reductive pseudo-Riemannian space (in particular symmetric spaces) can be endowed with an ad-invariant metric.
Semisimple Lie algebras are examples of Lie algebras admitting an ad-invariant metric since the Killing form is nondegenerate. In the solvable case, the Killing form is degenerate so one must search for another bilinear form with the ad-invariance property. The first investigations concerning general Lie algebras with ad-invariant metrics appeared in [@FS; @MR]. They get structure results proposing a method to construct these Lie algebras recursively. This enables a classification of nilpotent Lie algebras admitting ad-invariant metrics of dimension $\leq 7$ in [@FS] and a determination of the Lorentzian Lie algebras in [@Me]. The point is that by this recursive method one can reach the same Lie algebra starting from two non-isomorphic Lie algebras. This fact difficulties the classification in higher dimensions. More recently a new proposal for the classification problem is presented in [@KO] and this is applied in [@Ka] to get the nilpotent Lie algebras with ad-invariant metrics of dimension $\leq 10$.
However the basic question whether a non-semisimple Lie algebra admits such a metric is still opened. In the present paper we deal with this problem in the family of free nilpotent and free metabelian nilpotent Lie algebras.
[**Theorem**]{} \[t1\]. [*Let ${\mathfrak n }_{m,k}$ be the free $k$-step nilpotent Lie algebra in $m$ generators. Then ${\mathfrak n }_{m,k}$ admits an ad-invariant metric if and only if $(m,k)=(3,2)$ or $(m,k)=(2,3)$.*]{}
The techniques for the proof do not make use of the extension procedures mentioned before, but properties of free nilpotent Lie algebras which combined with the ad-invariance condition enable the deduction of the Lie algebras ${\mathfrak n }_{2,3}$ and ${\mathfrak n }_{3,2}$. We note that for $k=2,3$ the free and free metabelian $k$-step nilpotent Lie algebras coincide. For the free metabelian case the first approach lies in the fact that 2-step solvable Lie algebras admitting ad-invariant metrics are nilpotent and at most 3-step (Lemma \[le11\]). Thus working out we get the next result.
[**Theorem**]{} \[t2\]. [*Let $\tilde{{\mathfrak n }}_{m,k}$ be the free metabelian $k$-step nilpotent Lie algebra in $m$ generators. Then $\tilde{{\mathfrak n }}_{m,k}$ admits an ad-invariant metric if and only if $(m,k)=(3,2)$ or $(m,k)=(2,3)$.*]{}
These two Lie algebras have been studied since a long time in sub-Riemannian geometry [@Mo]. Thus ${\mathfrak n }_{2,3}$ is associated to the Carnot group distribution (see for instance [@BM]), which is related to the “rolling balls problem”, treated by Cartan in [@Ca]. The prolongation, representing -roughly speaking- the maximal possible symmetry of the distribution, in the case of ${\mathfrak n }_{2,3}$ is the exceptional Lie algebra ${\mathfrak g }_2$ [@BM]. The Lie algebra ${\mathfrak n }_{3,2}$ was studied more recently in [@My] in the context of the geometric characterization of the so-called Maxwell set, wave fronts and caustics (see also [@MA]).
We complete the work with a study of the group of automorphisms of the Lie algebras ${\mathfrak n }_{2,3}$ and ${\mathfrak n }_{3,2}$. The corresponding structure is described and in particular the subgroup of orthogonal automorphisms is determined.
Following [@FS] and the considerations above all Lie algebras here are over a field $K$ of characteristic 0, nevertheless some results in Section 3 could be still true for fields of a characteristic different from 2.
Free and free metabelian nilpotent Lie algebras
===============================================
Let ${\mathfrak g }$ denote a Lie algebra. The so-called central descending and ascending series of ${\mathfrak g }$, respectively $\{C^r({\mathfrak g })\}$ and $\{C_r({\mathfrak g })\}$ for all $r\geq 0$, are constituted by the ideals in ${\mathfrak g }$, which for non-negative integers $r$, are given by $$\begin{array}{rclrcl}
C^0({\mathfrak g })&=&{\mathfrak g }& C_0({\mathfrak g })&=&0 \\
C^r({\mathfrak g })&=&[{\mathfrak g },C^{r-1}({\mathfrak g })] & C_r({\mathfrak g })&=&\{x\in {\mathfrak g }:[x,{\mathfrak g }]\in C_{r-1}({\mathfrak g })\}.
\end{array}$$
Note that $C_1({\mathfrak g })$ is by definition the center of ${\mathfrak g }$, which will be denoted by ${\mathfrak z }({\mathfrak g })$.
A Lie algebra ${\mathfrak g }$ is called *k-step nilpotent* if $C^k({\mathfrak g })=\{0\}$ but $C^{k-1}({\mathfrak g })\neq \{0\}$ and clearly $C^{k-1}({\mathfrak g })\subseteq {\mathfrak z }({\mathfrak g })$.
Heisenberg algebras. Let $X_1, \hdots, X_n, Y_1, \hdots, Y_n$ denote a basis of the $2n$-dimensional real vector space $V$ and let $Z\notin V$. Define $[X_i, Y_j]=\delta_{ij} Z$ and $[Z, U]=0$ for all $U\in V$. Thus $\hh_n= V \oplus \RR Z$ is the Heisenberg Lie algebra of dimension $2n+1$, which is 2-step nilpotent.
We shall make use of the notation ${\mathfrak g }'=[{\mathfrak g }, {\mathfrak g }]$ and ${\mathfrak g }''=[{\mathfrak g }',{\mathfrak g }']$. A Lie algebra is called [*2-step solvable*]{} if its commutator is abelian, that is ${\mathfrak g }''=0$.
Let $\mathfrak f_m$ denotes the free Lie algebra on $m$ generators, with $m\geq 2$. (Notice that a unique element spans an abelian Lie algebra). Thus
- the free metabelian $k$-step nilpotent Lie algebra on $m$ generators is defined as $$\tilde{{\mathfrak n }}_{m,k}:= \mathfrak f_m/(C^{k+1}(\mathfrak f_m)+ \mathfrak f_m''),$$
- the *free $k$-step nilpotent* Lie algebra on $m$ generators ${\mathfrak n }_{m,k}$ is defined as the quotient algebra $${\mathfrak n }_{m,k}:=\mathfrak f_m/C^{k+1}(\mathfrak f_m).$$
In particular free metabelian nilpotent of any degree are 2-step solvable.
\[23\] For $k=2,3$ any $k$-step nilpotent Lie algebra is 2-step solvable, which follows from the Jacobi identity. Thus for the free nilpotent ones we get $\tilde{{\mathfrak n }}_{m,k}={\mathfrak n }_{m,k}$ for $k=2,3$.
Let ${\mathfrak n }_{m,k}$ be a free $k$-step nilpotent Lie algebra and let $\{e_1,\ldots,e_m\}$ be an ordered set of generators. The construction of a *Hall* basis associated to this sets of generators is explained below (see [@Ha; @GG]).
Start by defining the *length* $\ell$ of each generator as $1$. Take the Lie brackets $[e_i, e_j]$ for $i>j$, which by definition satisfy $\ell([e_i,e_j])=2$. Now the elements $e_1, \dots , e_m$, $[e_i, e_j]$, $i>j$ belong to the Hall basis. Define a total order in that set by extending the order of the set of generators and so that $E>F$ if $\ell(E)>\ell(F)$. They allow the construction of the elements of length $3$ and so on.
Recursively each element of the Hall basis of ${\mathfrak n }_{m,k}$ is defined as follows. The generators $e_1,\ldots,e_m$ are elements of the basis of length 1. Assume we have defined basic elements of lengths $1,\ldots, r-1\leq k-1$, with a total order satisfying $E > F$ if $\ell(E) > \ell(F)$.
If $\ell(E) =s$ and $\ell(F) = t$ and $r = s + t\leq k$, then $[E, F]$ is a basic element of length $r$ if both of the following conditions hold:
- $E$ and $F$ are basis elements and $E>F$, and
- if $\ell(E)>1$ and $E=[G,H]$ is the unique decomposition with $G,H$ basic elements, then $F\geq H$.
This gives rise to a natural graduation of ${\mathfrak n }_{m,k}$: $${\mathfrak n }_{m,k}=\bigoplus_{s=1}^k \mathfrak p(m,s),$$ where $\mathfrak p(m,s)$ denotes the subspace spanned by the elements of the Hall basis of length $s$. Notice that
- $C^r({\mathfrak n }_{m,k})=\oplus_{s=r+1}^k\mathfrak p(m,s)$,
- $\mathfrak p(m,k) = {\mathfrak z }({\mathfrak n }_{m,k})$
The first assertion follows from the fact that every bracket of $r + 1$ elements of ${\mathfrak n }_{m,k}$, is a linear combination of brackets of $r + 1$ elements in the Hall basis (see proof of Theorem 3.1 in [@Ha]). This implies $C^r({\mathfrak n }_{m,k})\subseteq \oplus_{s=r+1}^k \mathfrak p(m, s)$; the other inclusion is obvious. In particular, $\mathfrak p(m, k) = C^{k-1}({\mathfrak n }_{m,k}) \subseteq \mathfrak z({\mathfrak n }_{m,k})$. Now let $x\in \mathfrak z({\mathfrak n }_{m,k})$ and let $e$ be a generator and assume $x\notin C^{k-1}({\mathfrak n }_{m,k})$. Recall that ${\mathfrak n }_{m,k}$ is homomorphic image of the free Lie algebra $\mathfrak f_m$ so that there exist $X,E\in \mathfrak f_m$ such that $X\to x$ and $E \to e$, being $E$ a generator of $\mathfrak f_m$. Since $[x, e]=0$ then $[X,E]=0$ which says $X$and $E$ are proportional, which is impossible (see for instance Ch. 2 in [@Ba]). Thus $\mathfrak p(m, k) = \mathfrak z({\mathfrak n }_{m,k})$.
Denote as $d_m(s)$ the dimension of $\mathfrak p(m,s)$. Inductively one gets [@Se] $$\label{dim}s\cdot d_m(s)= m^s- \sum_{r|s, r<s} r\cdot d_m(r),\qquad s\geq 1.$$
Hence for a fixed $m$, one has $d_m(1)=m$ and $d_m(2)=m(m-1)/2$.
\[ex:1\]Given an ordered set of generators $e_1,\ldots, e_m$ of a free $2$-step nilpotent Lie algebra ${\mathfrak n }_{m,2}$, a Hall basis is $$\label{ba2}
\mathcal B=\{e_i,\,[e_j,e_k]:\,i=1,\ldots,m,\, 1\leq k< j\leq m\}.$$ Equation (\[dim\]) asserts that $\dim{\mathfrak n }_{m,2}=d_m(1)+d_m(2)=m+m(m-1)/2$. Since $\mathfrak z({\mathfrak n }_{m,2}) = \mathfrak p(m, 2)$, we have $\dim \mathfrak z({\mathfrak n }_{m,2})= m(m-1)/2$.
\[ex:2\] For the free $3$-step nilpotent Lie algebra on $m$ generators ${\mathfrak n }_{m,3}$ a Hall basis of a set of generators as before has the form $$\label{ba3}
\mathcal B=\{e_i, \,[e_j,e_k],\,[[e_r,e_s],e_t]:\,i=1,\ldots,m,\,1\leq k< j\leq m,\, 1\leq s<r\leq m,\,t\geq s\}.$$
It holds ${\mathfrak z }({\mathfrak n }_{m,3})=\mathfrak p(m,3)$, and so $$\dim\, {\mathfrak z }({\mathfrak n }_{m,3})= d_m(3)=m(m^2-1)/3.$$
Free and free metabelian nilpotent Lie algebras and ad-invariant metrics
========================================================================
In this section we determine free nilpotent and free metabelian nilpotent Lie algebras admitting ad-invariant metrics.
Let ${\mathfrak g }$ denote a Lie algebra equipped with an ad-invariant metric $\la\,,\,\ra$, see (\[adme\]). If $\mm \subseteq \ggo$ is a subset, then we denote by $\mm^{\perp}$ the linear subspace of ${\mathfrak g }$ given by $$\mm^{\perp}=\{x\in \ggo, \la x, v\ra=0 \mbox{ for all } v\in \mm\}.$$ In particular $\mm$ is called
- [*isotropic*]{} if $\mm \subseteq \mm^{\perp}$,
- [*totally isotropic*]{} if $\mm=\mm^{\perp}$, and
- [*nondegenerate*]{} if and only if $\mm \cap \mm^{\perp}=\{0\}$.
The proof of the next result follows easily from an inductive procedure.
\[le1\] Let $(\ggo, \la\,,\,\ra)$ denote a Lie algebra equipped with an ad-invariant metric.
- If $\hh$ is an ideal of $\ggo$ then $\hh^{\perp}$ is also an ideal in $\ggo$.
- $C^r(\ggo)^{\perp}=C_r(\ggo)$ for all $r$.
Thus on any Lie algebra admitting an ad-invariant metric the next equality holds $$\label{e2}
\dim \ggo=\dim C^r(\ggo) + \dim C_r(\ggo).$$ For the case $r=1$ one obtains $$\label{e1}
\dim \ggo=\dim \zz(\ggo) + \dim C^1(\ggo).$$
Let ${\mathfrak n }$ denote a $2$-step nilpotent Lie algebra equipped with an ad-invariant metric. Assume $\zz({\mathfrak n })=C^1({\mathfrak n })$, then by (\[e1\]) the metric is neutral and $\dim {\mathfrak n }= 2 \dim \zz({\mathfrak n })$. As a consequence the Heisenberg Lie algebra $\hh_n$ cannot be equipped with any ad-invariant metric.
Examples of nilpotent Lie algebras satisfying the equality (\[e2\]) above for every $r$ arise by considering the semidirect product of a nilpotent Lie algebra $\nn$ with its dual space via de coadjoint representation $\nn \ltimes \nn^*$. The natural neutral metric on $\nn \ltimes \nn^*$ is ad-invariant.
Nevertheless, condition (\[e1\]) (and hence (\[e2\])) is not sufficient for a 2-step nilpotent Lie algebra to admit an ad-invariant metric as shown for instance in [@Ov2].
\[le11\] Let ${\mathfrak g }$ denote a 2-step solvable Lie algebra provided with an ad-invariant metric, then ${\mathfrak g }$ is nilpotent and at most 3-step.
Let $\la\,,\,\ra$ denote an ad-invariant metric on ${\mathfrak g }$. Since ${\mathfrak g }$ is 2-step solvable for all $x,y \in {\mathfrak g }'$ one has $[x,y]=0$, which is equivalent to $$\begin{array}{rcll}
0 & = & \la [x,y], u\ra \qquad & \mbox{ for all } u \in {\mathfrak g }\\
& = & \la [u, x], y\ra \qquad & \mbox{ for all } y\in {\mathfrak g }'
\end{array}$$ thus $[u,x]\in [{\mathfrak g }, {\mathfrak g }]^{\perp}=\zz({\mathfrak g })$, and since $x\in {\mathfrak g }'$ can be written as $x=[v,w]$, then $[u,[v,w]]\subseteq \zz({\mathfrak g })$ for all $u,v,w\in {\mathfrak g }$, that is $C^4({\mathfrak g })=0$ and so ${\mathfrak g }$ is at most 3-step nilpotent.
\[corofree\] Let $\tilde{{\mathfrak n }}_{m,k}$ denote a free metabelian nilpotent Lie algebra admitting an ad-invariant metric, then $k\leq 3$.
Remark \[23\] and the previous result says that a free metabelian nilpotent Lie algebra with an ad-invariant metric if it exists, is free nilpotent. Below we determine which free nilpotent Lie algebra admits such a metric.
Whenever ${\mathfrak n }_{m,k}$ is free nilpotent we have that $\dim {\mathfrak n }_{m,k}/C^1({\mathfrak n }_{m,k}) =m$ so that $$\label{e3}
\dim {\mathfrak n }_{m,k}=m+\dim C^1({\mathfrak n }).$$
Hence Equations (\[e1\]) and (\[e3\]) show that if ${\mathfrak n }_{m,k}$ admits an ad-invariant metric then $$\label{e4}
\dim {\mathfrak z }({\mathfrak n }_{m,k})=m.$$
\[p1\] If $\nn_{m,2}$ is a free 2-step nilpotent Lie algebra endowed with an ad-invariant metric, then $m=3$.
Let ${\mathfrak n }_{m,2}$ be the free 2-step nilpotent Lie algebra on $m$ generators. As we showed in Example \[ex:1\] its center has dimension $m(m-1)/2$. Now if Equation (\[e4\]) holds then $m=3$.
\[p2\] Let $\nn_{m,3}$ be a free 3-step nilpotent Lie algebra provided with an ad-invariant metric, then $m=2$.
As shown in Example \[ex:2\] the center ${\mathfrak z }({\mathfrak n }_{m,3})$ has dimension $d_m(3)=m(m^2-1)/3$. From straightforward calculations, if Equation (\[e4\]) is satisfied then $m= 2$.
\[p3\] No free $k$-step nilpotent Lie algebra ${\mathfrak n }_{m,k}$ on $m$ generators with $k \geq 4$ can be endowed with an ad-invariant metric.
*$\bullet$ 4-step nilpotent case:* In this case $\mathfrak p(m,4)= {\mathfrak z }({\mathfrak n }_{m,4})$, thus from (\[dim\]): $$\dim{\mathfrak z }({\mathfrak n }_{m,4})\geq d_m(4)=\frac{1}{4}(m^4-d_m(1)-2d_m(2))= \frac{m^2(m^2-1)}{4}.$$ Notice that for $m\geq 2$, one has $m^2(m^2-1)/4>m$.
*$\bullet$ General case, $k\geq 5$:* Let ${\mathfrak n }_{m,k}$ denote the free $k$-step nilpotent Lie algebra in $m$ generators. The goal here is to show that for every $m$ and $k\geq 5$ the dimension of the center of ${\mathfrak n }_{m,k}$ is greater than $m$. In order to give a lower bound for $\dim{\mathfrak z }({\mathfrak n }_{m,k})$ we construct elements of length $k$ in a Hall basis $\mathcal B$.
Let $\{e_1,\ldots,e_m\}$ a set of generators of ${\mathfrak n }_{m,k}$ and consider the set $$\begin{aligned}
\mathcal U &=&\{[[[e_i,e_j],e_k],e_m]: 1\leq j<i\leq m,\,k\geq j\}.\nonumber $$
Any element in $\mathcal U$ is basic and of length 4. Given $x\in \mathcal U$, the bracket $$[x,e_m]^{(s)}:= [[[x,\overbrace{e_m],e_m]\cdots , e_m]}^{s} \;\;s\geq 1$$ is an element in the Hall basis if $\ell([x,e_m]^{(s)})\leq k$.
In fact if $s=1$ then
- both $x=[[[e_i,e_j],e_k],e_m] \in \mathcal U$ and $e_m$ are elements of the Hall basis, and $x >e_m$ because of their length;
- also $x=[G,H]$ with $G=[[e_i,e_j],e_k]$ and $H=e_m$ and we have $e_m\geq H$.
So both conditions of the Hall basis definition are satisfied and hence $[x,e_m]^{(1)}\in \mathcal B$ and it belongs to $ C^4({\mathfrak n }_{m,k})$.
Inductively suppose $[x,e_m]^{(s-1)}\in\mathcal B$, then clearly $[[[x,e_m],e_m]\cdots ], e_m]^{(s-1)}>e_m$ and it is possible to write $[x,e_m]^{(s-1)}=[G,H]$ with $H=e_m$. Thus $[x,e_m]^{(s)}\in\mathcal B$. Notice that $[x,e_m]^{(s)}\in C^{s+3}({\mathfrak n }_{m,k})$.
We construct the new set $$\widetilde{\mathcal U}:=\{[x,e_m]^{(k-4)}:x\in \mathcal U\}\subseteq C^{k-1}({\mathfrak n }_{m,k}),$$ which is contained in the center of ${\mathfrak n }_{m,k}$ and it is linearly independent. Therefore $$\label{u}
\dim {\mathfrak z }({\mathfrak n }_{m,k}) \geq |\widetilde{\mathcal U}|.$$
Clearly $\widetilde{\mathcal U}$ and $\mathcal U$ have the same cardinal. Also, $|\mathcal U|=\sum_{j=1}^m(m-j+1)(m-j)$ since for every fixed $j=1,\ldots,m$, the amount of possibilities to choose $k\geq j$ and $i>j$ is $(m-j+1)$ and $(m-j)$ respectively.
Straightforward computations give $ |\widetilde{\mathcal U}|=1/3\, m^3+m^2+2/3\,m$ which combined with (\[u\]) proves that for any $m$ and $k\geq 5$ $$\dim {\mathfrak z }({\mathfrak n }_{m,k})\geq 1/3 \,m^3+m^2+2/3\,m .$$ The right hand side is greater than $m$ for all $m\geq 2$. According to (\[e4\]), the free $k$-step nilpotent Lie algebra ${\mathfrak n }_{m,k}$ does not admit an ad-invariant metric if $k\geq 5$.
\[t1\] Let ${\mathfrak n }_{m,k}$ be the free $k$-step nilpotent Lie algebra in $m$ generators. Then ${\mathfrak n }_{m,k}$ admits an ad-invariant metric if and only if $(m,k)=(3,2)$ or $(m,k)=(2,3)$.
Propositions (\[p1\]) (\[p2\]) and (\[p3\]) prove that if ${\mathfrak n }_{m,k}$ admits an ad-invariant metric then $(m,k)=(3,2)$ or $(m,k)=(2,3)$. Let us show the resting part of the proof.
The Lie algebra ${\mathfrak n }_{3,2}$ has a basis $\{e_1,e_2,e_3,e_4,e_5,e_6\}$ with non zero brackets $$\label{b32}
[e_1,e_2]=e_4,\qquad [e_1,e_3]=e_5,\qquad [e_2,e_3]=e_6.$$ Since $C^1({\mathfrak n }_{3,2})=\zz({\mathfrak n }_{3,2})$, if the metric $\la \,,\,\ra$ is ad-invariant the center of ${\mathfrak n }_{3,2}$ is totally isotropic, so it must hold $$\la e_i, e_j\ra=0 \qquad \mbox { for all }i,j=4,5,6.$$
The ad-invariance property says $$\la e_4, e_1\ra=\la [e_1,e_2], e_1\ra=0$$ and similarly $\la e_4,e_2\ra=0$, therefore $\la e_4,e_3\ra \neq 0$. Analogously, $\la e_5, e_2\ra\neq 0$ and $\la e_6, e_1\ra\neq 0$. Moreover $$\alpha=\la [e_1, e_2], e_3\ra=-\la e_2, e_5, \ra=\la e_1, e_6\ra.$$
Thus in the ordered basis $\{e_1,e_2,e_3,e_4,e_5,e_6\}$, a matrix of the form $$\label{m32}
\left(
\begin{matrix}
a_{11} & a_{12} & a_{13} & 0 & 0 & \alpha \\
a_{12} & a_{22} & a_{23} & 0 & -\alpha & 0 \\
a_{13} & a_{23} & a_{33} & \alpha & 0 & 0 \\
0 & 0 & \alpha & 0 & 0 & 0 \\
0 & -\alpha & 0 & 0 & 0 & 0 \\
\alpha & 0 & 0 & 0 & 0 & 0
\end{matrix}
\right) \qquad \mbox{ with } a_{ij} \in K, \forall i,j=1,2,3 \mbox{ and } \alpha \neq 0$$
corresponds to an ad-invariant metric on ${\mathfrak n }_{3,2}$.
The Lie algebra ${\mathfrak n }_{2,3}$ has a basis $\{e_1,e_2,e_3,e_4,e_5\}$ with non-zero Lie brackets $$\label{b23}
[e_1,e_2]=e_3,\qquad [e_1,e_3]=e_4,\qquad [e_2,e_3]=e_5.$$ Let $\la \,,\,\ra$ be an ad-invariant metric on ${\mathfrak n }_{2,3}$. Then $\zz({\mathfrak n }_{3,2})=C^2({\mathfrak n }_{2,3})=span\{e_4, e_5\}$ while $C^1({\mathfrak n }_{2,3})=C^2({\mathfrak n }_{2,3}) \oplus K e_3$.
The ad-invariance property also says that $$0=\la e_4, e_3\ra =\la e_4, e_4\ra= \la e_4, e_5\ra=\la e_5, e_3\ra=\la e_5, e_5\ra.$$ Moreover $$\la e_1, e_3\ra= \la e_1, [e_1, e_2]\ra =0 \quad \mbox { and } \quad \la e_2, e_3\ra=\la e_2, [e_1, e_2] \ra=0;$$ $$\la e_1, e_4\ra=\la e_1, [e_1, e_3]\ra=0 \quad \mbox{ and } \quad \la e_2, e_5\ra=\la e_2, [e_2, e_3]\ra=0.$$ Therefore $$\la e_3, e_3\ra=\la [e_1,e_2], e_3\ra=-\la e_2, e_4\ra= \la e_1, e_5\ra =\alpha \neq 0,$$ which amounts to the following matrix for $\la \,,\,\ra$ in the ordered basis above: $$\label{m23}
\left(
\begin{matrix}
a_{11} & a_{12} & 0 & 0 & \alpha \\
a_{12} & a_{22} & 0 & - \alpha & 0 \\
0 & 0 & \alpha & 0 & 0 \\
0 & -\alpha & 0 & 0 & 0 \\
\alpha & 0 & 0 & 0 & 0
\end{matrix}
\right)
\qquad \mbox{ with } a_{ij} \in K, \forall i,j=1,2 \mbox{ and } \alpha \neq 0.$$
Remark \[23\], Corollary \[corofree\] and the previous theorem imply the next result.
\[t2\] Let $\tilde{{\mathfrak n }}_{m,k}$ be the free metabelian $k$-step nilpotent Lie algebra in $m$ generators. Then $\tilde{{\mathfrak n }}_{m,k}$ admits an ad-invariant metric if and only if $(m,k)=(3,2)$ or $(m,k)=(2,3)$.
The free nilpotent Lie algebras above can be constructed as extensions of abelian Lie algebras. This is the way in which they appear in [@FS], where Favre and Santharoubarne obtained the classification of the nilpotent Lie algebras of dimension $\leq 7$ admitting an ad-invariant metric. According to their results, any of the Lie algebras equipped with an ad-invariant metric ${\mathfrak n }_{3,2}$ or ${\mathfrak n }_{2,3}$ as above is equivalent to one of the followings
$$\label{m3223}
({\mathfrak n }_{3,2}, B_{3,2}) : \left(
\begin{matrix}
0 & 0 & 0 & 0 & 0 & 1 \\
0 & 0 & 0 & 0 & -1 & 0 \\
0 & 0 & 0 & 1 & 0 & 0 \\
0 & 0 & 1 & 0 & 0 & 0 \\
0 & -1 & 0 & 0 & 0 & 0 \\
1 & 0 & 0 & 0 & 0 & 0
\end{matrix}
\right) \;\;
({\mathfrak n }_{2,3}, B_{2,3}) :
\left(
\begin{matrix}
0 & 0 & 0 & 0 & 1 \\
0 & 0 & 0 & - 1 & 0 \\
0 & 0 & 1 & 0 & 0 \\
0 & -1 & 0 & 0 & 0 \\
1 & 0 & 0 & 0 & 0
\end{matrix}
\right).$$
The automorphism groups of ${\mathfrak n }_{3,2}$ and ${\mathfrak n }_{2,3}$
============================================================================
Here we study the automorfisms of the Lie algebras in Theorem \[t1\]. This is indeed a topic of active research (see for instance [@DF; @DG] and references therein). Our goal is to write explicitly the algebraic structure, in terms of the actions and representations of the different subgroups or subalgebras. We also distinguish the subgroup of orthogonal automorphism (resp. the Lie algebra of skew-symmetric derivations) in presence of the ad-invariant metric fixed in (\[m3223\]).
Recall that a derivation of a Lie algebra $\mathfrak g$ is a linear map $t: \mathfrak g \to \mathfrak g$ satisfying $$t[x,y]=[tx, y] + [x, ty ] \qquad \mbox{ for all } x,y \in \mathfrak g.$$
Whenever $\mathfrak g$ is endowed with a metric $\la \,,\,\ra$ a skew-symmetric derivation of ${\mathfrak g }$ is a derivation $t$ such that $$\label{antisim}
\la t a, b\ra=-\la a, tb\ra \quad \text{ for all }\; a,b\in{\mathfrak g }.$$
We denote by $\Der({\mathfrak g })$ the Lie algebra of derivations of ${\mathfrak g }$, which is the Lie algebra of the group of automorphisms of $\ggo$, $\Aut({\mathfrak g })$. Let $\Dera({\mathfrak g })$ denote the subalgebra of $\Der({\mathfrak g })$ consisting of skew-symmetric derivations of $({\mathfrak g }, \la\,,\,\ra)$. Thus $\Dera({\mathfrak g })$ is the Lie algebra of the group of orthogonal automorphisms denoted by $\Auto({\mathfrak g })$: $$\Auto({\mathfrak g })=\{ \alpha \in \Aut({\mathfrak g }) : \la \alpha x, \alpha y \ra=\la x,y \ra \quad \forall x,y \in {\mathfrak g }\}.$$
For an arbitrary Lie algebra ${\mathfrak g }$, each $t \in \Aut({\mathfrak g })$ leaves invariant both the commutator ideal $C^1({\mathfrak g })$ and the center $\zz({\mathfrak g })$. Thus $t$ induces automorphisms of the quotient Lie algebras ${\mathfrak g }/\zz({\mathfrak g })$ and ${\mathfrak g }/C^1({\mathfrak g })$.
In particular if ${\mathfrak g }$ is solvable, ${\mathfrak g }/C^1({\mathfrak g })$ is a nontrivial abelian Lie algebra. Hence any $t\in\Aut({\mathfrak g })$ induces an element $s \in \Aut({\mathfrak g }/C^1({\mathfrak g }))$ which as an automorphism of an abelian Lie algebra, $s \in \GL(p,K)$ where $p = \dim( {\mathfrak g }/C^1({\mathfrak g }))$. Note that if ${\mathfrak g }$ is free nilpotent, $p$ coincides with the number of generators of ${\mathfrak g }$.
Below we proceed to the study in each case.
The Lie algebra ${\mathfrak n }_{2,3}$
--------------------------------------
Let $t$ denote an automorphism of ${\mathfrak n }_{2,3}$ and let $e_1,e_2,e_3,e_4,e_5$ be the basis given in (\[b23\]). Denote as $(t_{ij})$ the matrix of $t$ in that basis. Since $t$ leaves invariant the commutator and the center and $C^1({\mathfrak n }_{2,3})=span\,\{e_3,e_4,e_5\}$ and ${\mathfrak z }({\mathfrak n }_{2,3})=span\,\{e_4,e_5\}$, we have that $t_{ij}=0$ if $i=1,2$, $j=3,4,5$.
Notice that the Lie algebra ${\mathfrak n }_{2,3}/\zz({\mathfrak n }_{2,3})$ is isomorphic to the Heisenberg Lie algebra $\hh_1$, hence $t$ induces a automorphism $\bar{t}\in \Aut(\hh_1)$. Let $\bar{x}$ denote the image of an arbitrary element $x\in {\mathfrak n }_{2,3}$ by the canonical epimorphism ${\mathfrak n }_{2,3} \to {\mathfrak n }_{2,3}/\zz({\mathfrak n }_{2,3})$, thus $\bar{t}(\bar{x})=\overline{tx}$. From the computation $[\bar{t} \bar{e}_1, \bar{t}\bar{e}_2]=\overline{te_3}$ one gets $$\label{det}
t_{11} t_{22}- t_{12} t_{21}= t_{33}.$$
We introduce the notation for the submatrices $$A:=\left(\begin{array}{cc}t_{11}&t_{12}\\ t_{21}& t_{22}\end{array}\right)\qquad B:=\left(\begin{array}{cc}t_{44}&t_{45}\\ t_{54}& t_{55}\end{array}\right)
.$$ Note that the second matrix is non-singular since $t$ is non-singular.
By compairing (\[det\]) with the computations $[t e_i, t e_j]= t[e_i,e_j]$ we get the conditions $$t_{33}=\det A,\qquad B=\det(A) A, \qquad
\left(\begin{array}{c}t_{43}\\ t_{53}\end{array}\right)=A \left(\begin{array}{c}t_{32}\\ -t_{31}\end{array}\right).$$
So for any $t\in \Aut({\mathfrak n }_{2,3})$, $(t_{ij})_{i,j}$ has the form $$\label{t}
\left(\begin{array}{ccc}
A & \begin{array}{cc}0\\0\end{array} & \begin{array}{cc}0&0\\0&0\end{array} \\
\begin{array}{cc} t_{31}&t_{32}\end{array} & \det(A) & \begin{array}{cc}0&0\end{array} \\
\begin{array}{cc}t_{41}&t_{42}\\t_{51}&t_{52}\end{array} &\begin{array}{cc}t_{11}t_{32}-t_{31}t_{12}\\t_{21}t_{32}-t_{31}t_{22}\end{array} & \det(A) A\end{array}
\right),$$ where $A\in \GL(2,K)$.
Consider the following matrices in $\Aut({\mathfrak n }_{2,3})$: $$\mathcal G=\left\{ \tilde{A}=\left(\begin{array}{ccc}
A & \begin{array}{cc}0\\0\end{array} & \begin{array}{cc}0&0\\0&0\end{array} \\
\begin{array}{cc} 0&0\end{array} & \det(A) & \begin{array}{cc}0&0\end{array} \\
\begin{array}{cc}0&0\\0&0\end{array} &\begin{array}{cc}0\\0\end{array} &\det(A)A\end{array}
\right), \, A\in \GL(2,K)\right\}$$ $$\mathcal H=\left\{ h_{(x,y,z)}=\left(
\begin{array}{ccccc}
1&0&0&0&0 \\
0&1&0&0&0\\
y&x&1&0&0\\
z+\frac{1}{2} x y&\frac{1}{2} x^2&x&1&0\\
-\frac{1}{2} y^2&z-\frac{1}{2} xy&-y&0&1\end{array}\right),\, (x,y,z)\in K^3\right\}$$
$$\mathcal R= \left\{ r_{(u,v,w)}=\left(
\begin{array}{ccccc}
1&0&0&0&0 \\
0&1&0&0&0\\
0&0&1&0&0\\
v&w&0&1&0\\
u&-v&0&0&1\end{array}\right),\, (u,v,w)\in K^3\right\}.$$
Note that $\mathcal{G}$, $\mathcal{R}$ and $\mathcal{H}$ are subgroups of $\Aut({\mathfrak n }_{2,3})$.
Moreover every $t$ of the form (\[t\]) can be written as a product of matrices $$t=\tilde{A}\cdot r_{(u,v,w)} \cdot h_{(x,y,z)},\qquad \mbox{ with } \tilde{A} \in \mathcal G,\; r_{(u,v,w)}\in \mathcal R,\; h_{(x,y,z)}\in \mathcal H$$
where $x=t_{32}/\det(A)$, $y=t_{31}/\det(A)$, $z=(t_{22}t_{41}-t_{12}t_{51}+t_{52}t_{11}-t_{42}t_{21})/2\det(A)^2$ and
$u=(2t_{51}t_{11}-2t_{41}t_{21}+t_{31}^2)/2 \det A^2$, $v=(t_{22}t_{41}-t_{31}t_{32}-t_{12}t_{51}+t_{42}t_{21}-t_{52}t_{11})/2 \det A$, $w=(2t_{42}t_{22}-t_{32}^2-2t_{52}t_{12})/2\det A^2$.
The elements of $\mathcal{H}$ commute with those of $\mathcal{R}$; also $\mathcal H \cap \mathcal R=\{I\}$. Hence $\mathcal{R}\cdot \mathcal{H}\simeq \mathcal{R}\times \mathcal{H}.$
It holds $h_{(x,y, z)} \cdot h_{(x',y',z')}=h_{(x+x', y+ y', z + z' + \frac12 (xy' -x' y))}$, where $\cdot$ denotes the product of matrices. Thus the map from the Heisenberg Lie group $H_1$ to $\Aut({\mathfrak n }_{2,3})$ given by $(x,y, z) \mapsto h_{(x,y,z)}$ is an isomorphism of groups.
Analogously $(u,v,w)\mapsto r_{(u,v,w)}$ is an isomorphism of groups from $K^3$ to $ \mathcal R\subseteq \Aut({\mathfrak n }_{2,3})$.
The action by conjugation of $\mathcal{G}$ preserves both $\mathcal{H}$ and $\mathcal{R}$. Thus the map $$\tau_{\tilde{A}}(r,h)=(\tilde{A} r\tilde{A}^{-1},\tilde{A} h\tilde{A}^{-1}).$$ defines a group homomorphism from $\mathcal{G}$ to $\Aut(\mathcal R \times \mathcal{H})$.
The subgroup $\mathcal{R}\times \mathcal{H}$ is normal in $\Aut({\mathfrak n }_{2,3})$ and $\mathcal{G}\cap (\mathcal{R}\times \mathcal{H})=\{I\}$, hence $\Aut({\mathfrak n }_{2,3})\simeq \mathcal G\ltimes_\tau (\mathcal{R}\times \mathcal{H})$ ([@Kn]). It is clear that $\mathcal{G}\simeq \GL(2,K)$, $\mathcal{R}\simeq K^3$ and $\mathcal{H}\simeq H_1$ and these isomorphisms preserve the action of $\GL(2,K)$ in $H_1$ and $K^3$. So the next result follows.
\[pro5\] The group of automorphisms of ${\mathfrak n }_{2,3}$ is $$\Aut({\mathfrak n }_{2,3}) \simeq \GL(2,K)\ltimes (K^3 \times H_1),$$ where $H_1$ denotes the Heisenberg Lie group of dimension three.
The aditional orthogonal condition leads the following result.
The group of orthogonal automorphisms of $({\mathfrak n }_{2,3},B_{2,3})$ is $$\Auto({\mathfrak n }_{2,3}) \simeq \mathcal S \ltimes H_1,$$ where $\mathcal S$ is the subgroup of $\mathcal G$ consisting of the matrices $\tilde{A}$ with $A\in GL(2,K),\,\det(A)=\pm1$. The action in the semidirect product is the restriction of the action of $\mathcal G$ in the Heisenberg Lie group $H_1$ described before.
The Lie algebra of derivations of ${\mathfrak n }_{2,3}$ is isomorphic to $$\mathfrak{gl}(2,K) \ltimes (\mathfrak h_1 \times K^3).$$ In particular, fix the ad-invariant metric $\la \,,\,\ra$ in $({\mathfrak n }_{2,3}, B_{2,3})$ given by the matricial representation as in (\[m3223\]).
The set of skew-symmetric derivations is represented by the Lie algebra $$\Dera({\mathfrak n }_{2,3}) \simeq \mathfrak{sl}(2,K) \ltimes \mathfrak h_1$$ while the set of inner derivations is (isomorphic to) $\hh_1$.
In the next paragraphs we describe $\Der({\mathfrak n }_{2,3})$ and $\Dera({\mathfrak n }_{2,3})$ explicitly as subalgebras of $\mathfrak{gl}(5,K)$ putting emphasis on the actions and representations.
By canonical computations one verifies that an element in $\Der({\mathfrak n }_{2,3})$ has the following matricial representation in the basis $e_1,e_2,e_3,e_4,e_5$
$$\left( \begin{matrix}
z_1+z_4 &-z_2 & 0 & 0 & 0 \\
-z_3 & z_1 -z_4& 0 & 0 & 0 \\
z_6& z_5 &2z_1 & 0 & 0 \\
z_7+z_9 & z_8 & z_5 & 3z_1+z_4 & -z_2\\
z_{10}& z_7-z_9 & -z_6 & -z_3 & 3z_1-z_4
\end{matrix}
\right).$$
As usual, denote by $E_{ij}$ the 5$\times$5 matrix which has a 1 at the place $ij$ and $0$ in the other places.
Consider the vector subspace of dimension three spanned by the matrices $$X = E_{32}+E_{43} \qquad Y= E_{31}- E_{53} \qquad Z=E_{41}+ E_{52}$$ which obey the Lie bracket relation $[X,Y]=Z$, thus this is a faithful representation of the Heisenberg Lie algebra $\hh_1$.
Also denote by $$U= E_{42}\qquad V=E_{41}-E_{52}\qquad W=E_{51}$$ a basis of the vector subspace which spans an abelian Lie algebra of dimension three.
The algebra $\mathfrak h_1 \times K^3$ is an ideal of the Lie algebra of derivations namely the radical. Denote by $E,F,H, T$ the matrices given by $$\begin{array}{rclrcl}
E & = & -E_{12}-E_{45} & F & = & -E_{21}-E_{54} \\
H & = & E_{11}-E_{22} + E_{44}-E_{55} & T & = & E_{11}+E_{22}+2E_{33}+3E_{44}+2E_{55}.
\end{array}$$
Thus $span \{E,F,H\}\simeq \mathfrak{sl}(2,K)$ and therefore $span \{E,F,H,T\}\simeq \mathfrak{gl}(2,K)=\mathfrak{sl}(2,K)\times K$. The action of $\mathfrak{gl}(2,K)$ on $\mathfrak h_1\times K^3$ preserves each of the ideals $\mathfrak h_1$ and $K^3$ respectively.
The action of $\mathfrak{sl}(2,K)$ on $\mathfrak h_1$ is is given by derivations, that is $t \in \mathfrak{sl}(2, K) \simeq \Der(\hh_1)$, in the basis $X,Y,Z$, is represented by $$\left(
\begin{matrix}
h & e & 0\\
f & -h & 0 \\
0 & 0 & 0
\end{matrix}
\right).$$
The action of $\mathfrak{sl}(2,K)$ on $K^3$ is given by its irreducible representation in dimension three (see [@Hu]) $$\ad(eE+fF+hH)|_{U,V,W} = \left( \begin{matrix}
2h & 2e & 0\\
f & 0 & e\\
0 & 2f & -2h
\end{matrix}
\right).$$
On the other hand the action of $T$ on $\mathfrak h_1$ is diagonal $$T \cdot X= X \qquad T \cdot Y = Y \qquad T \cdot Z = 2 Z;$$ and the action of $T$ on $K^3$ is twice the identity: $T \cdot A= 2 A$ for all $A \in K^3$.
The Lie algebra ${\mathfrak n }_{3,2}$
--------------------------------------
Let $e_1, e_2, e_3, e_4, e_5, e_6$ denote the basis of ${\mathfrak n }_{3,2}$ given in (\[b32\]) above. A derivation of this Lie algebra has a matrix as follows $$\left( \begin{matrix}
z_1+z_{2} & z_4 & z_6&0 & 0 & 0 \\
z_{5} & z_1+z_{3} & z_{8}&0 & 0 & 0 \\
z_{7}& z_9 &z_1 -z_{2}-z_3 &0 & 0 & 0 \\
-z_{12} & -z_{16}+z_{17} & z_{18} & 2z_1+z_{2}+z_3& z_{8}&-z_6\\
z_{14}+z_{15}& z_{10} & z_{16}+z_{17} & z_9 & 2z_1-z_3&z_4\\
z_{13} &z_{14}-z_{15}&z_{11}-z_{12}&-z_{7}&z_{5}&2z_1-z_{2}
\end{matrix}
\right).$$ Let $\mathfrak G$ and $\mathfrak R$ denote respectively the sets of matrices in $\Der({\mathfrak n }_{3,2})$ $$\mathfrak G=\left\{ \left( \begin{matrix}
z_1+z_{2} & z_4 & z_6&0 & 0 & 0 \\
z_{5} & z_1+z_{3} & z_{8}&0 & 0 & 0 \\
z_{7}& z_9 &z_1 -z_{2}-z_3 &0 & 0 & 0 \\
0 & 0 & 0 & 2z_1+z_{2}+z_3& z_{8}&-z_6\\
0 & 0 & 0 & z_9 & 2z_1-z_3&z_4\\
0 & 0 & 0&-z_{7}&z_{5}&2z_1-z_{2}
\end{matrix}
\right) \right\}$$ $$\mathfrak R= \left\{
\left( \begin{matrix}
0 &0 &0 &0 & 0 & 0 \\
0 & 0 & 0 &0 & 0 & 0 \\
0 & 0 & 0 &0 & 0 & 0 \\
-z_{12} & -z_{16}+z_{17} & z_{18} & 0 & 0 &0 \\
z_{14}+z_{15}& z_{10} & z_{16}+z_{17} & 0 & 0 &0 \\
z_{13} &z_{14}-z_{15}&z_{11}-z_{12}& 0 & 0 &0 \end{matrix}
\right)\right\}.$$
With the usual conventions, denote by $E_{ij}$ the 6$\times$6 matrix which has a $1$ in the file $i$ and column $j$ and $0$ otherwise. Let $T$ and $f_i$, $i=1, \hdots , 8$, be the following matrices
$$T = E_{11}+ E_{22} +E_{33}+2E_{44}+2E_{55}+2E_{66}$$ $$\begin{array}{cc}
\begin{array}{rcl}
f_1 & = & E_{11}-E_{33}+E_{44}-E_{66}\\
f_2 & = & E_{22}-E_{33}+E_{44}-E_{55}\\
f_3 & = & E_{12} + E_{56}\\
f_4 & = & E_{21} + E_{65}
\end{array}&
\begin{array}{rcl}
f_5 & = & E_{13} -E_{46}\\
f_6 & = & E_{31} - E_{64}\\
f_7 & = & E_{23} + E_{45}\\
f_8 & = & E_{32} + E_{54}
\end{array}
\end{array}
.$$
With the Lie bracket of matrices, the vector space spanned by $f_1, \hdots, f_8$ constitute a Lie algebra isomorphic to $\mathfrak{sl}(3,K)$, such that $[T,f_i]=0$ for all $i=1, \hdots , 8$. Hence one has the following isomorphism of Lie algebras
$$\mathfrak G \simeq \mathfrak{sl}(3,K) \times K = \mathfrak{gl}(3,K).$$
For $i=1, \hdots 9$, let $A_i$ denote the matrices $$\begin{array}{rclrclrcl}
A_1 & = & E_{52} & A_2 & = & E_{63} & A_3 & = & -E_{41}-E_{63}\\
A_4 & = & E_{61} & A_5 & = & E_{51}+E_{62} & A_6 & = & E_{51}-E_{62}\\
A_7 & = & -E_{42}+ E_{53} & A_8 & = & E_{42}+E_{53} & A_9 & = & E_{43}
\end{array}$$ which generate the abelian Lie algebra $\mathfrak R$ of dimension nine. Actually this is a faithful representation of minimal dimension of the Lie algebra $K^9$ (see for instance [@Bu]). Hence $$\mathfrak R \simeq K^9$$ and it coincides with the radical of the Lie algebra of derivations of ${\mathfrak n }_{3,2}$.
The action of $\mathfrak{sl}(3,K)$ on $K^9$ is given by the adjoint representation $f_i \cdot A_j=[f_i, A_j]$ for all $i,j$, while the action of $T$ on $K^9$ is represented by the identity map.
Let ${\mathfrak n }_{3,2}$ denote the free 2-step nilpotent Lie algebra in three generators. The Lie algebra of derivations of ${\mathfrak n }_{3,2}$ is isomorphic $\mathfrak{gl}(3,K) \ltimes K^9$.
For the ad-invariant metric $\la \,,\,\ra$ on $({\mathfrak n }_{3,2}, B_{3,2})$ as in (\[m3223\]) one has the next result.
The set of skew-symmetric derivations of ${\mathfrak n }_{3,2}$ with the metric $\la \,,\,\ra$ is given by the Lie algebra $$\Dera({\mathfrak n }_{3,2})\simeq \mathfrak{sl}(3,K)\ltimes K^3,$$ while the set of inner derivations is isomorphic to $K^3$.
One can easily check that $t\in \Dera({\mathfrak n }_{3,2})$ is skew-symmetric if and only if it belongs to the vector space spanned by $$\{f_i, i=1, \hdots 8,A_2 +\frac{1}{2} A_3, A_5, A_8\}$$
The elements $f_i$ generate a Lie algebra isomorphic to $\mathfrak{sl}(3,K)$ and $A_2 +\frac{1}{2} A_3, A_5, A_8$ span an abelian ideal, isomorphic to $K^3$. Furthermore $t \in \mathfrak{sl}(3)$ acts on $K^3$ as a linear transformation of $K^3$ (that is $s(x)$ for $x\in K^3$).
[**Acknoledgements.**]{} The authors are very grateful to A. Kaplan for useful suggestions and comments.
They also specially thank to an anonymous referee whose suggestions helped to improve the results in the paper.
[GGGG]{}
Y. A. Bahturin, Identical rlations in Lie algebras, VNU Science Press (1987).
G. Bor, R. Montgomery, $G_2$ and the rolling distribution, available http://count.ucsc.edu/$\sim$rmont/papers/R9A.pdf
D. Burde, A refinement of Ado’s Theorem. Archiv Math. 70 (1998), 118–127.
E. Cartan, Les systèmes de Pfaff à cinque variables et les équations aux dérivées partielles du second ordre, Ann. Sci. École Normale, [**27**]{}, no. 3 (1910), 109–-192. (Reprinted in Oeuvres completes, Partie III, vol. [**2**]{} 137–288.)
V. Cortes, Handbook of Pseudo-Riemannian Geometry and Supersymmetry, in: IRMA Lectures in Mathematics and Theoretical Physics, [**16**]{} (2010).
V. Drensky, S. Findik, Inner and outer automorphisms of free metabelian nilpotent Lie algebras, arXiv:1003:0350.
V. Drensky, C. K. Gupta, Automorphisms of free nilpotent Lie algebras, Can. J. Math. [**XLII**]{} 2 (1990), 259–279.
G. Favre, L. Santharoubane, Symmetric, invariant, non-degenerate bilinear form on a Lie algebra, J. Algebra, [**105**]{} (1987), 451–464.
M. Grayson, R. Grossman, Models for free nilpotent Lie algebras, J. Algebra [**35**]{} (1990), 177–191. See draft in http://users.lac.uic.edu/$\tilde{}$grossman/trees.htm
M. Hall, A basis for free Lie rings and higher commutators in free groups, Proc. Amer. Math. Soc. [**1**]{} (1950), 575–-581.
J. Humphreys, Introduction to Lie Algebras and Representation Theory, Springer Verlag (1972).
I. Kath, M. Olbricht, Metric Lie algebras with maximal isotropic centre, Math. Z. [**246**]{} 1-2 (2004), 23–53.
I. Kath, Nilpotent metric Lie algebras of small dimension, J. Lie Theory [**17**]{} (1) (2007), 41–61.
A. Knapp, Basic algebra, Birkhäuser (Boston-Basel-Berlin) (2006).
B. Kostant, On differential geometry and homogeneous spaces II, Proc. N.A.S. [**42**]{} (1956), 354–-357.
A. Medina, P. Revoy, Algèbres de Lie et produit scalaire invariant (French) \[Lie algebras and invariant scalar products\], Ann. Sci. École Norm. Sup. (4) [**18**]{} (3) (1985), 553–-561.
A. Medina, Groupes de Lie munis de métriques bi-invariantes, Tohoku Math. J., [**37**]{} (1985), 405–-421.
F. Monroy-Pérez, F., A. Anzaldo-Meneses, The step-2 nilpotent $(n,n(n+1)/2)$ sub-Riemannian geometry, J. Dyn. Control Syst. [**12**]{} (2) (2006), 185–216.
R. Montgomery, A tour of Subriemannian geometries, their geodesics and applications, Math. surveys and monographs, [**91**]{} Amer. Math. Soc. (2002).
O. Myasnichenko, Nilpotent $(3,6)$ sub-Riemannian problem, J. Dyn. Control Syst. [**8**]{} (4) (2002), 573–597.
G. P. Ovando, Naturally reductive pseudo-Riemannian spaces, J. of Geom. and Physics, [**61**]{} (1) (2010), 157–171.
G. P. Ovando, Two-step nilpotent Lie algebras with ad-invariant metrics and a special kind of skew-symmetric maps, J. Algebra and its Appl., [**6**]{} (6) (2007), 897–917.
J. P. Serre, Lie algebras and Lie groups, Lecture Notes in Math. [**1500**]{}, Springer Verlag (1992).
[^1]: Partially supported by Secyt-UNC and SCyT-UNR
[^2]: Keywords: Free nilpotent Lie algebra, free metabelian nilpotent Lie algebra, ad-invariant metrics, automorphisms and derivations. [^3]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'It is determined exactly which classifiable C\*-algebras have approximately inner flip. The answer includes a number of C\*-algebras with torsion in their K-theory, and a number of C\*-algebras that are self-absorbing but not strongly self-absorbing.'
address: |
Institute of Mathematics\
University of Aberdeen\
Aberdeen, UK AB24 3UE
author:
- Aaron Tikuisis
title: 'K-theoretic characterization of C\*-algebras with approximately inner flip'
---
[^1]
Introduction
============
The concept of approximately inner flip for C\*-algebras was first studied by Effros and Rosenberg in [@EffrosRosenberg]. The concept has a close connection to strongly self-absorbing algebras defined by Toms and Winter [@TomsWinter:ssa], a prominent idea in the Elliott classification programme [@BBSTWW; @DadarlatPennig; @DadarlatWinter:Trivialization; @ENST; @ElliottToms; @GongLinNiu; @HirshbergRordamWinter; @MatuiSato:comp; @MatuiSato:dr; @Rordam:Z; @SWW:Znucdim; @TW:Zdr; @TomsWinter:ZASH; @Winter:drZstable; @Winter:pure; @Winter:localizing].
Effros and Rosenberg showed that the class of C\*-algebras with approximately inner flip is fairly restricted: it is contained in the class of simple, nuclear C\*-algebras with at most one trace. By considering the case of AF algebras, they showed that, in fact, approximately inner flip entails considerably more restrictions than just simplicity, nuclearity, and at most one trace. Their result [@EffrosRosenberg Theorem 3.9] is that an AF algebra with approximately inner flip is stably isomorphic to a UHF algebra. This is easily reformulated as a K-theoretic characterization of AF algebras with approximately inner flip: an AF algebra $A$ has approximately inner flip if and only if $K_0(A)$ is a subset of ${\mathbb Q}$.
This article generalizes this K-theoretic analysis of approximately inner flip, subject to the Universal Coefficient Theorem (UCT) (without which, serious K-theoretic computations are hopeless). It is shown that if $A$ has approximately inner flip and satisfies the UCT, then $K_*(A):=K_0(A)\oplus K_1(A)$ is isomorphic (as an ungraded group) to one of the following groups:
1. $0$;
2. ${\mathbb Z}$;
3. ${\mathbb Q}_n$;
4. ${\mathbb Q}_m/{\mathbb Z}$;
5. ${\mathbb Q}_n \oplus {\mathbb Q}_m/{\mathbb Z}$,
where in (iii)-(v), $n$ and $m$ are supernatural numbers of infinite type such that $m$ divides $n$ (see Section \[sec:Notation\] for the definition of ${\mathbb Q}_n$). The result is tight, in that every one of these groups (with any ${\mathbb Z}_2$-grading and any unperforated ordering) does arise as $K_*(A)$ for some C\*-algebra with approximately inner flip. In fact, if $A$ is a classifiable C\*-algebra (in the sense of [@Phillips:piClass] and [@GongLinNiu]), then $A$ has approximately inner flip if and only if $K_*(A)$ is one of these groups and $A$ has at most one trace. Interestingly, this provides a significant number of C\*-algebras, even self-absorbing ones, with torsion in their K-theory (namely, from cases (iv) and (v)).
Some notable consequences of our results and methods are as follows, where $A$ is a classifiable C\*-algebra:
1. If $A \otimes A$ has approximately inner flip then so does $A$;
2. If $A$ has approximately inner flip then it has asymptotically inner flip;
3. If $A$ has approximately inner flip then $A \otimes A \otimes \mathcal K$ is self-absorbing.
(The first two of these are packaged into the main theorem, Theorem \[thm:MainThm\], while the third is Corollary \[cor:AAsa\].) This prompts the question: which of these three facts can be shown without assuming that $A$ is classifiable?
Let us compare the situation to that of strongly self-absorbing C\*-algebras. Strongly self-absorbing C\*-algebras have approximately inner flip, but have significantly more structure, allowing results such as (ii) to be proven without using classification, as done by Dadarlat and Winter [@DadarlatWinter:KKssa] (in fact, they show that any automorphism of a strongly self-absorbing C\*-algebra is asymptotically inner). Toms and Winter, in the same article that introduced the concept of strongly self-absorbing C\*-algebras, established which possible K-theories can arise for strongly self-absorbing C\*-algebra in the UCT class [@TomsWinter:ssa Proposition 5.1]; the main result of this article is a natural extension of their work.
Naturally, our main tool for analyzing the flip in terms of K-theory is the Künneth formula for C\*-algebras due to Schochet [@Schochet:KunnethThm], a short exact sequence relating $K_*(A\otimes A)$ to $K_*(A)\otimes K_*(A)$ and ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(A))$. However, one needs to know how the flip map $A \otimes A \to A \otimes A$ interacts with this exact sequence. In Section \[sec:KunnethFlip\], we solve this problem; the map between the Tor components comes from the natural isomorphism ${\mathrm{Tor}_1^{{\mathbb Z}}}(G,H) \cong {\mathrm{Tor}_1^{{\mathbb Z}}}(H,G)$, which we first describe in Section \[sec:TorFlipMap\].
In Section \[sec:Family\], we introduce a family of C\*-algebras (representatives for the groups listed above with possible gradings); having introduced notation for these C\*-algebras, we state the main result, Theorem \[thm:MainThm\]. In Section \[sec:Suff\], we show the C\*-algebras in this family each have approximately inner flip. This result is useful in establishing, in Section \[sec:Necessary\], that if $A$ satisfies the UCT and has approximately inner flip, then $K_*(A)$ is one of the groups above. Finally, in Section \[sec:Semigroup\], we explore the classifiable C\*-algebras with approximately inner flip, by describing the semigroup of isomorphism classes of such C\*-algebras, under the operation of $\otimes$.
Acknowledgments {#acknowledgments .unnumbered}
---------------
I would like to thank Bradd Hart, Ilijas Farah, and Mikael Rørdam for discussions that contributed to this article. In particular, Mikael Rørdam suggested the problem, and Ilijas Farah made suggestions that led to a simpler proof of Proposition \[prop:PruferFlip\]. I would moreover like to thank Sean Tilson for helping to explain a spectral sequence proof of Lemma \[lem:KunnethFlip\] for topological K-theory.
Notation {#sec:Notation}
--------
Let $A,B$ be C\*-algebras. Write $K_*(A):=K_0(A) \oplus K_1(A)$. The suspension of $A$ is $SA:=C_0((0,1),A)$. Write $A \otimes B$ to denote the minimal tensor product of $A$ and $B$. The C\*-algebras $A$ and $B$ are said to be stably isomorphic if they satisfy $A \otimes \mathcal K \cong B \otimes \mathcal K$ (in case $A$ and $B$ are separable, this is the same as Morita equivalence). Denote by $\sigma_{A,B}$ the flip isomorphism $A \otimes B \to B \otimes A$, defined on elementary tensors by $\sigma_{A,B}(a \otimes b) = b \otimes a$. Following [@EffrosRosenberg], we say that $A$ has **approximately inner flip** if there is a net $(u_\lambda)$ of unitaries on $A$ such that $$\lim_\lambda \|u_\lambda xu_\lambda^* - \sigma_{A,A}(x)\| = 0, \quad x \in A \otimes A.$$ When $A$ is separable and has approximately inner flip, the net $(u_\lambda)$ can be taken to be a sequence. If $A$ is separable, we say that $A$ has **asymptotically inner flip** if there is a continuous function $t \mapsto u_t$ from $[0,\infty)$ to the unitaries in $A \otimes A$, such that $$\lim_{t\to\infty} \|u_t xu_t^* - \sigma_{A,A}(x)\| = 0, \quad x \in A \otimes A.$$
A **Kirchberg algebra** is a simple, separable, purely infinite, nuclear C\*-algebra. The class of Kirchberg algebras which satisfy the UCT has been classified by K-theory, by Kirchberg and Phillips ([@Kirchberg:piClass; @Phillips:piClass]; see [@Rordam:ClassBook Chapter 8]). We use $\mathcal O^\infty$ to denote the unital Kirchberg algebra in the UCT class that satisfies $$K_0(\mathcal O^\infty) \cong 0,\quad K_1(\mathcal O^\infty) \cong {\mathbb Z}.$$
Let $G,H$ be abelian groups. For a prime number $p$, the group $G$ is a **$p$-group** if every $x \in G$ satisfies, for some $k\in{\mathbb{N}}$, $p^kx = 0$. Denote by $\sigma_{G,H}$ the flip isomorphism $G \otimes H \to G \otimes H$, defined on elementary tensors by $\sigma_{G,H}(g \otimes h) = g \otimes h$. (This should not be confused with the flip $\sigma_{A,B}$ on the C\*-algebraic tensor product $A \otimes B$, because it is absurd to treat a C\*-algebra as merely an abelian group.)
A **supernatural number** is a formal product $$\label{eq:SNproduct}
\prod_p p^{k_p},$$ where the product is taken over all primes and $k_p \in \{0,1,2,\dots,\infty\}$ for each $p$. Every natural number is a supernatural number. Supernatural numbers may be multiplied (even infinitely many times) and the multiplication operation is also used to define what is meant by one supernatural number, $m$, dividing another, $n$ (in symbols, $m|n$). A supernatural number $n$ is of infinite type if it is equal to its square (i.e., expressing $n$ as in , if $k_p \in \{0,\infty\}$ for each $p$); note that by this definition, $1$ is of infinite type.
If $n$ is a supernatural number, we define $${\mathbb Q}_n := \Big\{\frac pq \in {\mathbb Q}\Big| p \in \mathbb Z, q \in \mathbb N, q|n\Big\}.$$ Note when $n=k^\infty$, $k \in {\mathbb{N}}$, then ${\mathbb Q}_n = {\mathbb Z}[1/k]$. If $n$ is a supernatural number of infinite type, then ${\mathbb Q}_n/{\mathbb Z}\cong \bigoplus_p {\mathbb Q}_{p^\infty}/{\mathbb Z}$ where the direct sum is taken over all primes $p$ that divide $n$.
If $n$ is a supernatural number and $G$ is an abelian group, we say that a group $H$ is $n$-divisible if it is $p$-divisible for every prime $p$ which divides $n$.
\[lem:DivSplit\] Let $G$ be an abelian group and let $H \subset G$ be a subgroup, such that $G/H \cong {\mathbb Q}_{n}/{\mathbb Z}$ or ${\mathbb Q}_n$ for some supernatural number $n$. If $H$ is $n$-divisible then $G \cong H \oplus G/H$.
Assume that $H$ is $n$-divisible, and let us show that the exact sequence $$\xymatrix{
0 \ar[r] & H \ar[r] & G \ar[r]^-{\pi} & G/H \ar[r] & 0
}$$ splits. Identify $G/H$ with ${\mathbb Q}_n/{\mathbb Z}$ or ${\mathbb Q}_n$ for notational simplicity.
Let $n=m_1m_2\cdots$ where $m_i \in {\mathbb{N}}$ for each $i$, and set $n_i:=m_1\cdots m_i$ for each $i \geq 0$. Then $${\mathbb Q}_n \cong \langle a_0,a_1,a_2,\dots \mid m_ia_i = a_{i-1}\rangle,$$ and $${\mathbb Q}_n/{\mathbb Z}\cong \langle a_0,a_1,a_2,\dots \mid m_ia_i = a_{i-1}, a_0=0\rangle,$$ in both cases by identifying $\frac1{n_i}$ with $a_i$.
Therefore, to define a splitting of $\pi$, we need to find $b_i \in G$ such that $\pi(b_i) = \frac1{n_i}$ and $m_ib_i = b_{i-1}$ for all $i \geq 1$ and additionally, in case $G/H\cong {\mathbb Q}_n/{\mathbb Z}$, $b_0=0$. We do this recursively. If $G/H \cong {\mathbb Q}_n$, set $b_0$ equal to any lift of $1$; if $G/H \cong {\mathbb Q}_n/{\mathbb Z}$, set $b_0=0$.
Having defined $b_{i-1}$, choose any $c_i \in G$ such that $\pi(c_i)=\frac1{n_i}$. Thus $\pi(m_ic_i - b_{i-1}) = 0$ so that $m_ic_i-b_{i-1} \in H$. Since $H$ is $m_i$-divisible, there exists $z \in H$ such that $m_ic_i-b_{i-1}=m_iz$. Thus, we may set $b_i:=c_i-z$, so that $\pi(b_i)=\pi(c_i)=\frac1{n_i}$ and $m_ib_i = b_{i-1}$ as required.
A family of C\*-algebras {#sec:Family}
========================
For supernatural numbers $n$, $m_0$, and $m_1$, let $\mathcal E_{n,m_0,m_1}$ be the simple, separable, unital, $\mathcal Z$-stable, quasidiagonal C\*-algebra with unique trace that satisfies $$\begin{aligned}
K_0(\mathcal E_{n,m_0,m_1})&={\mathbb Q}_n \oplus {\mathbb Q}_{m_0}/{\mathbb Z}, \\
[1]_0&= 1 \oplus 0, \quad \text{and} \\
K_1(\mathcal E_{n,m_0,m_1})&={\mathbb Q}_{m_1}/{\mathbb Z}.\end{aligned}$$
For supernatural numbers $m_0,m_1$, let $\mathcal F_{m_0,m_1}$ be the unital Kirchberg algebra in the UCT class that satisfies $$\begin{aligned}
K_0(\mathcal F_{m_0,m_1})&={\mathbb Q}_{m_0}/{\mathbb Z}, \\
[1]_0&= 0, \quad \text{and} \\
K_1(\mathcal F_{m_0,m_1})&={\mathbb Q}_{m_1}/{\mathbb Z}.\end{aligned}$$
\[rmk:EFdef\] It is shown by Elliott in [@Elliott:ashrange Theorem 5.2.3.2] that the algebra $\mathcal E_{n,m_0,m_1}$ algebra exists, and Matui and Sato have shown that it is unique [@MatuiSato:dr Corollary 6.2] (which makes crucial use of the classification results of Winter [@Winter:RR0drClass; @Winter:localizing] and Lin-Niu [@LinNiu:Lifting]).
By Rørdam [@Rordam:pirange], the algebra $\mathcal F_{m_0,m_1}$ exists, and by Kirchberg and Phillips classification [@Phillips:piClass], it is unique.
Some of these algebras are important and/or already well-known:
1. $\mathcal E_{1,1,1} \cong \mathcal Z$ (the Jiang-Su algebra [@JiangSu]).
2. $\mathcal E_{n,1,1} \cong M_n$ (a UHF algebra) for any infinite supernatural number $n$.
3. $\mathcal F_{1,1} \cong \mathcal O_2$ (a Cuntz algebra [@Cuntz:On]).
4. Letting $m$ be the product of all primes infinitely many times, the C\*-algebra $T:= \mathcal F_{1,m}$ has the property that, if $A$ is a Kirchberg algebra in the UCT class, then $A$ is stably isomorphic to $A \otimes T$ if and only if $K_*(A)$ is a torsion group. This follows from the Künneth formula (see below) and Kirchberg-Phillips classification [@Kirchberg:piClass; @Phillips:piClass].
Our main result is the following.
\[thm:MainThm\] Let $A$ be a separable, unital, C\*-algebra with strict comparison, in the UCT class, which is either infinite or quasidiagonal. The following are equivalent.
1. $A$ has approximately inner flip;
2. $A \otimes A$ has approximately inner flip;
3. $A$ has asymptotically inner flip;
4. $A$ is simple, nuclear, has at most one trace and $K_*(A)$ (as an ungraded, unordered group) is isomorphic to one of $0$, ${\mathbb Z}$, ${\mathbb Q}_n$, ${\mathbb Q}_m/{\mathbb Z}$, or ${\mathbb Q}_n \oplus {\mathbb Q}_m/{\mathbb Z}$, where $n,m$ are supernatural numbers of infinite type and $m$ divides $n$;
5. $A$ is stably isomorphic to one of:
1. $\mathbb C$;
2. $\mathcal E_{n,m_0,m_1}$;
3. $\mathcal E_{n,m_0,m_1} \otimes \mathcal O_\infty$;
4. $\mathcal E_{n,m_0,m_1} \otimes \mathcal O^\infty$; or
5. $\mathcal F_{m_0,m_1}$,
where in $\mathrm{(b)}$-$\mathrm{(e)}$, $n$, $m_0$, and $m_1$ are supernatural numbers of infinite type such that $m_0,m_1$ are coprime and $m_0m_1|n$.
\[rmk:MainThmClass\] We call the C\*-algebras satisfying the hypotheses and equivalent conditions of Theorem \[thm:MainThm\] the separable unital **classifiable C\*-algebras with approximately inner flip**. Understanding that “classifiable” means “classifiable by K-theory and traces,” then the classification results to date (including Kirchberg-Phillips’ classification of purely infinite C\*-algebras [@Kirchberg:piClass; @Phillips:piClass] and the Gong-Lin-Niu classification of C\*-algebras of generalized tracial rank one [@GongLinNiu]) permit an extremely reasonable definition of a simple separable unital “classifiable C\*-algebra” as meaning a simple separable unital C\*-algebra which satisfies the UCT and is either purely infinite or has generalized tracial rank one (in the sense of [@GongLinNiu Definition 9.2]). Certainly, the aforementioned classification results show that the class of such C\*-algebras is classifiable and exhausts the range of the Elliott invariant; hence, this is a maximal classifiable class. The C\*-algebras satisfying the hypotheses and equivalent conditions of Theorem \[thm:MainThm\] are precisely the C\*-algebras in this classifiable class which have approximately inner flip (this is entailed by the theorem).
It is a long-standing open question whether there are simple nuclear C\*-algebras that (i) don’t satisfy the UCT or (ii) are stably finite but not quasidiagonal. Even for the much smaller and deeply studied class of strongly self-absorbing C\*-algebras, this question is open.
However, unlike the class of strongly self-absorbing C\*-algebras, it is unknown whether there exists a C\*-algebra with approximately inner flip which does not have strict comparison (equivalently, whether it is $\mathcal Z$-stable, by [@MatuiSato:comp]).
\(i) Does there exist a C\*-algebra with approximately inner flip which is not $\mathcal Z$-stable?
\(ii) Does there exist a C\*-algebra $A$ such that $A \otimes A$ is strongly self-absorbing, but $A$ is not $\mathcal Z$-stable?
A positive answer to (ii) would imply a positive answer to (i), since if $A \otimes A$ is strongly self-absorbing then all of its automorphisms are approximately inner.
The flip map on Tor(G,G) {#sec:TorFlipMap}
========================
Let $G_1,G_2$ be abelian groups. The flip isomorphism $\sigma_{G_1,G_2}: G_1 \otimes G_2 \to G_2 \otimes G_1$ induces a natural isomorphism $$\eta_{G_1,G_2}:{\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_2) \to {\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_1).$$
Here is a description of $\eta_{G_1,G_2}$. Fix a free abelian group $F_i$ that surjects onto $G_i$; the kernel $H_i$ of this surjection is also a free group, and we get an exact sequence $$\xymatrix{
0 \ar[r] & H_i \ar[r] & F_i \ar[r] & G_i \ar[r] & 0 }$$ (called a **free resolution** of $G_i$.)
This induces a double-complex with exact rows and columns:
$$\label{eq:etaDoubleComplex}
\xymatrix{
&&&& 0 \ar[d] & \\
&& 0 \ar[d] & 0 \ar[d] & {\mathrm{Tor}_1^{{\mathbb Z}}}(G_1, G_2) \ar[d] & \\
&0 \ar[r] & H_1 \otimes H_2 \ar[r]^-{\alpha_{11}} \ar[d]^-{\beta_{11}} & H_1 \otimes F_2 \ar[r]^-{\alpha_{12}} \ar[d]^-{\beta_{12}} & H_1 \otimes G_2 \ar[r] \ar[d]^-{\beta_{13}} & 0 \\
&0 \ar[r] & F_1 \otimes H_2 \ar[r]^-{\alpha_{21}} \ar[d]^-{\beta_{21}} & F_1 \otimes F_2 \ar[r]^-{\alpha_{22}} \ar[d]^-{\beta_{22}} & F_1 \otimes G_2 \ar[r] \ar[d]^-{\beta_{23}} & 0 \\
0 \ar[r] & {\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_1) \ar[r] & G_1 \otimes H_2 \ar[r]^-{\alpha_{31}} \ar[d] & G_1 \otimes F_2 \ar[r]^-{\alpha_{32}} \ar[d] & G_1 \otimes G_2 \ar[r]\ar[d] & 0 \\
&&0&0&0.&
}$$
In particular, we identify ${\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_2)$ and ${\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_1)$ with $\ker(\beta_{13})$ and $\ker(\alpha_{31})$ respectively. Let $x \in {\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_2) = \ker(\beta_{13})$. Let $x_{12} \in H_1 \otimes F_2$ be such that $\alpha_{12}(x_{12}) = x$. Then $$\alpha_{22}\circ \beta_{12}(x_{12}) = \beta_{13} \circ \alpha_{21}(x_{12}) = \beta_{13}(x) = 0,$$ i.e., $\beta_{12}(x_{12}) \in \ker(\alpha_{22})$. By exactness of the second row, this means that $\beta_{12}(x_{12}) = \alpha_{21}(x_{21})$ for some unique $x_{21} \in F_1 \otimes H_2$. Set $y:=\beta_{21}(x_{21})$. Then $$\alpha_{31}(y) = \alpha_{31} \circ \beta_{21}(x_{21}) = \beta_{22} \circ \alpha_{21}(x_{21}) = 0,$$ i.e., $y \in \ker(\alpha_{31}) = {\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_1)$. The element $y$ does not depend on the choice of $x_{12}$, and we have $$\eta_{G_1,G_2}(x) = y.$$
The flip map and the Künneth formula {#sec:KunnethFlip}
====================================
If $A$ is a C\*-algebra in the UCT class then Schochet’s Künneth Theorem [@Schochet:KunnethThm] provides an exact sequence for computing $K_*(A\otimes B)$. We will use the statement and exposition in [@Blackadar:Kbook Chapter 23] (specifically, [@Blackadar:Kbook Theorem 23.1.3]); the exact sequence is $$\label{eq:KunnethFormula}
\xymatrix{
0 \ar[r] & K_*(A) \otimes K_*(B) \ar[r]^-{\alpha_{A,B}} & K_*(A\otimes B) \ar[r]^-{\beta_{A,B}} & {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A), K_*(B)) \ar[r] & 0,}$$ where the maps involved are all natural, with $\alpha_{A,B}$ preserving the ${\mathbb Z}_2$-grading and $\beta_{A,B}$ reversing the ${\mathbb Z}_2$-grading.
In this section we prove the following.
\[lem:KunnethFlip\] The following commutes. $$\label{eq:KunnethFlipDiag}
\xymatrix{
0 \ar[r] & K_*(A) \otimes K_*(B) \ar[r]^-{\alpha_{A,B}}\ar[d]_-{\sigma_{K_*(A),K_*(B)}} & K_*(A\otimes B) \ar[r]^-{\beta_{A,B}}\ar[d]_-{K_*(\sigma_{A,B})} & {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A), K_*(B)) \ar[r]\ar[d]_-{\eta_{K_*(A),K_*(B)}} & 0 \\
0 \ar[r] & K_*(B) \otimes K_*(A) \ar[r]^-{\alpha_{B,A}} & K_*(B\otimes A) \ar[r]^-{\beta_{B,A}} & {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(B), K_*(A)) \ar[r] & 0.
}$$
For topological K-theory (equivalently, C\*-algebra K-theory restricted to the class of commutative C\*-algebras), a Künneth spectral sequence argument could be used to prove the above lemma, although the author was unable to find a precise reference for such an argument. To the author’s knowledge, the literature does not contain a proof of the above lemma for topological K-theory which extends easily to the case of C\*-algebra K-theory.
That the left square in commutes is well-known and not difficult. Although commutativitiy of the right square seems quite natural, it is not trivial to show and the proof requires some setup. Note that there are (at least) two natural isomorphisms ${\mathrm{Tor}_1^{{\mathbb Z}}}(G,H) \to {\mathrm{Tor}_1^{{\mathbb Z}}}(H,G)$, namely $\eta_{G,H}$ and $-\eta_{G,H}$; although one expects that one of these should fit into the commuting diagram , it is not a priori obvious which one.
The following setup comes from [@Blackadar:Kbook Section 23.5]. Since $K_*(A)$ is naturally isomorphic to $K_*(S^2A \otimes \mathcal K)$, we may assume (when proving Lemma \[lem:KunnethFlip\]) that $A$ is of the form $S^2A' \otimes \mathcal K$ for some C\*-algebra $A'$, and likewise for $B$. Under this assumption, by [@Blackadar:Kbook Proposition 23.5.1], there exists a separable commutative C\*-algebra $F_A$, whose spectrum consists of disjoint union of lines and planes, and a homomorphism $\phi_A:F_A \to B$ giving rise to a surjective map $K_*(F_A) \to K_*(A)$. With $C_A$ the mapping cone of this homomorphism, i.e., $$C_A := \{(f,g) \in F_A \oplus C_0((0,1],A) \mid \phi(f)=g(1)\},$$ we obtain an exact sequence $$\xymatrix{
0 \ar[r] & SA \ar[r]^-{\mu_A} & C_A \ar[r]^-{\nu_A} &F_A \ar[r] & 0,
}$$ whose 6-term exact sequence in K-theory becomes two short exact sequences, $$\label{eq:ResolutionK}
\xymatrix{
0 \ar[r] & K_*(C_A) \ar[r] & K_*(F_A) \ar[r]^-{\partial} &K_*(SA) \ar[r] & 0
}$$
Since $K_*(A) \cong K_*(SA)$ (by an isomorphism that reverses the grading), we may identify ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(B))$ with ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(SA),K_*(SB))$. In turn, we identify ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(SA),K_*(SB))$ with the kernel of $$(\nu_A)_* \otimes 1_{K_*(SB)}:K_*(C_A) \otimes K_*(SB) \to K_*(F_A) \otimes K_*(SB);$$ since $K_*(F_A)$ and $K_*(C_A)$ are free abelian, we have by the Künneth formula , a commuting diagram as follows: $$\xymatrix{
K_*(C_A) \otimes K_*(SB) \ar[r]^-{\cong} \ar[d]_-{(\nu_A)_* \otimes 1_{K_*(SB)}} &
K_*(C_A\otimes SB) \ar[d]^-{(\nu_A \otimes {\mathrm{id}}_{SB})_*} \\
K_*(F_A) \otimes K_*(SB) \ar[r]^-{\cong} & K_*(F_A \otimes SB).
}$$ Thus, we actually identify ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(B))$ with the kernel of $(\nu_{A} \otimes {\mathrm{id}}_{SB})_*$. Under this identification, $\beta_{A,B}:K_*(A\otimes B) \cong K_*(SA \otimes SB) \to {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(B))$ is precisely the map $$({\mathrm{id}}_{SA} \otimes \mu_B)_*:K_*(SA \otimes SB) \to K_*(SA \otimes C_B),$$ (by a 6-term exact sequence, the image of this map is indeed contained in the kernel of $({\mathrm{id}}_{SA} \otimes \nu_B)_*$) (see the proof of [@Blackadar:Kbook Proposition 23.6.1]).
By the same construction with $B$ in place of $A$, obtain $F_B$, $C_B$, $ \mu_B$, and $ \nu_B$.
\[lem:iotaKinj\] Let $C_A,C_B$ be as described above. Define $\iota_1:SA \otimes C_B \to SA \otimes C_B + C_A \otimes SB$ to be the inclusion. Then $(\iota_1)_*:K_*(SA \otimes C_B) \to K_*(SA \otimes C_B + C_A \otimes SB)$ is injective.
Consider the following commuting diagram with short exact rows: $$\label{eq:iotaKinjExactSeqs}
\xymatrix{
0 \ar[r] & SA \otimes C_B \ar[r]^-{\iota_1} \ar[d]_-{{\mathrm{id}}} & SA \otimes C_B + C_A \otimes SB \ar[r] \ar[d] & F_A \otimes SB \ar[r] \ar[d]^-{{\mathrm{id}}_{F_A} \otimes \mu_B} & 0\\
0 \ar[r] & SA \otimes C_B \ar[r]^-{\mu_A \otimes {\mathrm{id}}_{C_B}} & C_A \otimes C_B \ar[r]^-{\nu_A \otimes {\mathrm{id}}_{C_B}} & F_A \otimes C_B \ar[r] & 0.
}$$ The top row produces the 6-term exact sequence $$\xymatrix{
K_0(SA \otimes C_B) \ar[r]^-{(\iota_1)_*} & K_0(SA \otimes C_B + C_A \otimes SB) \ar[r] & K_0(F_A \otimes SB) \ar[d]^-{\partial_0} \\
K_1(F_A \otimes SB) \ar[u]_-{\partial_1} & K_1(SA \otimes C_B + C_A \otimes SB) \ar[l] & K_1(SA \otimes C_B). \ar[l]^-{(\iota_1)_*},
}$$ so that injectivity of $(\iota_1)_*$ is equivalent to $\partial_*:K_*(SA \otimes F_B) \to K_{*}(C_A \otimes SB)$ being the zero map.
The second row of produces a 6-term exact sequence which, by and the Künneth formula (since $K_*(C_B)$ is a free abelian group), becomes two short exact sequences, $$\xymatrix{
0 \ar[r] & K_i(C_A \otimes C_B) \ar[r]^-{(\nu_A \otimes {\mathrm{id}}_{C_B})_*} & K_i(F_A \otimes C_B) \ar[r]^-{\partial \otimes 1} &K_{1-i}(SA \otimes C_B) \ar[r] & 0,
}$$ $i=0,1$.
By and naturality of the 6-term exact sequence, the following commutes $$\label{eq:iotaKinjCommKDiag}
\xymatrix{
K_*(F_A \otimes SB) \ar[r]^-{\partial} \ar[d]_-{({\mathrm{id}}_{F_A} \otimes \mu_B)_*} & K_*(SA \otimes C_B) \ar[d]^-{1} \\
K_*(F_A \otimes C_B) \ar[r]^-{\partial \otimes 1} & K_*(SA \otimes C_B).
}$$
Using (with $B$ in place of $A$), and the Künneth formula (since $K_*(F_A)$ is a free abelian group), the map $$({\mathrm{id}}_{F_A} \otimes \mu_B)_*:K_*(F_A \otimes SB) \to K_*(F_A \otimes C_B)$$ is zero. Thus, by , the map $\partial:K_*(F_A \otimes SB) \to K_*(SA \otimes C_B)$ is zero, as required.
The map $\alpha_{A,B}$ is explicitly described in [@Blackadar:Kbook Section 23.1], and it is apparent from this description that $\alpha_{B,A} \circ \sigma_{K_*(A),K_*(B)} = K_*(\sigma_{A,B}) \circ \alpha_{A,B}$, i.e., the first square in commutes. Let us move on to the second square.
With $G_1:=K_*(A)$ and $G_2:=K_*(B)$, we use the description of $\eta_{K_*(A),K_*(B)}$ from Section \[sec:TorFlipMap\], making use of free resolutions of $K_*(A)$, $K_*(B)$ provided by (we assume that these exist by possibly replacing $A$, $B$ by $S^2A \otimes \mathcal K$, $S^2B \otimes \mathcal K$ respectively). Note that, since $K_*(F_A)$, $K_*(F_B)$, $K_*(C_A)$, and $K_*(C_B)$ are all free abelian groups, the Künneth formula turns the double complex into $$\scriptsize{
\xymatrix{
&&& {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A), K_*(B)) \ar[d] \\
& K_*(C_A \otimes C_B) \ar[r] \ar[d] & K_*(C_A \otimes F_B) \ar[r]^-{1 \otimes \partial} \ar[d]^-{(\nu_A \otimes {\mathrm{id}}_{F_B})_*} & K_*(C_A \otimes SB) \ar[d] \\
& K_*(F_A \otimes C_B) \ar[r]^-{({\mathrm{id}}_{F_A} \otimes \nu_B)_*} \ar[d]^-{\partial \otimes 1} & K_*(F_A \otimes F_B) \ar[r] \ar[d] & K_*(F_A \otimes SB) \ar[d]\\
{\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(B),K_*(A)) \ar[r] & K_*(SA \otimes C_B) \ar[r] & K_*(SA \otimes F_B) \ar[r] & K_*(SA) \otimes K_*(SB)\\
}}$$ (for space considerations, the zero terms are omitted).
Now, let $x \in K_*(SA \otimes SB) \cong K_*(A,B)$. We have $\beta_{A,B}(x)=({\mathrm{id}}_{SA} \otimes \nu_B)_*(x) \in {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(B))$. Using the description of $\eta$ from Section \[sec:TorFlipMap\], there exist $x_{12} \in K_*(C_A\otimes F_B)$ and $x_{21} \in K_*(F_A\otimes C_B)$ such that $$\begin{aligned}
\label{eq:KunnethFlipSetup1}
(1 \otimes \partial)(x_{12}) &= \beta_{A,B}(x) = ({\mathrm{id}}_{SA} \otimes \nu_B)_*(x), \\
\label{eq:KunnethFlipSetup2}
(\nu_A \otimes {\mathrm{id}}_{F_B})_*(x_{12}) &= ({\mathrm{id}}_{F_A} \otimes \nu_B)_*(x_{21}), \quad \text{and} \\
\label{eq:KunnethFlipSetup3}
(\partial \otimes 1)(x_{21}) &= \eta_{K_*(A),K_*(B)}(\beta_{A,B}(x)).\end{aligned}$$
Consider the following commuting diagram with short exact rows: $$\xymatrix{
0 \ar[r] & SA \otimes C_B \ar[r] \ar[d]_-{\iota_1} & C_A \otimes C_B \ar[r] \ar[d] & F_A \otimes C_B \ar[r] \ar[d]^-{{\mathrm{id}}_{F_A} \otimes \nu_B} & 0\\
0 \ar[r] & SA \otimes C_B+C_A \otimes SB \ar[r] & C_A \otimes C_B \ar[r] & F_A \otimes F_B \ar[r] & 0.
}$$ By naturality of the 6-term exact sequences, we obtain the following commuting diagram: $$\label{eq:KunnethFlipCommuting1}
\xymatrix{
K_*(F_A \otimes C_B) \ar[d]_-{({\mathrm{id}}_{F_A} \otimes \nu_B)_*} \ar[r]^-{\partial \otimes 1} & K_*(SA \otimes C_B) \ar[d]^-{(\iota_1)_*} \\
K_*(F_A \otimes F_B) \ar[r]^-{\partial} & K_*(SA \otimes C_B+C_A\otimes SB).
}$$ Likewise, the following also commutes $$\label{eq:KunnethFlipCommuting2}
\xymatrix{
K_*(C_A \otimes F_B) \ar[d]_-{(\nu_A \otimes {\mathrm{id}}_{F_B})_*} \ar[r]^-{1\otimes\partial} & K_*(C_A \otimes SB) \ar[d]^-{(\iota_2)_*} \\
K_*(F_A \otimes F_B) \ar[r]^-{\partial} & K_*(SA \otimes C_B+C_A\otimes SB),
}$$ where $\iota_2$ denotes the inclusion $C_A \otimes SB \to SA \otimes C_B + C_A \otimes SB$.
Note that $$\label{eq:KunnethFlipInclusions}
\iota_1 \circ (\nu_A \otimes {\mathrm{id}}_{SB})=\iota_2\circ ({\mathrm{id}}_{SA} \otimes \nu_B):SA \otimes SB \to SA \otimes C_B + C_A \otimes SB,$$ both maps being equal to the inclusion.
Putting these pieces together, we obtain $$\begin{aligned}
(\iota_1)_* \circ \eta_{K_*(A),K_*(B)} \circ \beta_{A,B}(x) &\stackrel{\eqref{eq:KunnethFlipSetup3}}=& (\iota_1)_* \circ (\partial \otimes 1)(x_{21}) \\
&\stackrel{\eqref{eq:KunnethFlipCommuting1}}=& \partial \circ ({\mathrm{id}}_{F_A} \otimes \nu_B)_*(x_{21}) \\
&\stackrel{\eqref{eq:KunnethFlipSetup2}}=& \partial \circ (\nu_A \otimes {\mathrm{id}}_{F_B})_*(x_{12}) \\
&\stackrel{\eqref{eq:KunnethFlipCommuting2}}=& (\iota_2)_* \circ (1 \otimes \partial)(x_{12}) \\
&\stackrel{\eqref{eq:KunnethFlipSetup1}}=& (\iota_2)_* \circ ({\mathrm{id}}_{SA} \otimes \nu_B)_*(x) \\
&\stackrel{\eqref{eq:KunnethFlipInclusions}}=& (\iota_1)_* \circ (\nu_A \otimes {\mathrm{id}}_{SB})_*(x).\end{aligned}$$ By Lemma \[lem:iotaKinj\], it follows that $$\eta_{K_*(A),K_*(B)} \circ \beta_{A,B}(x) = (\nu_A \circ {\mathrm{id}}_{SB})_*(x) = \beta_{B,A}\circ (\sigma_{A,B})_*(x),$$ as required.
Sufficient conditions for approximately inner flip {#sec:Suff}
==================================================
\[prop:PruferFlip\] Let $n$ be a supernatural number. Then $\eta_{{\mathbb Q}_n/{\mathbb Z},{\mathbb Q}_n/{\mathbb Z}}$ is the identity map.
For each natural number $m|n$, ${\mathbb Q}_m/{\mathbb Z}\subset {\mathbb Q}_n/{\mathbb Z}$, and this inclusion (used twice) produces a commutative diagram $$\xymatrix{
{\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_m/{\mathbb Z}, {\mathbb Q}_m/{\mathbb Z}) \ar[r]^-{\eta_{{\mathbb Q}_m/{\mathbb Z},{\mathbb Q}_m/{\mathbb Z}}} \ar[d] & {\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_m/{\mathbb Z}, {\mathbb Q}_m/{\mathbb Z}) \ar[d] \\
{\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_n/{\mathbb Z}, {\mathbb Q}_n/{\mathbb Z}) \ar[r]^-{\eta_{{\mathbb Q}_n/{\mathbb Z},{\mathbb Q}_n/{\mathbb Z}}} & {\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_n/{\mathbb Z}, {\mathbb Q}_n/{\mathbb Z}). }$$ We have ${\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_m/{\mathbb Z}, {\mathbb Q}_m/{\mathbb Z}) \cong {\mathbb Q}_m/{\mathbb Z}$ and ${\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_n/{\mathbb Z},{\mathbb Q}_n/{\mathbb Z}) \cong {\mathbb Q}_n/{\mathbb Z}$ by [@Fuchs:book 62.J], and these isomorphisms induce the natural inclusion ${\mathbb Q}_m/{\mathbb Z}\subset {\mathbb Q}_n/{\mathbb Z}$. Since ${\mathbb Q}_n/{\mathbb Z}$ is the union of such subgroups, it suffices to show that $\eta_{{\mathbb Q}_m/{\mathbb Z}, {\mathbb Q}_m/{\mathbb Z}}$ is the identity map.
For this, note that ${\mathbb Q}_m \cong {\mathbb Z}$, so we have a free resolution $$0 \to {\mathbb Z}\to {\mathbb Q}_m \to {\mathbb Q}_m/{\mathbb Z}\to 0.$$ Set $H:={\mathbb Z}, F:={\mathbb Q}_m, G:={\mathbb Q}_m/{\mathbb Z}$ and refer to the description of $\eta_{{\mathbb Q}_m/{\mathbb Z},{\mathbb Q}_m/{\mathbb Z}}$ in Section \[sec:TorFlipMap\]. Let $x \in {\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_m/{\mathbb Z},{\mathbb Q}_m/{\mathbb Z})$, which we identify with ${\mathbb Z}\otimes ({\mathbb Q}_m/{\mathbb Z})$ (because the map ${\mathbb Z}\otimes ({\mathbb Q}_m/{\mathbb Z}) \to {\mathbb Q}_m \otimes ({\mathbb Q}_m/{\mathbb Z})$ is the zero map); thereby write $x=1 \otimes (k/m + {\mathbb Z})$ for some $k \in {\mathbb Z}$. This lifts to $x_{12} = 1 \otimes (k/m) \in {\mathbb Z}\otimes {\mathbb Q}_m$, which is equal to $(k/m) \otimes 1$ in ${\mathbb Q}_m \otimes {\mathbb Q}_m$. Thus, $x_{21} = (k/m) \otimes 1$, so that $$\eta_{{\mathbb Q}_m/{\mathbb Z},{\mathbb Q}_m/{\mathbb Z}}(x) = (k/m + {\mathbb Z}) \otimes 1.$$ This establishes that $\eta_{{\mathbb Q}_m/{\mathbb Z},{\mathbb Q}_m/{\mathbb Z}}$ is the identity map.
\[thm:KKsuff\] Let $A$ be a separable $C^*$-algebra. Suppose that $K_*(A)$ is one of the following groups (ignoring the grading):
1. $0$;
2. ${\mathbb Z}$;
3. ${\mathbb Q}_n$, where $n$ is a supernatural number of infinite type;
4. ${\mathbb Q}_m/{\mathbb Z}$ where $m$ is a supernatural number of infinite type; or
5. ${\mathbb Q}_n \oplus {\mathbb Q}_m/{\mathbb Z}$, where $n,m$ are supernatural numbers of infinite type, and $m$ divides $n$.
Then the flip map $\sigma_{A,A}:A \otimes A \to A \otimes A$ has the same KK-class as the identity map.
Using the Künneth formula , one computes $K_*(A \otimes A)$:
1. $K_*(A \otimes A)=0$;
2. $K_0(A\otimes A)={\mathbb Z}$ and $K_1(A\otimes A)=0$;
3. $K_0(A \otimes A)={\mathbb Q}_n$ and $K_1(A\otimes A)=0$;
4. $K_0(A\otimes A)=0$ and $K_1(A\otimes A)={\mathbb Q}_m/{\mathbb Z}$;
5. $K_0(A\otimes A)={\mathbb Q}_n$ and $K_1(A\otimes A)={\mathbb Q}_m/{\mathbb Z}$.
By Lemma \[lem:KunnethFlip\] and (for cases (iv) and (v)) Proposition \[prop:PruferFlip\], the flip map $\sigma_{A,A}$ acts as the identity on $K_*(A\otimes A)$.
In cases (i)-(iv), ${\mathrm{Ext}_1^{{\mathbb Z}}}(K_i(A \otimes A), K_{1-i}(A \otimes A))=0$ for trivial reasons, for $i=0,1$. In case (v), ${\mathrm{Ext}_1^{{\mathbb Z}}}(K_i(A\otimes A), K_{1-i}(A\otimes A))=0$ by Lemma \[lem:DivSplit\]. Thus, by the UCT [@Blackadar:Kbook Theorem 23.1.1], it follows that $\sigma_{A,A}$ agrees with the identity in KK.
Using classification, we obtain the following,
\[cor:Suff\] Let $m_0,m_1,n$ be supernatural numbers of infinite type, such that $m_0,m_1$ are coprime and $m_0m_1$ divides $n$. Then $\mathcal E_{n,m_0,m_1}$ and $\mathcal F_{m_0,m_1}$ have asymptotically inner flip.
The algebras $\mathcal E_{n,m_0,m_1}$ and $\mathcal F_{m_0,m_1}$ satisfy the K-theoretic hypotheses of Theorem \[thm:KKsuff\], and therefore the flip map has the same KK-class as the identity. For $\mathcal F_{m_0,m_1}$, it follows from Kirchberg-Phillips classification (see [@Rordam:ClassBook Theorem 8.2.1 (ii) or 8.3.3 (iii)]) that the flip is asymptotically inner.
For $A:=\mathcal E_{n,m_0,m_1}$ (where $n\neq 1$), first note that $A$ is a simple AH algebra with real rank zero, by [@ElliottGong:RR0II Theorem 4.18] (see also Remark \[rmk:EFdef\]); also, the $K_0$-group is unperforated. Hence by [@Lin:AHRR0 Theorem 2.1], it has tracial rank zero. We shall appeal to Lin’s result [@Lin:AsympEquiv Theorem 10.7], which says that $\sigma_{A,A}$ and ${\mathrm{id}}_{A \otimes A}$ are asymptotically unitarily equivalent provided that they have the same KK-class, they agree on traces, and the rotation map $\tilde\eta_{\sigma_{A,A},{\mathrm{id}}_{A \otimes A}}$ vanishes. The first hypothesis has already been verified; the latter two hypotheses are true for trivial reasons, as follows. Since $A \otimes A$ has unique trace, all automorphisms must agree on this trace. The condition $\tilde\eta_{\sigma_{A,A},id_{A \otimes A}}=0$ means that a certain map $K_1(A \otimes A) \to \mathrm{Aff}(T(A \otimes A))$ is the zero map (see [@Lin:AsympEquiv Definition 3.4]); since $K_1(A \otimes A)$ is a torsion group, this holds automatically.
Finally, in the case $n=1$, $A=\mathcal E_{1,1,1} \cong \mathcal Z$, which has asymptotically inner flip by [@DadarlatWinter:KKssa Theorem 2.2].
Necessary conditions {#sec:Necessary}
====================
In this section, we shall prove (ii) $\Rightarrow$ (iv) of Theorem \[thm:MainThm\]. In fact, this implication is proven in potentially weaker generality (not assuming the conditions of strict comparison and infinite or quasidiagonal), as follows:
\[thm:Necessary\] Let $A$ be a $C^*$-algebra in the UCT class, such that $A \otimes A$ has approximately inner flip. Then $K_*(A)$ is isomorphic to one of the following groups (ignoring the grading):
1. $0$;
2. ${\mathbb Z}$;
3. ${\mathbb Q}_n$, where $n$ is a supernatural number of infinite type;
4. ${\mathbb Q}_m/{\mathbb Z}$ where $m$ is a supernatural number of infinite type; or
5. ${\mathbb Q}_n \oplus {\mathbb Q}_m/{\mathbb Z}$, where $n,m$ are supernatural numbers of infinite type, and $m$ divides $n$.
This result is derived from the following, a simple consequence of Lemma \[lem:KunnethFlip\].
\[lem:BasicRestrictions\] Let $A$ be a C\*-algebra in the UCT class with approximately inner flip. If $K_*(A)$ contains direct summands $G_1$ and $G_2$ (irrespective of the grading) then:
1. $G_1 \otimes G_2 = 0$; and
2. ${\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_2)=0$.
Since the flip $\sigma_{A,A}:A\otimes A \to A\otimes A$ is approximately inner, it must agree with the identity map on K-theory. By Lemma \[lem:KunnethFlip\], this implies that $\sigma_{K_*(A),K_*(A)}$ and $\eta_{K_*(A),K_*(A)}$ are both the identity map.
(i): Let $K_*(A)=G_1 \oplus G_2 \oplus G_3$. We have $$K_*(A) \otimes K_*(A) = (G_1 \otimes G_1) \oplus (G_1 \otimes G_2) \oplus (G_2 \otimes G_1) \oplus (G_2 \otimes G_2) \oplus H,$$ (where $H$ involves $G_3$) and $\sigma_{K_*(A),K_*(A)}$ sends $G_1 \otimes G_2$ to $G_2 \otimes G_1$ (by the flip isomorphism). Therefore, if $G_1 \otimes G_2 \neq 0$, then $\sigma_{K_*(A),K_*(A)}$ cannot be the identity map, which is a contradiction.
\(ii) is essentially the same argument: we have $$\begin{aligned}
{\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(A)) = &{\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_1) \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_2) \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_1) \\
&\qquad \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_2) \oplus H',\end{aligned}$$ and $\eta_{K_*(A),K_*(A)}$ sends ${\mathrm{Tor}_1^{{\mathbb Z}}}(G_1,G_2)$ to ${\mathrm{Tor}_1^{{\mathbb Z}}}(G_2,G_1)$ (by $\eta_{G_1,G_2}$).
\[lem:TensorKComp\] Let $A$ be a $C^*$-algebra in the UCT class which has approximately inner flip. Then $$\begin{aligned}
K_0(A \otimes A) &\cong \big(K_0(A) \otimes K_0(A)\big) \oplus \big(K_1(A) \otimes K_1(A)\big) \quad \text{and} \\
K_1(A \otimes A) &\cong {\mathrm{Tor}_1^{{\mathbb Z}}}(K_0(A),K_0(A)) \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}(K_1(A),K_1(A)).\end{aligned}$$
For C\*-algebras $A,B$, the exact sequence from the Künneth formula for $A \otimes B$ can be expressed as two exact sequences, $$\begin{aligned}
0 &\to K_0(A) \otimes K_0(B) \oplus K_1(A) \otimes K_1(B) \to K_0(A \otimes B) \\
&\qquad \to {\mathrm{Tor}_1^{{\mathbb Z}}}(K_0(A),K_1(B)) \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}(K_1(A),K_0(B)) \to 0, \\
0 &\to K_0(A) \otimes K_1(B) \oplus K_0(A) \otimes K_1(B) \to K_1(A \otimes B) \\
&\qquad \to {\mathrm{Tor}_1^{{\mathbb Z}}}(K_0(A),K_0(B)) \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}(K_1(A),K_1(B)) \to 0.\end{aligned}$$
By Lemma \[lem:BasicRestrictions\], $K_0(A) \otimes K_1(A) = 0$ and ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_0(A),K_1(A)) = 0$. Putting these together yield the result.
\[lem:NoCycSummand\] Let $A$ be a C\*-algebra in the UCT class and let $G_p$ be a direct summand of $K_*(A)$ which is a nonzero $p$-group for some prime $p$. Suppose that
1. $A$ has approximately inner flip; or
2. If $A \otimes A$ has approximately inner flip and $K_*(A)=G_p$.
Then $G_p \cong {\mathbb Q}_{p^\infty}/{\mathbb Z}$.
(i): We assume that $A$ has approximately inner flip. First let us show that $G_p$ is directly indecomposable, i.e., that it cannot be expressed as a direct sum of two nontrivial groups. If $G_p=H_1 \oplus H_2$ and $H_1,H_2$ are both nonzero then both are $p$-groups. Hence $H_1$ contains a subgroup isomorphic to ${\mathbb Z}/p{\mathbb Z}$ and so, by the Cartan-Eilenberg exact sequence for Tor, ${\mathrm{Tor}_1^{{\mathbb Z}}}(H_1,H_2)$ contains a subgroup isomorphic to ${\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Z}/p{\mathbb Z},H_2) \neq 0$. By Lemma \[lem:BasicRestrictions\] (ii), this contradicts that $A$ has approximately inner flip.
Therefore, $G_p$ is directly indecomposable and by [@Fuchs:book Corollary 27.4], $G_p$ is either cyclic or isomorphic to ${\mathbb Q}_{p^\infty}/{\mathbb Z}$.
To rule out the case that $G_p$ is cyclic, suppose for a contradiction that $G_p\cong {\mathbb Z}/n{\mathbb Z}$ ($n$ is necessarily a power of $p$). Since $G_p$ is directly indecomposable, it occurs in $K_i(A)$ for either $i=0$ or $1$. Then we see that ${\mathbb Z}/n{\mathbb Z}$ is a direct summand of both $K_i(A) \otimes K_i(A)$ and of ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_i(A),K_i(A))$. By Lemma \[lem:TensorKComp\], ${\mathbb Z}/n{\mathbb Z}$ is a direct summand of both $K_0(A \otimes A)$ and $K_1(A \otimes A)$, so that $K_*(A \otimes A)$ contains two direct summands isomorphic to ${\mathbb Z}/n{\mathbb Z}$. Since $({\mathbb Z}/n{\mathbb Z}) \otimes ({\mathbb Z}/n{\mathbb Z}) \neq 0$ and $A \otimes A$ has approximately inner flip, this contradicts Lemma \[lem:BasicRestrictions\] (i).
(ii): We now assume that $A \otimes A$ has approximately inner flip and that $K_*(A)=G_p$. By the Künneth formula , $K_*(A \otimes A) \cong {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(A))$. By the same argument used in (i) to show that $G_p$ is directly indecomposable, if $K_*(A)$ is not directly indecomposable, then neither is ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(A))$, which cannot be the case by (i). Hence $G_p$ is directly indecomposable, so by [@Fuchs:book Corollary 27.4], $G_p$ is either cyclic or isomorphic to ${\mathbb Q}_{p^\infty}/{\mathbb Z}$ Since ${\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(A)) \cong {\mathbb Q}_{p^\infty}/{\mathbb Z}$ (by (i)), only the latter case is possible.
Let $G=K_*(A)$ and let $T_G$ denote the torsion subgroup of $G$. Our steps are as follows:
1. $G/T_G$ has rank at most one;
2. $G$ splits as a direct sum $T_G \oplus G/T_G$;
3. the theorem.
(i): Let $\mathcal Q$ be the universal UHF algebra, so that $A \otimes A \otimes \mathcal Q$ has approximately inner flip. By the Künneth formula , $K_*(A \otimes \mathcal Q) \cong G \otimes {\mathbb Q}$, which is a rational vector space, and likewise $K_*(A \otimes A \otimes \mathcal Q) \cong (G \otimes {\mathbb Q})^{\otimes 2}$, so by Lemma \[lem:BasicRestrictions\] (i), $G \otimes {\mathbb Q}$ must be either $0$ or ${\mathbb Q}$. Note that $G \otimes {\mathbb Q}\cong (G/T_G) \otimes {\mathbb Q}$ so that $G/T_G$ has rank at most one.
(ii): Since $G/T_G$ is a torsion free group of rank at most one, it is either $0$ or a subgroup of ${\mathbb Q}$. If $G/T_G=0$, there is nothing more to show for (ii).
If $G/T_G\cong {\mathbb Z}$ then the exact sequence $0 \to T_G \to G \to G/T_G \to 0$ splits and (ii) is established.
Otherwise, $G/T_G \cong {\mathbb Q}_n$ for some supernatural number $n$ of infinite type. In order to use Lemma \[lem:DivSplit\] to show that $G\cong T_G \oplus G/T_G$, we need to show that $T_G$ is $n$-divisible. According to [@Fuchs:book Theorem 8.4], $T_G$ is a direct sum of $p$-components $T_p$, over all primes $p$. For $p$ coprime with $n$, $T_p$ is $n$-divisible; we need to show that $T_p$ is also $n$-divisible when $p$ divides $n$.
Let $p$ be a prime which divides $n$. By Corollary \[cor:Suff\], $\mathcal F_{1,p^\infty}$ has approximately inner flip, whence so does $(A \otimes \mathcal F_{1,p^\infty})^{\otimes 2}$. We see that $K_*(\mathcal F_{1,p^\infty})\otimes T_G=0$ (since $K_*(\mathcal F_{1,p^\infty})$ is divisible) and $K_*(\mathcal F_{1,p^\infty}) \otimes G/T_G=0$ (since $G/T_G\cong {\mathbb Q}_n$ is $p$-divisible). Since $$K_*(\mathcal F_{1,p^\infty}) \otimes T_G \to K_*(\mathcal F_{1,p^\infty}) \otimes G \to K_*(\mathcal F_{1,p^\infty}) \otimes G/T_G$$ is exact, it follows that $K_*(\mathcal F_{1,p^\infty}) \otimes G = 0$. Therefore, by the Künneth formula , $$K_*(A \otimes \mathcal F_{1,p^\infty}) \cong {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),K_*(\mathcal F_{1,p^\infty})) \cong {\mathrm{Tor}_1^{{\mathbb Z}}}(K_*(A),{\mathbb Q}_{p^\infty}/{\mathbb Z}).$$ This is isomorphic to $T_p$ by [@Fuchs:book 62.J]. By Lemma \[lem:NoCycSummand\] (ii) (applied to $A \otimes \mathcal F_{1,p^\infty}$ in place of $A$), $T_p$ is either $0$ or ${\mathbb Q}_{p^\infty}/{\mathbb Z}$; in either case, it is $p$-divisible.
This establishes that $T_G$ is $n$-divisible, so by Lemma \[lem:DivSplit\], $G\cong T_G \oplus G/T_G$.
(iii): Now we know that $G\cong T_G \oplus G/T_G$, and $G/T_G$ has rank at most $1$. If $G/T_G \cong {\mathbb Z}$ then by Lemma \[lem:BasicRestrictions\], $T_G=0$.
Otherwise, $G/T_G$ is either ${\mathbb Q}_n$ for some supernatural number $n$ of infinite type or $0$ (in which case we set $n=0$). The summand $T_G$ is the direct sum of its $p$-components $T_p$ over all primes $p$, and the argument in (ii) shows that, for each $p|n$, $T_p$ is either $0$ or ${\mathbb Q}_{p^\infty}/{\mathbb Z}$. For each prime $p$ that doesn’t divide $n$ (in the case $G/T_G \cong {\mathbb Q}_n$), we have ${\mathbb Q}_n \otimes T_p \cong T_p$, and therefore by Lemma \[lem:BasicRestrictions\], $T_p = 0$.
Let $m$ be the supernatural number given by taking the product of all primes $p$ for which $T_p\neq 0$. Thus, $m|n$ and $T_G \cong {\mathbb Q}_m/{\mathbb Z}$, so that $$G \cong \begin{cases} {\mathbb Q}_n \oplus {\mathbb Q}_m/{\mathbb Z}, \quad &G/T_G \cong {\mathbb Q}_n; \\ {\mathbb Q}_m/{\mathbb Z}, \quad &G/T_G=0. \end{cases}$$
\(iii) $\Rightarrow$ (i) $\Rightarrow$ (ii) are immediate.
\(ii) $\Rightarrow$ (iv): By [@EffrosRosenberg Proposition 2.10], $A \otimes A$ has a unique trace, and therefore so does $A$. Moreover, since $A$ has approximately inner half-flip, it is simple and nuclear [@KirchbergPhillips:ExactEmbedding Lemma 3.10] (alternatively, one could use [@EffrosRosenberg Propositions 2.7, 2.8] on $A\otimes A$ here). The K-theoretic restrictions follow from Theorem \[thm:Necessary\].
\(iv) $\Rightarrow$ (v): There exists a C\*-algebra $B$ isomorphic to one of $\mathcal E_{n,m_0,m_1}$, $E_{n,m_0,m_1} \otimes \mathcal O_\infty$, $E_{n,m_0,m_1} \otimes \mathcal O^\infty$, or $\mathcal F_{m_0,m_1}$, which has the same Elliott invariant (consisting of graded, ordered K-theory, the class of the unit in $K_0$, the space of traces, and the pairing between traces and $K_0$) as $A$, except possibly for the $K_0$-class of the unit. If $A$ is type I then it must be stably isomorphic to $\mathbb C$. Otherwise, it is $\mathcal Z$-stable (by [@MatuiSato:comp] in the finite case or [@KirchbergPhillips:ExactEmbedding] in the infinite case), and either quasidiagonal or purely infinite; in either case classification results show it is stably isomorphic to $B$ (see Remark \[rmk:EFdef\])
Finally, (v) $\Rightarrow$ (iii) follows from Corollary \[cor:Suff\].
Semigroup structure under the tensor product operation {#sec:Semigroup}
======================================================
For a supernatural number $n$ of infinite type, let $P_n$ denote the set of all primes which divide $n$.
\[prop:EtensE\] Let $n^{(1)},m^{(1)}_0,m^{(1)}_1,n^{(2)},m^{(2)}_0,m^{(2)}_1$ be supernatural numbers of infinite type, such that $m^{(i)}_0,m^{(i)}_1$ are coprime and $m^{(i)}_0m^{(i)}_1$ divides $n^{(i)}$ for $i=1,2$. Then $\mathcal E_{n^{(1)},m^{(1)}_0,m^{(1)}_1} \otimes \mathcal E_{n^{(2)},m^{(2)}_0,m^{(2)}_1} \cong \mathcal E_{n,m_0,m_1}$ where:
1. $n:=n^{(1)}n^{(2)}$;
2. $m_0$ is the product of the following set of primes, each taken infinitely many times $$(P_{m^{(1)}_0} {\backslash}P_{n^{(2)}}) \cup (P_{m^{(2)}_0} {\backslash}P_{n^{(1)}}) \cup (P_{m^{(1)}_0} \cap P_{m^{(2)}_1}) \cup (P_{m^{(2)}_0} \cap P_{m^{(1)}_1}); \quad \text{and}$$
3. $m_1$ is the product of the following set of primes, each taken infinitely many times $$(P_{m^{(1)}_1} {\backslash}P_{n^{(2)}}) \cup (P_{m^{(2)}_1} {\backslash}P_{n^{(1)}}) \cup (P_{m^{(1)}_1} \cap P_{m^{(2)}_1}) \cup (P_{m^{(2)}_0} \cap P_{m^{(1)}_0}).$$
Set $A:=\mathcal E_{n^{(1)},m^{(1)}_0,m^{(1)}_1} \otimes \mathcal E_{n^{(2)},m^{(2)}_0,m^{(2)}_1}$. This is a simple, separable, unital, $\mathcal Z$-stable, quasidiagonal C\*-algebra that satisfies the UCT. By the Künneth formula , we obtain the following short exact sequence $$\begin{aligned}
\label{eq:EtensEexseq}
\notag
&0 \rightarrow ({\mathbb Q}_{n^{(1)}} \oplus {\mathbb Q}_{m^{(1)}_0}/{\mathbb Z}) \otimes ({\mathbb Q}_{n^{(2)}} \oplus {\mathbb Q}_{n^{(2)}}/{\mathbb Z}) \\
&\qquad \rightarrow K_0(A) \rightarrow {\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_{m^{(1)}_0},{\mathbb Q}_{m^{(2)}_1}) \oplus {\mathrm{Tor}_1^{{\mathbb Z}}}({\mathbb Q}_{m^{(1)}_1},{\mathbb Q}_{m^{(2)}_0}) \to 0.\end{aligned}$$ The first term in this exact sequence is equal to $${\mathbb Q}_{n^{(1)}n^{(2)}} \oplus {\mathbb Q}_{k_1^\infty}/{\mathbb Z}\oplus {\mathbb Q}_{k_2^\infty}/{\mathbb Z}$$ where $k_1$ is the product of the primes in $(P_{m^{(1)}_0} {\backslash}P_{n^{(2)}})$ and $k_2$ is the product of the primes in $(P_{m^{(2)}_0} {\backslash}P_{n^{(1)}})$. Since this term is a divisible group, it is a direct summand of $K_0(A)$.
By [@Fuchs:book 62.J], the last term in the exact sequence is equal to $${\mathbb Q}_{k_3^\infty}/{\mathbb Z}\oplus {\mathbb Q}_{k_4^\infty}/{\mathbb Z},$$ where $k_3$ is the product of the primes in $(P_{m^{(1)}_0} \cap P_{m^{(2)}_1})$ and $k_4$ is the product of the primes in $(P_{m^{(2)}_0} \cap P_{m^{(1)}_1})$.
Let us argue that $k_1,\dots,k_4$ are (pairwise) coprime. First, since $m^{(2)}_0m^{(2)}_1$ divides $n^{(2)}$, it follows that $\gcd(k_1,k_2)=\gcd(k_2,k_3)=\gcd(k_1,k_4)=1$. Likewise, since $m^{(1)}_0m^{(1)}_1$ divides $n^{(1)}$, it follows that $\gcd(k_2,k_3)=\gcd(k_2,k_4)=1$. Finally, since $m^{(1)}_0,m^{(1)}_1$ are coprime, it follows that $\gcd(k_3,k_4)=1$. Hence, $k_1,\dots,k_4$ are coprime. Consequently, $${\mathbb Q}_{k_1^\infty}/{\mathbb Z}\oplus \cdots \oplus {\mathbb Q}_{k_4^\infty}/{\mathbb Z}\cong {\mathbb Q}_{(k_1\cdots k_4)^\infty}/{\mathbb Z}= {\mathbb Q}_{m_0}/{\mathbb Z},$$ and therefore, $$K_0(A) \cong {\mathbb Q}_n \oplus {\mathbb Q}_{m_0}/{\mathbb Z}.$$
Essentially the same argument shows that $K_1(A) \cong {\mathbb Q}_{m_1}/{\mathbb Z}$. Also $$[1_A]_0 = [1_{\mathcal E_{n^{(1)},m^{(1)}_0,m^{(1)}_1}}]_0 \otimes [\mathcal E_{n^{(1)},m^{(1)}_0,m^{(1)}_1}]_0 = 1 \oplus 0 \in {\mathbb Q}_n \oplus {\mathbb Q}_{m_0}/{\mathbb Z}.$$ Hence, by classification (see Remark \[rmk:EFdef\]), $A \cong \mathcal E_{n,m_0,m_1}$.
Essentially the same arguments can be used to derive the next two computations.
Let $n,m^{(1)}_0,m^{(1)}_1,m^{(2)}_0,m^{(2)}_1$ be supernatural numbers of infinite type, such that $m^{(i)}_0,m^{(i)}_1$ are coprime for $i=1,2$ and $m^{(1)}_0m^{(1)}_1$ divides $n$. Then $\mathcal E_{n,m^{(1)}_0,m^{(1)}_1} \otimes \mathcal F_{m^{(2)}_0,m^{(2)}_1} \cong \mathcal F_{m_0,m_1}$ where:
1. $m_0$ is the product of the following set of primes, each taken infinitely many times $$(P_{m^{(2)}_0} {\backslash}P_{n}) \cup (P_{m^{(1)}_0} \cap P_{m^{(2)}_1}) \cup (P_{m^{(2)}_0} \cap P_{m^{(1)}_1}); \quad \text{and}$$
2. $m_1$ is the product of the following set of primes, each taken infinitely many times $$(P_{m^{(2)}_1} {\backslash}P_{n}) \cup (P_{m^{(1)}_1} \cap P_{m^{(2)}_1}) \cup (P_{m^{(2)}_0} \cap P_{m^{(1)}_0}).$$
Let $m^{(1)}_0,m^{(1)}_1,m^{(2)}_0,m^{(2)}_1$ be supernatural numbers of infinite type, such that $m^{(i)}_0,m^{(i)}_1$ are coprime for $i=1,2$. Then $\mathcal F_{m^{(1)}_0,m^{(1)}_1} \otimes \mathcal F_{m^{(2)}_0,m^{(2)}_1} \cong \mathcal F_{m_0,m_1}$ where:
1. $m_0$ is the product of the following set of primes, each taken infinitely many times $$(P_{m^{(1)}_0} \cap P_{m^{(2)}_1}) \cup (P_{m^{(2)}_0} \cap P_{m^{(1)}_1}); \quad \text{and}$$
2. $m_1$ is the product of the following set of primes, each taken infinitely many times $$(P_{m^{(1)}_1} \cap P_{m^{(2)}_1}) \cup (P_{m^{(2)}_0} \cap P_{m^{(1)}_0}).$$
Let $A$ be a classifiable C\*-algebra with approximately inner flip (recall from Remark \[rmk:MainThmClass\] that these are precisely C\*-algebras in Theorem \[thm:MainThm\] with approximately inner flip). Then, by considering the various cases from Theorem \[thm:MainThm\] (v), and using the above computations as appropriate, $A \otimes A$ is stably isomorphic to one of
1. $\mathcal Z$;
2. $\mathcal E_{n,1,m}$ where $m,n$ are supernatural numbers of infinite type and $m$ divides $n$;
3. $\mathcal E_{n,1,m} \otimes \mathcal O_\infty$ where $m,n$ are supernatural numbers of infinite type and $m$ divides $n$ (this occurs in both cases (b) or (c) of Theorem \[thm:MainThm\] (v)); or
4. $\mathcal F_{1,m}$.
Each of these C\*-algebras is self-absorbing (isomorphic to its tensor product with itself), which establishes the corollary below. Among them, $\mathcal Z$, $\mathcal E_{n,1,1}$, $\mathcal E_{n,1,1} \otimes \mathcal O_\infty$, and $\mathcal F_{1,1}$ are strongly self-absorbing (recall that $\mathcal E_{n,1,1} \cong M_n$ and $\mathcal F_{1,1} \cong \mathcal O_2$), hence each of the others is not isomorphic to their own infinite tensor product (by [@TomsWinter:ssa Proposition 1.9]). In fact, without much effort, one can show using classification that $\mathcal E_{n,1,m}^{\otimes \infty} \cong M_n$ and $\mathcal F_{1,m}^{\otimes \infty} \cong \mathcal O_2$.
\[cor:AAsa\] Let $A$ be a classifiable C\*-algebra with approximately inner flip. Then $A \otimes A \otimes$ is stably self-absorbing (i.e., $A^{\otimes 2} \otimes \mathcal K \cong A^{\otimes 4} \otimes \mathcal K$).
We may organize the self-absorbing classifiable C\*-algebras with approximately inner flip (listed above) by ordering them according to absorption ($A \prec B$ if $B \cong A \otimes B$). I would like to thank Mikael Rørdam for suggesting to do so. The following diagram captures this ordering (lower algebras absorb the higher algebras). $$\xymatrix{
& \mathcal E_{1,1,1} \cong \mathbb C \ar@{-}[d] & \\
& \mathcal Z \ar@{-}[dr] \ar@{-}[dl] & \\
\mathcal O_\infty \ar@{-}[dr] && \mathcal E_{n,1,m} \ar@{-}[dl] & \\
&\mathcal E_{n,1,m} \otimes \mathcal O_\infty \ar@{-}[d] & \\
&\mathcal F_{1,m}. &
}$$ Within the family of algebras of the form $\mathcal E_{n,1,m}$, we have $\mathcal E_{n,1,m} \cong \mathcal E_{n,1,m} \otimes \mathcal E_{n',1,m'}$ iff $n'|n$ and $m|m'$. Hence, locally the ordering on this family looks like the following, where $m,n,p,q$ are supernatural numbers of infinite type such that $m,p$ are coprime, $n,q$ are coprime, and $m$ and $p$ both divide $n$. $$\xymatrix{
& \mathcal E_{n,1,mp} \ar@{-}[dr] \ar@{-}[dl] & \\
\mathcal E_{nq,1,mp} \ar@{-}[dr] && \mathcal E_{n,1,m} \ar@{-}[dl] & \\
& \mathcal E_{nq,1,m}. &
}$$ The ordering within the families of algebras of the form $\mathcal E_{n,1,m} \otimes \mathcal O_\infty$ and $\mathcal F_{1,m}$ are similar.
[10]{}
Bruce Blackadar. , volume 5 of [ *Mathematical Sciences Research Institute Publications*]{}. Cambridge University Press, Cambridge, second edition, 1998.
Joan Bosa, Nathaniel Brown, Yasuhiko Sato, Aaron Tikuisis, Stuart White, and Wilhelm Winter. Covering dimension of [[C\*]{}]{}-algebras and 2-coloured classification, 2014. Preprint available at http://homepages.abdn.ac.uk/a.tikuisis/BBSTWW.pdf.
Joachim Cuntz. Simple [[C\*]{}]{}-algebras generated by isometries. , 57(2):173–185, 1977.
Marius Dadarlat and Ulrich Pennig. A [D]{}ixmier-[D]{}ouady theory for strongly self-absorbing [[C\*]{}]{}-algebras, 2013. arXiv preprint math.OA/1302.4468.
Marius Dadarlat and Wilhelm Winter. Trivialization of [$C(X)$]{}-algebras with strongly self-absorbing fibres. , 136(4):575–606, 2008.
Marius Dadarlat and Wilhelm Winter. On the [$KK$]{}-theory of strongly self-absorbing [[C\*]{}]{}-algebras. , 104(1):95–107, 2009.
Edward G. Effros and Jonathan Rosenberg. -algebras with approximately inner flip. , 77(2):417–443, 1978.
George Elliott, Zhuang Niu, Luis Santiago, and Aaron Tikuisis. Decomposition rank of approximately subhomogeneous [[C\*]{}]{}-algebras, 2014. Preprint available at http://homepages.abdn.ac.uk/a.tikuisis/ENSTW.pdf.
George A. Elliott. An invariant for simple [[C\*]{}]{}-algebras. In [*Canadian [M]{}athematical [S]{}ociety. 1945–1995, [V]{}ol. 3*]{}, pages 61–90. Canadian Math. Soc., Ottawa, ON, 1996.
George A. Elliott and Guihua Gong. On the classification of [[C\*]{}]{}-algebras of real rank zero. [II]{}. , 144(3):497–610, 1996.
George A. Elliott and Andrew S. Toms. Regularity properties in the classification program for separable amenable [[C\*]{}]{}-algebras. , 45(2):229–245, 2008.
L[á]{}szl[ó]{} Fuchs. . Pure and Applied Mathematics, Vol. 36. Academic Press, New York-London, 1970.
Guihua Gong, Huaxin Lin, and Zhuang Niu. Classification of finite simple amenable [$\mathcal Z$]{}-stable [[C\*]{}]{}-algebras. arXiv preprint math.OA/1501.00135, 2015.
Ilan Hirshberg, Mikael R[ø]{}rdam, and Wilhelm Winter. -algebras, stability and strongly self-absorbing [[C\*]{}]{}-algebras. , 339(3):695–732, 2007.
Xinhui Jiang and Hongbing Su. On a simple unital projectionless [[C\*]{}]{}-algebra. , 121(2):359–413, 1999.
Eberhard Kirchberg. The classification of purely infinite [[C\*]{}]{}-algebras using kasparov’s theory, 1994. Unpublished preprint.
Eberhard Kirchberg and N. Christopher Phillips. Embedding of exact [[C\*]{}]{}-algebras in the [C]{}untz algebra [$\mathcal O_2$]{}. , 525:17–53, 2000.
Huaxin Lin. Simple [$AH$]{}-algebras of real rank zero. , 131(12):3813–3819 (electronic), 2003.
Huaxin Lin. Asymptotically unitary equivalence and asymptotically inner automorphisms. , 131(6):1589–1677, 2009.
Huaxin Lin and Zhuang Niu. Lifting [$KK$]{}-elements, asymptotic unitary equivalence and classification of simple [[C\*]{}]{}-algebras. , 219(5):1729–1769, 2008.
Hiroki Matui and Yasuhiko Sato. Strict comparison and [$\mathcal {Z}$]{}-absorption of nuclear [$\mathrm C^*$]{}-algebras. , 209(1):179–196, 2012.
Hiroki Matui and Yasuhiko Sato. Decomposition rank of [UHF]{}-absorbing [[C\*]{}]{}-algebras. , 163(14):2687–2708, 2014.
N. Christopher Phillips. A classification theorem for nuclear purely infinite simple [[C\*]{}]{}-algebras. , 5:49–114 (electronic), 2000.
Mikael R[ø]{}rdam. Classification of certain infinite simple [[C\*]{}]{}-algebras. , 131(2):415–458, 1995.
Mikael R[ø]{}rdam. The stable and the real rank of [$\mathcal Z$]{}-absorbing [[C\*]{}]{}-algebras. , 15(10):1065–1084, 2004.
Mikael R[ø]{}rdam and Erling St[ø]{}rmer. , volume 126 of [*Encyclopaedia of Mathematical Sciences*]{}. Springer-Verlag, Berlin, 2002. Operator Algebras and Non-commutative Geometry, 7.
Yasuhiko Sato, Stuart White, and Wilhelm Winter. Nuclear dimension and [$\mathcal Z$]{}-stability, 2014. Invent. Math., to appear. arXiv:1403.0747.
Claude Schochet. Topological methods for [$C^{\ast} $]{}-algebras. [II]{}. [G]{}eometric resolutions and the [K]{}ünneth formula. , 98(2):443–458, 1982.
Aaron Tikuisis and Wilhelm Winter. Decomposition rank of [${\mathcal Z}$]{}-stable [[C\*]{}]{}-algebras. , 7(3):673–700, 2014.
Andrew S. Toms and Wilhelm Winter. Strongly self-absorbing [[C\*]{}]{}-algebras. , 359(8):3999–4029, 2007.
Andrew S. Toms and Wilhelm Winter. -stable [ASH]{} algebras. , 60(3):703–720, 2008.
Wilhelm Winter. On the classification of simple [$\mathcal Z$]{}-stable [[C\*]{}]{}-algebras with real rank zero and finite decomposition rank. , 74(1):167–183, 2006.
Wilhelm Winter. Decomposition rank and [$\mathcal Z$]{}-stability. , 179(2):229–301, 2010.
Wilhelm Winter. Nuclear dimension and [$\mathcal{Z}$]{}-stability of pure [[C\*]{}]{}-algebras. , 187(2):259–342, 2012.
Wilhelm Winter. Localizing the [E]{}lliott conjecture at strongly self-absorbing [[C\*]{}]{}-algebras. , 692:193–231, 2014.
[^1]: The author is supported by an NSERC PDF
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study families of objects in Fukaya categories, specifically ones whose deformation behaviour is prescribed by the choice of an odd degree cohomology class. This leads to invariants of symplectic manifolds, which we apply to blowups along symplectic mapping tori.'
author:
- Paul Seidel
date: August 2011
title: |
**Abstract analogues of flux\
as symplectic invariants**
---
Introduction {#introduction .unnumbered}
============
[**Motivation.**]{} An interesting invariant of a closed symplectic manifold $M$ is its flux group, a subgroup of $H^1(M;{\mathbb{R}})$ obtained from the topology of loops of symplectic automorphisms. This can be effectively studied using Floer cohomology, one of the notable insights being that the flux group is always discrete [@ono06]. Now consider the following question:
> *Are there flux-type subgroups in $H^{2k-1}(M;{\mathbb{R}})$, for $k>1$, which can be nontrivial for manifolds with $H^1(M;{\mathbb{R}}) = 0$?*
The last clause excludes one obvious direction, which is to consider the action of loops of symplectic automorphisms on $H^*(M;{\mathbb{R}})$ (that action vanishes if $H^1(M;{\mathbb{R}}) = 0$ by the rigidity theorem [@lalonde-mcduff-polterovich97]). Really, what the question is aiming for is a formalism in which higher degree differential forms replace the closed one-forms in their usual relation to symplectic vector fields, so anything related to symplectic automorphism groups can’t really be the answer. This clarification may make the whole endeavour seem quixotic. Still, if one looks at it from the point of view of quantum cohomology $\mathit{QH}^*(M)$, the situation is less clear-cut. Passage to quantum cohomology generally reduces the grading to ${\mathbb{Z}}/2$, putting all odd degree cohomology formally on equal footing (but degree one classes retain a more direct connection to geometry, because their quantum product remains equal to the classical cup product). In that vein, it turns out that one can give a partially positive answer to the question above, at least if one is willing to settle for an invariant which is somewhat more obscure, lacking the simplicity and geometric elegance of the flux group.
[**The examples.**]{} As an application, we consider a particular pair of $28$-dimensional simply-connected symplectic manifolds (the following is only an outline of the construction, omitting many details and assumptions). Let $K$ be a $K3$ surface, and $T \subset K$ a symplectically embedded two-torus. Take $K^7$, the product of seven copies of $K$, and blow up the $12$-dimensional submanifold $T^2 \times K^2$. Denote the outcome by $B^{\mathit{triv}}$. This has a more interesting cousin $B$, defined in the same way but where the blowup locus is a product of $T$ and the symplectic mapping torus of a certain automorphism of $K \times K$ (embedded into $K^7$ by using the $h$-principle). It is known that the symplectic automorphism group has many connected components which are not detected by classical topological means (see for instance [@seidel04b], or [@seidel-thomas99] for a mirror symmetry viewpoint). Based on that, one can ensure that $B^{\mathit{triv}}$ is diffeomorphic to $B$, and that their symplectic structures are deformation equivalent. Nevertheless, for a specific choice of automorphism, we will show:
> *$B$ and $B^{\mathit{triv}}$ are not symplectically isomorphic.*
The construction of these manifolds is similar to that of the first known examples of distinct but deformation equivalent symplectic structures [@mcduff87], which were also based on blowing up. That paper used (roughly speaking) a bordism-valued refinement of Gromov-Witten theory as an invariant. Because such refinements are hard to define and compute, we can’t say how they would behave in our situation, but intuitively it seems unlikely that they would be able to distinguish the two manifolds above. In any case, the approach taken in this paper is quite different.
[**The invariant.**]{} Let’s temporarily go back to the simpler case of symplectic mapping tori. The symplectic mapping torus of an automorphism $f$ is a symplectic fibration over $T$ which has trivial monodromy in one direction, and monodromy $f$ in the other direction. Let’s say for concreteness that $T = {\mathbb{R}}^2/{\mathbb{Z}}^2$ has coordinates $(p,q)$, and that the monodromy is trivial in $q$-direction, and $f$ in $p$-direction. The symplectic mapping torus contains plenty of Lagrangian submanifolds fibered over trivial circles $\{p\} \times S^1$. If one then moves such a Lagrangian submanifold by the time-one map of the symplectic vector field $\partial_p$, the effect is the same as applying $f$ fibrewise. For suitable examples of $f$, this allows one to show that $[dq]$ does not lie in the flux group, which distinguishes the mapping torus from the trivial one (a similar approach was used in [@seidel02b]). To make a more abstract version of the argument, we consider families of objects in the Fukaya category whose deformation is driven by the class $[dq]$. The idea of introducing families of Lagrangian submanifolds into Floer cohomology theory is due to Fukaya [@fukaya02c; @fukaya09]. Generally speaking, it shows much potential for giving fundamental insights as well as applications, but also encounters considerable foundational difficulties. Here, we bypass these issues by choosing a more constrained version of the notion of family, which has a straightforward basis in algebraic geometry, but is somewhat harder to connect to symplectic geometry.
The outcome is an invariant of a symplectic manifold $M$ which depends on the following auxiliary data. Take an element $\lambda$ of the (universal single-variable) Novikov field, and an idempotent $z$ in $\mathit{QH}^0(M)$. To these we associate an appropriate version of the Fukaya category $\mathit{Fuk}(M)_{\lambda,z}$ and its completion (split-closed triangulated envelope) $\mathit{Fuk}(M)_{\lambda,z}^{\mathit{perf}}$. Additionally, choose an elliptic curve $\bar{{\mathscr{S}}}$ over the Novikov field, together with a nonzero algebraic one-form on it. Given a class in odd degree quantum cohomology, $x \in \mathit{QH}^1(M)$, one can then ask whether it is [*periodic*]{}, which means whether objects of the Fukaya category can be extended to families over $\bar{{\mathscr{S}}}$ with deformation behaviour prescribed by $x$. In particular, we can apply this idea to symplectic mapping tori and recover some of the results ordinarily proved using flux. More interestingly, we can exploit existing ideas about the behaviour of Fukaya categories under blowups [@smith10], and thereby arrive at the result stated above. The main point is that the relevant part of the Fukaya category of $B^{\mathit{triv}}$ is well-understood, allowing us to prove that certain classes $x$ are periodic. The Fukaya category of $B$ is not known to the same extent, but partial computations are enough to determine that certain classes $x$ are not periodic, since that only requires finding a specific Lagrangian submanifold which can serve as a counterexample.
To conclude this discussion, we should mention that the basic idea is by no means new in homological algebra. From that elevated vantage point, what we are doing (in a rather ad hoc way) is to study algebraic one-parameter subgroups (the elliptic curves $\bar{{\mathscr{S}}}$ which appeared above) inside the derived Picard group [@yekutieli04; @keller04] of the Fukaya category. The idea is to think of elements of Hochschild cohomology as vector fields on an abstract “moduli space of objects”, and that we are asking which vector fields integrate to “periodic flows”. On an informal level, it is clear that this provides an algebraic counterpart to the geometric ideas underlying flux.
[**Structure of the paper.**]{} Section \[sec:theory\] sets up the algebraic theory of families of objects, much of it straightforward. The key to uniqueness results for families is the discussion surrounding Lemma \[th:its-projective\], which is then further developed for our intended applications in Section \[subsec:unique\]. An abstract version of our invariant is introduced in Definition \[th:per\]. Section \[sec:elliptic\] discusses the simplest example of the two-torus $T$. Of course, its Fukaya category has already been studied exhaustively, starting with [@polishchuk-zaslow98]. Still, we devote some energy to it in order to prepare for the case of mapping tori, which is the topic of Section \[sec:mapping-tori\] (following some preliminaries on Floer cohomology and Fukaya categories, in Section \[sec:automorphisms\]). At first, it will seem that our computations lead us further from the intended goal, since we choose Lagrangian submanifolds whose Floer cohomology is largely independent of the choice of automorphism underlying the mapping torus. However, we eventually do manage to recover some information about the automorphism, by a double covering trick which appears in Section \[subsec:covering-trick\]. Finally, most of Section \[sec:blowup\] is general discussion of blowups. The detailed construction of the manifolds $B$ and $B^{\mathit{triv}}$ is carried out in Section \[subsec:final\].
[**Acknowledgments.**]{} Section \[sec:blowup\] of this paper relies crucially on results of [@abouzaid10; @abouzaid-fukaya-oh-ohta-ono11; @smith10]. I am particularly indebted to Mohammed Abouzaid and Ivan Smith for explaining their insights to me. Conversations with Denis Auroux, Ludmil Katzarkov, Davesh Maulik, Tim Perutz, and Nick Sheridan were also helpful. I could not have carried out this project without the generous support of the Simons Center for Geometry and Physics. Further support was received through NSF grant DMS-1005288.
Families of objects\[sec:theory\]
=================================
Suppose that we are given an $A_\infty$-category $A$, and a class in its degree $1$ Hochschild cohomology $\mathit{HH}^1(A,A)$. This class determines one in $H^1(\mathit{hom}_A(X,X))$ for every object $X$, hence an infinitesimal first order deformation of $X$. Deformations coming from Hochschild cohomology have additional properties, for instance (in characteristic $0$) they can be extended to arbitrarily high orders in a formal parameter. However, instead of looking at the infinitesimal theory, we want to consider global deformations. For the sake of illustration, take the parameter space to be the affine line. One then looks for families ${\mathscr{X}}= \{{\mathscr{X}}_s\}$ depending on one algebraic variable $s$, whose fibre at the origin is fixed, ${\mathscr{X}}_0 {\cong}X$, and whose first order deformation behaviour at any value of $s$ is the element of $H^1(\mathit{hom}_A({\mathscr{X}}_s,{\mathscr{X}}_s))$ induced by our Hochschild class. We are mainly concerned with the uniqueness of such families. This, while not totally straightforward, turns out to be much easier than existence issues. An elementary parallel would be the question of integrating a vector field. Indeed, one might think of the original Hochschild cohomology class as determining a vector field on the “moduli spaces of objects in $A$” (making this rigorous requires machinery far beyond that deployed here; interested readers are referred to [@toen-vaquie07]).
We work over an algebraically closed field $R$ of characteristic $0$. While the condition of algebraic closedness is perhaps mostly for the sake of familiarity, the restriction on the characteristic is crucial, since we will be using differentiation (and in particular Lemma \[th:locally-free\]). Sign conventions for $A_\infty$-algebras and associated structures usually follow [@seidel04]. Those for twisted complexes are specifically as in [@seidel04 Remark 3.26]. All categories are assumed to be small.
$A_\infty$-categories
---------------------
Fix an $A_\infty$-category $A$ over $R$. This is assumed to be ${\mathbb{Z}}$-graded, strictly unital, and proper. The units (identity endomorphisms) are denoted by $e_X \in \mathit{hom}_A^0(X,X)$. Write $H(A)$ for the associated cohomology level category, and by $H^0(A)$ the version where only morphisms of degree $0$ are allowed. Properness means that the morphisms in $H(A)$ are graded vector spaces of finite total dimension. Two objects of $A$ are called quasi-isomorphic if they become isomorphic in $H^0(A)$. There are various canonical formal enlargements of $A$. Possibly the simplest one is the $A_\infty$-category of twisted complexes $A^{\mathit{tw}}$, introduced in [@bondal-kapranov91; @kontsevich94]. One can carry out this enlargement in two steps. First, consider the additive envelope $A^{\oplus}$, whose objects are formal expressions $$\label{eq:formal-sum}
X = \bigoplus_{i \in I} F^i \otimes X^i[-\sigma^i]$$ where $I$ is a finite set, the $X^i$ are objects of $A$, formally shifted by degrees $\sigma^i \in {\mathbb{Z}}$, and the $F^i$ are finite-dimensional vector spaces. Morphisms in $A^{\oplus}$ can be thought of as matrices, whose entries are morphisms in $A$ tensored with maps of vector spaces. Correspondingly, the $A_\infty$-products combine those of $A$ with composition of linear maps and matrix multiplication (with auxiliary signs due to the shift). In the second step, one defines a twisted complex to be an object $X \in \mathit{Ob}\,A^{\oplus}$ equipped with an additional differential. This differential $\delta_X \in \mathit{hom}_{A^{\oplus}}^1(X,X)$ is an endomorphism which is strictly decreasing with respect to some filtration of $X$ by sub-objects, and which satisfies the equation $$\label{eq:generalized-maurer-cartan}
\mu^1_{A^{\oplus}}(\delta_X) + \mu^2_{A^{\oplus}}(\delta_X,\delta_X) + \cdots = 0.$$ $A^{\mathit{tw}}$ is an $A_\infty$-category with the same general properties as $A$, and which contains $A$ as a full subcategory. It is closed under shifts and mapping cones, and is characterized up to quasi-equivalence as the minimal enlargement with this property. In our formulation, it also admits a canonical operation of tensoring a given object with a finite-dimensional vector space.
Suppose that we allow only trivial one-dimensional spaces $F^i = R$. The resulting objects, which can be written more concisely as $X = \bigoplus_{i \in I} X^i[-\sigma^i]$, form a quasi-equivalent full subcategory of $A^{\mathit{tw}}$. We prefer the more general form since it is better suited to later generalizations.
There is a different approach to formal enlargements, through $A_\infty$-modules [@keller99]. Write $C$ for the differential graded category of complexes of $R$-vector spaces whose cohomology is of finite total dimension. A (right) $A$-module with finite cohomology is an $A_\infty$-functor $A^{\mathit{opp}} \rightarrow C$. Concretely, such a module $M$ assigns to each $X \in \mathit{Ob}\,A$ a graded vector space $M(X)$, together with structure maps $$\label{eq:module}
\begin{aligned}
& \mu^1_M: M(X_0) \longrightarrow M(X_0)[1], \\
& \mu^2_M: M(X_1) \otimes \mathit{hom}_A(X_0,X_1) \longrightarrow M(X_0), \\
& \mu^3_M: M(X_2) \otimes \mathit{hom}_A(X_1,X_2) \otimes \mathit{hom}_A(X_0,X_1) \longrightarrow M(X_0)[-1], \\
& \dots
\end{aligned}$$ satisfying the $A_\infty$-module equations ([@keller99] or [@seidel04 Section 1j]), and such that $(M(X),\mu^1_M) \in \mathit{Ob}\,C$ for all $X$. We require $M$ to be strictly unital, which means that $\mu^2_M(m,e_{X_0}) = m$, and $\mu^d_M(m,a_{d-1},\dots,a_1) = 0$ whenever $d \geq 3$ and one of the $a_i$ is a unit. Such modules form an $A_\infty$-category $A^{\mathit{mod}}$ (this is in fact a differential graded category, but we prefer to view it as an $A_\infty$-category with trivial higher order products, which entails slightly different sign conventions). A morphism $b \in \mathit{hom}_{A^{\mathit{mod}}}(M_0,M_1)$ consists of $$\label{eq:module-map}
\begin{aligned}
& b^1: M_0(X_0) \longrightarrow M_1(X_0)[|b|], \\
& b^2: M_0(X_1) \otimes \mathit{hom}_A(X_0,X_1) \longrightarrow M_1(X_0)[|b|-1], \\
& \dots
\end{aligned}$$ Again, we require strict unitality, which means that $b^d(m,a_{d-1},\dots,a_1) = 0$ whenever one of the $a_i$ is a unit.
The Yoneda embedding ([@fukaya01b] or [@seidel04 Section 1l]) is a canonical $A_\infty$-functor $A \rightarrow A^{\mathit{mod}}$. On objects, it maps $Y$ to the module $Y^{\mathit{yon}}$ with $Y^{\mathit{yon}}(X) = \mathit{hom}_A(X,Y)$ and $\mu^d_{Y^{\mathit{yon}}} = \mu^d_A$. The first level map on morphisms is $$\label{eq:yoneda1}
\begin{aligned}
& \mathit{hom}_A(Y_0,Y_1) \longrightarrow \mathit{hom}_{A^{\mathit{mod}}}(Y_0^{\mathit{yon}},Y_1^{\mathit{yon}}), \\
& a \longmapsto a^{\mathit{yon}}, \quad a^{\mathit{yon},d}(a_d,\dots,a_1) = \mu^{d+1}_A(a,a_d,\dots,a_1).
\end{aligned}$$
\[th:yoneda\] The map is a quasi-isomorphism.
Consider the map in inverse direction, taking a module homomorphism $b$ to the element $a = b^1(e_{Y_0})$. Composing the two in one way yields the identity map on $\mathit{hom}_A(Y_0,Y_1)$. The other composition is chain homotopic to the identity: an explicit homotopy is $$\label{eq:h-homotopy}
\begin{aligned}
& h: \mathit{hom}_{A^{\mathit{mod}}}(Y_0^{\mathit{yon}},Y_1^{\mathit{yon}}) \longrightarrow \mathit{hom}_{A^{\mathit{mod}}}(Y_0^{\mathit{yon}},Y_1^{\mathit{yon}})[-1], \\
& h(b)^d(a_d,\dots,a_1) = b^{d+1}(e_{Y_0},a_d,\dots,a_1).
\end{aligned}$$
In other words, the Yoneda embedding is cohomologically full and faithful. Moreover, it canonically extends to a cohomologically full and faithful functor $A^{\mathit{tw}} \rightarrow A^{\mathit{mod}}$. The easiest way to see this is to think of it as the composition $$\label{eq:yoneda-restriction}
A^{\mathit{tw}} \longrightarrow (A^{\mathit{tw}})^{\mathit{mod}} \longrightarrow A^{\mathit{mod}}.$$ where the first arrow is the Yoneda embedding for $A^{\mathit{tw}}$, and the second is restriction of modules from $A^{\mathit{tw}}$ to $A$. Since objects of $A$ generate $A^{\mathit{tw}}$ by definition, the restriction functor is a quasi-equivalence, so is again cohomologically full and faithful.
Idempotent splittings\[subsec:splittings\]
------------------------------------------
Suppose that we have an object $Y$ of $A$ together with an endomorphism in the category $H^0(A)$ which is idempotent. One can always lift it to a [*homotopy idempotent*]{}, which is a sequence $p = \{p^d\}$ of elements $p^d \in \mathit{hom}_A(Y,Y)^{1-d}$ ($d \geq 1$) satisfying the equations $$\label{eq:homotopy-idempotent}
\sum_r \sum_{k_1 + \cdots + k_r = d} \mu_A^r(p^{k_r},\dots,p^{k_1}) =
\begin{cases} p^{d-1} & \text{$d$ even,} \\ 0 & \text{$d$ odd} \end{cases}$$ for any $d \geq 1$, and such that $p^1$ represents our original idempotent. This is proved in [@seidel04 Section 4], but it is maybe useful to summarize the argument in more elementary language. The choice of $p^1,p^2$ is straightforward, and the remaining process is inductive. Suppose that $p^1,\dots,p^{d-1}$ have been chosen satisfying the respective equations. Take the sum of all the terms on left hand side of which have $r \geq 2$. These give rise to a cocycle of bidegree $(d,2-d)$ in the following periodic complex of graded vector spaces: $$\begin{gathered}
\label{eq:trivial-hh}
\cdots \rightarrow \mathit{Hom}_{H(A)}(Y,Y) \xrightarrow{[a] \mapsto [(-1)^{|a|}\mu^2_A(p^1,a) - \mu^2_A(a,p^1)] } \mathit{Hom}_{H(A)}(Y,Y) \longrightarrow \\ \xrightarrow{[a] \mapsto [(-1)^{|a|+1}\mu^2_A(p^1,a) - \mu^2_A(a,p^1) + a]} \mathit{Hom}_{H(A)}(Y,Y) \rightarrow \cdots \qquad\end{gathered}$$ Since that complex is acyclic, we can modify $p^{d-1}$ by adding a $\mu^1_A$-cocycle, so that the same sum of terms represents the zero class in $\mathit{Hom}_{H(A)}^{2-d}(Y,Y)$, and then choose $p^d$ so that holds.
The homotopy idempotent can then be used to define an $A_\infty$-module $M = (Y,p)^{\mathit{yon}}$, which in the category $H^0(A^{\mathit{mod}})$ is the direct summand of $Y^{\mathit{yon}}$ associated to the Yoneda image of $[p^1]$ (hence, independent of the choice of the homotopy idempotent up to quasi-isomorphism). It consists of the spaces $$\label{eq:q-splitting}
M(X) = \mathit{hom}_A(X,Y)[q]$$ where $q$ is a formal variable of degree $-1$, and has differential $$\mu^1_M(aq^j) = \sum_{r \geq 0} \sum_{k_1 + \cdots + k_r \leq j} \mu^{r+1}_A(p^{k_r},\dots,p^{k_1},a)\, q^{j-k_1-\cdots-k_r}
+ \begin{cases} 0 & \text{$j$ even,} \\ aq^{j-1} & \text{$j$ odd.}
\end{cases}$$ We refer to [@seidel04 Section 4] for full details, including the definition of the remaining maps $\mu^d_M$. By the same kind of argument as in , idempotent summands of objects in $A^{\mathit{tw}}$ (or indeed in $A^{\mathit{mod}}$) can also be represented in $A^{\mathit{mod}}$. One defines $A^{\mathit{perf}}$, the category of [*perfect modules*]{}, to be the full subcategory of $A^{\mathit{mod}}$ consisting of all objects that are quasi-isomorphic to an idempotent summand of an object in $A^{\mathit{tw}}$. It is easy to see that $A^{\mathit{perf}}$ is again proper.
Hochschild cohomology
---------------------
Let $\mathit{CC}(A,A)$ be the (reduced) Hochschild complex of $A$, and $\mathit{HH}(A,A)$ its cohomology. A Hochschild cochain $g$ is a sequence of multilinear maps $$g^d: \mathit{hom}_A(X_{d-1},X_d) \otimes \cdots \otimes \mathit{hom}_A(X_0,X_1) \longrightarrow
\mathit{hom}_A(X_0,X_d)[|g|-d],$$ $d \geq 0$, which vanish if one of the inputs is an identity morphism. The Hochschild differential is $$\begin{aligned}
(\partial g)^d(a_d,\dots,a_1) = \sum_{i,j} (-1)^{(|g|-1)(|a_1|+\cdots+|a_i|-i)}
\mu^{d-j+1}_A(a_d,\dots,g^j(a_{i+j},\dots,a_{i+1}),\dots,a_1) \\
+ \sum_{i,j} (-1)^{|g|+|a_1|+\cdots+|a_i|-i} g^{d-j+1}(a_d,\dots,\mu_A^j(a_{i+j},\dots,a_{i+1}),\dots,a_1).
\end{aligned}$$ We need to remind the reader briefly of the (partial) functoriality properties of Hochschild cohomology. Let $G: A \rightarrow \tilde{A}$ be a (strictly unital) $A_\infty$-functor. Then there are canonical chain maps $$\label{eq:functoriality}
\mathit{CC}(A,A) \xrightarrow{G_*} \mathit{CC}(A,\tilde{A}) \xleftarrow{G^*} \mathit{CC}(\tilde{A},\tilde{A}),$$ where the middle term is the Hochschild cochain complex of $A$ with coefficients in the $A$-bimodule $\tilde{A}$. If $G$ is cohomologically full and faithful, the left hand map is a quasi-isomorphism. Less obviously (it is a form of Morita invariance), if $G$ is a quasi-equivalence, the right hand map is also a quasi-isomorphism. Informally one can think of these two maps as follows. If we deform the $A_\infty$-structure on either $A$ or $\tilde{A}$ infinitesimally, there will be a term measuring the failure of $G$ to be an $A_\infty$-functor for the deformed structure, which is the image of the corresponding deformation classes in $\mathit{HH}(A,\tilde{A})$.
We also need to know about the behaviour under formal enlargement. Thinking of Hochschild cohomology in terms of deformation theory, one expects deformations of $A$ to induce ones of $A^{\mathit{tw}}$. This can indeed be made rigorous, leading to a canonical map $$\label{eq:twisted-gamma}
\Gamma^{\mathit{tw}}: \mathit{CC}(A,A) \longrightarrow \mathit{CC}(A^{\mathit{tw}},A^{\mathit{tw}}).$$ To see an example of how this looks like explicitly, take a Hochschild cochain $g$, and extend it in the obvious way to a cochain $g^{\oplus}$ on $A^{\oplus}$. Then the image $g^{\mathit{tw}} = \Gamma^{\mathit{tw}}(g)$ has leading term $$\label{eq:twisted-gamma-0}
g^{\mathit{tw},0} = g^{\oplus,0} + g^{\oplus,1}(\delta_X) + g^{\oplus,2}(\delta_X,\delta_X) + \cdots \in \mathit{hom}_{A^{\mathit{tw}}}^{|g|}(X,X).$$ Maybe more obviously, restriction to the full subcategory $A \subset A^{\mathit{tw}}$ yields a map in reverse direction to . These two maps are inverse quasi-isomorphisms (strict inverses in one order, and inverses up to homotopy in the other), which is one form of the derived invariance of Hochschild cohomology. There is also an analogue for $A$-modules. Again thinking in terms of deformations, a deformation of $A$ induces a deformation of $A^{\mathit{mod}}$, but one which remains within the class of dg categories with curvature. In fact, the product $\mu^2_{A^{\mathit{mod}}}$ does not actually change, since it is defined in a way which does not involve $\mu^*_A$. Concretely, this means that we have a map $$\label{eq:gamma-for-modules}
\Gamma^{\mathit{mod}}: \mathit{CC}(A,A) \longrightarrow \mathit{CC}(A^{\mathit{mod}},A^{\mathit{mod}})$$ such that $g^{\mathit{mod}} = \Gamma^{\mathit{mod}}(g)$ has only two nontrivial components $$\label{eq:gamma-mod}
\begin{aligned}
& g^{\mathit{mod},0} \in \mathit{hom}_{A^{\mathit{mod}}}^{|g|}(M_0,M_0), \\
& g^{\mathit{mod},1}: \mathit{hom}_{A^{\mathit{mod}}}(M_0,M_1) \longrightarrow \mathit{hom}_{A^{\mathit{mod}}}(M_0,M_1)[|g|-1].
\end{aligned}$$ These are given by $$\label{eq:gamma-formula}
\begin{aligned}
(g^{\mathit{mod},0})^d(m,a_{d-1},\dots,a_1) & = \sum_{ij} (-1)^\ast \, \mu_{M_0}^{d-j+1}(m,\dots, g^j(a_{i+j},\dots,a_{i+1}),\dots,a_1), \\
(g^{\mathit{mod},1})(b)^d(m,a_{d-1},\dots,a_1) & = \sum_{ij} (-1)^\ast \, b^{d-j+1}(m,\dots,g^j(a_{i+j},\dots,a_{i+1}),\dots,a_1), \\
& \ast = (|g|-1)(|a_{i+1}|+\cdots+|a_{d-1}|+|m|+d-i-1)+|g|.
\end{aligned}$$
There is a restriction map from the Hochschild complex of $A^{\mathit{mod}}$ to that of its full subcategory $A^{\mathit{perf}}$. One can accordingly restrict and get a map $\Gamma^{\mathit{perf}}$, which is of course defined by the same formulae. Another manifestation of derived invariance of Hochschild cohomology says that $\Gamma^{\mathit{perf}}$ is a quasi-isomorphism. Next, $A^{\mathit{perf}}$ contains a full subcategory quasi-isomorphic to $A^{\mathit{tw}}$, and by restriction and the argument from , one gets a further map from the Hochschild cohomology of $A^{\mathit{perf}}$ to that of $A^{\mathit{tw}}$. The situation can be summarized in the commutative diagram $$\xymatrix{
\mathit{HH}(A,A)
\ar@/_1pc/[drr]^-{\Gamma^{\mathit{tw}}}_{{\cong}} \ar@/_3pc/[ddrr]^-{\Gamma^{\mathit{perf}}}_{{\cong}} \ar@/_6pc/[dddrr]^-{\Gamma^{\mathit{mod}}} \ar[rr]^-{\mathrm{id}} && \mathit{HH}(A,A) \\
&& \mathit{HH}(A^{\mathit{tw}},A^{\mathit{tw}}) \ar[u]_{{\cong}} \\
&& \mathit{HH}(A^{\mathit{perf}},A^{\mathit{perf}}) \ar[u]_{{\cong}} \\
&& \mathit{HH}(A^{\mathit{mod}},A^{\mathit{mod}}) \ar[u]
}$$ where the vertical arrows are restriction type maps. Other than derived invariance (which we will not discuss further, but see [@keller06 Section 5.4]), the only nontrivial point in this diagram is that the map $\Gamma^{\mathit{mod}}$ is compatible with restriction. To understand that, we have to look at for the Yoneda embedding: $$\label{eq:yoneda-functoriality}
\mathit{CC}(A,A) \longrightarrow \mathit{CC}(A,A^{\mathit{mod}}) \longleftarrow \mathit{CC}(A^{\mathit{mod}},A^{\mathit{mod}}).$$ Take $g \in \mathit{CC}(A,A)$, and consider its image under the first map in . For simplicity let’s look only at the constant term of this, which consists of an endomorphism $b \in \mathit{hom}_{A^{\mathit{mod}}}(Y^{\mathit{yon}},Y^{\mathit{yon}})$ for each $Y$, given by $$b^d(a_d,\dots,a_1) = \mu^{d+1}_A(g^0,a_d,\dots,a_1). \label{eq:1-cc}$$ On the other hand, we can take $g^{\mathit{mod}} \in \mathit{CC}(A^{\mathit{mod}},A^{\mathit{mod}})$ and pull it back to $\mathit{CC}(A,A^{\mathit{mod}})$ as in the second map in , which leads to another cochain $\mathit{CC}(A,A^{\mathit{mod}})$ with constant term $$\label{eq:2-cc}
\begin{aligned}
\tilde{b}^d(a_d,\dots,a_1) = & -\sum_{i+j<d} (-1)^\ast \mu^{d-j+1}_A(a_d,\dots,g^j(a_{i+j},\dots,a_{i+1}),\dots,a_1), \\ & \ast = (|g|-1)(|a_{i+1}|+\cdots+|a_d|+d-i).
\end{aligned}$$ Assuming that $g$ is a Hochschild cocycle, we can write $$\begin{aligned} &
\tilde{b}^d(a,a_{d-1},\dots,a_1) - b^d(a,a_{d-1},\dots,a_1) = \\ &
\qquad \sum_{i<d} (-1)^{(|g|-1)(|a_{i+1}| + \cdots + |a_d| + d - i)} \mu^{i+1}_A(g^{d-i}(a_d,\dots,a_{i+1}),\dots,a_1) \\ & \qquad + \sum_{i,j} (-1)^{|g| + |a_1| + \cdots + |a_i| - i + (|g|-1)(|a_1|+\cdots+|a_d|-d)} g^{d-j+1}(a_d,\dots,\mu_A^j(a_{i+j},\dots,a_{i+1}),\dots,a_1).
\end{aligned}$$ Quite obviously, this is the coboundary of another endomorphism of $Y^{\mathit{yon}}$ of degree $|g|-1$, given by $(a_d,\dots,a_1) \mapsto (-1)^{|g|(|a_1|+ \cdots + |a_d|-d)+1}g^d(a_d,\dots,a_1)$, hence zero on cohomology. The general computation is similar.
Bimodules\[subsec:bimodules\]
-----------------------------
$A_\infty$-bimodules have already made a brief appearance before, but we will now consider them in a little more detail. Let $A$ and $\tilde{A}$ be $A_\infty$-categories over $R$, with the usual unitality and properness assumptions. An [*$(A,\tilde{A})$-bimodule with finite cohomology*]{} $P$ assigns to any pair of objects $(X,\tilde{X}) \in \mathit{Ob}\, A \times \mathit{Ob}\,\tilde{A}$ a graded vector space $P(\tilde{X},X)$, which comes with structure maps $$\label{eq:bimodule-structure}
\begin{aligned}
& \mu_P^{s|1|t}: \mathit{hom}_A(X_{s-1},X_s) \otimes \cdots \otimes \mathit{hom}_A(X_0,X_1) \otimes P(\tilde{X}_t,X_0) \\ & \qquad \qquad \qquad \qquad \otimes \mathit{hom}_{\tilde{A}}(\tilde{X}_{t-1},\tilde{X}_t) \otimes \cdots \otimes \mathit{hom}_{\tilde{A}}(\tilde{X}_0,\tilde{X}_1) \longrightarrow P(\tilde{X}_0,X_s)[1-s-t]
\end{aligned}$$ for all $s,t \geq 0$, satisfying analogues of the $A_\infty$-module equation. We assume that the cohomology groups $H(P(\tilde{X},X),\mu_P^{0|1|0})$ are of finite total dimension, and also impose strict unitality properties. Bimodules form an $A_\infty$-category $(A,\tilde{A})^{\mathit{mod}}$ (just like in the case of modules, this has vanishing higher order compositions, hence could be considered as a dg category).
The standard example is the diagonal $A$-bimodule, $P(X,Y) = \mathit{hom}_A(X,Y)$ with $\mu^{s|1|t}_P = \mu^{s+1+t}_A$ (this also indirectly illustrates our sign conventions). We usually just write $P = A$. Another example is the one which appeared in : if $G: A \rightarrow \tilde{A}$ is an $A_\infty$-functor, one can define an $A$-bimodule $P(X,Y) = \mathit{hom}_{\tilde{A}}(GX,GY)$, with structure maps similar to the diagonal one but plugging in multiple copies of $G$. We again denote this by $P = \tilde{A}$, but always make sure to mention that the pullback to an $A$-bimodule is intended.
One traditional use of bimodules is as “kernels” defining “convolution functors” between categories of modules. The tensor product of an $A$-module $M$ and an $(A,\tilde{A})$-bimodule $P$ is an $\tilde{A}$-module $\tilde{M} = M \otimes_A P$, given by a bar construction $$\label{eq:ainfty-tensor}
\tilde{M}(\tilde{X}) = \bigoplus M(X_r) \otimes \mathit{hom}_A(X_{r-1},X_r)[1] \otimes \cdots \mathit{hom}_A(X_0,X_1)[1] \otimes P(\tilde{X},X_0),$$ where the sum is over all $r \geq 0$ and objects $(X_0,\dots,X_r)$. The induced differential is $$\begin{aligned}
& \mu^1_{\tilde{M}}(m \otimes a_r \otimes \cdots \otimes a_1 \otimes p) = \\
& \qquad \sum_i (-1)^{|p|+|a_1|+\cdots+|a_i|-i} \mu^{r-i+1}_M(m,a_r,\dots,a_{i+1}) \otimes a_i \cdots \otimes a_1 \otimes p \\ &
\qquad + \sum_{i,j} (-1)^{|p|+|a_1|+\cdots+|a_i|-i} m \otimes \cdots \otimes \mu_A^j(a_{i+j},\dots,a_{i+1}) \otimes a_i \otimes \cdots \otimes a_1 \otimes p \\ &
\qquad + \sum_i m \otimes \cdots \otimes a_{i+1} \otimes \mu_P^{i|1|0}(a_i,\dots,a_1,p).
\end{aligned}$$ There are similar formulae for $\mu^d_{\tilde{M}}$, $d > 1$, which are in fact simpler since they involve only $\mu_P^{*|1|d-1}$. Naturality of the tensor product with a fixed $P$ is expressed by a map $\mathit{hom}_{A^{\mathit{mod}}}(M_0,M_1) \rightarrow \mathit{hom}_{\tilde{A}^{\mathit{mod}}}(\tilde{M}_0,\tilde{M}_1)$ for $\tilde{M}_k = M_k \otimes_A P$. This takes $b$ to $b \otimes_A e_P$, given by $$\begin{aligned}
& (b \otimes_A e_P)^1(m_0 \otimes a_r \cdots \otimes a_1 \otimes p) = \\
& \qquad \qquad \qquad \sum_i (-1)^{|p|+|a_1|+\cdots+|a_i|-i} b^{r-i+1}(m_0,a_r,\dots,a_{i+1}) \otimes a_i \otimes \cdots \otimes p,
\end{aligned}$$ with vanishing higher order terms. The resulting convolution $A_\infty$-functor (in fact a dg functor, since it has no higher order terms) between module categories will be denoted by $K_P$. So far, we have skirted the issue of whether $M \otimes_A P$ is really an object of $A^{\mathit{mod}}$ as defined, meaning whether it satisfies the cohomological finiteness condition. This fails in general, but it will hold if $M$ is perfect (based on the fact that $X^{\mathit{yon}} \otimes_A P$ is quasi-isomorphic to the $\tilde{A}$-module $P(\cdot,X)$, which has finite cohomology by assumption on $P$). Hence, we always get a functor $K_P: A^{\mathit{perf}} \rightarrow \tilde{A}^{\mathit{mod}}$. If $P$ is itself [*right perfect*]{}, which means that $P(\cdot,X)$ is itself a perfect $\tilde{A}$-module for any $X$, then $K_P$ takes $A^{\mathit{perf}}$ to $\tilde{A}^{\mathit{perf}}$.
Let’s suppose, for the sake of concreteness, that $P$ is right perfect. There are natural chain maps $$\label{eq:one-sided-deformation}
\mathit{CC}(A,A) \longrightarrow \mathit{hom}_{(A,\tilde{A})^{\mathit{mod}}}(P,P) \longleftarrow
\mathit{CC}(\tilde{A},\tilde{A}).
$$ Informally, one can think of these as follows. Given an infinitesimal deformation of either $A$ or $\tilde{A}$, they measure the failure of $P$ to remain a bimodule with respect to the deformed structure. This may remind the reader of , and indeed one can define a map (represented by the dashed arrow below) which fits into a homotopy commutative diagram $$\label{eq:grrr}
\xymatrix{
\mathit{CC}(A,A) \ar[d]^{\Gamma^{\mathit{perf}}} \ar[r] & \mathit{hom}_{(A,\tilde{A})^{\mathit{mod}}}(P,P) \ar@{-->}[d] & \ar[l] \ar[d]^{\Gamma^{\mathit{perf}}} \mathit{CC}(\tilde{A},\tilde{A}) \\
\mathit{CC}(A^{\mathit{perf}},A^{\mathit{perf}}) \ar[r]^-{(K_P)_*} &
\mathit{CC}(A^{\mathit{perf}},\tilde{A}^{\mathit{perf}}) & \ar[l]_-{(K_P)^*}
\mathit{CC}(\tilde{A}^{\mathit{perf}},\tilde{A}^{\mathit{perf}}).
}$$
Connections
-----------
Fix a smooth affine algebraic curve ${\mathscr{S}}$ over $R$, with ${\mathscr{R}}= R[{\mathscr{S}}]$ its ring of functions. Recall that ${\mathscr{R}}$-modules correspond to quasi-coherent sheaves on ${\mathscr{S}}$; finitely generated modules to coherent sheaves; projective modules to vector bundles; and rank $1$ projective modules to line bundles. We will mostly use the algebraic language as it is more elementary, but the reader is encouraged to keep the geometric viewpoint in mind.
Denote by $\Omega^1_{\mathscr{R}}$ the module of K[ä]{}hler differentials, which is a rank $1$ projective module, and comes with its canonical derivation $d: {\mathscr{R}}\rightarrow \Omega^1_{\mathscr{R}}$. A connection on an ${\mathscr{R}}$-module ${\mathscr{F}}$ is a map $\nabla_{\mathscr{F}}: {\mathscr{F}}\rightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{F}}$ satisfying the Leibniz identity with respect to $d$. An equivalent viewpoint is as follows. For any ${\mathscr{F}}$ there is a canonical short exact sequence of modules $$\label{eq:pullback-sequence}
0 \rightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{F}}\longrightarrow J^1({\mathscr{F}}) \longrightarrow {\mathscr{F}}\rightarrow 0,$$ where ${\mathscr{J}}^1({\mathscr{F}})$ is the one-jet module [@atiyah57]. Connections correspond to splittings of this sequence. ${\mathscr{F}}$ admits a connection if and only if its Atiyah class $\mathit{At}({\mathscr{F}}) \in \mathit{Ext}^1_{\mathscr{R}}({\mathscr{F}}, \Omega^1_{\mathscr{R}}\otimes {\mathscr{F}})$, which is the extension class of , vanishes. In particular, projective modules always admit connections (this can also be proved directly). In the other direction, one has [@bernstein-dmodules Lecture 2]:
\[th:locally-free\] Let ${\mathscr{F}}$ be a finitely generated ${\mathscr{R}}$-module which admits a connection. Then it is necessarily projective.
Let $R({\mathscr{S}})$ be the field of rational functions on our curve, which is the quotient field of ${\mathscr{R}}$. Tautologically, all projective modules of the same rank become isomorphic after tensoring with $R({\mathscr{S}})$. However, it is intuitively clear that the situation for modules with connections is quite different, which is indeed confirmed by:
\[th:rational-section\] Let ${\mathscr{F}}$ be a finitely generated ${\mathscr{R}}$-module with a connection. Then, if $f \in {\mathscr{F}}\otimes_{\mathscr{R}}R({\mathscr{S}})$ is a rational solution of $\nabla_{\mathscr{F}}(f) = 0$, it automatically lies in ${\mathscr{F}}$ itself.
Suppose that $f \neq 0$, and choose a point on our curve. Take a function $r \in {\mathscr{R}}$ which vanishes (to order $1$) at this point. There is a unique $m \in {\mathbb{Z}}$ such that $r^m f$ takes a nonzero finite value at our point. By assumption $$\nabla_{\mathscr{F}}( r^m f) = m (r^{-1} dr) r^m f,$$ but while the left hand side is regular, the right hand one has a pole unless $m = 0$.
One can extend the notion of connection to the derived category (see for instance [@markarian09; @huybrechts-thomas10]), as follows. Any complex of modules ${\mathscr{F}}$ sits in a short exact sequence generalizing . Define a [*homotopy connection*]{} to be a splitting ${\mathscr{F}}\rightarrow J^1({\mathscr{F}})$ in the derived category. If this exists, it induces connections on all the cohomology modules ${\mathscr{H}}^i({\mathscr{F}})$. The obstruction to the existence of a homotopy connection is the morphism completing the short exact sequence to an exact triangle in the derived category, which we again call the Atiyah class $\mathit{At}({\mathscr{F}}) \in \mathit{Hom}_{D({\mathscr{R}})}({\mathscr{F}}, \Omega^1_{\mathscr{R}}\otimes {\mathscr{F}}[1])$. To get a more hands-on description, assume that each of the modules ${\mathscr{F}}^i$ forming our complex already comes equipped with a connection. We call this a [*pre-connection*]{} on ${\mathscr{F}}$, and denote it by ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}}$. Its failure to commute with the differential $d_{\mathscr{F}}$ gives rise to a chain map $$\label{eq:cochain-atiyah}
\begin{aligned}
& \mathit{at}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}}): {\mathscr{F}}\longrightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{F}}[1], \\
& \mathit{at}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}})({\mathscr{f}}) = (\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes d_{\mathscr{F}})({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{F}}} {\mathscr{f}}) - {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{F}}}(d_{\mathscr{F}}{\mathscr{f}}),
\end{aligned}$$ which is the boundary homomorphism for , hence represents $\mathit{At}({\mathscr{F}})$. We call $\mathit{at}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}})$ the [*Atiyah cocycle*]{} of ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}}$. Suppose that ${\mathscr{F}}_0,{\mathscr{F}}_1$ are complexes of modules, with ${\mathscr{F}}_0$ consisting of projective modules. Then we have a short exact sequence [@dold Theorem VI.10.11] $$\label{eq:dold}
0 \rightarrow \mathit{Ext}^1(H({\mathscr{F}}_0),H({\mathscr{F}}_1))[-1] \longrightarrow H(\mathit{hom}({\mathscr{F}}_0,{\mathscr{F}}_1)) \longrightarrow \mathit{Hom}(H({\mathscr{F}}_0),H({\mathscr{F}}_1)) \rightarrow 0.$$ In particular:
\[th:proj\] If ${\mathscr{F}}_0$ is a complex of projective ${\mathscr{R}}$-modules, and ${\mathscr{F}}_1$ an acyclic complex of ${\mathscr{R}}$-modules, then $\mathit{hom}({\mathscr{F}}_0,{\mathscr{F}}_1)$ is again acyclic.
In the terminology of [@spaltenstein88], this means that (unbounded) complexes of projective ${\mathscr{R}}$-modules are $K$-projective.
\[th:homotopy-idempotent\] Let ${\mathscr{F}}$ be a complex of projective ${\mathscr{R}}$-modules, and $c: {\mathscr{F}}\rightarrow {\mathscr{F}}$ a chain map which is chain homotopic to its square, and which induces an isomorphism on cohomology. Then $c$ is chain homotopic the identity.
Consider with ${\mathscr{F}}_0 = {\mathscr{F}}_1 = {\mathscr{F}}$. Left composition with $c$ induces isomorphisms on the left and right terms of that sequence, hence also on the middle one. Since the identity and $c$ become homotopic after composition with $c$, they must have been homotopic in the first place.
Lemma \[th:proj\] has implications for homotopy connections, as follows. Take a complex of projective modules ${\mathscr{F}}$, and assume that $\mathit{At}({\mathscr{F}}) = 0$. Then the map $\mathit{at}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}})$ must be nullhomotopic, which means that one can modify the given pre-connection ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{F}}$ so that it becomes compatible with the differential. The result should then be properly called a connection on the complex ${\mathscr{F}}$, and we reserve the notation $\nabla_{{\mathscr{F}}}$ for those.
\[th:geometric\] We want to briefly consider the extension of the theory to non-affine bases. Let ${\mathscr{S}}$ be a smooth quasi-projective curve over $R$, and $\Omega^1_{\mathscr{S}}$ the line bundle of differentials. For any quasi-coherent sheaf ${\mathscr{F}}$ we have an analogue of , which can be used to define connections and Atiyah classes $\mathit{At}({\mathscr{F}}) \in \mathit{Ext}^1_{{\mathscr{S}}}({\mathscr{F}}, \Omega^1_{\mathscr{S}}\otimes {\mathscr{F}})$. The same holds for complexes, except that projective resolutions do not exist, and need to be replaced by injective quasi-coherent ones, which do. The analogue of Lemma \[th:proj\] says the following: if ${\mathscr{F}}_1$ is a complex of injective quasi-coherent sheaves, and ${\mathscr{F}}_0$ an acyclic complex of quasi-coherent sheaves, the complex of vector spaces $\mathit{hom}({\mathscr{F}}_0,{\mathscr{F}}_1)$ is again acyclic.
Let ${\mathscr{F}}_0,{\mathscr{F}}_1$ be complexes of injective quasi-coherent sheaves, and $\mathscr{hom}({\mathscr{F}}_0,{\mathscr{F}}_1)$ the complex of hom sheaves. By choosing pre-connections as in , one sees that $\mathit{At}(\mathscr{hom}({\mathscr{F}}_0, {\mathscr{F}}_1))$ is the difference between left multiplication with $\mathit{At}({\mathscr{F}}_1)$ and right multiplication with $\mathit{At}({\mathscr{F}}_0)$ (compare [@markarian09], which gives a similar formula for the Atiyah class of a tensor product). Now fix some $\gamma \in H^1({\mathscr{S}}, \Omega^1_{\mathscr{S}})$, and restrict attention to complexes ${\mathscr{F}}$ with bounded coherent cohomology, and such that $$\label{eq:central}
\mathit{At}({\mathscr{F}}) = \gamma \otimes \mathit{id}_{\mathscr{F}}.$$ For any two such complexes, $\mathit{At}(\mathscr{hom}({\mathscr{F}}_0,{\mathscr{F}}_1)) = 0$. By applying Lemma \[th:locally-free\], it then follows that the cohomology sheaves ${\mathscr{H}}(\mathscr{hom}({\mathscr{F}}_0,{\mathscr{F}}_1))$ are vector bundles. This is a simple illustration of the ideas that will play an important role later on (starting with Lemma \[th:its-projective\]).
\[th:higher-dimensions\] All we have said so far generalizes to higher-dimensional smooth varieties. The higher-dimensional analogue of Lemma \[th:locally-free\] can be derived from the case of curves, which is indeed what happens in [@bernstein-dmodules]. The generalization of Lemma \[th:proj\] to higher-dimensional affine varieties is [@dold60 Satz 3.1] (there is a spectral sequence which replaces , and which can be used to generalize Lemma \[th:homotopy-idempotent\]). For injective quasi-coherent sheaves on affine quasi-projective varieties, one has [@krause05 Example 3.10]. However, we have no real use for higher-dimensional bases in the present paper.
Families of objects\[subsec:families\]
--------------------------------------
Take an $A_\infty$-category $A$ as before, and denote by ${\mathscr{A}}$ the constant family of $A_\infty$-categories over ${\mathscr{S}}$ with fibre $A$. This has the same objects as $A$, and its morphisms and $A_\infty$-structure are obtained by extending constants to ${\mathscr{R}}$ in the obvious way: $$\mathit{hom}_{\mathscr{A}}(X_0,X_1) = {\mathscr{R}}\otimes_R hom_A(X_0,X_1).$$
Objects of ${\mathscr{A}}$ can be thought of as constant families. To get more interesting ones, we again have to introduce formal enlargements. First, there is an additive enlargement ${\mathscr{A}}^{\oplus}$, whose objects are finite formal sums $$\label{eq:xx}
{\mathscr{X}}= \bigoplus_{i \in I} {\mathscr{F}}^i \otimes X^i[-\sigma^i]$$ where the ${\mathscr{F}}^i$ are finitely generated projective ${\mathscr{R}}$-modules, the $X^i$ are objects of $A$, and the $\sigma^i$ integers. The $A_\infty$-structure is extended to such sums exactly as for $A^{\oplus}$. One then defines a [*family of twisted complexes*]{} to be a pair $({\mathscr{X}},\delta_{\mathscr{X}})$ where ${\mathscr{X}}\in \mathit{Ob}\, {\mathscr{A}}^{\oplus}$, and $\delta_{\mathscr{X}}\in \mathit{hom}_{{\mathscr{A}}^{\oplus}}^1({\mathscr{X}},{\mathscr{X}})$ is strictly decreasing with respect to some filtration of , and satisfies the analogue of . The $A_\infty$-category ${\mathscr{A}}^{\mathit{tw}}$ of twisted complexes obtained in this way allows the operations of shifts, mapping cones, and tensoring with a finitely generated projective ${\mathscr{R}}$-module.
Let ${\mathscr{C}}$ be the dg category of complexes of projective ${\mathscr{R}}$-modules with bounded finitely generated cohomology. A [*family of $A$-modules with finite cohomology*]{} is an $A_\infty$-functor $A^{\mathit{opp}} \rightarrow {\mathscr{C}}$. Concretely, such a family is given by a complex ${\mathscr{M}}(X)$ of ${\mathscr{R}}$-modules for each $X \in \mathit{Ob}(A)$, with structure maps as in but which are ${\mathscr{R}}$-linear, hence extend to $$\mu^d_{\mathscr{M}}: {\mathscr{M}}(X_d) \otimes \mathit{hom}_{\mathscr{A}}(X_{d-1},X) \otimes \cdots \otimes \mathit{hom}_{\mathscr{A}}(X_0,X_1) \longrightarrow {\mathscr{M}}(X_0)[2-d].$$ We impose the same strict unitality conditions as before. Such modules form an $A_\infty$-category over ${\mathscr{R}}$, denoted by ${\mathscr{A}}^{\mathit{mod}}$. The following statement is well-known in the case of $A_\infty$-categories over a field, see for instance [@keller99 Section 4], but slightly less so in the current framework:
\[th:acyclic-modules\] If the chain complexes $({\mathscr{M}}(X),\mu^1_{\mathscr{M}})$ are acyclic for all $X$, ${\mathscr{M}}$ is quasi-isomorphic to zero in ${\mathscr{A}}^{\mathit{mod}}$.
The length filtration of $\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}},{\mathscr{M}})$ gives rise to a spectral sequence, whose starting page is $$\label{eq:length-sequence}
E_1^{p\bullet} = \prod_{X_0,\dots,X_p} H(\mathit{Hom}({\mathscr{M}}(X_p) \otimes \mathit{hom}_A(X_{p-1},X_p) \otimes \cdots \otimes \mathit{hom}_A(X_0,X_1),{\mathscr{M}}(X_0))).$$ Even though that spectral sequence does not converge in general, one can apply comparison and vanishing arguments to it. Since ${\mathscr{M}}(X_p) \otimes \mathit{hom}_A(X_{p-1},X_d) \otimes \cdots \otimes \mathit{hom}_A(X_0,X_1)$ is a complex of projective ${\mathscr{R}}$-modules, and ${\mathscr{M}}(X_0)$ is acyclic, it follows from Lemma \[th:proj\] that the $E_1$ page vanishes.
One has a Yoneda functor ${\mathscr{A}}\rightarrow {\mathscr{A}}^{\mathit{mod}}$ as well as its extension ${\mathscr{A}}^{\mathit{tw}} \rightarrow {\mathscr{A}}^{\mathit{mod}}$, which are cohomologically full and faithful for the same reason as before. Moreover, given ${\mathscr{Y}}\in \mathit{Ob}\,{\mathscr{A}}^{\mathit{tw}}$ and an idempotent endomorphism on the cohomology level, one can find an object of ${\mathscr{A}}^{\mathit{mod}}$ representing the associated direct summand of the Yoneda image ${\mathscr{Y}}^{\mathit{yon}}$. To see that, one goes through the construction in Section \[subsec:splittings\], which defines a homotopy idempotent ${\mathscr{p}}$ over ${\mathscr{R}}$ as well as an associated family of modules ${\mathscr{M}}= ({\mathscr{Y}},{\mathscr{p}})^{\mathit{yon}}$. A noteworthy technical point is that ${\mathscr{M}}(X) = \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}(X,{\mathscr{Y}})[q]$ is still a complex of projective ${\mathscr{R}}$-modules for any $X$, and has finitely generated cohomology since $H({\mathscr{M}}(X))$ is a direct summand of $H(\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}(X,{\mathscr{Y}}))$. One then defines the full subcategory of [*perfect families of modules*]{}, ${\mathscr{A}}^{\mathit{perf}} \subset {\mathscr{A}}^{\mathit{mod}}$, to consist of all objects quasi-isomorphic to direct summands of families of twisted complexes. This category is proper, in the sense that $\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1)$ is a chain homotopy retract of a bounded complex of finitely generated projective ${\mathscr{R}}$-modules. In particular, the cohomology $H(\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1))$ is itself bounded and finitely generated over ${\mathscr{R}}$ in each degree.
For any $R$-point of ${\mathscr{S}}$, with associated map ${\mathscr{R}}\rightarrow R$, we can define restriction functors $$\begin{aligned}
& {\mathscr{A}}^{\mathit{tw}} \longrightarrow A^{\mathit{tw}}, \\
& {\mathscr{A}}^{\mathit{mod}} \longrightarrow A^{\mathit{mod}}, \\
& {\mathscr{A}}^{\mathit{perf}} \longrightarrow A^{\mathit{perf}}.
\end{aligned}$$ The first one takes an ${\mathscr{X}}$ as in and passes to the fibres $F^i = R \otimes_{\mathscr{R}}{\mathscr{F}}^i$ to get an ordinary twisted complex. The second one is similarly given by $M(X) = R \otimes_{\mathscr{R}}{\mathscr{M}}(X)$. In either case, the morphism spaces again get specialized to the given point, which is unproblematic from a homological algebra viewpoint since they consist of projective ${\mathscr{R}}$-modules for ${\mathscr{A}}^{\mathit{tw}}$, and at least of flat ${\mathscr{R}}$-modules for ${\mathscr{A}}^{\mathit{mod}}$ (see for instance [@cartan-eilenberg p. 122, Exercise 4]). Clearly, these two functors are compatible with Yoneda embeddings, which ensures that the third functor is well-defined.
Twisted complexes with connections\[subsec:twisted-connections\]
----------------------------------------------------------------
Let ${\mathscr{X}}$ be an object of ${\mathscr{A}}^{\mathit{tw}}$, written as in . A [*pre-connection on ${\mathscr{X}}$*]{} is a pair ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}= (\{\nabla_{{\mathscr{F}}^i}\},\alpha_{\mathscr{X}})$ consisting of an ordinary connection $\nabla_{{\mathscr{F}}^i}$ on each ${\mathscr{R}}$-module ${\mathscr{F}}^i$, together with an element $\alpha_{\mathscr{X}}\in \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}^0({\mathscr{X}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{X}}) = \Omega^1_{\mathscr{R}}\otimes \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}^0({\mathscr{X}},{\mathscr{X}})$. This becomes more meaningful if one writes it as a formal expression $$\label{eq:pre-connection-on-tw}
{\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}= \bigoplus_i \nabla_{{\mathscr{F}}^i} \otimes e_{X^i[-\sigma^i]} + \alpha_{\mathscr{X}}$$ (recall that due to sign conventions, the identity for the shifted object $X^i[-\sigma^i]$ is $e_{X^i[-\sigma^i]} = (-1)^{\sigma^i} e_{X^i}$). There is some redundancy in this description: given elements ${\mathscr{f}}_i \in \mathit{Hom}({\mathscr{F}}_i,\Omega^1_{\mathscr{R}}\otimes {\mathscr{F}}_i)$, one can change $\nabla_{{\mathscr{F}}^i} \rightarrow \nabla_{{\mathscr{F}}^i} + {\mathscr{f}}_i$, and simultaneously $\alpha_{\mathscr{X}}\rightarrow \alpha_{{\mathscr{X}}} - \bigoplus_i {\mathscr{f}}_i \otimes e_{X^i[-\sigma^i]}$, and the result is still considered to be the same pre-connection. With that in mind, pre-connections form an affine space over $\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}^0({\mathscr{X}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{X}})$.
Now we include the differential $\delta_{\mathscr{X}}$ in our discussion. Its compatibility with a pre-connection is measured by the [*deformation cocycle*]{} $$\label{eq:deformation-cocycle}
\begin{aligned}
& \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}) \in \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}^1({\mathscr{X}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{X}}), \\
& \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}) = -\Big( \textstyle\bigoplus_{i,j} \nabla_{\mathit{Hom}({\mathscr{F}}^i,{\mathscr{F}}^j)} \otimes \mathit{id}_{\mathit{hom}_A(X^i[-\sigma^i],X^j[-\sigma^j])} \Big) (\delta_{\mathscr{X}}) + \mu^1_{{\mathscr{A}}^{\mathit{tw}}}(\alpha_{\mathscr{X}}).
\end{aligned}$$ Here, $\nabla_{\mathit{Hom}({\mathscr{F}}^i,{\mathscr{F}}^j)}$ is the connection induced by $\nabla_{{\mathscr{F}}^i}$ and $\nabla_{{\mathscr{F}}^j}$. The formula becomes clearer if one thinks in terms of components $$\begin{aligned}
& \alpha_{{\mathscr{X}}}^{ji} \in \mathit{Hom}({\mathscr{F}}^i,\Omega^1_{\mathscr{R}}\otimes {\mathscr{F}}^j) \otimes \mathit{hom}_A(X^i,X^j)[\sigma^i-\sigma^j], \\
& \delta_{{\mathscr{X}}}^{ji} \in \mathit{Hom}({\mathscr{F}}^i,{\mathscr{F}}^j) \otimes \mathit{hom}_A(X^i,X^j)[\sigma^i-\sigma^j]
\end{aligned}$$ Then, the components of the deformation cocycle are $$\label{eq:def-components}
\begin{aligned}
\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}})^{ji} = & -(\nabla_{\mathit{Hom}({\mathscr{F}}^i,{\mathscr{F}}^j)} \otimes \mathit{id})(\delta_{\mathscr{X}}^{ji}) + (-1)^{\sigma^i} \mu^1_{{\mathscr{A}}}(\alpha_{\mathscr{X}}^{ji}) \\
& + (-1)^{\sigma^i} \sum_k \mu^2_{{\mathscr{A}}}(\alpha_{\mathscr{X}}^{jk},\delta_{\mathscr{X}}^{ki}) + \mu^2_{{\mathscr{A}}}(\delta_{\mathscr{X}}^{jk},\alpha_{\mathscr{X}}^{ki}) + \cdots
\end{aligned}$$
\[th:def-cocycle\] $\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}})$ is a cocycle, whose cohomology class $\mathit{Def}({\mathscr{X}})$ (the [*deformation class*]{} of ${\mathscr{X}}$) is independent of the choice of pre-connection.
There is a simple trick which facilitates these computations, namely to temporarily forget that morphisms are supposed to be ${\mathscr{R}}$-linear. In this weakened sense, one can consider ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}$ itself as being a morphism from ${\mathscr{X}}$ to $\Omega^1_{\mathscr{R}}\otimes {\mathscr{X}}$, and then the formula actually represents $$\mu^2_{{\mathscr{A}}^{\oplus}}(\delta_{\Omega^1_{\mathscr{R}}\otimes {\mathscr{X}}},{\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}-\alpha_{\mathscr{X}}) + \mu^2_{{\mathscr{A}}^{\oplus}}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}-\alpha_{\mathscr{X}},\delta_{\mathscr{X}}) + \mu^1_{{\mathscr{A}}^{\mathit{tw}}}(\alpha_{\mathscr{X}}) = \mu^1_{{\mathscr{A}}^{\mathit{tw}}}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}})$$ (this takes into accounts cancellations which arise from the fact that the components of ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}- \alpha_{\mathscr{X}}$ are multiples of the identity $e_{X^i[-\sigma^i]}$). The desired statements now follow directly.
Take two families of twisted complexes ${\mathscr{X}}_k = \bigoplus_{i \in I_k} {\mathscr{F}}^i_k \otimes X^i_k[-\sigma^i_k]$ ($k = 0,1$) equipped with pre-connections ${\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}_k}$, written in the analogous way. This induces a pre-connection on the chain complex $\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)$, namely $$\label{eq:hom-pre-connection}
\begin{aligned}
{\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}}) = & \bigoplus_{i,j} \Big(\nabla_{\mathit{Hom}({\mathscr{F}}^i_0,{\mathscr{F}}^j_1)} \otimes \mathit{id}_{\mathit{hom}_A(X^i_0,X^j_1)} \Big)({\mathscr{a}}) \\ & + (-1)^{|{\mathscr{a}}|} \mu^2_{{\mathscr{A}}^{\mathit{tw}}}(\alpha_{{\mathscr{X}}^1},{\mathscr{a}}) - \mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}},\alpha_{{\mathscr{X}}^0}).
\end{aligned}$$ By a computation similar to the one in Lemma \[th:def-cocycle\], this gives a formula for the Atiyah cocycle of the chain complex $\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)$: $$\label{eq:atiyah-def}
\begin{aligned}
& \mathit{at}({\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)})({\mathscr{a}}) =
\mu^1_{{\mathscr{A}}^{\mathit{tw}}}({\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}})) - {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}(\mu_{{\mathscr{A}}^{\mathit{tw}}}^1({\mathscr{a}})) \\ & \qquad \qquad
= \mu^2_{{\mathscr{A}}^{\mathit{tw}}}(\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}^1}),{\mathscr{a}}) + \mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}},\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}^0})).
\end{aligned}$$ From here on, the obvious development would be the following one. Define a [*connection*]{} on ${\mathscr{X}}$ to be a pre-connection for which vanishes. If ${\mathscr{X}}_0$ and ${\mathscr{X}}_1$ carry connections, implies that $\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)$ carries a connection. However, families with vanishing deformation class are close to constant ones and therefore not terribly interesting.
Instead, we want to introduce a relative version, as follows. Let $\mathit{CC}({\mathscr{A}}, {\mathscr{A}})$ be the (reduced) Hochschild complex of the constant family ${\mathscr{A}}$, and $HH({\mathscr{A}},{\mathscr{A}})$ its cohomology. More relevant for us is a twisted version, with coefficients in the bimodule $\Omega^1_{\mathscr{R}}\otimes {\mathscr{A}}$. The effect of twisting is fairly trivial, both on the cochain and cohomology level: $$\begin{aligned}
& \mathit{CC}({\mathscr{A}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{A}}) {\cong}\Omega^1_{\mathscr{R}}\otimes \mathit{CC}({\mathscr{A}},{\mathscr{A}}), \\
& \mathit{HH}({\mathscr{A}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{A}}) {\cong}\Omega^1_{\mathscr{R}}\otimes \mathit{HH}({\mathscr{A}},{\mathscr{A}}).
\end{aligned}$$ As before, there is a chain map $\Gamma^{\mathit{tw}}$ between the twisted Hochschild chain complex and its analogue for ${\mathscr{A}}^{\mathit{tw}}$.
\[th:deformation-field\] A cohomology class $[\gamma] \in \mathit{HH}^1({\mathscr{A}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{A}})$ is called a [*deformation field*]{}. Let ${\mathscr{X}}$ be a family of twisted complexes. We say that ${\mathscr{X}}$ [*follows the deformation field*]{} if the image $\gamma^{\mathit{tw}} = \Gamma^{\mathit{tw}}(\gamma)$ satisfies $$\mathit{Def}({\mathscr{X}}) = [\gamma^{\mathit{tw},0}] \in \mathit{H}^1(\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{X}})).$$
In almost every situation where we use deformation fields, a choice of cocycle representative $\gamma$ is assumed to have been made (the resulting theory is always independent of that choice up to quasi-isomorphism). If ${\mathscr{X}}$ follows $[\gamma]$, we can choose a pre-connection whose deformation cocycle is exactly $\gamma^{\mathit{tw,0}}$. Call these [*relative connections*]{}, and denote them by $\nabla_{\mathscr{X}}$. Given two objects ${\mathscr{X}}_k$ with relative connections, we can introduce a modified version of , namely $$\label{eq:hom-connection}
\begin{aligned}
\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}}) = {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}}) + \gamma^{\mathit{tw},1}({\mathscr{a}}).
\end{aligned}$$ This is an actual connection, since the cocycle equation for $\gamma^{\mathit{tw}}$ says that $$\mu^1_{{\mathscr{A}}^{\mathit{tw}}}(\gamma^{\mathit{tw},1}({\mathscr{a}})) -
\gamma^{\mathit{tw},1}(\mu^1_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}})) =
- \mu^2_{{\mathscr{A}}^{\mathit{tw}}}(\gamma^{\mathit{tw},0},{\mathscr{a}}) -
\mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}},\gamma^{\mathit{tw},0}),$$ which one can add to to get the desired property.
We would like to study the behaviour of under composition of morphisms in ${\mathscr{A}}^{\mathit{tw}}$. Let’s temporarily forget about Hochschild cohomology and just assume that we have families of twisted complexes ${\mathscr{X}}_k$ ($k = 0,1,2$) equipped with pre-connections. Along the same lines as in one finds that for any $\mu^1_{{\mathscr{A}}^{\mathit{tw}}}$-cocycles ${\mathscr{a}}_k \in \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_{k-1},{\mathscr{X}}_k)$ ($k=1,2$), $$\begin{aligned}
& - {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_2)}(\mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}}_2,{\mathscr{a}}_1)) \\ & \qquad \qquad +
\mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_1,{\mathscr{X}}_2)}({\mathscr{a}}_2),{\mathscr{a}}_1)
+ \mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}}_2,{\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}}_1)) \\ &
= \mu^3_{{\mathscr{A}}^{\mathit{tw}}}(\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}_2}),{\mathscr{a}}_2,{\mathscr{a}}_1) +
\mu^3_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}}_2,\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}_1}),{\mathscr{a}}_1) +
\mu^3_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}}_2,{\mathscr{a}}_1,\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}_0})) \\ & \qquad \qquad + \text{\it (coboundary)}.
\end{aligned}$$ Assuming that the ${\mathscr{X}}_k$ follow $[\gamma]$ and come with relative connections, one adds correction terms and gets $$\label{eq:added-product}
\begin{aligned}
& -\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_2)}(\mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}}_2,{\mathscr{a}}_1)) \\ & \qquad \qquad +
\mu^2_{{\mathscr{A}}^{\mathit{tw}}}(\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_1,{\mathscr{X}}_2)}({\mathscr{a}}_2),{\mathscr{a}}_1)
+ \mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}}_2,\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}}_1)) \\ &
= -\mu^1_{{\mathscr{A}}^{\mathit{tw}}}(\gamma^{\mathit{tw},2}({\mathscr{a}}_2,{\mathscr{a}}_1)) + \text{\it (same coboundary as before)},
\end{aligned}$$ which means that the cohomology level connections act as derivations with respect to the product.
We have to confess that the framework introduced above, even though natural and accessible, is not satisfactory, for two reasons. The first (technical) reason is that the definition of pre-connection makes sense only under the assumption of strict unitality, which we have imposed so far but will want to relax eventually. The second (conceptual) reason is that families of twisted complexes are far too restrictive to be useful in general – as one can see by observing that if ${\mathscr{X}}$ is such a family, its fibre at any point of ${\mathscr{S}}$ represents the same class in the $K$-theory of $A^{\mathit{tw}}$. With this in mind, we will now switch to $A_\infty$-modules and carry out the corresponding developments there.
Modules with connections
------------------------
Let ${\mathscr{M}}$ be a family of $A$-modules. A [*pre-connection*]{} ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}$ is a sequence of maps $$\label{eq:pre-connection-components}
\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^1: {\mathscr{M}}(X_0) \longrightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{M}}(X_0), \\
& {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^2: {\mathscr{M}}(X_1) \otimes \mathit{hom}_A(X_0,X_1) \longrightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{M}}(X_0)[-1], \\
& {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^3: {\mathscr{M}}(X_2) \otimes \mathit{hom}_A(X_1,X_2) \otimes \mathit{hom}_A(X_0,X_1) \longrightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{M}}(X_0)[-2], \\
& \dots
\end{aligned}$$ where: the maps ${\mathscr{m}}\mapsto (-1)^{|{\mathscr{m}}|}{\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^1({\mathscr{m}})$ are connections in the standard sense; and the higher order terms ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^d$, $d>1$, are ${\mathscr{R}}$-linear. Clearly, pre-connections form an affine space over $\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}^0({\mathscr{M}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{M}}) {\cong}\Omega^1_{\mathscr{R}}\otimes \mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}^0({\mathscr{M}},{\mathscr{M}})$. The [*deformation cocycle*]{} $$\label{eq:module-cocycle}
\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}) \in \mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}^1({\mathscr{M}}, \Omega^1_{\mathscr{R}}\otimes {\mathscr{M}})$$ of a pre-connection is obtained by applying $\mu^1_{\mathit{{\mathscr{A}}}^{\mathit{mod}}}$ to . This makes sense every though ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^1$ is not ${\mathscr{R}}$-linear. The same observation shows that:
$\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}})$ is closed, and its cohomology class $\mathit{Def}({\mathscr{M}})$ is independent of the choice of pre-connection.
Suppose that our pre-connection has vanishing higher order terms ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^d = 0$, $d>1$, hence is just given by a family of connections $\nabla_{{\mathscr{M}}(X)}({\mathscr{m}}) = (-1)^{|{\mathscr{m}}|} {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^1({\mathscr{m}})$ on the graded ${\mathscr{R}}$-modules ${\mathscr{M}}(X)$. Then the deformation cocycle is given by $$\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}})^d({\mathscr{m}},a_{d-1},\dots,a_1) = (\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes \mu^d_{\mathscr{M}})({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}(X_d)}({\mathscr{m}}),\dots,a_1) - {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}(X_0)}(\mu^d_{\mathscr{M}}({\mathscr{m}},a_d,\dots,a_1))$$ for ${\mathscr{m}}\in {\mathscr{M}}(X_d)$, $a_k \in \mathit{hom}_A(X_{k-1},X_k)$. This is just the covariant derivative of the module structure of ${\mathscr{M}}$, measuring its failure to be compatible with the connections.
Given two families of modules ${\mathscr{M}}_k$ ($k = 0,1$) equipped with pre-connections ${\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_k}$, consider the map $$\label{eq:hom-pre-connection-2}
\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}: \mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1) \longrightarrow \Omega^1_{\mathscr{R}}\otimes \mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1), \\
& {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}({\mathscr{b}}) = (-1)^{|{\mathscr{b}}|} \mu^2_{\mathit{{\mathscr{A}}}^{\mathit{mod}}}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_1},{\mathscr{b}}) - \mu^2_{\mathit{{\mathscr{A}}}^{\mathit{mod}}}(\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes {\mathscr{b}},{\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_0}).
\end{aligned}$$ Spelled out, this means that ${\mathscr{c}}= {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}({\mathscr{b}})$ is given by $$\begin{aligned}
& {\mathscr{c}}^1({\mathscr{m}}) = (-1)^{|{\mathscr{b}}^1({\mathscr{m}})|} {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_1}^1({\mathscr{b}}^1({\mathscr{m}})) - (-1)^{|{\mathscr{m}}|} (\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes {\mathscr{b}}^1)({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_0}^1({\mathscr{m}})), \\
& {\mathscr{c}}^2({\mathscr{m}},a) = (-1)^{|{\mathscr{b}}^2({\mathscr{m}},a)|} {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_1}^1({\mathscr{b}}^2({\mathscr{m}},a)) + (-1)^{|{\mathscr{b}}^1({\mathscr{m}})|} {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_1}^2({\mathscr{b}}^1({\mathscr{m}}),a) \\ & \qquad \qquad - (-1)^{|{\mathscr{m}}|+|a|-1}
(\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes {\mathscr{b}}^1)({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_0}^2({\mathscr{m}},a)) - (-1)^{|{\mathscr{m}}|}
(\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes {\mathscr{b}}^2)({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_0}^1({\mathscr{m}}),a), \\
& \dots \\
& {\mathscr{c}}^d({\mathscr{m}},a_{d-1},\dots,a_1) = (-1)^{|{\mathscr{b}}^d({\mathscr{m}},a_{d-1},\dots,a_1)|} {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_1}^1({\mathscr{b}}^d({\mathscr{m}},a_{d-1},\dots,a_1)) \\ & \qquad \qquad - (-1)^{|{\mathscr{m}}|} (\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes {\mathscr{b}}^d({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}_0}^1({\mathscr{m}}),a_{d-1},\dots,a_1) \\ & \qquad \qquad + \text{\it (terms involving the higher order parts of the pre-connections on ${\mathscr{M}}_0,{\mathscr{M}}_1$)}.
\end{aligned}$$ This shows that is a pre-connection on the chain complex of $\mathit{hom}$’s. It follows from the definition and the $A_\infty$-equations on ${\mathscr{A}}^{\mathit{mod}}$ that $$\label{eq:atiyah-def-2}
\begin{aligned}
& \mu^1_{{\mathscr{A}}^{\mathit{mod}}}({\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}({\mathscr{b}})) - {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}(\mu_{{\mathscr{A}}^{\mathit{mod}}}^1({\mathscr{b}})) = \\ & \qquad \qquad
\mu^2_{{\mathscr{A}}^{\mathit{mod}}}(\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}^1}),{\mathscr{b}}) + \mu^2_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{b}},\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{M}}^0})).
\end{aligned}$$
Given a deformation field represented by $\gamma \in \mathit{CC}^1({\mathscr{A}},\Omega^1_{{\mathscr{R}}} \otimes {\mathscr{A}})$, write $\gamma^{\mathit{mod}} = \Gamma^{\mathit{mod}}(\gamma)$. In parallel with Definition \[th:deformation-field\], we say that a family of modules ${\mathscr{M}}$ [*follows*]{} $[\gamma]$ if $$\mathit{Def}({\mathscr{M}}) = [\gamma^{\mathit{mod},0}] \in H^1(\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{M}})).$$ If this holds, one can equip ${\mathscr{M}}$ with a [*relative connection*]{} $\nabla_{\mathscr{M}}$, by which we again mean a pre-connection whose deformation cocycle is exactly $\gamma^{\mathit{mod},0}$. Given two modules equipped with relative connections, one can modify to get an actual connection on that chain complex, as in : $$\label{eq:hom-connection-2}
\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)} =
{\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)} + \gamma^{\mathit{mod},1}.$$ Moreover, by essentially the same computation as in , these connections satisfy $$\label{eq:multiplicative-connections}
\begin{aligned}
& -\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_2)}(\mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{b}}_2,{\mathscr{b}}_1)) \\ & \qquad \qquad +
\mu^2_{{\mathscr{A}}^{\mathit{tw}}}(\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_1,{\mathscr{X}}_2)}({\mathscr{b}}_2),{\mathscr{b}}_1)
+ \mu^2_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{b}}_2,\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{b}}_1)) \\ &
\qquad \qquad = \text{\it (coboundary)}
\end{aligned}$$ for any cocycles ${\mathscr{b}}_k \in \mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_{k-1},{\mathscr{M}}_k)$.
Unsurprisingly, all these notions are compatible with their counterparts from Section \[subsec:twisted-connections\] via the Yoneda embedding. If ${\mathscr{Y}}= \bigoplus_i {\mathscr{F}}^i \otimes Y^i[-\sigma^i]$ is a family of twisted complexes with a pre-connection ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{Y}}$ as in , then the module ${\mathscr{Y}}^{\mathit{yon}}$ inherits a pre-connection: $$\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{Y}}^{\mathit{yon}}}^1({\mathscr{a}}) = (-1)^{|{\mathscr{a}}|} (\textstyle\bigoplus_i \nabla_{{\mathscr{F}}^i} \otimes \mathit{id})({\mathscr{a}}) + \mu^2_{\mathscr{A}}(\alpha_{{\mathscr{Y}}},{\mathscr{a}}) \\
& \qquad \qquad \text{for $\textstyle{\mathscr{a}}\in {\mathscr{Y}}^{\mathit{yon}}(X) = \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}(X,{\mathscr{Y}}) = \bigoplus_i {\mathscr{F}}^i \otimes \mathit{hom}_A(X,Y^i)[-\sigma^i]$, and } \\
& {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{Y}}^{\mathit{yon}}}^d({\mathscr{a}}_d,a_{d-1},\dots,a_1) = \mu^{d+1}_{\mathscr{A}}(\alpha_{{\mathscr{Y}}},{\mathscr{a}}_d,a_{d-1},\dots,a_1) \\
& \qquad \qquad \text{where ${\mathscr{a}}_d \in \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}(X_d,{\mathscr{Y}})$, and $a_k \in \mathit{hom}_A(X_{k-1},X_k)$ for $k =1,\dots,d-1$.}
\end{aligned}$$ The deformation cocycle of ${\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{Y}}^{\mathit{yon}}}$ is the image of that of ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{Y}}$ under the Yoneda functor. One can take this comparison further to relative connections, but we will not need that.
We want to highlight one simple consequence:
\[th:its-projective\] Suppose that ${\mathscr{M}}_0,{\mathscr{M}}_1$ are families of modules (as always, with finite cohomology) following $[\gamma]$, and where ${\mathscr{M}}_0$ is perfect. Then $H(\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1))$ is a finitely generated graded projective ${\mathscr{R}}$-module.
Finite generation follows from the fact that ${\mathscr{M}}_0$ is perfect. On the other hand, the module carries a connection, hence Lemma \[th:locally-free\] applies.
By construction, relative connections on a given family ${\mathscr{M}}$ form an affine space over the space of cocycles inside $\mathit{hom}^0_{\mathit{{\mathscr{A}}}^{\mathit{perf}}}({\mathscr{M}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{M}})$. If we change the relative connections on ${\mathscr{M}}_k$ ($k =0,1$) to $\nabla_{{\mathscr{M}}_k}' = \nabla_{{\mathscr{M}}_k} + {\mathscr{c}}_k$, the induced connection on the morphism spaces changes by $$\label{eq:modify-connection}
\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}'({\mathscr{b}}) =
\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{mod}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}({\mathscr{b}}) + (-1)^{|{\mathscr{b}}|} \mu^2_{\mathit{{\mathscr{A}}}^{\mathit{mod}}}({\mathscr{c}}_1,{\mathscr{b}}) - \mu^2_{\mathit{{\mathscr{A}}}^{\mathit{mod}}}(\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes {\mathscr{b}},{\mathscr{c}}_0).$$ In particular, if we are only interested in the connection on the cohomology level, relative connections which differ by coboundaries yield the same result, so the space of relevant choices is an affine space over $\mathit{Hom}_{H^0(\mathit{{\mathscr{A}}}^{\mathit{perf}})}({\mathscr{M}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{M}})$.
If ${\mathscr{M}}$ is a family of modules with a pre-connection ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}$, and ${\mathscr{F}}$ a finitely generated projective ${\mathscr{R}}$-module with its own connection $\nabla_{\mathscr{F}}$, the tensor product ${\mathscr{F}}\otimes {\mathscr{M}}$ inherits a pre-connection, defined by $$\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{F}}\otimes {\mathscr{M}}}^1({\mathscr{f}}\otimes {\mathscr{m}}) = (-1)^{|{\mathscr{m}}|} (\nabla_{\mathscr{F}}{\mathscr{f}}) \otimes {\mathscr{m}}+ {\mathscr{f}}\otimes {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^1({\mathscr{m}}), \\
& {\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{F}}\otimes {\mathscr{M}}}^d({\mathscr{f}}\otimes {\mathscr{m}},a_{d-1},\dots,a_1) = {\mathscr{f}}\otimes {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}^d({\mathscr{m}},a_{d-1},\dots,a_1) \quad \text{for $d>1$.}
\end{aligned}$$ The associated deformation cocycle is $$\label{eq:tensor-deformation}
\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{F}}\otimes {\mathscr{M}}}) = \mathit{id}_{\mathscr{F}}\otimes \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}).$$ In particular, if ${\mathscr{M}}$ follows a given deformation field $[\gamma]$, then so does ${\mathscr{F}}\otimes {\mathscr{M}}$. One could generalize this slightly by allowing ${\mathscr{F}}$ to be a complex of projective ${\mathscr{R}}$-modules with a pre-connection, in which case the associated Atiyah cocycle would appear in an additional summand in .
Functoriality\[subsec:functor\]
-------------------------------
Let $G: A \rightarrow \tilde{A}$ be a (strictly unital) $A_\infty$-functor. We want to study the action of $G$ on families of objects, over a fixed base space ${\mathscr{S}}$. For expository reasons, we temporarily return to the framework of twisted complexes. It is well-known (to the man on the street) that $G$ induces an $A_\infty$-functor $G^{\mathit{tw}}: A^{\mathit{tw}} \rightarrow \tilde{A}^{\mathit{tw}}$. The same formulae applied to families define an $A_\infty$-functor ${\mathscr{G}}^{\mathit{tw}}: {\mathscr{A}}^{\mathit{tw}} \rightarrow \tilde{{\mathscr{A}}}^{\mathit{tw}}$.
A pre-connection on ${\mathscr{X}}\in \mathit{Ob}\,{\mathscr{A}}^{\mathit{tw}}$ induces one on its image $\tilde{{\mathscr{X}}} = {\mathscr{G}}^{\mathit{tw}}({\mathscr{X}})$: $$\label{eq:pre-connection-image}
\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{\tilde{\mathscr{X}}} = \bigoplus_i \nabla_{{\mathscr{F}}^i} \otimes e_{G(X^i)[-\sigma^i]} + {\mathscr{G}}^{\mathit{tw},1}(\alpha_{\mathscr{X}}), \\
& \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\tilde{\mathscr{X}}}) = {\mathscr{G}}^{\mathit{tw},1}(\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}})).
\end{aligned}$$ If ${\mathscr{X}}_k$ ($k = 0,1$) are families with pre-connections, and $\tilde{{\mathscr{X}}}_k$ their images under ${\mathscr{G}}^{\mathit{tw}}$ equipped with the induced pre-connections, then for any cocycle ${\mathscr{a}}$ we have $$\label{eq:functoriality-fails}
\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}(\tilde{{\mathscr{X}}}_0,\tilde{{\mathscr{X}}}_1)}({\mathscr{G}}^{\mathit{tw},1}({\mathscr{a}})) =
{\mathscr{G}}^{\mathit{tw},1}({\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}})) \\ & \qquad \qquad
- {\mathscr{G}}^{\mathit{tw},2}(\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}_1}),{\mathscr{a}})
- {\mathscr{G}}^{\mathit{tw},2}({\mathscr{a}},\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{{\mathscr{X}}_0})) + \text{\it (coboundary)}.
\end{aligned}$$
\[th:matching-deformations\] Let $[\gamma]$ and $[\tilde\gamma]$ be deformation fields for ${\mathscr{A}}$ and $\tilde{{\mathscr{A}}}$, respectively. Suppose that there is a $\beta \in \mathit{CC}^0({\mathscr{A}},\Omega^1_{\mathscr{R}}\otimes \tilde{\mathscr{A}})$ such that $$\label{eq:beta-diff}
\partial \beta= {\mathscr{G}}^*(\tilde\gamma) - {\mathscr{G}}_*(\gamma).$$
In the definition of $\beta$, we consider $\tilde{\mathscr{A}}$ as an ${\mathscr{A}}$-bimodule by pullback through ${\mathscr{G}}$. As in , $\beta$ induces a cochain $\beta^{\mathit{tw}} \in \mathit{CC}^0({\mathscr{A}}^{\mathit{tw}},\Omega^1_{\mathscr{R}}\otimes \tilde{{\mathscr{A}}}^{\mathit{tw}})$, which satisfies the analogue of . Concretely, this means that $$\label{eq:relative-field}
\begin{aligned}
& \mu^1_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}(\beta^{\mathit{tw},0}) = \tilde\gamma^{\mathit{tw},0} - {\mathscr{G}}^{\mathit{tw},1}(\gamma^{\mathit{tw},0}) \in \mathit{hom}_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}({\mathscr{G}}^{\mathit{tw}}({\mathscr{X}}_0),{\mathscr{G}}^{\mathit{tw}}({\mathscr{X}}_0)), \\
& \mu^1_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}(\beta^\mathit{tw,1}({\mathscr{a}})) + \mu^2_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}({\mathscr{G}}^{\mathit{tw},1}({\mathscr{a}}),\beta^{\mathit{tw},0}) +
(-1)^{|{\mathscr{a}}|-1} \mu^2_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}(\beta^{\mathit{tw},0},{\mathscr{G}}^{\mathit{tw},1}({\mathscr{a}})) \\ & \qquad \qquad +
\beta^{\mathit{tw},1}(\mu^1_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{a}})) = \tilde\gamma^{\mathit{tw},1}({\mathscr{G}}^{\mathit{tw},1}({\mathscr{a}})) - {\mathscr{G}}^{\mathit{tw},1}(\gamma^{\mathit{tw},1}({\mathscr{a}})) \\ & \qquad \qquad \qquad \qquad
- {\mathscr{G}}^{\mathit{tw},2}(\gamma^{\mathit{tw},0},{\mathscr{a}}) - {\mathscr{G}}^{\mathit{tw},2}({\mathscr{a}},\gamma^{\mathit{tw},0}), \\
& \dots
\end{aligned}$$
A first consequence of is that if ${\mathscr{X}}$ follows $[\gamma]$, then $\tilde{\mathscr{X}}= {\mathscr{G}}^{\mathit{tw}}({\mathscr{X}})$ follows $[\tilde{\gamma}]$. Indeed, if $\nabla_{\mathscr{X}}= {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{X}}$ is a relative connection with respect to $\gamma$, then $$\label{eq:image-relative-connection}
\nabla_{\tilde{\mathscr{X}}} = {\nabla\mkern-12mu/\mkern2mu}_{\tilde{\mathscr{X}}} + \beta^{\mathit{tw},0}$$ is a relative connection for $\tilde\gamma$. Suppose that ${\mathscr{X}}_k$ ($k = 0,1$) are families with relative connections, and we equip their images $\tilde{{\mathscr{X}}}_k$ with the induced relative connections as in . From and it then follows that for any cocycle ${\mathscr{a}}$, $$\label{eq:functoriality-of-connections}
\begin{aligned}
& \nabla_{\mathit{hom}_{\tilde{{\mathscr{A}}}^{\mathit{tw}}}(\tilde{{\mathscr{X}}}_0,\tilde{{\mathscr{X}}}_1)}({\mathscr{G}}^{\mathit{tw},1}({\mathscr{a}})) =
{\mathscr{G}}^{\mathit{tw},1}(\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}}_0,{\mathscr{X}}_1)}({\mathscr{a}})) + \text{\it (coboundary)}.
\end{aligned}$$ This explains the sense in which, under Assumption \[th:matching-deformations\], the cohomology level connections on $\mathit{hom}$ spaces are functorial.
Let’s turn to the corresponding question for $A_\infty$-modules. Take $\tilde{A}$, considered as an $(A,\tilde{A})$-bimodule by $G$-pullback on the left side only. This is right perfect, since $\tilde{A}(\cdot,X) = G(X)^{\mathit{yon}}$, hence gives rise to a convolution functor $$G^{\mathit{perf}} = K_{\tilde{A}}: A^{\mathit{perf}} \longrightarrow \tilde{A}^{\mathit{perf}}.$$ In the same way, one can define an analogue ${\mathscr{G}}^{\mathit{perf}} = {\mathscr{K}}_{\tilde{A}}$ acting on families. Suppose that ${\mathscr{M}}$ is a perfect family of modules carrying a pre-connection ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}$. Then there is an induced pre-connection on $\tilde{{\mathscr{M}}} = {\mathscr{G}}^{\mathit{perf}}({\mathscr{M}}) = {\mathscr{M}}\otimes_A \tilde{A}$: $$\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{\tilde{{\mathscr{M}}}}^0({\mathscr{m}}\otimes a_r \otimes \cdots \otimes a_1 \otimes \tilde{a}) =
\textstyle\sum_i {\nabla\mkern-12mu/\mkern2mu}_{\tilde{{\mathscr{M}}}}^{r-i+1}({\mathscr{m}},a_r,\dots,a_{i+1}) \otimes a_i \otimes \cdots \otimes \tilde{a}, \\
& {\nabla\mkern-12mu/\mkern2mu}_{\tilde{{\mathscr{M}}}}^d = 0 \quad \text{for all $d>0$,} \\
& \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\tilde{{\mathscr{M}}}}) = {\mathscr{G}}^{\mathit{perf},1}(\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}})).
\end{aligned}$$ If ${\mathscr{M}}_k$ ($k = 0,1$) are families with pre-connections, and $\tilde{{\mathscr{M}}}_k$ their images under ${\mathscr{G}}^{\mathit{perf}}$ equipped with the induced pre-connections, the following simpler analogue of holds: $${\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{\tilde{{\mathscr{A}}}^{\mathit{perf}}}(\tilde{{\mathscr{M}}}_0,\tilde{{\mathscr{M}}}_1)}({\mathscr{G}}^{\mathit{perf},1}({\mathscr{b}})) = {\mathscr{G}}^{\mathit{perf},1}({\nabla\mkern-12mu/\mkern2mu}_{\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}({\mathscr{b}})).$$
Now suppose again that Assumption \[th:matching-deformations\] holds. Write $\gamma^{\mathit{perf}} \in \mathit{CC}^1({\mathscr{A}}^{\mathit{perf}}, \Omega^1_{\mathscr{R}}\otimes {\mathscr{A}}^{\mathit{perf}})$ for the element induced by $\gamma$ as in , and similarly for $\tilde\gamma^{\mathit{perf}}$. Then, there is a corresponding element $\beta^{\mathit{perf}} \in \mathit{CC}^0({\mathscr{A}}^{\mathit{perf}},\Omega^1_{\mathscr{R}}\otimes \tilde{{\mathscr{A}}}^{\mathit{perf}})$ which satisfies $$\label{eq:beta-perf}
\partial \beta^{\mathit{perf}} = ({\mathscr{G}}^{\mathit{perf}})^*(\tilde\gamma^{\mathit{perf}}) - {\mathscr{G}}^{\mathit{perf}}_*(\gamma^{\mathit{perf}}).$$ Instead of attempting to define $\beta^{\mathit{perf}}$ by a direct formula, it seems more reasonable to argue by restriction to the images of the Yoneda embeddings ${\mathscr{A}}^{\mathit{tw}} \rightarrow {\mathscr{A}}^{\mathit{perf}}$, $\tilde{{\mathscr{A}}}^{\mathit{tw}} \rightarrow \tilde{{\mathscr{A}}}^{\mathit{perf}}$. This restriction induces quasi-isomorphisms on the relevant Hochschild complexes, and it essentially reduces this situation to the previously discussed case of twisted complexes. also has similar consequences as before: if $\nabla_{\mathscr{M}}= {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{M}}$ is a relative connection for $\gamma$, then $$\label{eq:image-relative-connection-for-modules}
\nabla_{\tilde{\mathscr{M}}} = {\nabla\mkern-12mu/\mkern2mu}_{\tilde{\mathscr{M}}} + \beta^{\mathit{perf},0}$$ is a relative connection for $\tilde\gamma$, and moreover these relative connections satisfy a simplified version of : $$\label{eq:strict-functoriality-of-connections}
\nabla_{\mathit{hom}_{\tilde{{\mathscr{A}}}^{\mathit{perf}}}(\tilde{{\mathscr{M}}}_0,\tilde{{\mathscr{M}}}_1)}({\mathscr{G}}^{\mathit{perf},1}({\mathscr{b}})) =
{\mathscr{G}}^{\mathit{perf},1}(\nabla_{\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1)}({\mathscr{b}})).$$
Thinking in terms of modules naturally accomodates a generalization, in which we do not start with a functor $G$, but instead with a general right perfect $(A,\tilde{A})$-bimodule $P$, and its convolution functor ${\mathscr{K}}_P$ for families. As in , we then have a homotopy commutative diagram $$\label{eq:grrr-2}
\xymatrix{
\mathit{CC}({\mathscr{A}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{A}}) \ar[d] \ar[r] & \mathit{hom}_{({\mathscr{A}},\tilde{{\mathscr{A}}})^{\mathit{mod}}}(P,\Omega^1_{\mathscr{R}}\otimes P) \ar[d] & \ar[l] \ar[d] \mathit{CC}(\tilde{{\mathscr{A}}},\Omega^1_{\mathscr{R}}\otimes \tilde{{\mathscr{A}}}) \\
\mathit{CC}({\mathscr{A}}^{\mathit{perf}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{A}}^{\mathit{perf}}) \ar[r] &
\mathit{CC}({\mathscr{A}}^{\mathit{perf}},\Omega^1_{\mathscr{R}}\otimes \tilde{{\mathscr{A}}}^{\mathit{perf}}) & \ar[l]
\mathit{CC}(\tilde{{\mathscr{A}}}^{\mathit{perf}},\Omega^1_{\mathscr{R}}\otimes \tilde{{\mathscr{A}}}^{\mathit{perf}}),
}$$ where $P$ is considered as a constant family of bimodules over ${\mathscr{R}}$ (the general theory of such families will be our next topic of discussion, but it is easy to see what we mean in this special case). The natural analogue of Assumption \[th:matching-deformations\] in this context is therefore:
\[th:matching-deformations-2\] Suppose that we have deformation fields $[\gamma]$ and $[\tilde\gamma]$ for ${\mathscr{A}}$ and $\tilde{{\mathscr{A}}}$, respectively, whose images in $H^1(\mathit{hom}_{({\mathscr{A}},\tilde{{\mathscr{A}}})^{\mathit{mod}}}(P,\Omega^1_{\mathscr{R}}\otimes P))$ agree.
If this is the case, one can apply the same argument as before to ${\mathscr{K}}_P$, meaning that relative connections on perfect families of modules can be pushed forward, and the analogue of will hold.
Existence\[subsec:existence\]
-----------------------------
In our discussion of functoriality, we have used the tensor product of a family of modules and a fixed bimodule. The other combination, where the module is fixed but the bimodule varies, is also useful. A [*family of bimodules with finite cohomology*]{} over ${\mathscr{S}}$ associates to any $(X,\tilde{X}) \in \mathit{Ob}\, A \times \mathit{Ob} \, \tilde{A}$ a complex ${\mathscr{P}}(\tilde{X},X)$ of projective ${\mathscr{R}}$-modules, which comes with structure maps as in , and such that the cohomology of $({\mathscr{P}}(\tilde{X},X),\mu^{0|1|0}_{{\mathscr{P}}})$ is bounded and finitely generated in each degree. Such bimodules form an $A_\infty$-category over ${\mathscr{R}}$, denoted by $({\mathscr{A}},\tilde{\mathscr{A}})^{\mathit{mod}}$. The elementary theory of families of modules, as developed in Section \[subsec:families\], carries over to this situation without any complications.
Let ${\mathscr{P}}$ be a family of $(A,\tilde{A})$-bimodules. A pre-connection ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}$ is a sequence of maps $$\label{eq:bimod-pre-connection-components}
\begin{aligned}
& {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}^{s|1|t}: \mathit{hom}_A(X_{s-1},X_s) \otimes \cdots \otimes \mathit{hom}_A(X_0,X_1) \otimes {\mathscr{P}}(\tilde{X}_t,X_0) \\ & \qquad \qquad \qquad \qquad \otimes \mathit{hom}_{\tilde{A}}(\tilde{X}_{t-1},\tilde{X}_t) \otimes \cdots \otimes \mathit{hom}_{\tilde{A}}(\tilde{X}_0,\tilde{X}_1) \longrightarrow \Omega^1_{\mathscr{R}}\otimes {\mathscr{P}}(\tilde{X}_0,X_s)[1-s-t],
\end{aligned}$$ where the maps ${\mathscr{p}}\mapsto (-1)^{|{\mathscr{p}}|}{\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}^{0|1|0}({\mathscr{p}})$ are connections in the standard sense, while all the other terms are ${\mathscr{R}}$-linear. Each pre-connection has a deformation cocycle $$\label{eq:bimodule-cocycle}
\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}) \in \mathit{hom}_{({\mathscr{A}},\tilde{{\mathscr{A}}})^{\mathit{mod}}}^1({\mathscr{P}}, \Omega^1_{\mathscr{R}}\otimes {\mathscr{P}}),$$ obtained by applying $\mu^1_{({\mathscr{A}},\tilde{{\mathscr{A}}})^{\mathit{mod}}}$ to . As usual, the cohomology class $\mathit{Def}({\mathscr{P}})$ represented by is independent of the choice of pre-connection.
Take a perfect $A$-module $M$. Then $\tilde{{\mathscr{M}}} = M \otimes_A {\mathscr{P}}$, defined as in , is a family of $\tilde{A}$-modules with finite cohomology. If we assume in addition that ${\mathscr{P}}$ is right perfect, then $\tilde{{\mathscr{M}}}$ is again a perfect family. Moreover, a pre-connection on ${\mathscr{P}}$ defines one on $\tilde{{\mathscr{M}}}$, formally defined by taking the identity on $M$ and tensoring it with ${\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}$. We have an obvious correspondence between deformation cocycles: $$\label{eq:def-of-a-tensor}
\mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\tilde{{\mathscr{M}}}}) = e_M \otimes_A \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}),$$ where $e_M$ is the identity endomorphism. Now suppose that our target category $\tilde{A}$ comes with a deformation field represented by $\tilde{\gamma} \in \mathit{CC}^1(\tilde{\mathscr{A}},\Omega^1_{{\mathscr{R}}} \tilde\otimes {\mathscr{A}})$. In a slight generalization of , we have a canonical chain map $$\label{eq:bimodule-induced}
\mathit{CC}(\tilde{{\mathscr{A}}},\Omega^1_{\mathscr{R}}\otimes\tilde{{\mathscr{A}}}) \longrightarrow
\mathit{hom}_{({\mathscr{A}},\tilde{{\mathscr{A}}})^{\mathit{mod}}}({\mathscr{P}},\Omega^1_{\mathscr{R}}\otimes {\mathscr{P}}).$$ As usual, we say that ${\mathscr{P}}$ follows $\tilde{\gamma}$ if its deformation class is the image of $[\tilde{\gamma}]$ under , and define the notion of relative connection by requiring equality on the cocycle level. It follows from and the explicit formula for that if ${\mathscr{P}}$ follows $[\tilde{\gamma}]$, then so do all the families $\tilde{{\mathscr{M}}} = M \otimes_{\mathscr{A}}{\mathscr{P}}$. The induced connection on the space of morphisms between two such families is given by , which one can write as $$\nabla_{\mathit{hom}_{\tilde{\mathscr{A}}^{\mathit{mod}}}(\tilde{\mathscr{M}}_0,\tilde{\mathscr{M}}_1)}(\tilde{\mathscr{b}}) =
(-1)^{|\tilde{{\mathscr{b}}}|} \mu^2_{\tilde{\mathscr{A}}^{\mathit{mod}}}(e_{M_1} \otimes_A {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}, \tilde{\mathscr{b}}) - \mu^2_{\tilde{\mathscr{A}}^{\mathit{mod}}}(\mathit{id}_{\Omega^1_{\mathscr{R}}} \otimes_A \tilde{\mathscr{b}}, e_{M_0} \otimes_A {\nabla\mkern-12mu/\mkern2mu}_{\mathscr{P}}) + \gamma^{\mathit{mod},1}(\tilde{{\mathscr{b}}}).$$ In particular, if $\tilde{{\mathscr{b}}} = b \otimes_A e_{\mathscr{P}}$, then the first two terms cancel, while the last one vanishes by inspection of . The application we are aiming for is this:
\[th:universal-family\] Take an $A_\infty$-category $A$ with a deformation field $[\gamma]$. Suppose that there is a family of $A$-bimodules ${\mathscr{P}}$ which is right perfect, follows $[\gamma]$, and whose fibre at some base point $s \in {\mathscr{S}}$ is quasi-isomorphic to the diagonal bimodule. Then, for any $M \in \mathit{Ob}\,A^{\mathit{perf}}$, there is a perfect family of modules ${\mathscr{M}}$ which follows $[\gamma]$, and with ${\mathscr{M}}_s$ to quasi-isomorphic to $M$. Moreover, any two such families satisfy $$H^0(\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1)) {\cong}{\mathscr{R}}\otimes H^0(\mathit{hom}_{A^{\mathit{perf}}}(M_0,M_1)),$$ and (for a suitable choice of relative connection) the induced connections on these morphism spaces are trivial.
Define ${\mathscr{M}}= M \otimes_A {\mathscr{P}}$. By our previous discussion, this follows $[\gamma]$ and has the required behaviour at the fibre over $s$. Moreover, if we make the obvious choice of relative connections, for each morphism $[b] \in H^0(\mathit{hom}_{A^{\mathit{perf}}}(M_0,M_1))$ we have a covariantly constant section $[b \otimes e_{{\mathscr{P}}}] \in H^0(\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1))$, which specializes to $[b]$ at the point $s$. This establishes the remaining part of the statement.
Uniqueness\[subsec:unique\]
---------------------------
As before, we work with a fixed deformation field $[\gamma]$. As an aid to intuition, we will increasingly use geometric language. Take two points $s,s' \in {\mathscr{S}}$. One can envisage a process of [*moving objects of $A^{\mathit{perf}}$ along the deformation field*]{} from $s$ to $s'$. Namely, start with some object $M$, and suppose that there is a perfect family of modules ${\mathscr{M}}$ following $[\gamma]$, whose fibre at $s$ is quasi-isomorphic to $M$. Then, take the fibre $M' = {\mathscr{M}}_{s'}$. Generally speaking, no such family may exist, making it impossible to carry out the process at all. However, assuming existence, there is a good uniqueness statement at least for a certain class of objects $M$. Suppose from now on that the following holds:
\[th:augmented\] $H^0(\mathit{hom}_{A^{\mathit{perf}}}(M,M))$ is a commutative ring. Moreover, the ideal of nilpotent elements in that ring has codimension $1$.
\[th:uniqueness-1\] If $M$ satisfies Assumption \[th:augmented\] and can be moved along the deformation field from $s$ to $s'$, the outcome $M'$ is unique up to quasi-isomorphism (which means independent of the family ${\mathscr{M}}$).
The proof is based on a number of elementary observations. For the sake of brevity, let’s write ${\mathscr{H}}= H^0(\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}},{\mathscr{M}}))$. Choose a relative connection $\nabla_{{\mathscr{M}}}$ on our family, and write $\nabla_{{\mathscr{H}}}$ for the induced connection on ${\mathscr{H}}$.
${\mathscr{H}}$ is a commutative ${\mathscr{R}}$-algebra.
By , the image of the commutator map $${\mathscr{H}}\otimes_{\mathscr{R}}{\mathscr{H}}\longrightarrow {\mathscr{H}}, \;\; x \otimes y \longmapsto xy-yx$$ is a subsheaf of ${\mathscr{H}}$ invariant under $\nabla_{\mathscr{H}}$, which therefore must be locally free. By looking at the point $s$, one sees that this sheaf must be zero.
Note that, because of the commutativity and , $\nabla_{{\mathscr{H}}}$ is actually independent of the choice of relative connection on ${\mathscr{M}}$.
Consider the ideal ${\mathscr{H}}^{\mathit{nil}} \subset {\mathscr{H}}$ of nilpotent elements. Then ${\mathscr{H}}^{\mathit{nil}}$ is preserved by $\nabla_{\mathscr{H}}$, and ${\mathscr{H}}/\mathit{{\mathscr{H}}}^{\mathit{nil}}$ is the trivial line bundle.
Choose a tangent vector field $\xi$ on ${\mathscr{S}}$. For any element $h \in {\mathscr{H}}$ and any $m>0$, we have $$\nabla_{{\mathscr{H}},\xi}^m (h^m) \in m! (\nabla_{{\mathscr{H}},\xi} h)^m + h{\mathscr{H}}.$$ Choosing $h \in {\mathscr{H}}^{\mathit{nil}}$ and $m$ large, one sees that ${\mathscr{H}}^{\mathit{nil}}$ is closed under $\nabla_{{\mathscr{H}}}$. By the same reasoning as before, ${\mathscr{H}}/{\mathscr{H}}^{\mathit{nil}}$ must be a line bundle. But the identity endomorphism yields a nowhere vanishing section, which provides a trivialization.
Now suppose that we have two perfect families ${\mathscr{M}}_+$ and ${\mathscr{M}}_-$ both following $[\gamma]$ and whose fibres at $s$ are quasi-isomorphic to the same object $M$, which still satisfies Assumption \[th:augmented\]. Working on the cohomology level as before, we denote by ${\mathscr{H}}_{\pm}$ the endomorphism rings of these objects, by ${\mathscr{H}}_{\pm}^{\mathit{nil}}$ the ideals of nilpotent endomorphisms, and by ${\mathscr{H}}_{-+}$, respectively ${\mathscr{H}}_{+-}$, the space of morphisms from ${\mathscr{M}}_+$ to ${\mathscr{M}}_-$, and vice versa. We choose relative connections on ${\mathscr{M}}_{\pm}$, equipping all these morphism spaces with the induced connections.
The multiplication maps $$\begin{aligned}
\label{eq:compose-1} & {\mathscr{H}}_{-+} \otimes {\mathscr{H}}_{+}^{\mathit{nil}} \longrightarrow {\mathscr{H}}_{-+}, \\
\label{eq:compose-2} & {\mathscr{H}}_{-}^{\mathit{nil}} \otimes {\mathscr{H}}_{-+} \longrightarrow {\mathscr{H}}_{-+}\end{aligned}$$ both have the same image, which we denote by ${\mathscr{H}}_{-+}^{\mathit{nil}}$. This is preserved by the connection, and ${\mathscr{H}}_{-+}/{\mathscr{H}}_{-+}^{\mathit{nil}}$ is a line bundle.
Let’s first define ${\mathscr{H}}_{-+}^{\mathit{nil}}$ to be the image of the first map . By compatibility with connections, ${\mathscr{H}}_{-+}/{\mathscr{H}}^{\mathit{nil}}_{-+}$ must be locally free, hence in view of the behaviour at the point $s$ a line bundle. Now take and compose it with projection to ${\mathscr{H}}_{-+}/{\mathscr{H}}_{-+}^{\mathit{nil}}$. Again by the same argument, the composition vanishes identically, hence the image of is contained in that of . Running the argument the other way yields the required equality.
Let’s define ${\mathscr{H}}_{+-}^{\mathit{nil}} \subset {\mathscr{H}}_{+-}$ in the same way. It follows directly from the definition that the composition ${\mathscr{H}}_{+-} \otimes {\mathscr{H}}_{-+} \rightarrow {\mathscr{H}}_+$ takes ${\mathscr{H}}_{+-} \otimes {\mathscr{H}}_{-+}^{\mathit{nil}}$ and ${\mathscr{H}}_{+-}^{\mathit{nil}} \otimes {\mathscr{H}}_{-+}$ to ${\mathscr{H}}_+^{\mathit{nil}}$, and the same is true in the other order.
Multiplication induces isomorphisms $$\label{eq:multiplication-mod-nilpotents}
\begin{aligned}
& ({\mathscr{H}}_{+-}/{\mathscr{H}}_{+-}^{\mathit{nil}}) \otimes ({\mathscr{H}}_{-+}/{\mathscr{H}}_{-+}^{\mathit{nil}}) \longrightarrow {\mathscr{H}}_{+}/{\mathscr{H}}_{+}^{\mathit{nil}}, \\
& ({\mathscr{H}}_{-+}/{\mathscr{H}}_{-+}^{\mathit{nil}}) \otimes ({\mathscr{H}}_{+-}/{\mathscr{H}}_{+-}^{\mathit{nil}}) \longrightarrow {\mathscr{H}}_{-}/{\mathscr{H}}_{-}^{\mathit{nil}}.
\end{aligned}$$
We already established the well-definedness of these maps. Both sides are line bundles and carry connections, which are compatible with the maps, and at the fibre at $s$ we get isomorphisms.
There is a line bundle ${\mathscr{F}}$ such that ${\mathscr{M}}_-$ is quasi-isomorphic to ${\mathscr{F}}\otimes {\mathscr{M}}_+$.
Since we are free to tensor ${\mathscr{M}}_+$ with a line bundle, we may assume without loss of generality that $({\mathscr{H}}_{-+}/{\mathscr{H}}_{-+}^{\mathit{nil}})$ is the trivial line bundle. By , the same must then be true for $({\mathscr{H}}_{+-}/{\mathscr{H}}_{+-}^{\mathit{nil}})$. Choose trivializations and lift them to sections of ${\mathscr{H}}_{-+}$ and ${\mathscr{H}}_{+-}$, respectively (recall that we are working over an affine curve, so there is no problem in doing this). The product of these in either order yields invertible elements of ${\mathscr{H}}_+$ and ${\mathscr{H}}_-$.
The last-mentioned claim clearly establishes Proposition \[th:uniqueness-1\].
Here is a slightly weaker uniqueness statement, which does not require Assumption \[th:augmented\]. Suppose that ${\mathscr{M}}_+$ and ${\mathscr{M}}_-$ are perfect families following $[\gamma]$, and whose fibres at $s$ are quasi-isomorphic to $M$. The composition maps ${\mathscr{H}}_{+-} \otimes {\mathscr{H}}_{-+} \rightarrow {\mathscr{H}}_+$ and ${\mathscr{H}}_{-+} \otimes {\mathscr{H}}_{+-} \rightarrow {\mathscr{H}}_-$ are onto. Specializing to any other fibre $s'$, one finds that ${\mathscr{M}}_{+,s'}$ is quasi-isomorphic to a direct summand of a finite direct sum of copies of ${\mathscr{M}}_{-,s'}$, and vice versa.
We want to take a similar approach to morphisms. Let $M_k$ ($k = 0,1$) be objects of $A^{\mathit{perf}}$, and $B \subset \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_0,M_1)$ a one-dimensional subspace. Suppose that ${\mathscr{M}}_k$ ($k = 0,1$) are perfect families which follow $[\gamma]$, and with ${\mathscr{M}}_{k,s} {\cong}M_k$. Choose relative connections on them. Suppose also that there is a line bundle ${\mathscr{B}}\subset \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_0, {\mathscr{M}}_1)$ whose fibre at $s$ equals $B$, and which is invariant under the induced connection (if such a ${\mathscr{B}}$ exists, it is unique). We can then restrict the given data to the fibre $s'$, yielding $B' \subset \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_0',M_1')$. We say that $B'$ is obtained from $B$ [*by parallel transport*]{}. Obviously, this can’t be unique unless Proposition \[th:uniqueness-1\] applies to both objects $M_k$, but in fact we will need more than that:
\[th:augmented-plus\] Both $M_k$ satisfy Assumption \[th:augmented\], and the two multiplication maps $$\label{eq:swap-sides}
\begin{aligned}
& B \otimes \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_0,M_0) \longrightarrow \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_0,M_1), \\
& \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_1,M_1) \otimes B \longrightarrow \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_0,M_1)
\end{aligned}$$ have the same image.
\[th:uniqueness-2\] Suppose that $(M_0,M_1,B)$ satisfy Assumption \[th:augmented-plus\], and that parallel transport to $s'$ yields $(M_0',M_1',B')$. Then, this is unique up to quasi-isomorphism (independent of the choice of families, and of the relative connections).
We can apply some preliminary simplifications. One can change any of the families ${\mathscr{M}}_k$ by tensoring it with a line bundle ${\mathscr{F}}_k$ (equipped with a connection), and then take the corresponding line bundle ${\mathscr{F}}_1 \otimes {\mathscr{F}}_0^\vee \otimes {\mathscr{B}}\subset \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{F}}_0 \otimes {\mathscr{M}}_0, {\mathscr{F}}_1 \otimes {\mathscr{M}}_1)$. This does not affect the outcome of parallel transport. With this and the results of the previous proof in mind, the choice of families is indeed irrelevant, so we can consider some fixed choices ${\mathscr{M}}_k$.
The two multiplication maps $$\label{eq:swap-sides-2}
\begin{aligned}
& {\mathscr{B}}\otimes \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_0,{\mathscr{M}}_0) \longrightarrow \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_0,{\mathscr{M}}_1), \\
& \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_1,{\mathscr{M}}_1) \otimes {\mathscr{B}}\longrightarrow \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_0,{\mathscr{M}}_1)
\end{aligned}$$ have the same image.
The proof is routine. We denote the image by ${\mathscr{J}}$.
${\mathscr{J}}$ is preserved by the connection, and this remains true if we change the relative connections on ${\mathscr{M}}_k$.
The first statement is obvious from the definition and the corresponding property of ${\mathscr{B}}$. The second one follows from this and , because left and right multiplication with endomorphisms of ${\mathscr{M}}_k$ preserves ${\mathscr{J}}$.
Now, suppose that we have made different choices of relative connections $\nabla_{{\mathscr{M}}_0,\pm}$ and $\nabla_{{\mathscr{M}}_1,\pm}$, leading to two different line bundles ${\mathscr{B}}_\pm$, and associated subbundles ${\mathscr{J}}_{\pm}$.
In fact, ${\mathscr{J}}_+ = {\mathscr{J}}_-$.
Choose one of the two connections on $\mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_0,{\mathscr{M}}_1)$ arising from our choices. Both ${\mathscr{J}}_+$ and ${\mathscr{J}}_-$ are invariant under this connection, and they agree at one point.
Specializing to the fibres at $s'$, this means that $B_-'$ is contained in the image of the multiplication map $\mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_1',M_1') \otimes B_+' \rightarrow \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_0',M_1')$, and vice versa. Hence, one can write $B_-' = x_{-+}' B_+'$ and $B_+' = x_{+-}' B_-'$ for some $x_{-+}',x_{+-}' \in \mathit{Hom}_{H^0(A^{\mathit{perf}})}(M_1',M_1')$. Since the endomorphism ring is commutative and $B_+' = x_{+-}'x_{-+}' B_+'$, none of the $x_{-+}',x_{+-}'$ can be nilpotent. Arguing as in the proof of Proposition \[th:uniqueness-1\], they must then be invertible. What we have shown is that $B_-'$ can be obtained from $B_+'$ by applying automorphisms of the $M_k'$, which is indeed what was claimed in Proposition \[th:uniqueness-2\].
Periods\[subsec:flux\]
----------------------
Fix a smooth elliptic curve $\bar{{\mathscr{S}}}$ over $R$, together with a differential $\bar\theta \in H^0(\bar{{\mathscr{S}}},\Omega^1_{\bar{{\mathscr{S}}}})$. Given any nonempty affine open subset ${\mathscr{S}}\subset \bar{{\mathscr{S}}}$, we consider the one-form $\theta = \bar\theta|{\mathscr{S}}$.
\[th:per\] A class $[g] \in \mathit{HH}^1(A,A)$ is called [*periodic*]{} if the following holds. For every $X \in \mathit{Ob}\,A^{\mathit{perf}}$ there is a subset ${\mathscr{S}}$ as before, as well as a perfect family ${\mathscr{M}}$ over ${\mathscr{S}}$ which follows $[\gamma] = \theta \otimes [g]$, and whose fibre at some point $s \in {\mathscr{S}}$ is quasi-isomorphic to $X$. Moreover, given any two objects $(X_0,X_1)$, one can choose families $({\mathscr{M}}_0,{\mathscr{M}}_1)$ as before with quasi-isomorphisms ${\mathscr{M}}_{k,s} {\cong}X_k$, so that the bundle of cohomology level morphisms is trivial: $$\label{eq:triv-h0}
H^0(\mathit{hom}_{{\mathscr{A}}^{\mathit{perf}}}({\mathscr{M}}_0,{\mathscr{M}}_1)) {\cong}{\mathscr{R}}\otimes_R H^0(\mathit{hom}_A(X_0,X_1)).$$ In addition, one should be able to choose relative connections on the families so that the induced connection is trivial, which means compatible with a trivialization . The subset of periodic classes is denoted by $$\label{eq:periodic-set}
\mathit{Per}(A,\bar{{\mathscr{S}}},\bar\theta) \subset \mathit{HH}^1(A,A).$$
Note that this is really an invariant of $A^{\mathit{perf}}$ up to quasi-isomorphism. By pulling back a given family ${\mathscr{X}}$ by the $n$-th power map $\bar{{\mathscr{S}}} \rightarrow \bar{{\mathscr{S}}}$ (defined by taking $s$ to be the origin) for some $n \in {\mathbb{Z}}$, one gets a family which follows the restriction of $n[(\bar\theta|{\mathscr{S}}) \otimes g]$. This proves that is closed under multiplication by integers.
\[th:downsides\] The notion introduced above is called “working definition” in view of its numerous shortcomings. These deserve some discussion, even though they do not stand in the way of our immediate application.
The first and most obvious point is the object-by-object approach we’ve taken. This violates categorical common sense and manners, and is likely to be the reason why we can’t prove that the set of periodic classes is an abelian group. However, it is not difficult to envisage a more universal approach, based on Corollary \[th:universal-family\]; it is maybe more appropriate to think of it as an $A_\infty$-version of the [*derived Picard group*]{} [@yekutieli99; @yekutieli04; @keller04].
The next point is the use of a priori undetermined open affine subsets ${\mathscr{S}}\subset \bar{{\mathscr{S}}}$, which essentially means that we are working (Zariski) locally around the base point $s$. This reflects the insufficient technical sophistication of our definitions of families, which are not local over the base. This is a major inconvenience, but does not lead to a decisive loss of information (see Lemma \[th:rational-section\]).
There is one more issue which is conceptually by far the most important one. Asking for families parametrized by an elliptic curve amounts to a double periodicity requirement, but single periodicity seems a more fundamental notion. Instead of a “torus” one would then want a “thin annulus” as a parameter space. This makes sense in analytic geometry, either over ${\mathbb{C}}$ or over a non-archimedean field. Such a theory would require extensive reworking of the foundations. On the other hand, if successful, it might allow substantial simplifications and extensions of the main arguments in this paper, bringing them in line with existing ideas about convergence in Floer cohomology [@fukaya02c; @fukaya09].
The techniques from Section \[subsec:existence\] can be used to show that a given Hochschild class is periodic. On the other hand, if one wants to show that $[g]$ is not periodic, the arguments from Section \[subsec:unique\], in combination with the following trick, can be useful.
Suppose that for some ${\mathscr{S}}\subset \bar{{\mathscr{S}}}$ we have a smooth affine curve $\tilde{{\mathscr{S}}}$ together with a morphism $\tilde{\mathscr{S}}\rightarrow {\mathscr{S}}$, and two points $\tilde{s}_\pm \in \tilde{{\mathscr{S}}}$ whose image is the same $s \in {\mathscr{S}}$. Let $\tilde\theta$ be the pullback of $\theta$. Suppose also that we have perfect families $\tilde{{\mathscr{M}}}_0,\tilde{{\mathscr{M}}}_1$ on $\tilde{{\mathscr{S}}}$ following $\tilde\theta \otimes [g]$. Equip them with relative connections, and suppose further that $\tilde{\mathscr{B}}\subset \mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}(\tilde{{\mathscr{M}}}_0, \tilde{{\mathscr{M}}}_1)$ is a line bundle preserved by the induced connection. By restriction to the fibres at $\tilde{s}_{\pm}$, we get objects $M_{0,\pm}$ and $M_{1,\pm}$, as well as morphism subspaces $B_{\pm}$. We require that $(M_{0,+},M_{1,+},B_+)$ should satisfy Assumption \[th:augmented-plus\].
\[th:contradiction\] In the situation set up above, assume additionally that $(M_{0,+},M_{1,+},B_+)$ is not isomorphic to $(M_{0,-},M_{1,-},B_-)$. Then $[g]$ is not periodic (for the original $\bar{{\mathscr{S}}}$ and $\bar{\theta}$).
Assume that $[g]$ is in fact periodic. Then we can find families ${\mathscr{M}}_0$ and ${\mathscr{M}}_1$ over some open subset ${\mathscr{S}}$, whose fibre at some point $s$ is $M_{0,+}$ and $M_{1,+}$, respectively. A priori, the open subset and the point do not have to coincide with those that appeared in the statement of the Lemma. However, that discrepancy can be removed by using the group structure of the elliptic curve $\bar{{\mathscr{S}}}$ (which yields translations acting transitively on points, and preserving $\bar\theta$), and by making the open subsets smaller if necessary. Having resolved that issue, we continue the discussion: by definition, the families can be chosen so that $\mathit{Hom}_{H^0({\mathscr{A}}^{\mathit{perf}})}({\mathscr{M}}_0,{\mathscr{M}}_1)$ is the trivial bundle and carries the trivial connection. This allows one to find a line bundle ${\mathscr{B}}$ inside that morphism space, which is compatible with the connection and has fibre $B_+$ at $s$. Now pull back those families to $\tilde{{\mathscr{S}}}$. Comparison with $(\tilde{\mathscr{M}}_0,\tilde{\mathscr{M}}_1,\tilde{\mathscr{B}})$ shows that Proposition \[th:uniqueness-2\] is violated, since the two choices of families are isomorphic at $s_+$ but not at $s_-$.
Relaxing the assumptions
------------------------
To conclude the abstract part of our discussion, we want to enlarge the existing framework in two minor ways. The first one is to pass from ${\mathbb{Z}}$-gradings to $({\mathbb{Z}}/2)$-gradings. We then need a version of Lemma \[th:proj\] for ${\mathbb{Z}}/2$-graded complexes ${\mathscr{F}}_0$ and ${\mathscr{F}}_1$, but that is unproblematic: one passes to the associated ${\mathbb{Z}}$-graded periodic complexes and applies the original form of the Lemma to those, thereby deriving the desired result.
The other generalization is to allow $A_\infty$-categories $A$ which are only cohomologically unital (but still proper). The first effect of this is on twisted complexes, where we have to prove:
$A^{\mathit{tw}}$ and ${\mathscr{A}}^{\mathit{tw}}$ are cohomologically unital.
Let $X$ be a twisted complex written as in . There is a spectral sequence (convergent after finitely many steps) which leads to $H(\mathit{hom}_{A^{\mathit{tw}}}(X,X))$ and starts with $$E_1 = \bigoplus_{ij} \mathit{Hom}(F^i,F^j) \otimes H(\mathit{hom}_A(X^i,X^j))[\sigma^i-\sigma^j].$$ Moreover, this spectral sequence is multiplicative, which implies that the identity element in the $E_1$ page always survives. This yields a degree zero endomorphism $[u_X]$, with the property (by a spectral sequence comparison theorem) that the maps $$\begin{aligned}
& [a] \longmapsto [u_X] \cdot [a]: H(\mathit{hom}_{A^{\mathit{tw}}}(Y,X)) \longrightarrow H(\mathit{hom}_{A^{\mathit{tw}}}(Y,X)), \\
& [a] \longmapsto [a] \cdot [u_X]: H(\mathit{hom}_{A^{\mathit{tw}}}(X,Y)) \longrightarrow H(\mathit{hom}_{A^{\mathit{tw}}}(X,Y))
\end{aligned}$$ are isomorphisms for any $Y$. In particular, there is an $[e_X]$ which satisfies $[e_X] \cdot [u_X] = [u_X]$, and then automatically also $[u_X] \cdot [e_X] = [u_X]$. One easily checks that this is the required cohomological identity. This argument extends to families without any problems.
\[th:unit-h-unit\] In the case of families of twisted complexes, $\mu^2_{{\mathscr{A}}^{\mathit{tw}}}(\cdot,e_{\mathscr{X}}): \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}},{\mathscr{Y}}) \rightarrow \mathit{hom}_{{\mathscr{A}}^{\mathit{tw}}}({\mathscr{X}},{\mathscr{Y}})$ induces the identity on cohomology, and is chain homotopic to its own square (by the $A_\infty$-equations). Therefore, it is actually chain homotopic to the identity, by Lemma \[th:homotopy-idempotent\], and the same holds on the other side. This is a slightly stronger property than cohomological unitality, and generally more appropriate for $A_\infty$-categories defined over a ring.
Unfortunately, the theory of pre-connections on twisted complexes does not generalize to the cohomologically unital context in an obvious way, so we’ll have to be careful to use that only for strictly unital $A_\infty$-categories.
The situation for modules is slightly different, since unitality requirements enter into the definition of the objects and morphisms themselves. Given a cohomologically unital $A$, one defines $A^{\mathit{mod}}$ by taking cohomologically unital modules with finite cohomology as objects, and arbitrary module homomorphisms as morphisms. If $A$ was strictly unital, this would yield a category quasi-equivalent to the previously considered version using strictly unital modules and homomorphisms [@lefevre Section 3.3]. The same construction for families defines ${\mathscr{A}}^{\mathit{mod}}$, and Lemma \[th:acyclic-modules\] still holds.
The cohomological unitality condition on a family of modules ${\mathscr{M}}$ says that if $e_X \in \mathit{hom}_A^0(X,X)$ is a representative for the cohomology unit in $A$, then $\mu^2_{\mathscr{M}}(\cdot,e_X): {\mathscr{M}}(X) \longrightarrow {\mathscr{M}}(X)$ induces the identity on cohomology. Arguing as in Remark \[th:unit-h-unit\], one sees that this map is in fact chain homotopic to the identity, which would again be the more natural condition in general (but as we’ve seen, turns out to be equivalent in our context).
The Yoneda embeddings $A \rightarrow A^{\mathit{mod}}$ and ${\mathscr{A}}\rightarrow {\mathscr{A}}^{\mathit{mod}}$ are quasi-isomorphisms.
Take the same maps $$\mathit{hom}_A(Y_0,Y_1) \longrightarrow \mathit{hom}_{A^{\mathit{mod}}}(Y^{\mathit{yon}}_0,Y_1^{\mathit{yon}}) \longrightarrow \mathit{hom}_A(Y_0,Y_1)$$ as in Lemma \[th:yoneda\], using any representative $e_{Y_0}$ for the cohomological unit. Composition in the given order is the endomorphism $a \mapsto \mu^2_A(a,e_{Y_0})$, which by definition acts as the identity on cohomology. Take the composition in the opposite order and add the coboundary of the homotopy . The outcome is the map $$\label{eq:k-homotopy}
\begin{aligned}
& k: \mathit{hom}_{A^{\mathit{mod}}}(Y_0^{\mathit{yon}},Y_1^{\mathit{yon}}) \longrightarrow \mathit{hom}_{A^{\mathit{mod}}}(Y_0^{\mathit{yon}},Y_1^{\mathit{yon}}), \\
& k(b)^d(a_d,\dots,a_1) = \sum_i (-1)^{|a_{i+1}| + \cdots + |a_d| + d - i + 1} b^{i+1}(\mu_A^{d-i+1}(e_{Y_0},a_d,\dots, a_{i+1}),a_i,\dots,a_1).
\end{aligned}$$ By looking at the length filtration , one sees that this is a quasi-isomorphism. The same thing applies to constant families.
As before, for a strictly unital $A$ we now have a two version of ${\mathscr{A}}^{\mathit{mod}}$, one defined in a strictly unital context, and the other by treating $A$ as cohomology unital. Unfortunately, the previously quoted result from [@lefevre] does not immediately extend to this context, since it relies on minimal models for modules, which only exist over a field (and the alternative approach from [@seidel04 Section 2] has not been extended to modules so far). It seems highly plausible that the two versions are still quasi-equivalent, but we will allow ourselves to sweep the issue under the rug. In fact, in all our applications what counts are the subcategories of perfect families ${\mathscr{A}}^{\mathit{perf}}$, where this problem does not arise (because in both contexts they are split-generated by constant families).
Finally, in the cohomologically unital context, one similarly wants to adjust the notion of bimodule, and use the full Hochschild complex instead of the reduced one.
The two-torus\[sec:elliptic\]
=============================
In this section we consider a specific finite-dimensional algebra, together with its $A_\infty$-deformations. The algebra occurs geometrically in connection with degree $2$ line bundles on elliptic curves, and its $A_\infty$-deformations yield one possible model for the derived category of coherent sheaves on such curves. In view of homological mirror symmetry (see [@kontsevich94] for the general statement, and [@polishchuk-zaslow98; @kontsevich-soibelman00; @fukaya02b; @abouzaid-smith09; @lekili-perutz11] for the case relevant here), the same structure describes the Fukaya category of the two-torus. Our aim is to construct a particular family of objects, which in terms of the elliptic curve is the tautological family of structure sheaves of its points, and in terms of the Fukaya category is a family of parallel lines on the torus (the connection between the two could be made directly via SYZ transformations, as in [@kontsevich-soibelman00; @fukaya02c]). The point of the exercise is to see how this fits in with the technical notions of family given in the previous section. This is not entirely straightforward, since the $K$-theory class varies, which precludes a description as family of twisted complexes.
Initially, we will work over an arbitrary field $R$ of characteristic $0$. Later on, when considerations become more geometric, we will re-introduce the added assumption that $R$ be algebraically closed. In the last parts, we will specialize this further to the simplest (one-variable) Novikov field from Floer theory, namely $$\label{eq:novikov-field}
R = \left\{
\begin{aligned} & u = c_0 \hbar^{m_0} + c_1 \hbar^{m_1} + \cdots, \\
& \text{where } c_k \in {\mathbb{C}}, \;\; m_k \in {\mathbb{R}}, \text{ and } \lim_{k \rightarrow \infty} m_k = +\infty \end{aligned}\right\}.$$ Note that, since the coefficient field ${\mathbb{C}}$ is algebraically closed, so is $R$ [@fukaya-oh-ohta-ono10 Appendix]. The sign conventions used in constructing the Fukaya category of the two-torus follow [@seidel04 Section 13].
Koszul algebras\[subsec:koszul\]
--------------------------------
Let $W$ be a finite-dimensional graded $R$-vector space. A [*quadratic algebra*]{} is an associative unital graded $R$-algebra of the form $A = TW/J$, where $TW$ is the tensor algebra, and $J \subset W \otimes W$ is a graded linear subspace. Let $J^\perp \subset W^\vee \otimes W^\vee$ be the orthogonal complement of $J$ with respect to the canonical pairing $$\begin{aligned}
& W^\vee \otimes W^\vee \otimes W \otimes W \longrightarrow R, \\
& w_2^\vee \otimes w_1^\vee \otimes w_1 \otimes w_2 \longmapsto (-1)^{|w_2|} w_2^\vee(w_2)w_1^\vee(w_1).
\end{aligned}$$ The [*quadratic dual*]{} of $A$ is defined as $A^! = T(W^\vee[-1])/J^\perp[-2]$. The Koszul complex is the graded vector space $A^! \otimes A$ with differential $$x^! \otimes x \longmapsto \sum_r (-1)^{|x|} x^! w_r^\vee \otimes w_r x,$$ where $\{w_r\}$ is a basis of $W$, and $\{w_r^\vee\}$ the dual basis. One says that $A$ is a [*Koszul algebra*]{} if the Koszul complex is acyclic.
There is also a more abstract formulation. Consider the abelian category of graded left $A$-modules, and in it the simple module $R$. We then have a bigraded group $\mathit{Ext}^i_A(R,R[j])$, where $i$ is the cohomological grading and $j$ the internal one, inherited from the grading of $A$ itself. For instance, $$\label{eq:low-ext}
\begin{aligned}
& \mathit{Ext}^0_A(R,R[j]) = \begin{cases} R & j = 0, \\ 0 & \text{otherwise,} \end{cases} \\
& \mathit{Ext}^1_A(R,R[j]) {\cong}(W^\vee)^j, \\
& \mathit{Ext}^2_A(R,R[j]) {\cong}(J^\vee)^j.
\end{aligned}$$ In our case $A$ has an extra grading by pathlength, and this induces another grading on each $\mathit{Ext}^i_A(R,R[j])$. For low values of $i$, one sees from that $\mathit{Ext}^i$ is concentrated in path length $i$. Then, $A$ is Koszul if and only if the same holds for all $i$ (the original reference is [@priddy70]; for more recent expositions in slightly varying degrees of generality, see [@beilinson-ginsburg-schechtman88; @froberg89; @beilinson-ginzburg-soergel96]).
\[th:double-grading\] Even though we have kept track of some signs arising from the grading of $W$, these are actually irrelevant for the purpose of determining whether $A$ is Koszul or not, as the following trick shows. Let $\tilde{A}$ be the algebra obtained from $A$ by multiplying the given grading of $W$ by 2. The category of graded $A$-modules embeds fully and faithfully into that of graded $\tilde{A}$-modules in the same way, which we denote by $M \mapsto \tilde{M}$. This doubles the amount of shift, $$\label{eq:double-shift}
\widetilde{M[j]} {\cong}\widetilde{M}[2j].$$ Using that and any projective resolution, one sees that $$\mathit{Ext}^i_{\tilde{A}}(\tilde{M},\tilde{N}[j]) = \begin{cases} \mathit{Ext}^i_{A}(M,N[j/2]) & \text{if $j$ is even}, \\
0 & \text{otherwise.}
\end{cases}$$ Moreover, the isomorphism is compatible with path length. Hence, $A$ is Koszul if and only if $\tilde{A}$ is. Once one has done this change, it is clear that the same will hold even if one changes the grading to be trivial (concentrated in degree zero), since the Koszul complex is not affected by even changes in the grading.
The Hochschild cohomology of a Koszul algebra $A$ can be computed [@green-hartman-marcos-solberg05] as the cohomology of $A^! \otimes A$ with a modified differential $$\label{eq:koszul-hochschild}
x^! \otimes x \longmapsto \sum_r (-1)^{|x|} x^! w_r^\vee \otimes w_r x - (-1)^{(|w_r|+1)(|x|+|x^!|)} w_r^\vee x^! \otimes x w_r.$$ More precisely, the Hochschild cohomology of any graded algebra $A$ is a bigraded vector space $\mathit{HH}^i(A,A[j])$, where again $i$ is the cohomological grading, and $j$ the internal one. For Koszul algebras, $i+j$ is the natural grading of $A^! \otimes A$, whereas $i$ measures path lengths in the $A^!$ piece.
\[th:truncated\] Take a free algebra $A = R\langle w \rangle$, where $w$ has odd degree $d$. The Koszul dual is a truncated polynomial algebra $A^! = R[w^\vee]/(w^\vee)^2$, where $w^\vee$ has even degree $1-d$. The Hochschild cohomology has a basis consisting of $1 \otimes w^i$ with $i$ even, and $w^\vee \otimes w^i$ with $i$ odd. This contrasts with the case of even $d$, where the differential vanishes.
It is useful to consider this example in the context of Addendum \[th:double-grading\]. In the category of graded $A$-bimodules, we have $$\label{eq:hh-ext}
\mathit{HH}^i(A,A[j]) = \mathit{Ext}_{A \otimes A^{\mathit{opp}}}^i(A,A[j])$$ where the shifted space $A[j]$ has the bimodule structure which is given by ordinary multiplication on the left side, and twisted multiplication $(-1)^{j|x|} xy$ on the right side. If we take $A$ and double its given grading to $\tilde{A}$, the category of graded $A$-bimodules embeds into that of $\tilde{A}$-bimodules in the obvious way. However, this embedding fails to be compatible with shifts, so that the analogue of for the groups holds only if $j$ is even.
A Hochschild cohomology computation\[subsec:hh\]
------------------------------------------------
Consider the graded path algebra associated to the quiver $$\label{eq:quiver}
\xymatrix{\stackrel{1}{\bullet} \ar@<.5em>@/^1pc/[rr]^-{w_1} \ar@/^1pc/[rr]_-{w_2} && \ar@/^1pc/[ll]_-{w_3} \ar@<.5em>@/^1pc/[ll]^-{w_4} \stackrel{2}{\bullet}}$$ where $w_1,w_2$ have degree $0$, and $w_3,w_4$ degree $1$. Composition of paths will be written from right to left, so the path $w_3w_1$ means going first along $w_1$ and then $w_3$.
\[th:q-algebra\] The graded algebra $Q$ is the quotient of the path algebra of obtained by imposing the relations $$\label{eq:quadratic-relations}
w_3w_2 + w_4w_1 = 0, \quad w_1w_4 + w_2w_3 = 0, \quad w_3w_1 = w_4w_2 = 0.$$
Here’s an alternative description. Let $e_1,e_2 \in Q$ be the idempotents associated to the length $0$ paths. Take $V = R^2$. One can identify $$\label{eq:identifications} \left\{
\begin{aligned}
& e_2 Q e_1 = V && \text{using $w_1,w_2$ as a basis,} \\
& e_1 Q e_2 = V && \text{using $w_3,w_4$ as a basis,} \\
& e_1 Q e_1 = \Lambda^{\mathit{even}}V = \Lambda^0 V \oplus \Lambda^2 V && \text{using $e_1,q_1 = w_3w_2$ as a basis,} \\
& e_2 Q e_2 = \Lambda^{\mathit{even}}V = \Lambda^0 V \oplus \Lambda^2 V && \text{using $e_2,q_2 = w_1w_4$ as a basis.}
\end{aligned} \right.$$ where of course $\Lambda^0 V = R$. With respect to these identifications, both nontrivial multiplications $$\label{eq:wedge}
\begin{aligned}
& e_1 Q e_2 \otimes e_2 Q e_1 \longrightarrow e_1 Q e_1, \\
& e_2 Q e_1 \otimes e_1 Q e_2 \longrightarrow e_2 Q e_2
\end{aligned}$$ equal the ordinary wedge product $V \otimes V \rightarrow \Lambda^2 V$.
The entire theory of Koszul algebras can be carried out over any semisimple base algebra, such as $R_2 = R e_1 \oplus R e_2$ (as already noticed in [@beilinson-ginsburg-schechtman88]). With respect to the exposition in Section \[subsec:koszul\], the main change needed is that all tensor products should be taken over the base algebra. Our original description shows that $Q$ is quadratic in these terms, and in fact:
$Q$ is Koszul.
Consider the ${\mathbb{Z}}/2$ action on $V$ where the nontrivial element as by $-1$. It follows from that $$\label{eq:lambda-q}
Q {\cong}\Lambda(V) \rtimes {\mathbb{Z}}/2$$ as an algebra over $R[{\mathbb{Z}}/2] {\cong}R_2$. Of course, this isomorphism is not compatible with the grading of $Q$ and the natural grading of the exterior algebra. However, by Addendum \[th:double-grading\] the discrepancy is irrelevant for deciding whether $Q$ is Koszul or not. On the other hand, the Koszulness of $\Lambda(V) \rtimes {\mathbb{Z}}/2$ is well-known, being a minor variation on the basic case of the exterior algebra itself (for the dual case of polynomial algebras twisted by finite groups, see [@gordon08 Section 2.7]).
The quadratic dual $Q^!$ of $Q$ (again, taken over $R_2$) is based on the quiver $$\label{eq:dual-quiver}
\xymatrix{\stackrel{1}{\bullet} \ar@<.5em>@/^2pc/[rr]^-{w_3^\vee} \ar@/^2pc/[rr]_-{w_4^\vee} && \ar@/^2pc/[ll]_-{w_1^\vee} \ar@<.5em>@/^2pc/[ll]^-{w_2^\vee} \stackrel{2}{\bullet}}$$ where $w_1^\vee,w_2^\vee$ have degree $0$, and $w_3^\vee,w_4^\vee$ degree $1$. This is actually isomorphic to but we avoid making the identification, and in any case the relations defining $Q^!$ are different: $$w_3^\vee w_2^\vee - w_4^\vee w_1^\vee = 0, \quad w_1^\vee w_4^\vee - w_2^\vee w_3^\vee = 0.$$ As in , if one forgets the grading then $Q^! {\cong}\mathit{Sym}(V^\vee) \rtimes {\mathbb{Z}}/2$.
\[th:hh-computation-1\] $\mathit{HH}^i(Q,Q[2-i])$ vanishes for $i \geq 5$, and $\mathit{HH}^i(Q,Q[3-i])$ vanishes for $i \geq 7$.
We need to adapt the previous discussion slightly to the framework over $R_2$. The relevant complex computing the Hochschild cohomology is now $$(Q^! \otimes_{R_2} Q)_{diag} = \bigoplus_{i,j=1,2} e_i Q^! e_j \otimes_R e_j Q e_i,$$ where the differential remains the same as before. [*If $i$ is even*]{}, any path in $Q^!$ of length $i$ has degree $i/2$, which implies that $\mathit{HH}^i(Q,Q[j]) = 0$ for $i+j < i/2$. [*If $i$ is odd*]{}, the minimal degree of a path of length $i$ in $Q^!$ is $(i-1)/2$, but the paths having this minimal degree all lie in $e_2 Q^! e_1$, whereas $e_1 Q e_2$ is concentrated in degree $1$. Therefore, $\mathit{HH}^i(Q,Q[j]) = 0$ for $i+j < (i+1)/2$.
\[th:hh-computation-2\] $\mathit{HH}^3(Q,Q[-1]) = 0$, and $\mathit{HH}^4(Q,Q[-2]) {\cong}\mathit{Sym}^4(V^\vee)$.
The $\mathit{HH}^3$ case can be carried out by an explicit calculation, which we omit (software for doing such calculations is available from the author’s homepage). For $\mathit{HH}^4$ we can follow a more conceptual path. The argument from Example \[th:truncated\] shows that $$\label{eq:hh-lambda}
\mathit{HH}^4(Q,Q[-2]) {\cong}\mathit{HH}^4(\Lambda(V) \rtimes {\mathbb{Z}}/2, \Lambda(V) \rtimes {\mathbb{Z}}/2[-4]).$$ The Hochschild cohomology of $\Lambda(V) \rtimes {\mathbb{Z}}/2$ is isomorphic (with a suitable adjustment in the bigrading) to that of its Koszul dual $Sym(V^\vee) \rtimes {\mathbb{Z}}/2$. The particular group corresponds to the length $4$ piece of the center of the Koszul dual, which is just $Sym^4(V^\vee)$.
Deformations
------------
An $A_\infty$-deformation of $Q$ is an $A_\infty$-structure $\{\mu^*\}$ which respects the grading and $R_2$-bimodule structure, and whose starting terms are $$\mu^1(x) = 0, \;\; \mu^2(x_2,x_1) = (-1)^{|x_1|} x_2 x_1.$$ In particular, $\{\mu^*\}$ is necessarily cohomologically unital. As part of the higher order product structure, we then have maps $$\label{eq:mu4}
\begin{aligned}
& \mu^4_{[12121]}: e_1 Q e_2 \otimes e_2 Q e_1 \otimes e_1 Q e_2 \otimes e_2 Q e_1 {\cong}V^{\otimes 4} \longrightarrow R e_1 {\cong}R, \\
& \mu^4_{[21212]}: e_2 Q e_1 \otimes e_1 Q e_2 \otimes e_2 Q e_1 \otimes e_1 Q e_2 {\cong}V^{\otimes 4} \longrightarrow R e_2 {\cong}R.
\end{aligned}$$ where the identifications have been applied. Suppose that $\mu^3 = 0$. Because of the $A_\infty$-equations, the two order $4$ expressions in have the same symmetric part, which we denote by $$\label{eq:p-poly}
p(v) = \mu^4_{[12121]}(v,v,v,v) = \mu^4_{[21212]}(v,v,v,v) \in Sym^4(V^\vee).$$
\[th:classify-deformations\] (i) $A_\infty$-deformations of $Q$ satisfying $\mu^3 = 0$ are classified up to isomorphism by : any polynomial can occur, and it determines the isomorphism class of the deformation. (ii) Any $A_\infty$-deformation of $Q$ is isomorphic to one which is strictly unital, and has $\mu^3 = 0$, $\mu^5 = 0$.
Most of this follows from Lemmas \[th:hh-computation-1\] and \[th:hh-computation-2\], together with the general classification theory of $A_\infty$-deformations [@kadeishvili88] (compare also the discussion in [@seidel03b Section 3]). The vanishing of $\mathit{HH}^3(Q,Q[-1])$ tells us that any $A_\infty$-deformation is equivalent to one with $\mu^3 = 0$. For general reasons, $\mu^4$ then defines a class in $\mathit{HH}^4(Q,Q[-2])$. We know that this group is isomorphic to $\mathit{Sym}^4(V^\vee)$, and one can check (by explicitly comparing the standard Hochschild complex with the one coming from Koszul duality) that this isomorphism takes $[\mu^4]$ to the polynomial $p$ defined above. Since $\mathit{HH}^i(Q,Q[2-i])$ vanishes for all $i>4$, $[\mu^4]$ determines the isomorphism class of the $A_\infty$-deformation completely. The obstructions to existence are given by maps $$\mathit{HH}^4(Q,Q[-2])^{\otimes d} \longrightarrow \mathit{HH}^{3+2d}(Q,Q[-2d]),$$ $d \geq 2$, of which the first one is the Gerstenhaber bracket (the others come from a suitably defined $L_\infty$-structure on Hochschild cohomology). In our case, all the spaces in which these take value vanish. The last statement follows by inspection of the inductive procedure which solves the obstruction problem.
For a given $p$, we denote by $Q_p$ the $A_\infty$-structure obtained by equipping $Q$ with the higher order products from Proposition \[th:classify-deformations\].
$Q_p$ is an $A_\infty$-algebra over $R_2$, or equivalently an $A_\infty$-category with two objects $X_1,X_2$. By construction, it is strictly unital and has $\mu^3_{Q_p} = \mu^5_{Q_p} = 0$. Of course, it is unique only up to $A_\infty$-isomorphism.
There is also an $A_\infty$-isomorphism $Q_p {\cong}Q_{\gamma^2 p}$ for any $\gamma \in R^\times$, obtained (for suitable choices on both sides) by multiplying the degree $k$ part of the algebra with $\gamma^k$. In particular, if $R$ is algebraically closed (or at least contains square roots), knowing $p$ up to nonzero multiples is sufficient to determine the $A_\infty$-algebra.
Let’s briefly consider the Hochschild cohomology $\mathit{HH}^*(Q_p,Q_p)$ (unlike that of a graded algebra like $Q$, this carries a single grading). The length filtration of the Hochschild complex yields a spectral sequence, starting with $E_1^{ij} = \mathit{HH}^i(Q,Q[j])$. Since $\mu^3_{Q_p}$ vanishes, the first potentially nontrivial differential is $d^3: E_1^{ij} \rightarrow E_1^{i+3,j-2}$, which is the Gerstenhaber bracket with $\mu^4_{Q_p}$. Moreover, since $\mu^5_{Q_p} = 0$, the next potentially nonzero differential is $d^5$. Here is a picture of all the nonzero entries in the lines $i+j = 1$ and $i+j = 2$, obtained by the same techniques as Lemmas \[th:hh-computation-1\] and \[th:hh-computation-2\]: $$\begin{array}{r|ccccc}
j = 2 & 0 \\
j = 1 & \Lambda^2(V) \oplus \Lambda^2(V) & 0 \\
j = 0 && \mathit{End}(V) & \mathit{Sym}^2(V^\vee) \\
j = -1 &&& 0 & 0 \\
j = -2 &&&& 0 & \mathit{Sym}^4(V^\vee) \\
\hline
& i = 0 & i = 1 & i = 2 & i = 3 & i = 4
\end{array}$$ The only possibly nontrivial differential in this picture is $d^3: \mathit{End}(V) \rightarrow \mathit{Sym}^4(V^\vee)$. This can be interpreted as the action of linear vector fields on the polynomial $p$, $d^3(Z) = L_Zp$. In particular:
\[th:generic\] Suppose that there is no nontrivial linear vector field which acts trivially on $p$. Then $\mathit{HH}^1(Q_p,Q_p) {\cong}\Lambda^2(V) \oplus \Lambda^2(V)$ is two-dimensional.
\[th:quasi-generators\] Inspection of the argument above allows one to approximately determine the form of two generators $[g_1]$, $[g_2]$ of $\mathit{HH}^1(Q_p,Q_p)$. Each of them is represented by a Hochschild cochain whose leading order term $g_k^0$ is a nontrivial element of $(e_k Q e_k)^1 {\cong}\Lambda^2(V)$, let’s say the standard elements $q_1 = w_3w_2 = -w_4w_1$ (for $k = 1$) and $q_2 = w_1w_4 = -w_2w_3$ (for $k = 2$). Moreover, the next order term $g_k^1$ can be chosen to be zero.
Some twisted complexes\[subsec:twisted\]
----------------------------------------
$Q$ can be thought of as a linear graded category with two objects $X_i$ corresponding to the vertices of the quiver, so that for instance $\mathit{hom}_Q(X_1,X_2) = e_2 Q e_1 {\cong}V$. The $A_\infty$-deformation $Q_p$ can the be viewed as an $A_\infty$-category with the same objects. The aim of the following discussion is to understand how the choice of $p$ affects the structure of the formal enlargement $Q_p^{\mathit{tw}}$ (and $Q_p^{\mathit{perf}}$). For that, it is useful to require $R$ to be algebraically closed, which we will do from now on.
For any nonzero $v \in V$, consider the twisted complex $C_v = \mathit{Cone}(v: X_1 \rightarrow X_2)$. We have $$\label{eq:endomorphism-of-c}
\mathit{hom}_{Q_p^{\mathit{tw}}}(C_v,C_v) = \begin{pmatrix} \Lambda^0(V) \oplus \Lambda^2(V)[-1] & V \\
V[-1] & \Lambda^0(V) \oplus \Lambda^2(V)[-1] \end{pmatrix}.
$$ The matrix notation here stands for taking the direct sum of the four graded vector spaces involved. Taking into account the fact that $\mu^1_{Q_p}$ and $\mu^3_{Q_p}$ vanish, the differential on is $$\begin{pmatrix} x_{11} & x_{12} \\ x_{21} & x_{22}
\end{pmatrix} \stackrel{\mu^1_{Q_p^{\mathit{tw}}}}{\longmapsto}
\begin{pmatrix}
x_{12} \wedge v & 0 \\
-v \wedge x_{11} - x_{22} \wedge v & v \wedge x_{12}
\end{pmatrix}$$ and the product is $$\begin{aligned} &
\begin{pmatrix} y_{11} & y_{12} \\ y_{21} & y_{22}
\end{pmatrix} \otimes
\begin{pmatrix} x_{11} & x_{12} \\ x_{21} & x_{22}
\end{pmatrix} \stackrel{\mu^2_{Q_p^{\mathit{tw}}}}{\longmapsto}
\begin{pmatrix}
y_{11} \wedge x_{11} + y_{12} \wedge x_{21} & y_{11} \wedge x_{12} + y_{12} \wedge x_{22} \\
y_{21} \wedge x_{11} + y_{22} \wedge x_{21} & y_{22} \wedge x_{22} + y_{21} \wedge x_{12}
\end{pmatrix} \\
& + \begin{pmatrix} \mu^4_{Q_p}(y_{12},v,x_{12},v) & \mu^4_{Q_p}(v,y_{12},v,x_{11}) \\
\mu^4_{Q_p}(y_{22},v,x_{12},v) + \mu^4_{Q_p}(v,y_{12},x_{22},v) + \mu^4_{Q_p}(v,y_{11},x_{12},v) & \mu^4_{Q_p}(v,y_{12},v,x_{12})
\end{pmatrix}.
\end{aligned}$$ Representative cochains for a basis in the cohomology of $\mu^1_{Q_p^{\mathit{tw}}}$ are $$\label{eq:endomorphism}
e = \begin{pmatrix} -1 & 0 \\ 0 & 1 \end{pmatrix}, \;\;
t = \begin{pmatrix} 0 & v \\ 0 & 0 \end{pmatrix}, \;\;
q = \begin{pmatrix} v \wedge v^* & 0 \\ 0 & v \wedge v^* \end{pmatrix}, \;\;
u = \begin{pmatrix} 0 & 0 \\ v^* & 0 \end{pmatrix},$$ where $v^* \in V$ satisfies $v^* \wedge v \neq 0$. The first generator is the identity element, with the sign due to convention. Some explicit products of the other generators are $$\label{eq:qp-products}
\begin{aligned}
& \mu^2_{Q_p^{\mathit{tw}}}(u,t) = \begin{pmatrix} -v \wedge v^* & 0 \\ 0 & 0 \end{pmatrix} =
{{\textstyle\frac{1}{2}}}\mu^1_{Q_p^{\mathit{tw}}}\begin{pmatrix} 0 & v^* \\ 0 & 0 \end{pmatrix} - {{\textstyle\frac{1}{2}}}q,
\\ &
\mu^2_{Q_p^{\mathit{tw}}}(t,u) = \begin{pmatrix} 0 & 0 \\ 0 & v \wedge v^* \end{pmatrix}
= {{\textstyle\frac{1}{2}}}\mu^1_{Q_p^{\mathit{tw}}}\begin{pmatrix} 0 & v^* \\ 0 & 0 \end{pmatrix} + {{\textstyle\frac{1}{2}}}q,
\\
& \mu^2_{Q_p^{\mathit{tw}}}(u,u) = 0, \\ &
\mu^2_{Q_p^{\mathit{tw}}}(t,t) = p(v)e.
\end{aligned}$$ On the cohomology level, this implies that $$\label{eq:cv-endomorphism}
H^0(\mathit{hom}_{Q_p^{\mathit{tw}}}(C_v,C_v)) {\cong}R[t]/(t^2-p(v)).$$ We have therefore shown:
\[th:cone-splitting\] $C_v$ splits into two summands in $Q_p^{\mathit{perf}}$ if and only if $p(v) \neq 0$. This allows one to reconstruct $p$ up to a scalar multiple from categorical data.
\[th:hochschild-cohomology-for-cones\] Just like the $A_\infty$-structure itself, a Hochschild cochain $g \in \mathit{CC}(Q_p,Q_p)$ has components $g^d_{[i_d\dots i_0]}$ for $d \geq 0$ and $i_0,\dots,i_d \in \{1,2\}$. The induced cochain $g^{\mathit{tw}}$ as in has in particular $$\label{eq:explicit-gamma}
\begin{aligned}
& g^{\mathit{tw},0} \in \mathit{hom}_{Q_p^{\mathit{tw}}}(C_v,C_v), \\
& g^{\mathit{tw},0} = \begin{pmatrix} -g^0_{[1]} & 0 \\ -g^1_{[21]}(v) & g^0_{[2]} \end{pmatrix}.
\end{aligned}$$ Suppose that $p$ satisfies the assumptions of Lemma \[th:generic\], and consider the generators $g_1,g_2$ from Addendum \[th:quasi-generators\]. The notation here is potentially confusing: $g_k$ is a whole Hochschild cocycle, whose components would be written as $(g_k)^d_{[i_d\dots i_0]}$. Suppose that $v^*$ is chosen in such a way that $v \wedge v^* = (g_k)^0_{[k]}$. By comparing with , one sees that $$\label{eq:anti-sign}
(g_1)^{\mathit{tw},0} = -(g_2)^{\mathit{tw},0} = \mu^2_{Q_p^{\mathit{tw}}}(u,t).$$
By carefully inspecting the argument leading to Lemma \[th:cone-splitting\], we can sharpen it to a criterion that determines $p$ on the nose, and also works in slightly more general circumstances. Suppose that $\tilde{Q}$ is an $A_\infty$-category with objects $\tilde{X}_1,\tilde{X}_2$, together with a fixed isomorphism $H(\tilde{Q}) {\cong}Q$ on the cohomology level. We will use the triangulated structure of $H^0(\tilde{Q}^{\mathit{tw}})$, following [@seidel04 Section 3] for the sign conventions used in establishing exact triangles.
\[th:exactly-complete-triangle\] Given $v \in V {\cong}\mathit{Hom}_{H^0(\tilde{Q}^{\mathit{tw}})}(\tilde{X}_1,\tilde{X}_2)$, complete it to an exact triangle $$\label{eq:tilde-exact-triangle}
\tilde{X}_1 \stackrel{v}{\longrightarrow} \tilde{X}_2 \longrightarrow \tilde{C}_v \longrightarrow \tilde{X}_1[1].$$ There is a unique (degree $0$) endomorphism $\tilde{t}$ of $\tilde{C}_v$ with the following two properties. The composition $$\label{eq:recover-v}
\tilde{X}_2 \longrightarrow \tilde{C}_v \stackrel{\tilde{t}}{\longrightarrow} \tilde{C}_v \longrightarrow \tilde{X}_1[1]$$ (where the first and last map are taken from the exact triangle) equals $v$; and $\tilde{t}^2$ is a multiple of the identity endomorphism. Moreover, that multiple is necessarily given by $\tilde{t}^2 = p(v)$, where $Q_p$ is the $A_\infty$-deformation of $Q$ quasi-isomorphic to $\tilde{Q}$.
The object $\tilde{C}_v$ is unique up to (non-canonical) isomorphisms which commute with the maps from $\tilde{X}_2$ and to $\tilde{X}_1[1]$. Hence, the statement is independent of the specific choice of $\tilde{C}_v$. Without loss of generality, we can assume that $\tilde{Q} = Q_p$ for some $p$. By definition, $$\label{eq:archetypal-triangle}
X_1 \stackrel{v}{\longrightarrow} X_2 \xrightarrow{(0,e_{X_2})} C_v \xrightarrow{(-e_{X_1},0)} X_1[1]$$ is an exact triangle. With this and , one checks easily that $t$ is the unique endomorphism satisfying .
A perfect family\[subsec:tautological\]
---------------------------------------
It is a natural step to let the parameter $v$ vary. For simplicity, we will use the affine line (rather than the projective line) as a parameter space, setting $v = (1,s_2)$ with $s_2 \in R$. Take the double cover of the affine line ramified at the zero-locus of $p(1,s_2)$, and then remove the branch points. The outcome is a smooth curve ${\mathscr{S}}$ whose ring of functions is $$\label{eq:punctured-rr}
{\mathscr{R}}= R[s_1,s_1^{-1},s_2]/(s_1^2 - p(1,s_2)).$$ We equip this with the nowhere vanishing one-form $$\label{eq:punctured-rr-form}
\theta = -{{\textstyle\frac{1}{2}}}s_1^{-1} ds_2 \in \Omega^1_{\mathscr{R}}.$$
Let ${\mathscr{Q}}_p$ be the constant family of $A_\infty$-structures over ${\mathscr{S}}$ associated to $Q_p$. Consider the object of ${\mathscr{Q}}_p^{\mathit{tw}}$ given by ${\mathscr{C}}= \mathit{Cone}((1,s_2): X_1 \rightarrow X_2)$. By the same computation as in , taking the natural choice $v^* = (0,1)$ of generator linearly independent of $v$, we have $$H(\mathit{hom}_{Q_p^{\mathit{tw}}}({\mathscr{C}},{\mathscr{C}})) {\cong}{\mathscr{R}}[t,u]/(t^2-s_1^2).$$ After lifting the idempotent endomorphism $$\label{eq:projection}
{{\textstyle\frac{1}{2}}}(1 + s_1^{-1} t) \in H^0(\mathit{hom}_{Q_p^{\mathit{tw}}}({\mathscr{C}},{\mathscr{C}}))$$ to a homotopy idempotent, one associates to this a family ${\mathscr{M}}$ of perfect modules, which is a direct summand of the Yoneda image of ${\mathscr{C}}$. Use the generator $g_2$ from Addendum \[th:quasi-generators\] to define a deformation field $$\label{eq:elliptic-deformation-field}
\gamma = -2\theta \otimes g_2 \in \mathit{HH}^1({\mathscr{Q}}_p, \Omega^1_{\mathscr{R}}\otimes {\mathscr{Q}}_p).$$
\[th:example-family\] ${\mathscr{M}}$ follows $[\gamma]$ (in the sense of Definition \[th:deformation-field\]).
The deformation cocycle of ${\mathscr{C}}$ can be determined by applying to the trivial pre-connection. On the other hand, $\gamma^{\mathit{tw},0}$ can be computed as in Addendum \[th:hochschild-cohomology-for-cones\]. The result is $$\begin{aligned}
& \mathit{def}({\nabla\mkern-12mu/\mkern2mu}_{\mathscr{C}}) = -\partial_{s_2}(\delta_{\mathscr{C}}) ds_2 = 2\theta \otimes s_1 \partial_{s_2}(\delta_{\mathscr{C}}) = 2\theta \otimes s_1 u, \\
& \gamma^{\mathit{tw},0} = 2\theta \otimes \mu^2_{Q_p^{\mathit{tw}}}(u,t).
\end{aligned}$$ The corresponding elements for ${\mathscr{M}}$, at least on the cohomology level, can be computed by applying the projection (it doesn’t matter on which side, since there are no $\mathit{Hom}$s from one summand of ${\mathscr{C}}$ to the other), which indeed yields the same result in both cases: $${{\textstyle\frac{1}{2}}}(1 + s_1^{-1} t) ut = {{\textstyle\frac{1}{2}}}(s_1^{-1} t^2 + t) u = {{\textstyle\frac{1}{2}}}(s_1 + t) u = {{\textstyle\frac{1}{2}}}(1 + s_1^{-1}t) s_1 u.$$
We could add any multiple of $g_1 + g_2$ to our deformation field and still obtain the same result, in view of .
Elliptic curves\[subsec:elliptic\]
----------------------------------
The algebra $Q$ from Definition \[th:q-algebra\] arises in the following algebro-geometric context. Take some $p \in \mathit{Sym}^4(V^\vee)$ which is [*simple*]{}, meaning that it has four distinct zeros. This gives rise to a double branched cover $\pi: Y_p \rightarrow \mathbb{P}(V)$, whose total space is a smooth elliptic curve, embedded into the total space of the bundle ${\mathscr{O}}_{\mathbb{P}(V)}(2)$. The sheaf $\pi_*{\mathscr{O}}_{Y_p}$ decomposes into $\pm 1$ eigenspaces for the action of the covering transformation. These can be identified with $$\begin{aligned}
& (\pi_* {\mathscr{O}}_{Y_p})_{+1} {\cong}{\mathscr{O}}_{\mathbb{P}(V)}, \\
& (\pi_* {\mathscr{O}}_{Y_p})_{-1} {\cong}{\mathscr{O}}_{\mathbb{P}(V)}(-2) {\cong}\Omega^1_{\mathbb{P}(V)} \otimes \Lambda^2(V), \label{eq:anti-invariant}\end{aligned}$$ where the second part is obtained by taking functions linear on the fibres of ${\mathscr{O}}_{\mathbb{P}(V)}(2)$ and restricting them to $Y_p$. If $E_1,E_2$ are locally free sheaves on $\mathbb{P}(V)$, we have canonical isomorphisms $$\label{eq:pushforward}
\begin{aligned}
\mathit{Ext}^*_{Y_p}(\pi^*E_1,& \pi^*E_2) {\cong}H^*(Y_p,\pi^*E_1^\vee \otimes \pi^*E_2) \\
& {\cong}H^*(Y_p,\pi^*(E_1^\vee \otimes E_2)) \\
& {\cong}H^*(\mathbb{P}(V),(E_1^\vee \otimes E_2) \otimes \pi_*{\mathscr{O}}_{Y_p}) \\ &
{\cong}H^*(\mathbb{P}(V),E_1^\vee \otimes E_2) \oplus H^*(\mathbb{P}(V), E_1^\vee \otimes E_2 \otimes \Omega_{\mathbb{P}(V)}^1) \otimes \Lambda^2(V) \\ &
{\cong}\mathit{Ext}^*_{\mathbb{P}(V)}(E_1,E_2) \oplus \mathit{Ext}^{1-*}_{\mathbb{P}(V)}(E_2,E_1)^\vee \otimes \Lambda^2(V),
\end{aligned}$$ where the last isomorphism uses Serre duality on $\mathbb{P}(V)$. Consider in particular $E_1 = {\mathscr{O}}_{\mathbb{P}(V)}$, $E_2 = {\mathscr{O}}_{\mathbb{P}(V)}(1) \otimes \Lambda^2(V)$, which has $\mathit{Hom}_{P(V^\vee)}(E_1,E_2) = V^\vee \otimes \Lambda^2(V) {\cong}V$ by definition. The computation above (with $E_1$ and $E_2$ exchanged) shows that $\mathit{Ext}^1_{Y_p}(\pi^*E_2,\pi^*E_1) {\cong}V^\vee \otimes \Lambda^2(V) {\cong}V$ as well. Using this and similar arguments (compare [@seidel-thomas99 Section 3c]) one sees that:
\[th:q-geometric\] We have an isomorphism of graded algebras, $\mathit{Ext}^*_{Y_p}(\pi^*E_1 \oplus \pi^*E_2,\pi^*E_1 \oplus \pi^*E_2) {\cong}Q$ (if one thinks of $Q$ as defined in , the isomorphism is canonical).
Let $D^b\mathit{Coh}(Y_p)$ be a suitable dg enhancement of the standard bounded derived category of coherent sheaves on $Y_p$. This category is closed under shifts, mapping cones and direct summands (the last-mentioned fact follows from the characterization of its objects as compact objects in a larger category [@bondal-vandenbergh02 Theorem 3.1.1]). Lemma \[th:q-geometric\] says that the subcategory of $D^b\mathit{Coh}(Y_p)$ with objects $\pi^*E_1,\pi^*E_2$ is quasi-isomorphic to an $A_\infty$-deformation of $Q$, which by Lemma \[th:classify-deformations\] can be chosen to be $Q_{\tilde{p}}$ for some polynomial $\tilde{p}$. We then have a cohomologically full and faithful $A_\infty$-functor from $Q_{\tilde{p}}$ to $D^b\mathit{Coh}(Y_p)$, taking $X_i$ to $\pi^*E_i$. Moreover, since the $\pi^*E_i$ are split-generators of $D^b\mathit{Coh}(Y_p)$, this functor extends to a quasi-equivalence $$\label{eq:quiver-embedding}
Q_{\tilde{p}}^{\mathit{perf}} \stackrel{{\simeq}}{\longrightarrow} D^b\mathit{Coh}(Y_p).$$ Unsurprisingly,
\[th:p-p\] The polynomial $\tilde{p}$ in is a nonzero constant multiple of $p$.
Any $A_\infty$-functor take cones to cones, up to quasi-isomorphism. In particular, $C_v$ maps to the cone of the morphism $\pi^*E_1 \rightarrow \pi^*E_2$ corresponding to $v$. That cone, which we denote by $\tilde{C}_v$, is isomorphic to the structure sheaf of the scheme-theoretic fibre $\pi^{-1}([v])$; more canonically, it can be written as that structure sheaf tensored with the one-dimensional vector space $V/Rv$. If $p(v) \neq 0$, the fibre consists of two closed points, hence has a nontrivial idempotent endomorphism. On the other hand, there are four points for which $p(v) = 0$, and where the scheme-theoretic fibre is a single fat point. Lemma \[th:cone-splitting\] then yields the desired result.
We can refine this observation slightly. One can compute geometrically that $$\begin{aligned}
& \mathit{Hom}_{Y_p}(\tilde{C}_v,\tilde{C}_v) {\cong}R \oplus (Rv)^{\otimes 2}\label{eq:geom-cone-end}, \\
& \mathit{Hom}_{Y_p}(\pi^*E_2,\tilde{C}_v) {\cong}R \oplus (Rv)^{\otimes 2} \label{eq:in-cone}, \\
& \mathit{Ext}^1_{Y_p}(\tilde{C}_v,\pi^*E_1) {\cong}(Rv)^{\vee} \oplus Rv \label{eq:out-cone}.\end{aligned}$$ In the ring structure of , the first summand is generated by the identity endomorphism, and the square of $v \otimes v$ in the second summand is exactly $p(v)$ times the identity. The action of $v \otimes v$ on by left multiplication is given by $(1,0) \mapsto (0,v \otimes v)$ (and correspondingly $(0,v \otimes v) \mapsto (p(v),0)$, so as to satisfy the given relation). Both groups and contain canonical elements, which are parts of the obvious exact triangle involving $\tilde{C}_v$, and those are just the generators of the first summands (in the case of , this is the generator of $(Rv)^\vee$ dual to $v \in Rv$). Finally, the composition $$\mathit{Ext}^1_{Y_p}(\tilde{C}_v,\pi^*E_1) \otimes \mathit{Hom}_{Y_p}(\pi^*E_2,\tilde{C}_v) \longrightarrow \mathit{Ext}^1_{Y_p}(\pi^*E_2,\pi^*E_1)$$ is given by the obvious maps $(Rv)^{\vee} \otimes (Rv)^{\otimes 2} \rightarrow Rv \subset V$, $Rv \otimes R \rightarrow Rv \subset V$. By putting together those facts, one sees that taking $\tilde{t} = v \otimes v$ exactly satisfies the assumptions of Lemma \[th:exactly-complete-triangle\] (modulo tedious sign verifications, which we have omitted), and therefore that:
The constant from Lemma \[th:p-p\] is trivial, meaning that $p = \tilde{p}$.
This geometric interpretation also throws some light on Lemma \[th:generic\]. In view of the derived invariance of Hochschild cohomology, one has $$\label{eq:hh-geometric-1}
\mathit{HH}^d(Q_p,Q_p) {\cong}\mathit{HH}^d(D^b\mathit{Coh}(Y_p)) {\cong}\mathit{HH}^d(Y_p) {\cong}\bigoplus_{i+j=d} H^i(Y_p,\Lambda^j TY_p),$$ where $TY_p$ is the tangent bundle. In particular, $\mathit{HH}^1(Q_p,Q_p) {\cong}H^1(Y_p,{\mathscr{O}}_{Y_p}) \oplus H^0(Y_p,T_{Y_p})$ is indeed two-dimensional.
We make a slight digression, whose aim is to explain our original computations of $\mathit{HH}^*(Q,Q)$ geometrically. One can associate to an arbitrary $p \in \mathit{Sym}^4(V^\vee)$ a subscheme $Y_p$ of the total space of ${\mathscr{O}}_{\mathbb{P}(V)}(2)$, and Lemma \[th:q-geometric\] still holds. So does once one replaces the derived category of coherent sheaves with its subcategory of perfect complexes. In particular, we can set $p = 0$, in which case the “double branched cover” $Y_0$ is the first order infinitesimal neighbourhood of the zero-section in ${\mathscr{O}}_{\mathbb{P}(V)}(2)$. Using the action of $R^\times$ by fibrewise rescaling, one can show that the resulting $A_\infty$-structure on $Q$ is formal (this is a well-known idea, in a sense going back to [@deligne-griffiths-morgan-sullivan75], see [@seidel-solomon10 Remark 7.6] and [@lekili-perutz11] for recent occurrences). Hence, $$\label{eq:hh-geometric}
\bigoplus_{i+j = d} \mathit{HH}^i(Q,Q[j]) {\cong}\mathit{HH}^d(Y_0),$$ where the right hand side can be written as the hypercohomology of a complex of sheaves on $\mathbb{P}(V)$ (locally quasi-isomorphic to the Hochschild complex of the ring $R[t_1,t_2]/t_2^2$), making it easily amenable to computation. Moreover, the equivariant version of the same Hochschild cohomology recovers the bigrading on the left hand side of .
Identify $V = R^2$ with coordinates $(v_1,v_2)$. Consider the affine chart for the total space of ${\mathscr{O}}_{\mathbb{P}(V)}(2)$ with coordinates $(s_1,s_2)$, which is such that for $s_2 = v_2/v_1$ for the underlying point $[v_1:v_2] \in \mathbb{P}(V)$, and the section $s_1 = 1$ corresponds to the quadratic polynomial $v_1^2$. In this chart, $Y_p$ has equation $s_1^2 = p(1,s_2)$. The object $\tilde{C}_v$ constructed above, for $v = (1,s_2)$, is the structure sheaf of the ideal obtained by additionally setting $s_2$ to a specific value, and its endomorphism $\tilde{t}$ is multiplication by $s_1$. If $p(1,s_2) \neq 0$, the idempotent endomorphism $\frac{1}{2}(1 + s_1^{-1} \tilde{t})$ of $\tilde{C}_v$ singles out a direct summand, which is the structure sheaf at the point $(s_1,s_2)$. Applying Lemma \[th:example-family\] to this, we get a perfect family of sheaves on $Y_p$ parametrized by the curve ${\mathscr{S}}$ from , which is itself an affine open part of $Y_p$, and such that the fibre of the family at $(s_1,s_2)$ is isomorphic to the structure sheaf of that point. This justifies calling it a “tautological family”.
The canonical bundle of the total space of ${\mathscr{O}}_{\mathbb{P}(V)}(2)$ is the pullback of ${\mathscr{O}}_{\mathbb{P}(V)}(-2) \otimes \Omega^1_{\mathbb{P}(V)} {\cong}{\mathscr{O}}_{\mathbb{P}(V)}(-4) \otimes \Lambda^2(V)^\vee$. Hence, fixing a symplectic form on the vector space $V$ singles out a two-form with poles exactly along $Y_p$, whose residue is then a nowhere vanishing one-form on $Y_p$. Returning to the identification $V = R^2$ and taking the symplectic form to be the standard form $dv_1 \wedge dv_2$, one finds that the restriction of the associated one-form to ${\mathscr{S}}\subset Y_p$ is precisely , since that satisfies $\theta \wedge d(s_1^2 - p(1,s_2)) = ds_1 \wedge ds_2$.
A universal construction\[subsec:universal-bimodule\]
-----------------------------------------------------
We will now give an alternative construction of the universal family and relates ones. Any $N \in \mathit{Ob}\, D^b\mathit{Coh}(Y_p \times Y_p)$ defines a Fourier-Mukai functor $K_N$ from $D^b\mathit{Coh}(Y_p)$ to itself. Its action on objects is $K_N(E) = (q_2)_*(q_1^*E \otimes N)$, where $q_k: Y_p \times Y_p \rightarrow Y_p$ are the projections. To explain the interaction of this with the description of the derived category, we find it convenient to reverse directions of the arrows, which means to consider the pullback functor $$\label{eq:pullback-sheaf-1}
D^b\mathit{Coh}(Y_p) \longrightarrow Q_p^{\mathit{mod}}.$$ The image of an object $F$ is a module $N$ with $H(N(X_i)) = \mathit{Hom}^*_{Y_p}(E_i,F)$. One shows easily that is cohomologically full and faithful, and in fact a quasi-equivalence to the subcategory $Q_p^{\mathit{perf}}$ of perfect modules, which is inverse to . There is a similar pullback functor $$D^b\mathit{Coh}(Y_p \times Y_p) \longrightarrow (Q_p,Q_p)^{\mathit{mod}}.$$ This maps $N$ to a bimodule $P$ whose cohomology is $H(P(X_i,X_j)) = \mathit{Hom}^*_{Y_p \times Y_p}(E_j \boxtimes E_i^\vee,N) {\cong}\mathit{Hom}^*_{Y_p}(E_i,(q_2)_*(q_1^*E_j \otimes N))$. For instance, the structure sheaf of the diagonal maps to the diagonal bimodule. Note also that if we consider $P(\cdot,X_j)$ just as a right $Q_p$-module, it is quasi-isomorphic to the image of the sheaf $(q_2)_*(q_1^*E_j \otimes N)$ under . This implies that $P$ is always right perfect. Finally, the following diagram is commutative up to homotopy: $$\label{eq:fm-versus-convolution}
\xymatrix{
\ar[d] D^b\mathit{Coh}(Y_p) \ar[rr]^-{K_N} && D^b\mathit{Coh}(Y_p) \ar[d] \\
Q_p^{\mathit{mod}} \ar[rr]^-{K_P} && Q_p^{\mathit{mod}}
}$$ In the top row of this, $K_N$ is the Fourier-Mukai functor, whereas on the bottom row we have the tensor product functor $K_P$. We will in fact only need to know commutativity of this on the level of quasi-isomorphism classes of objects, which is somewhat easier than the full statement.
The same observations hold for families. Let ${\mathscr{Q}}_p$ be the constant family of $A_\infty$-structures over ${\mathscr{S}}$ with fibre $Q_p$. There are functors $$\begin{aligned}
\label{eq:family-modules} & D^b\mathit{Coh}({\mathscr{S}}\times Y_p) \longrightarrow {\mathscr{Q}}_p^{\mathit{mod}}, \\
& D^b\mathit{Coh}({\mathscr{S}}\times Y_p \times Y_p) \longrightarrow ({\mathscr{Q}}_p,{\mathscr{Q}}_p)^{\mathit{mod}}, \label{eq:kernels-to-bimodules-2}\end{aligned}$$ of which the first is an equivalence to the subcategory of perfect families, and the second one at least lands in the subcategory of right perfect families of bimodules. Any object ${\mathscr{N}}$ of $D^b\mathit{Coh}({\mathscr{S}}\times Y_p \times Y_p)$ defines a functor $K_{\mathscr{N}}: D^b\mathit{Coh}(Y_p) \longrightarrow D^b\mathit{Coh}({\mathscr{S}}\times Y_p)$, which can be thought of as a family of Fourier-Mukai functors parametrized by ${\mathscr{S}}$. If ${\mathscr{P}}$ is the image of ${\mathscr{N}}$ under , we have a tensor product functor $K_{\mathscr{P}}: Q_p^{\mathit{mod}} \longrightarrow {\mathscr{Q}}_p^{\mathit{mod}}$, already considered (without the notation) in Section \[subsec:existence\]. The analogue of is $$\label{eq:fm-versus-convolution-2}
\xymatrix{
\ar[d] D^b\mathit{Coh}(Y_p) \ar[rr]^-{K_{\mathscr{N}}} && D^b\mathit{Coh}({\mathscr{S}}\times Y_p) \ar[d] \\
Q_p^{\mathit{mod}} \ar[rr]^-{K_{\mathscr{P}}} && {\mathscr{Q}}_p^{\mathit{mod}}
}$$ where the vertical arrows are . So far the discussion has been essentially limited to abstract nonsense, but now we want to draw some consequences more specific to the case of elliptic curves.
Fix a point $s \in {\mathscr{S}}\subset Y_p$, and give $Y_p$ its unique structure of an elliptic curve with $s$ as the neutral element. The graph of the addition morphism $\Sigma: Y_p \times Y_p \rightarrow Y_p$ restricts to a smooth subvariety $\{(s,y_1,y_2) \in {\mathscr{S}}\times Y_p \times Y_p \, : \, y_2 = \Sigma(s,y_1)\}$. Let ${\mathscr{N}}$ be the structure sheaf of that subvariety, and ${\mathscr{P}}$ its image under . This is a right perfect family of $Q_p$-bimodules parametrized by ${\mathscr{S}}$, whose fibre at $s$ is quasi-isomorphic to the diagonal bimodule.
\[th:fm-follows\] ${\mathscr{P}}$ follows the deformation field $[\gamma]$ from .
At any point $s \in {\mathscr{S}}$, ${\mathscr{P}}_s$ is the graph of an autoequivalence of $Q_p^{\mathit{perf}}$. This implies that the maps $\mathit{HH}^*(Q_p,Q_p) \longrightarrow H^*(\mathit{hom}_{(Q_p,Q_p)^{\mathit{mod}}}({\mathscr{P}}_s,{\mathscr{P}}_s))$ considered in are both isomorphisms. The same then holds in the entire family, which means that is an isomorphism. By definition, we have therefore shown that ${\mathscr{P}}$ follows [*some*]{} deformation field $[\tilde{\gamma}] \in \mathit{HH}^1({\mathscr{Q}}_p, \Omega^1_{\mathscr{R}}\otimes {\mathscr{Q}}_p)$.
We will now argue indirectly based on Corollary \[th:universal-family\] and the Hochschild cohomology computation in Lemma \[th:generic\]. The sheaf $E_1 = {\mathscr{O}}_{Y_p}$ is invariant under translations, which means that its Fourier-Mukai convolution with ${\mathscr{N}}$ yields a constant family. By , this implies that $X_1 \otimes_{{\mathscr{Q}}_p} {\mathscr{P}}$ is a constant family, hence that the component $\tilde{\gamma}^0_{[1]} \in e_1{\mathscr{Q}}_p e_1$ must be zero. If we now take the skyscraper sheaf at the point $s$ as our starting object, the outcome of $K_{\mathscr{N}}$ is again a “tautological” family, meaning that the fibre at any point of ${\mathscr{S}}$ is isomorphic to the skyscraper sheaf at that point. Since ${\mathscr{S}}$ is affine, any two such families are actually isomorphic. By comparing this with Lemma \[th:example-family\], one sees that $\tilde{\gamma}^0_{[2]} \in e_2{\mathscr{Q}}_p e_2$ must be equal to $-2\theta$.
The category $Q_p$ has a nonempty set of periodic elements, in the sense of Definition \[th:per\]. More specifically, if $[g_2] \in \mathit{HH}^1(Q_p,Q_p)$ is as in Addendum \[th:quasi-generators\], then $[-2g_2] \in \mathit{Per}(Q_p,\bar{{\mathscr{S}}},\bar{\theta})$. Here, $\bar{{\mathscr{S}}}$ is the smooth closure of ${\mathscr{S}}$ (obviously isomorphic to $Y_p$), and $\bar\theta$ the extension of $\theta$ to that closure.
This is a direct consequence of Lemma \[th:fm-follows\] and Corollary \[th:universal-family\]. Now recall that the category $Q_p^{\mathit{perf}} {\simeq}D^b\mathit{Coh}(Y_p)$ carries an action of $\mathit{SL}_2({\mathbb{Z}})$ (in a weak sense, and ignoring even shifts) [@mukai81 Remark 3.15]. This acts on Hochschild cohomology and also maps families to families. From this, one gets the following:
\[th:periodic-lattice\] The set $\mathit{Per}(Q_p,\bar{{\mathscr{S}}},\bar{\theta}) \subset \mathit{HH}^1(Q_p,Q_p)$ contains $m_1[g_1] + m_2[g_2]$ for all $m_1 \in {\mathbb{Z}}$, $m_2 \in m_1 + 2{\mathbb{Z}}$.
Theta functions
---------------
For the remainder of this section, we work exclusively over the field $R$ from . Elliptic curves over fields like $R$ can be studied using methods of non-archimedean analytic geometry. The comparison with ordinary algebraic geometry over $R$ is provided by a suitable GAGA theorem (see [@papikian05; @conrad08] for introductory accounts). We will borrow the intuition from there, but otherwise proceed by direct computation, avoiding abstract tools as much as possible.
Let $F$ be the ring of Laurent series over $R$ in one variable $t$, and which have “infinite convergence radius” (the non-archimedean analogue of holomorphic functions on ${\mathbb{C}}^*$). Explicitly, this means that $$\label{eq:tate}
F = \left\{
\begin{aligned}
& f(t) = c_0 \hbar^{m_0} t^{n_0} + c_1 \hbar^{m_1} t^{n_1} + \cdots, \\
& c_k \in {\mathbb{C}}, \;\; m_k \in {\mathbb{R}}, \;\; n_k \in {\mathbb{Z}}, \;\; \lim_{k \rightarrow \infty} m_k + A n_k = +\infty \text{ for any $A \in {\mathbb{R}}$}
\end{aligned}
\right\}.$$ We will be particularly interested in the theta-functions (a standard reference is [@mumford85], but our notation is a little different) $$\label{eq:theta-functions}
\vartheta_{n,k}(t) = \sum_{i \in n{\mathbb{Z}}+k} \hbar^{i^2/2n} t^i,$$ where $n$ is a positive integer, and $k \in {\mathbb{Z}}/n{\mathbb{Z}}$. These functions satisfy $$\begin{aligned}
\label{eq:periodicity}
& \vartheta_{n,k}(\hbar t) = \hbar^{-n/2} t^{-n} \vartheta_{n,k}(t) && \text{(periodicity)}, \\
\label{eq:fractional-period} & \vartheta_{n,k}(\hbar^{1/n} t) = \hbar^{-1/2n} t^{-1} \vartheta_{n,k+1}(t) && \text{(fractional periodicity)}, \\
& \vartheta_{n,k}(t^{-1}) = \vartheta_{n,n-k}(t) && \text{(symmetry)}. \label{eq:theta-symmetry}\end{aligned}$$ The simplest example, $\vartheta_{1,1}(t) = \vartheta(t)$, is the common or garden Jacobi theta function. On the next level $n = 2$, one has two functions $\vartheta_{2,1}(t)$ and $\vartheta_{2,2}(t)$, which besides are related by the addition formula $$\label{eq:theta-factorization}
\vartheta_{2,1}(u) \vartheta_{2,1}(t) + \vartheta_{2,2}(u) \vartheta_{2,2}(t) = \vartheta(u t)\vartheta(u^{-1} t).$$ As a consequence of that formula, we have $$\label{eq:special-doubling}
\begin{aligned}
& \vartheta_{2,2}(\hbar^{1/2})\vartheta_{2,1}(t) - \vartheta_{2,1}(\hbar^{1/2})\vartheta_{2,2}(t) = -\hbar^{-1/4}\vartheta(-t)^2, \\
& \vartheta_{2,2}(-\hbar^{1/2})\vartheta_{2,1}(t) - \vartheta_{2,1}(-\hbar^{1/2})\vartheta_{2,2}(t) = \hbar^{-1/4}\vartheta(t)^2, \\
& \vartheta_{2,2}(1)\vartheta_{2,1}(t) - \vartheta_{2,1}(1)\vartheta_{2,2}(t) =
\hbar^{1/4} t \,\vartheta(-\hbar^{1/2}t)^2, \\
& \vartheta_{2,2}(-1)\vartheta_{2,1}(t) - \vartheta_{2,1}(-1)\vartheta_{2,2}(t) =
\hbar^{1/4}t \, \vartheta(\hbar^{1/2}t)^2.
\end{aligned}$$ We will need the duplication formula [@whittaker-watson p. 488] $$\label{eq:duplication}
\begin{aligned}
\vartheta(t)\vartheta(-t)\vartheta(\hbar^{1/2}t)\vartheta(-\hbar^{1/2}t) & = {{\textstyle\frac{1}{2}}}\vartheta(1) \vartheta(-1) \vartheta(\hbar^{1/2}) \vartheta(-\hbar^{1/2}t^2) \\
& = {{\textstyle\frac{1}{2}}}\hbar^{-1/8}t^{-1}\, \vartheta(1)\vartheta(-1) \vartheta(\hbar^{1/2}) (\vartheta_{4,1}(t) - \vartheta_{4,3}(t)).
\end{aligned}$$ We will also need an identity for the derivatives (in $t$-direction), $$\label{eq:derivative-identity}
\begin{aligned}
t(\vartheta_{2,2}'(t)\vartheta_{2,1}(t) - \vartheta_{2,1}'(t)\vartheta_{2,2}(t)) & = \vartheta_{4,1}(t)\vartheta_{4,3}'(1) + \vartheta_{4,3}(t)\vartheta_{4,1}'(1) \\
& = \vartheta_{4,3}'(1)(\vartheta_{4,1}(t) - \vartheta_{4,3}(t)).
\end{aligned}$$
\[th:unit-torus-polynomial\] The [*unit torus polynomial*]{} $p \in R[v_1,v_2]$ is $$\begin{aligned}
\label{eq:torus-polynomial}
& \begin{aligned}
p(v_1,v_2) & = c\,(\vartheta_{2,2}(\hbar^{1/2})v_2 - \vartheta_{2,1}(\hbar^{1/2})v_1)(\vartheta_{2,2}(-\hbar^{1/2})v_2 - \vartheta_{2,1}(-\hbar^{1/2})v_1)\\ & \qquad \qquad \qquad \cdot (\vartheta_{2,2}(1)v_2 - \vartheta_{2,1}(1)v_1)(\vartheta_{2,2}(-1)v_2 - \vartheta_{2,1}(-1)v_1) \\
& = c\,(\vartheta_{2,2}(\hbar^{1/2})^2 v_2^2 - \vartheta_{2,1}(\hbar^{1/2})^2 v_1^2)(
\vartheta_{2,2}(1)^2 v_2^2 - \vartheta_{2,1}(1)^2 v_1^2),
\end{aligned}
\intertext{where}
& \begin{aligned}
c & = -\hbar^{1/4} \vartheta(1)^{-2}\vartheta(-1)^{-2}\vartheta(\hbar^{1/2})^{-2} \vartheta_{4,3}'(1)^2.
\end{aligned}\end{aligned}$$
One can associate to this polynomial a smooth elliptic curve over $R$, by the double branched cover method from Section \[subsec:elliptic\]. For the most part, we will only look at the affine part ${\mathscr{S}}= \mathit{Spec}({\mathscr{R}})$ of that curve as in , and equip that with the one-form $\theta$ from (this extends to a regular one-form on the entire elliptic curve). We call this the [*unit torus curve*]{}. Points of ${\mathscr{S}}$ have a transcendental parametrization by $u \in R^\times$, called the [*theta parametrization*]{}: $$\label{eq:theta-parametrization}
\begin{aligned}
s_2 & = \vartheta_{2,1}(u)\,\vartheta_{2,2}(u)^{-1}, \\
s_1 & = -{{\textstyle\frac{1}{2}}}u s_2'(u) = {{\textstyle\frac{1}{2}}}\, u \vartheta_{2,2}(u)^{-2}(\vartheta_{2,2}'(u)\vartheta_{2,1}(u) - \vartheta_{2,1}'(u)\vartheta_{2,2}(u)), \\
& = {{\textstyle\frac{1}{2}}}\vartheta_{2,2}(u)^{-2} \vartheta_{4,3}'(1) (\vartheta_{4,1}(u) - \vartheta_{4,3}(u)).
\end{aligned}$$ The fact that these points satisfy the equation for ${\mathscr{S}}$ follows from the relations between theta-functions listed above. Note that, due to the periodicity property, $u$ and $\hbar^k u$ ($k \in {\mathbb{Z}}$) describe the same point (if we had set up the general machinery properly, this would yield the Tate uniformization of the closure of ${\mathscr{S}}$, which of course is the original $Y_p$, as a quotient $R^\times/h^{\mathbb{Z}}$). To be precise, we have to exclude the values $u = \pm i h^{k+1/2}$, which is where $\vartheta_{2,2}$ vanishes (these would be mapped to points at infinity) as well as $u = \pm \hbar^{k/2}$, which is where $\vartheta_{4,1}-\vartheta_{4,3}$ vanishes, see (those points would be mapped to the branch points which do not lie in ${\mathscr{S}}$ by definition). The involution $u \mapsto \hbar^{1/2}u$ corresponds to $(s_1,s_2) \mapsto (-s_1s_2^{-2},s_2^{-1})$, whereas $u \mapsto u^{-1}$ corresponds to $(s_1,s_2) \mapsto (-s_1,s_2)$. Note also that in this parametrization, the one-form is $\theta = u^{-1} du$.
\[th:double-cover\] Suppose that we change variables from $\hbar$ to $\hbar^2$ (an automorphism of the field $R$). Apply this to the coefficients of $p$ and denote the resulting polynomial by $\tilde{p}$, with its associated curve $\tilde{{\mathscr{S}}}$ and one-form $\tilde{\theta}$. Our claim is that then, the projective closure of $\tilde{{\mathscr{S}}}$ is an étale $m$-fold cover of that of ${\mathscr{S}}$, and that the given one-forms are compatible with the covering map. In fact, the covering transformation is the abovementioned involution $(\tilde{s}_1,\tilde{s}_2) \mapsto (-\tilde{s}_1\tilde{s}_2^{-2},\tilde{s}_2^{-1})$. This is obvious set-theoretically by comparing the theta-parametrizations, and one can derive from the abstract theory (or at least in principle, check by hand) that the resulting map is indeed algebraic.
A nonarchimedean model\[subsec:nonarchimedean\]
-----------------------------------------------
Nonarchimedean analytic geometry appears naturally in the context of mirror symmetry, as shown for torus fibrations in [@kontsevich-soibelman00]. Here, we will spell out a version of their construction for the case of elliptic curves. The ring $F$ from comes with an action of ${\mathbb{Z}}$, generated by the translation $$\label{eq:translation-action}
(Tf)(t) = f(\hbar t).$$ We want to consider equivariant $T$-modules, or equivalently (right) modules over the semidirect product algebra $F \rtimes {\mathbb{Z}}$ (whose generators are $f(t)$ and $\tau$, with relations $\tau f = T(f) \tau$). For any $d \in {\mathbb{Z}}$ one has the equivariant module $F(d)$, which is $F$ itself with the twisted ${\mathbb{Z}}$-action generated by $$(T(d)f)(t) = t^d f(\hbar t).$$ $\mathit{Hom}_{F \rtimes {\mathbb{Z}}}(F(d_0),F(d_1))$ is isomorphic to the subspace of those elements of $F$ which are invariant under $T(d_1-d_0)$. Similarly, using the projective resolution (the map is left multiplication) $$\label{eq:f-resolution}
\tilde{F}(d) = \Big\{F \rtimes {\mathbb{Z}}\xrightarrow{\mathit{id} - \tau t^{-d}} F \rtimes {\mathbb{Z}}\Big\}$$ of $F(d)$, one sees that $\mathit{Ext}^1_{F \rtimes {\mathbb{Z}}}(F(d_0),F(d_1))$ is isomorphic to the space of coinvariants for the same action. In fact, we have a subcomplex of $\mathit{hom}_{F \rtimes {\mathbb{Z}}}(\tilde{F}(d_0),\tilde{F}(d_1))$ quasi-isomorphic to the whole thing, and which is of the form $$\label{eq:t-subcomplex}
\Big\{F \xrightarrow{\mathit{id} - T(d_1-d_0)} F\Big\}[-1].$$
Consider $Z_1 = F$ and $Z_2 = F(2)$. Using the computational ideas which we have just explained, one easily shows that $$\label{eq:q-algebra-again}
\mathit{Ext}^*_{F \rtimes {\mathbb{Z}}}(Z_1 \oplus Z_2, Z_1 \oplus Z_2) {\cong}Q$$ is the graded algebra from Definition \[th:q-algebra\]. Hence, the underlying cochain level algebraic structure is quasi-isomorphic to $Q_p$ for some $p$. One can use the subcomplexes to write down a relatively simple (if still infinite-dimensional) dga model. In principle, the Homological Perturbation Lemma could be applied to that, as in [@polishchuk09], which would give a way of determining $p$ directly. However, there is also a more categorical approach, in parallel with that from Section \[subsec:elliptic\]. By , an explicit basis of $\mathit{Ext}^*_{F \rtimes {\mathbb{Z}}}(Z_1,Z_2)$ is given by the functions $$f_k(t) = \vartheta_{2,k}(\hbar^{-1/2}t) = t^{-1}\vartheta_{2,k+1}(t),$$ ($k = 1,2$). Given $u \in R^\times$, consider the linear combination $$\vartheta_{2,2}(u)f_1 + \vartheta_{2,1}(u)f_2 : Z_1 \longrightarrow Z_2.$$ This is always injective. If $u \notin \{\pm \hbar^{k/2}\, : \, k \in {\mathbb{Z}}\}$, its cokernel splits into a direct sum of two nontrivial objects. This is a consequence of , which shows that the map can be written as the product of two elements which generate distinct (and ${\mathbb{Z}}$-invariant) principal ideals. One can also check that this fails for the remaining values of $u$, where the cokernel is indecomposable. By applying Lemma \[th:cone-splitting\], it then follows that the homogeneous polynomial $p$ relevant to our situation must vanish at the points $(\vartheta_{2,2}(\pm \hbar^{1/2}),\vartheta_{2,1}(\pm \hbar^{1/2}))$ and $(\vartheta_{2,2}(\pm 1),\vartheta_{2,1}(\pm 1))$. Hence, it must be of the form for some nonzero constant $c$. One can adjust the isomorphism to make that constant equal to $1$, which shows:
\[th:nonarch\] The full subcategory of the (chain level) derived category of modules over $F \rtimes {\mathbb{Z}}$ with objects $F(0)$, $F(2)$ is quasi-isomorphic to $Q_p$, where $p$ is the unit torus polynomial.
The two-torus\[subsec:twotorus\]
--------------------------------
Take the unit area torus $$\begin{aligned}
& T = {\mathbb{R}}^2/{\mathbb{Z}}^2, \\
& \omega_T = dp \wedge dq
\end{aligned}$$ equipped with the constant holomorphic volume form $dz = dp + i\, dq$. The Fukaya category $\mathit{Fuk}(T)$ is a ${\mathbb{Z}}$-graded, cohomological unitally $A_\infty$-category over the field $R$ from . This is related by mirror symmetry to the derived category of an elliptic curve over $R$, see [@polishchuk-zaslow98; @abouzaid-smith09]. More directly, [@kontsevich-soibelman00] relates it to the ring $F \rtimes {\mathbb{Z}}$ described above. One could use either of these theorems to derive the desired results, but we prefer to argue by direct geometric computation. This means that our exposition is a little ad hoc, which is hopefully forgivable in view of the relative simplicity of the target geometry (for more details of the approach to Fukaya categories used here, see Section \[subsec:fukaya\] below; of course there is by now extensive literature on this topic).
Objects of $\mathit{Fuk}(T)$ are simple closed curves $L \subset T$, equipped with a grading with respect to the complex one-form $dz$, and with a local coefficient system $\xi \rightarrow L$ whose fibre is $R^r$ for some $r$, and whose holonomy lies in the subgroup $$\begin{gathered}
\label{eq:gl-subgroup}
\mathit{GL}_0(r,R) \stackrel{\mathrm{def}}{=} \{A = A_0 + A_1 \hbar^{m_1} + \cdots, \; \\ A_0 \in \mathit{GL}(r,{\mathbb{C}}), \text{and $A_i \in \mathit{Mat}(r,{\mathbb{C}})$, $m_i > 0$ for $i \geq 1$}\} \subset \mathit{GL}(r,R).\end{gathered}$$
The grading of $L$ induces an orientation. In the general definition of the Fukaya, one has an additional choice of [*Spin*]{} structure. Changing the [*Spin*]{} structure is the same as tensoring $\xi$ with a line bundle having holonomy $-1$. Hence, it is enough for us to consider circles with trivial [*Spin*]{} structures (ones which are compatible with the trivialization of $\mathit{TL}$).
The spaces of morphisms between two objects $L_0$ and $L_1$ is the Floer cochain space $\mathit{CF}^*(L_0,L_1)$. In the case where $L_0$ intersects $L_1$ transversally, one can set it up so that generators correspond bijectively to points $x \in L_0 \cap L_1$. More precisely, each such point has an absolute index $\mathrm{deg}(x) \in {\mathbb{Z}}$, which depends on the gradings. This has the property that $(-1)^{\mathrm{deg}(x)+1}$ is the local intersection number. Then, $x$ contributes a copy of $\mathit{Hom}(\xi_0,\xi_1)_x$ to $\mathit{CF}^*(L_0,L_1)$ in degree $\mathrm{deg}(x)$.
The general theory dictates that the contribution of $x$ is a copy of $\mathit{Hom}(\xi_0,\xi_1)_x \otimes o_x$, where $o_x$ is the orientation space, identified with $R$ in a way which is unique only up to sign. However, in the specific case where the target space is a surface, we know that [@seidel04 Equation (13.6)] $o_x {\cong}R$ canonically if $\mathrm{deg}(x)$ is even, and $o_x {\cong}(TL_1)_x \otimes_{\mathbb{R}}R$ otherwise. In the second case, we use the orientation of $L_1$ given by the grading to pick a preferred generator for $o_x$.
(0,0)![\[fig:abcd1\]](abcd1.pstex "fig:")
\#1\#2\#3\#4\#5[ @font ]{}
(1227,1408)(1189,-4607) (2301,-4561)[(0,0)\[lb\]]{} (1681,-4561)[(0,0)\[lb\]]{}
The first two objects relevant for our argument are $$L_1 = \{q = 0\}, \quad L_2 = \{q = -2 p\},$$ We equip them with gradings so that $\mathit{HF}^*(L_1,L_2)$ is concentrated in degree $0$, and so that the induced orientations are as shown in Figure \[fig:abcd1\]. Both should carry trivial local systems. Write $w_1, -w_2 \in \mathit{CF}^0(L_1,L_2)$ for the generators coming from the two intersection points $({{\textstyle\frac{1}{2}}},0)$ and $(0,0)$. We also write $w_4, w_3 \in \mathit{CF}^1(L_2,L_1)$ for the generators coming from the same two points. By introducing a perturbed version of one of our two $L_k$ and counting triangles, one shows that each of the products $$\label{eq:w-product}
\begin{aligned}
& [w_4] \cdot [w_1] \in \mathit{HF}^1(L_1,L_1) {\cong}H^1(L_1;R), \\
& [w_1] \cdot [w_4] \in \mathit{HF}^1(L_2,L_2) {\cong}H^1(L_2;R), \\
& -[w_2] \cdot [w_3] \in \mathit{HF}^1(L_2,L_2) {\cong}H^1(L_2;R), \\
& -[w_3] \cdot [w_2] \in \mathit{HF}^1(L_1,L_1) {\cong}H^1(L_1;R)
\end{aligned}$$ equals the generator of $H^1$ given by our choice of orientations. Hence:
\[th:product\] The cohomology level product satisfies the relations . This means that as a graded algebra, $\bigoplus_{i,j=1}^2 \mathit{HF}^*(L_i,L_j) {\cong}Q$.
Take any $u \in R^\times$, written as $u = \hbar^{m_0}a$ for some $m_0 \in {\mathbb{R}}$ and $a \in \mathit{GL}_0(1,R)$. We associate to this another object of the Fukaya category, as follows. The underlying curve is $$L_{3,u} = \{p = m_0\}.$$ It comes equipped with the grading such that $\mathit{HF}^*(L_1,L_{3,u})$, and then also $\mathit{HF}^*(L_2,L_{3,u})$, is concentrated in degree $0$ (the induced orientation is in negative $q$-direction). The rank $1$ local system $\xi_u$ on $L_{3,u}$ should have holonomy $a$ when going around the curve in positive $q$-direction. Both chain level morphism spaces $\mathit{CF}^0(L_1,L_{3,u})$ and $\mathit{CF}^0(L_2,L_{3,u})$ are canonically isomorphic to the fibre of this local system at the unique intersection points, which are $(m_0,0)$ and $(m_0,-2m_0)$, respectively. We identify these with $R$ as follows: $$\label{eq:xi-trivializations}
\parbox{35em}{
Pick an arbitrary trivialization $(\xi_u)_{(m_0,0)} {\cong}R$. Then, pick a trivialization of $(\xi_u)_{(m_0,-2m_0)}$ in such a way that parallel transport along the path $\{(m_0,-2m_0 t), \; : \; 0 \leq t \leq 1\}$ is multiplication with $\hbar^{m_0^2}$.
}$$ Denote the resulting generators by $z_{1,u} \in \mathit{CF}^0(L_1,L_{3,u})$, $z_{2,u} \in \mathit{CF}^0(L_2,L_{3,u})$. A triangle count (compare [@polishchuk-zaslow98 Section 4]) determines the product $$\label{eq:theta-product-1}
\begin{aligned}
& \mathit{HF}^0(L_2,L_{3,u}) \otimes \mathit{HF}^0(L_1,L_2) \longrightarrow \mathit{HF}^0(L_1,L_{3,u}), \\
& [z_{2,u}] \cdot [w_1] = \vartheta_{2,1}(u) [z_{1,u}], \\
& [z_{2,u}] \cdot [w_2] = -\vartheta_{2,2}(u) [z_{1,u}],
\end{aligned}$$ where the negative sign comes from our choice of $w_2$, rather than from any geometric aspect of the computation.
The intersection point $(m_0,0)$ contributes a copy of $(\xi_u^\vee)_{(m_0,0)}$ to $\mathit{CF}^1(L_{3,u},L_1)$. Take $y_{1,u}$ to be the generator dual to $z_{1,u}$. In the same way, we define a generator $y_{2,u} \in \mathit{CF}^1(L_{3,u},L_2)$ dual to $z_{2,u}$. Then, the cohomology level products $$\begin{aligned}
& [y_{1,u}] \cdot [z_{1,u}] \in \mathit{HF}^1(L_1,L_1) {\cong}H^1(L_1;R), \\
& [z_{1,u}] \cdot [y_{1,u}] \in \mathit{HF}^1(L_{3,u},L_{3,u}) {\cong}H^1(L_{3,u};R), \\
& [y_{2,u}] \cdot [z_{2,u}] \in \mathit{HF}^1(L_2,L_2) {\cong}H^1(L_2;R), \\
& [z_{2,u}] \cdot [y_{2,u}] \in \mathit{HF}^1(L_{3,u},L_{3,u}) {\cong}H^1(L_{3,u};R)
\end{aligned}$$ each equal the generator of $H^1$ singled out by the given orientations. Using this, , and the associativity of the product (on the cohomology level), one computes the following: $$\label{eq:cyclic-mu2}
\begin{aligned}
& \mathit{HF}^0(L_1,L_2) \otimes \mathit{HF}^1(L_{3,u},L_1) \longrightarrow \mathit{HF}^1(L_{3,u},L_2), \\
& [w_1] \cdot [y_{1,u}] = \vartheta_{2,1}(u) [y_{2,u}], \\
& [w_2] \cdot [y_{1,u}] = -\vartheta_{2,2}(u) [y_{2,u}],
\end{aligned}$$ and $$\label{eq:more-products}
\begin{aligned}
& \mathit{HF}^1(L_{3,u},L_1) \otimes \mathit{HF}^0(L_2,L_{3,u}) \longrightarrow \mathit{HF}^1(L_2,L_1), \\
& [y_{1,u}] \cdot [z_{2,u}] = \vartheta_{2,2}(u)[w_3] + \vartheta_{2,1}(u)[w_4].
\end{aligned}$$
From now on, assume that $u \notin \{\pm \hbar^{k/2} \, : \, k \in {\mathbb{Z}}\}$. Then $L_{3,u}$ and $L_{3,u^{-1}}$ are mutually orthogonal objects (which means that the Floer cohomology from one to the other is zero). The computations above, together with , show that the composition of any two of the following morphisms vanishes: $$\label{eq:123-triangle}
\xymatrix{L_1 \ar[rrrr]^-{\frac{\vartheta_{2,2}(u)[w_1] + \vartheta_{2,1}(u)[w_2]}{\vartheta_{4,3}'(1)(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))}} &&&& L_2 \ar[dll]^{\;\;\;\;\;\;\; ([z_{2,u}], [z_{2,u^{-1}}])} \\
&& L_{3,u} \oplus L_{3,u^{-1}} \ar[ull]_{[1]}^{([y_{1,u}], -[y_{1,u^{-1}}])\;\;\;\;\;\;\;} &&
}$$
\[th:exact-triangle\] The diagram is an exact triangle in the category $H^0(\mathit{Fuk}(T)^{tw})$.
To prove this, we need one part of the higher order $A_\infty$-structure, namely: $$\label{eq:mu3}
\begin{aligned}
& \mu^3_{\mathit{Fuk}(T)}: \mathit{CF}^0(L_2,L_{3,u}) \otimes \mathit{CF}^0(L_1,L_2) \otimes \mathit{CF}^1(L_{3,u},L_1) \longrightarrow \mathit{CF}^0(L_{3,u},L_{3,u}), \\
& \mu^3_{\mathit{Fuk}(T)}(z_{2,u},w_1,y_{1,u}) = \left( - u \vartheta_{2,1}'(u)\, + b \, \vartheta_{2,1}(u) \right) e_{3,u}, \\
& \mu^3_{\mathit{Fuk}(T)}(z_{2,u},w_2,y_{1,u}) = \left( u \vartheta_{2,2}'(u) - b \, \vartheta_{2,2}(u) \right) \, e_{3,u}. \\
\end{aligned}$$ Here, $e_{3,u} \in \mathit{CF}^0(L_{3,u},L_{3,u})$ represents the unit element, $\vartheta_{n,k}'(t)$ is the derivative of $\vartheta_{n,k}(t)$ with respect to the $t$ variable, and $b \in R$ is a constant depending on the details of the setup. This may require some more explanation. One adopts a Morse-Bott approach (see Section \[subsec:clean\] below for more explanations and references) in which generators of $\mathit{CF}^*(L_{3,u},L_{3,u})$ correspond to the unique minimum and maximum of a Morse function on $L_{3,u}$. Then is determined by counting triangles with sides on $(L_{3,u},L_1,L_2)$ and with an additional marked boundary point which goes through the minimum point. Figure \[fig:mu3\] shows the universal cover $\tilde{T} = {\mathbb{R}}^2$, with three triangles. If we choose the minimum to be the [*white*]{} dot, the resulting coefficient in $\mu^3_{\mathit{Fuk}(T)}(z_{2,u},w_1,y_{1,u})$ is $$\cdots - \hbar^{1/4} u + \hbar^{1/4} u^{-1} + 3 \hbar^{9/4} u^{-3} + \cdots = -u\vartheta_{2,1}'(u).$$ If we move the minimum to the [*black*]{} dot, the coefficient changes to $$\cdots -2 \hbar^{1/4} u + 0 \hbar^{1/4} u^{-1} + 2 \hbar^{9/4} u^{-2} + \cdots = -u \vartheta_{2,1}'(u) - \vartheta_{2,1}(u). $$ This explains the ambiguity in . The constant depends on our choice of minimum, but crucially, it remains the same in both $\mu^3$ computations.
(0,0)![\[fig:mu3\]](mu3.pstex "fig:")
\#1\#2\#3\#4\#5[ @font ]{}
(3510,5274)(1003,-4948) (2386,-2161)[(0,0)\[lb\]]{}
Because the underlying chain complexes have trivial differentials, one actually knows that the composition of any two maps in is zero on the chain level. Write $v \in \mathit{CF}^0(L_1,L_2)$ for the morphism appearing in . Then the maps $$\begin{aligned}
& (0,z_{2,u}) \in \mathit{hom}_{\mathit{Fuk}(T)^{\mathit{tw}}}^0(C_v,L_{3,u}) = \mathit{CF}^{-1}(L_1,L_{3,u}) \oplus \mathit{CF}^0(L_2,L_{3,u}), \\
& (y_{1,u},0) \in \mathit{hom}_{\mathit{Fuk}(T)^{\mathit{tw}}}^0(L_{3,u},C_v) = \mathit{CF}^1(L_{3,u},L_1) \oplus \mathit{CF}^0(L_{3,u},L_2)
\end{aligned}$$ are cocycles. From and one sees that their composition in one direction is $$\mu^2_{\mathit{Fuk}(T)^{\mathit{tw}}}((0,z_{2,u}),(y_{1,u},0)) =
\mu^3_{\mathit{Fuk}(T)}(z_{2,u},v,y_{1,u}) = e_{3,u}.$$ The analogous properties hold for the maps $$\begin{aligned}
& (0,z_{2,u^{-1}}) \in \mathit{hom}_{\mathit{Fuk}(T)^{\mathit{tw}}}^0(C_v,L_{3,u^{-1}}), \\
& (-y_{1,u^{-1}},0) \in \mathit{hom}_{\mathit{Fuk}(T)^{\mathit{tw}}}^0(L_{3,u^{-1}},C_v). \\
\end{aligned}$$ This shows that $L_{3,u} \oplus L_{3,u^{-1}}$ is a direct summand of $C_v$, but then a comparison of the sizes of the endomorphism rings, with one side computed as in , shows that the two must actually be quasi-isomorphic. Moreover, these quasi-isomorphisms fit in with .
\[th:twist-sequence\] Suppose that $u = \pm \hbar^{m_0}$ for some $m_0 \notin {{\textstyle\frac{1}{2}}}{\mathbb{Z}}$, which means that $L_{3,u^{\pm}}$ is a curve equipped with the nontrivial [*Spin*]{} structure. From the well-known exact triangle associated to a Dehn twist, and the Hamiltonian isotopy $$L_2 {\simeq}\tau_{L_{3,u^{-1}}}\tau_{L_{3,u}}(L_1),$$ one can derive the existence of a diagram involving the same objects as in . This can be extended to the more general case as follows. For $v \in R^\times$, let $\theta_v: \mathit{Fuk}(T) \rightarrow \mathit{Fuk}(T)$ be the functor obtained by tensoring with a flat $R$-line bundle on $M$ which has monodromy $v$ in $q$-direction (and trivial monodromy in $p$-direction). Then we have quasi-isomorphisms in $\mathit{Fuk}(T)^{\mathit{itw}}$, $$\begin{aligned}
L_2 & {\simeq}\theta_{-u^{-1}} \tau_{L_{3,u^{-1}}} \theta_{u^2} \tau_{L_{3,u}}(L_1), \\
& {\simeq}\theta_{u^{-1}} \tau_{L_{3,u^{-1}}}\theta_{u^2}(\mathit{Cone}(\theta_{-u^{-1}}(L_{3,u})[-1] \rightarrow L_1)) \\
& {\simeq}\theta_{-u^{-1}} \tau_{L_{3,u^{-1}}}(\mathit{Cone}(\theta_{-u}(L_{3,u})[-1] \rightarrow L_1)) \\
& {\simeq}\theta_{-u^{-1}}(\mathit{Cone}(\theta_{-u} L_{3,u} [-1] \oplus \theta_{-u} L_{3,u^{-1}}[-1] \rightarrow L_1)) \\ & {\simeq}\mathit{Cone}(L_{3,u}[-1] \oplus L_{3,u^{-1}}[-1] \rightarrow L_1).
\end{aligned}$$ However, from that point of view it is not easy to write down explicitly the maps involved, in particular the horizontal one (which would be given by counting holomorphic sections of a Lefschetz fibration).
By Lemma \[th:product\], we know that $\mathit{Fuk}(T)$ induces an $A_\infty$-structure on $Q$, which is necessarily quasi-isomorphic to $Q_{\tilde{p}}$ for some $\tilde{p}$.
\[th:p-p2\] $\tilde{p}$ is a nonzero constant multiple of the unit torus polynomial $p$.
The considerations above show that the cones $C_v$ split into orthogonal direct summands for all $v = (\vartheta_{2,2}(u),\vartheta_{2,1}(u)) \in V$, as long as $u \notin \pm h^{k/2}$. Hence $\tilde{p}$ is nonzero at all those points, by Lemma \[th:cone-splitting\].
We can use a symmetry trick to derive a little bit of additional information from the geometry. Consider translation by $({{\textstyle\frac{1}{2}}},0)$, which is a free symplectic involution of $T$ preserving $dz$. This maps each $L_i$ to itself, hence induces an action on $\mathit{HF}^*(L_i,L_j)$. The action on $\mathit{HF}^*(L_i,L_i) {\cong}H^*(L_i;R)$ is trivial, whereas that on $\mathit{HF}^*(L_1,L_2)$ maps $w_1 \rightarrow - w_2$ and vice versa. It is not hard to lift this to an action on $\mathit{Fuk}(T)$, and there is an equivariant analogue of Proposition \[th:classify-deformations\], which implies that the $\tilde{p}$ must be invariant under $(v_1,v_2) \mapsto (-v_2,-v_1)$. Another automorphism of the Fukaya category is the tensor product operation $\theta_{-1}$ which already appeared in Remark \[th:twist-sequence\]. This preserves both our $L_i$. The induced action on $\mathit{HF}^*(L_i,L_i)$ is trivial, whereas that on $\mathit{HF}^*(L_1,L_2)$ preserves one of the two generators, and reverses the sign of the other one (exactly which one this is depends on how one trivializes the restriction of the line bundle to our Lagrangian submanifolds). In the same way as before, this implies that $\tilde{p}$ is invariant under $(v_1,v_2) \mapsto (-v_1,v_2)$. As a consequence, its order of vanishing of at the four points $(\vartheta_{2,2}(\pm 1), \vartheta_{2,1}(\pm 1))$, $(\vartheta_{2,2}(\pm \hbar^{1/2}),\vartheta_{2,2}(\pm \hbar^{1/2}))$ must be the same. Since $\tilde{p}$ is nonzero everywhere else, it must have simple zeros at all the four points, which implies the desired result.
\[th:fukaya-torus\] The constant from Lemma \[th:p-p2\] is trivial, meaning that $\tilde{p} = p$.
Consider the degree zero endomorphism of $L_{3,u} \oplus L_{3,u^{-1}}$ given by $$\label{eq:geometric-t-tilde}
\tilde{t} = \frac{[e_{3,u}] \oplus [-e_{3,u^{-1}}]}{2 \vartheta_{4,3}'(1)(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))} \in \mathit{HF}^0(L_{3,u},L_{3,u}) \oplus \mathit{HF}^0(L_{3,u^{-1}}, L_{3,u^{-1}}).$$ Its square is clearly a multiple of the identity, and by we have $$\label{eq:dual-triple-product}
([y_{1,u}], [-y_{1,u^{-1}}]) \tilde{t} ([z_{2,u}],[z_{2,u^{-1}}]) = \frac{\vartheta_{2,2}(u)[w_3] + \vartheta_{2,1}(u)[w_4]}{\vartheta_{4,3}'(1)(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))},$$ which means that $\tilde{t}$ precisely satisfies the assumptions of Lemma \[th:exactly-complete-triangle\]. As a result, we get the following information concerning $\tilde{p}$: $$\label{eq:determine-ptilde}
\begin{aligned}
& \tilde{p}\left (1,\frac{\vartheta_{2,1}(u)}{\vartheta_{2,2}(u)}\right) \\
& = \frac{\vartheta_{4,3}'(1)^4(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))^4}{\vartheta_{2,2}(u)^4} \tilde{p}\left(\frac{\vartheta_{2,2}(u)}{\vartheta_{4,3}'(1)(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))}, \frac{\vartheta_{2,1}(u)}{\vartheta_{4,3}'(1)(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))}\right) \\
& = \frac{\vartheta_{4,3}'(1)^4(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))^4}{\vartheta_{2,2}(u)^4} \tilde{t}^2
= \frac{\vartheta_{4,3}'(1)^2(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))^2}{4 \vartheta_{2,2}(u)^4}.
\end{aligned}$$ In terms of the parametrization , this shows that $s_1^2 = \tilde{p}(1,s_2)$, which of course implies that $\tilde{p} = p$.
What does this say about the actual $A_\infty$-products in the Fukaya category? By definition, $\mu^4_{Q_p}(w_3,w_1,w_3,w_1)$ is given by the coefficient of $v_1^4$ in $p(v_1,v_2)$, which is $$\begin{aligned}
& -\hbar^{1/4} \vartheta(1)^{-2}\vartheta(-1)^{-2}\vartheta(\hbar^{1/2})^{-2} \vartheta_{4,3}'(1)^2 \vartheta_{2,1}(\hbar^{1/2})^2 \vartheta_{2,1}(1)^2 \\
& = -\textstyle{\frac{1}{4}}\hbar^{1/4} \vartheta(1)^{-2}\vartheta(-1)^{-2}\vartheta(\hbar^{1/2})^2 \vartheta_{4,3}'(1)^2
= -\hbar^{1/2} - 4 \hbar^{3/2} + \cdots
\end{aligned}$$ At least on the two leading orders we’ve written down, this agrees with the result of counting holomorphic squares with vertices on $(w_3,w_1,w_3,w_1)$ and which go through an additional generic marked point of $L_1$. However, a direct attempt to compute all of $\mu^4$ directly in $\mathit{Fuk}(T)$ is tricky, because the moduli spaces of constant holomorphic discs mapping to points of $L_1 \cap L_2$ are not regular. This difficulty is avoided in the approach we’ve chosen here.
\[th:whole-fukaya-torus\] $Q_p^{\mathit{perf}}$ is quasi-equivalent to $\mathit{Fuk}(T)^{\mathit{perf}}$.
From Lemma \[th:fukaya-torus\] we get a full and faithful functor $Q_p^{\mathit{perf}} \rightarrow \mathit{Fuk}(T)^{\mathit{perf}}$. The only additional fact needed is that the objects $L_1,L_2$ which are in the image of this functor split-generate the Fukaya category. They clearly split-generate the object $L_{3,u}$ for generic $u$, by the previous argument. On the other hand, one can use [@seidel04 Corollary 5.8] (together with suitable tensor product functors) to show that $L_1,L_{3,u}$ split-generate the Fukaya category. For alternative approaches, see [@abouzaid-smith09] or the review in Section \[subsec:split-generate\] below.
The Fukaya category comes with a canonical [*open-closed string map*]{} (which has a long history going back to [@kontsevich94], see also Section \[subsec:define-open-closed\] below) $$\label{eq:open-closed-t2}
H^*(T;R) \longrightarrow \mathit{HH}^*(\mathit{Fuk}(T),\mathit{Fuk}(T)).$$
\[th:t2-family\] Let ${\mathscr{S}}= \mathit{Spec}({\mathscr{R}})$ be the affine curve associated to the unit torus polynomial $p$, and $\theta$ its standard one-form. There is a perfect family of modules over $\mathit{Fuk}(T)$ parametrized by ${\mathscr{S}}$, which follows the image of $$\label{eq:explicit-current}
\theta \otimes [dq] \in H^0({\mathscr{S}},\Omega^1_{\mathscr{S}}) \otimes H^1(T;R)$$ under the open-closed string map. The fibre of this family associated to a point $(s_1,s_2) \in {\mathscr{S}}$ is isomorphic to $L_{3,u}$, where $u \in R^\times/\hbar^{{\mathbb{Z}}}$ satisfies .
The constant term of the open-closed string map, for any object $L$, is the standard map $H^*(T;R) \rightarrow H^*(L;R)$. We restrict this to the subcategory consisting of the objects $L_1,L_2$, so that it lands in $\mathit{HH}^*(Q_p,Q_p)$. With respect to the basis from Addendum \[th:quasi-generators\], $[dp] \in H^1(T;{\mathbb{R}})$ maps to $[g_1]+[g_2]$, and $[dq]$ to $-2[g_2]$. We take the family ${\mathscr{M}}$ from Section \[subsec:tautological\] and carry it over to the Fukaya category through the equivalence from Corollary \[th:whole-fukaya-torus\]. By construction this follows the deformation field $[\gamma]$ from , which is indeed the image of .
By definition, the object of this family associated to a point $(s_1,s_2)$ is the direct summand of $C_{(1,s_2)}$ associated to the projection ${{\textstyle\frac{1}{2}}}(e + s_1^{-1} t)$. Reversing the rescaling applied in , one finds that this is quasi-isomorphic to the direct sumand of $C_v$ associated to the projection $${{\textstyle\frac{1}{2}}}(e + \vartheta_{2,2}(u)^{-2} \vartheta_{4,3}'(1)^2 (\vartheta_{4,1}(u)-\vartheta_{4,3}(u))^2 s_1^{-1} t) = {{\textstyle\frac{1}{2}}}e + \vartheta_{4,3}'(1) (\vartheta_{4,1}(u) - \vartheta_{4,3}(u)) t.$$ Under the isomorphism $C_{(\vartheta_{2,2}(u),\vartheta_{2,1}(u))} {\cong}L_{3,u} \oplus L_{3,u^{-1}}$, $t$ goes to the endomorphism $\tilde{t}$ from , so the corresponding projection is $${{\textstyle\frac{1}{2}}}([e_{3,u}] \oplus [e_{3,u^{-1}}]) + \vartheta_{4,3}'(1) (\vartheta_{4,1}(u) - \vartheta_{4,3}(u)) \tilde{t} = [e_{3,u}],$$ which indeed picks out the summand $L_{3,u}$.
Symplectic automorphisms\[sec:automorphisms\]
=============================================
The symplectic automorphism group of a manifold of dimension $\geq 4$ often has many connected components which map to the identity component of the diffeomorphism group (see for instance [@seidel97; @seidel04b]). One way to detect this phenomenon is by using fixed point Floer cohomology. Through the connection with the Lagrangian Floer cohomology of graphs, this also provides interesting examples of Lagrangian submanifolds in products. In this section, we discuss both versions of Floer theory (with emphasis on computational methods that will be useful later in the paper), and then consider some specific examples of automorphisms obtained as compositions of Dehn twists.
We will work in a “symplectic Calabi-Yau” context, in which Floer cohomology groups are defined over the Novikov field $R$ and carry absolute ${\mathbb{Z}}$-gradings. As a side-effect, this makes the definition of fixed point Floer cohomology technically simpler. We will impose additional restrictions on Lagrangian submanifolds, which rule out bubbling of holomorphic discs, hence permit a similar simplification to take place in the construction of Lagrangian Floer cohomology.
Fixed point Floer cohomology\[subsec:fixed-point-floer\]
--------------------------------------------------------
Let $M^{2n}$ be a (connected) closed symplectic manifold, satisfying
\[th:calabi-yau\] $c_1(M) = 0$. In fact, we want to fix a trivialization of the anticanonical line bundle $K_M^{-1} = \Lambda^n_{\mathbb{C}}(TM)$ (for some compatible almost complex structure).
The choice of trivialization allows one to define the notion of graded symplectic automorphism [@seidel99]. Fixed point Floer cohomology [@dostoglou-salamon94; @seidel97; @seidel04b] associates to each graded symplectic automorphism $f$ a ${\mathbb{Z}}$-graded $R$-vector space $\mathit{HF}^*(f)$, whose Euler characteristic is the Lefschetz number of $f$ (up to a sign which depends on the choice of grading). This invariant comes with a rich structure of operations, among which we list the basic ones.
[*The pair-of-pants product.*]{} This is an associative multiplication $\mathit{HF}^*(f_2) \otimes \mathit{HF}^*(f_1) \rightarrow \mathit{HF}^*(f_2f_1)$, which comes with a two-sided unit element in $\mathit{HF}^*(\mathit{id})$. The standard example is $f_2 = f_1 = \mathit{id}$, where the canonical isomorphism [@piunikhin-salamon-schwarz94] $H^*(M;R) = \mathit{QH}^*(M) {\cong}\mathit{HF}^*(\mathit{id})$ identifies the pair-of-pants product with the small quantum product (in particular, the unit is the standard one in $H^0(M;R) = R$). As a consequence, any fixed point Floer cohomology group $\mathit{HF}^*(f)$ inherits the structure of a $\mathit{QH}^*(M)$-module [@floer88; @schwarz96].
[*Duality.*]{} There is a distinguished co-unit $\mathit{HF}^*(\mathit{id}) \rightarrow R[-2n]$, which together with the pair-of-pants product gives rise to a nondegenerate pairing $\mathit{HF}^*(f) \otimes \mathit{HF}^*(f^{-1}) \rightarrow R[-2n]$ (in terms of the isomorphism with quantum cohomology, the co-unit is just the standard integration map).
[*Continuation elements.*]{} Let $\{f_t\}$ be a Hamiltonian isotopy of graded symplectic automorphisms, with $f_0 = f$ and $f_1 = \mathit{id}$. This determines an element $I_{\{f_t\}} \in \mathit{HF}^*(f)$. The pair-of-pants product with such elements is used to prove Hamiltonian isotopy invariance of general fixed point Floer cohomology groups. [*Conjugation isomorphisms.*]{} These are maps $C_{f_2,f_1}: \mathit{HF}^*(f_1) \longrightarrow \mathit{HF}^*(f_2f_1f_2^{-1})$. Besides their general functoriality properties, which say that $C_{f_3,f_2f_1f_2^{-1}}C_{f_2,f_1} = C_{f_3f_2,f_1}$ and $C_{\mathit{id},f} = \mathit{id}$, we have the self-conjugation identity $$\label{eq:self-conjugation}
C_{f,f} = \mathit{id}.$$ This implies that $C_{f,f^m}$ generates an action of ${\mathbb{Z}}/m$ on $\mathit{HF}^*(f^m)$.
\[th:tcft\] All these operations have chain map realizations on the level of the Floer complexes $\mathit{CF}^*(f)$, and the relations between them are given by appropriate chain homotopies. Here is a more systematic way to approach the formal description of the theory [@seidel97; @seidel04b]. Write $\mathit{Aut}^{\mathrm{gr}}(M)$ for the group of graded symplectic automorphisms, equipped with the Hamiltonian topology (in which only Hamiltonian isotopies are continuous). Fixed point Floer cohomology can be viewed as a $(1+1)$-dimensional TCFT (topological conformal field theory) with target space $B\mathit{Aut}^{\mathrm{gr}}(M)$, which means a TCFT for surfaces equipped with graded Hamiltonian fibrations. In this framework, we view a symplectic automorphism as giving rise to its mapping torus $$\label{eq:ordinary-mapping-torus}
Z_f = {\mathbb{R}}\times M\;\; / \;\; (p,x) \sim (p-1,f(x)),$$ which is an $\mathit{Aut}^{\mathrm{gr}}(M)$-fibration over a circle $S^1 = {\mathbb{R}}/{\mathbb{Z}}$, and $\mathit{CF}^*(f)$ is the invariant associated to that fibration. The fibrewise action of $f_2$ yields an isomorphism $Z_{f_1} \rightarrow Z_{f_2f_1f_2^{-1}}$, for which there is an associated chain map $c_{f_2,f_1}$ inducing the previously introduced conjugation maps. This for instance explains : even though the fibrewise action of $f$ on $Z_f$ itself is nontrivial, it can be deformed continuously to the identity through rotations of the base, and this gives rise to a chain homotopy between $c_{f,f}$ and the identity. Similarly, $c_{f,f^m}$ is chain homotopic to the $m$-periodic automorphism $$\begin{aligned}
& Z_{f^m} \longrightarrow Z_{f^m},
& (p,x) \longmapsto (p-1/m,f(x)).
\end{aligned}$$
The actual definition of fixed point Floer cohomology is a mild generalization of the better-known Hamiltonian Floer cohomology. For simplicity, assume that $f$ has nondegenerate fixed points. The graded $R$-vector space $\mathit{CF}^*(f)$ is the direct sum of one-dimensional spaces $o_x$ associated to fixed points $f(x) = x$. Each such point has an absolute Conley-Zehnder index $\mathrm{deg}(x) \in {\mathbb{Z}}$, which determines the degree in which $o_x$ is placed. Take a family $J_f = (J_{f,t})$ of almost complex structures, parametrized by $t \in {\mathbb{R}}$ and satisfying $$\label{eq:j-periodicity}
J_{f,t-1} = f_*(J_{f,t}).$$ The differential $d: \mathit{CF}^*(f) \rightarrow \mathit{CF}^{*+1}(f)$ counts solutions of $$\label{eq:closed-floer}
\left\{
\begin{aligned}
& u: {\mathbb{R}}\times {\mathbb{R}}\longrightarrow M, \\
& u(s,t-1) = f(u(s,t)), \\
& \partial_s u + J_{f,t}(u)\,\partial_t u = 0
\end{aligned}
\right.$$ asymptotic to fixed points as $s \rightarrow \pm\infty$, with powers $\hbar^{E(u)}$ given by their energies $E(u) = \int_{{\mathbb{R}}\times [0,1]} u^*\omega_M$. The technical trick is to avoid bubbling off of holomorphic spheres, which can be done by a dimension-counting argument as in [@hofer-salamon95]. The outcome is independent of the choice of almost complex structure up to quasi-isomorphism. These quasi-isomorphisms are defined through continuation maps, and “essentially canonical” (which means unique up to chain homotopies, which can be extended to a system of higher homotopies; this is what makes it possible to omit the almost complex structures from a formal description as in Remark \[th:tcft\]). One special application which deserves some discussion is loop rotation. Suppose that we have chosen a family $J_{f,+}$ of almost complex structures as in . Fix some constant $t_0 \in {\mathbb{R}}$, and set $J_{f,-}(t) = J_{f,+}(t-t_0)$. If $u_+(s,t)$ is a solution of for $J_{f,+}$, then $u_-(s,t) = u_+(s,t-t_0)$ is a solution of the corresponding equation for $J_{f,-}$. This implies that the associated differentials $d_{\pm}$ agree. Note that on the other hand, we have a quasi-isomorphism $(\mathit{CF}^*(f),d_+) \rightarrow (\mathit{CF}^*(f),d_-)$ defined through continuation maps, which means solutions of $$\label{eq:cont-floer}
\left\{
\begin{aligned}
& u: {\mathbb{R}}\times {\mathbb{R}}\longrightarrow M, \\
& u(s,t-1) = f(u(s,t)), \\
& \partial_s u + J_{\mathrm{cont},s,t}(u)\,\partial_t u = 0
\end{aligned}
\right.$$ where $J_{\mathrm{cont}} = (J_{\mathrm{cont},s,t})$ is a two-parameter family with the same periodicity in $t$-direction, and such that $J_{\mathrm{cont},s,t} = J_{f,\pm,t}$ for $\pm s \gg 0$. There is a parametrized moduli problem which involves varying $t_0$ in an interval, and this yields a chain homotopy showing that the continuation map is homotopic to the identity. Suppose for instance that we take $t_0 = 1$, in which case $J_{f,-} = f_*J_{f,+}$. Then, the argument we have just outlined explains .
Another part of the theory for which we’ll need an explicit expression is the structure of $\mathit{HF}^*(f)$ as a module over $\mathit{QH}^*(M)$, sometimes called the [*quantum cap product*]{}. Fix a Morse function $h$ on $M$ (whenever we do that, we also tacitly choose a Riemannian metric, which is used to form $\nabla h)$, and let $\mathit{CM}^*(h)$ be the resulting Morse cochain complex. Suppose that we have fixed the almost complex structure $J_f$ defining the Floer differential. Then, choose a family $J_{\mathrm{cap}}$ in a similar way as for $J_{\mathrm{cont}}$, but where now the behaviour on both ends $s \rightarrow \pm\infty$ is given by $J_f$. Choose also a family $h_{\mathrm{cap},s}$, $s \in [0,\infty)$, of functions (with their associated metrics), such that $h_{\mathrm{cap},s} = h$ for $s \gg 0$. Then, consider pairs $(u_1,u_2)$ as follows: $$\label{eq:cap-floer}
\left\{
\begin{aligned}
& u_1: {\mathbb{R}}\times {\mathbb{R}}\longrightarrow M, \\
& u_1(s,t-1) = f(u_1(s,t)), \\
& \partial_s u_1 + J_{\mathrm{cap},s,t}(u_1)\,\partial_t u_1 = 0, \\
& u_2: [0,\infty) \longrightarrow M, \\
& du_2/ds + \nabla h_{\mathrm{cap},s}(u_2) = 0,
& u_1(0,0) = u_2(0).
\end{aligned}
\right.$$ $u_1$ should be asymptotic to fixed points of $f$ at both ends, and $u_2$ is asymptotic to a critical point of $h$. A count of the number of solutions of yields a chain map representing the quantum cap product [@schwarz96]: $$\label{eq:quantum-cap}
\mathit{CM}^*(h) \otimes \mathit{CF}^*(f) \longrightarrow \mathit{CF}^*(f).$$
It is in fact possible to choose $J_{\mathrm{cap},s,t} = J_{f,t}$ and $h_{\mathrm{cap},s} = h$, and this leads to the more familiar picture of “cutting down moduli spaces”. However, the greater freedom allowed above is more natural, and also technically useful.
Clean intersections\[subsec:clean\]
-----------------------------------
For $M$ as before (Assumption \[th:calabi-yau\]), we will consider Lagrangian submanifolds with the following added properties and structure.
\[th:adiscic\] Each Lagrangian submanifold $L$ is equipped with a grading (and hence an orientation), a [*Spin*]{} structure, as well as a local coefficient system with holonomy in . Moreover, it comes with a compatible almost complex structure $J_L$ with the following property. The subset of points of $L$ which lie either on a non-constant $J_L$-holomorphic sphere ${\mathbb{C}}P^1 \rightarrow M$, or on the boundary of a non-constant $J_L$-holomorphic disc $(D,\partial D) \rightarrow (M,L)$, has dimension $\leq n-3$.
Here, by a subset of dimension $\leq k$, we mean one that is contained in the image of a smooth map from a (possibly noncompact) manifold of dimension $\leq k$ to $L$. The Floer cohomology of two submanifolds satisfying Assumption \[th:adiscic\] is fairly straightforward to define. To make later computations easier, we adopt a Morse-Bott approach [@pozniak; @bourgeois02; @biran-cornea09; @biran-cornea09c; @johns08; @sheridan11]. A small amount of technicalities will be included, but without any attempt at completeness or full justification.
Take $(L_0,L_1)$ as in Assumption \[th:adiscic\], which have clean intersection [@pozniak]. Choose a Morse function $h_{L_0,L_1}$ on $L_0 \cap L_1$. The Morse-Bott type Floer cochain complex is a modification of the Morse cochain space of that function, more precisely: $$\label{eq:hom-c}
\mathit{CF}^*(L_0,L_1) \stackrel{\mathrm{def}}{=} \bigoplus_C \mathit{CM}^{*-\mathrm{deg}(C)}(h_{L_0,L_1}|C;\mathit{Hom}(\xi_0,\xi_1)|C \otimes o_C),$$ where the direct sum is over connected components $C$ of $L_0 \cap L_1$; the dimension offset $\mathrm{deg}(C) \in {\mathbb{Z}}$ is an absolute Maslov index, which depends on the gradings; $\xi_k$ are the given local systems on $L_k$, which we restrict to $L_0 \cap L_1$; and there is an additional local system $o_C \rightarrow C$ with holonomy $\pm 1$, which depends on the [*Spin*]{} structures. Choose also a a family $J_{L_0,L_1} = (J_{L_0,L_1,t})$ of almost complex structures, parametrized by $t \in [0,1]$, and such that $J_{L_0,L_1,0} = J_{L_0}$, $J_{L_0,L_1,1} = J_{L_1}$. To define the Floer differential, one primarily considers holomorphic strips, which are non-constant solutions of $$\label{eq:strip-equation}
\left\{
\begin{aligned}
& u: {\mathbb{R}}\times [0,1] \longrightarrow M, \\
& u({\mathbb{R}}\times \{0\}) \subset L_0, \;\; u({\mathbb{R}}\times \{1\}) \subset L_1, \\
& \partial_s u + J_{L_0,L_1,t}(u) \partial_t u = 0,
\end{aligned}
\right.$$ which are asymptotic to points of $L_0 \cap L_1$ as $s \rightarrow \pm\infty$. However, these have to be combined with Morse theory in an appropriate way, which we now set out to describe (see the references above, especially [@biran-cornea09c], for other accounts of this).
\[th:pearly-chain\] A [*pearly chain*]{} $T$ is a decorated graph of the following kind. First, the graph itself has only two-valent vertices, and two ends, one of which is singled out (and called the root, the other being the leaf). This determines an orientation of the graph (from the root to the leaf). Each edge $e$ is decorated with a closed interval $I_e \subset {\mathbb{R}}$. This is unbounded below if and only if the edge contains the root, and unbounded above if and only if contains the other end. In the bounded case we allow the length to become zero, meaning that $I_e$ is a point (while still thinking of $e$ combinatorially as an edge). Finally, each vertex $v$ is equipped with the Riemann surface $S_v = {\mathbb{R}}\times [0,1]$.
Note that we allow one slightly exceptional case: namely, that $T$ has a single edge which is infinite in both directions, with $I_e = {\mathbb{R}}$, and no vertices. It is convenient to associate to $T$ a topological space $\bar{S}_T$, obtained by compactifying each $S_v$ to $\bar{S}_v = S_v \cup \{s = \pm \infty\}$ (the closed unit disc), then identifying the added points with the endpoints of the intervals $I_e$ (compatibly with the orientations), and finally adding two more points at infinity to the ends of the non-compact intervals. The two points added in the last step will be denoted by $\bar{z}_0$ (corresponding to the root) and $\bar{z}_1$.
\[th:chain-perturbation\] A [*perturbation datum*]{} on a pearly chain is given by a family $h_e = (h_{e,s})$ of functions on $L_0 \cap L_1$ (with their associated choices of metrics), parametrized by $s \in I_e$, for each edge $e$, subject to the following additional conditions. If $e$ is the edge containing the root, then $h_{e,s} = h_{L_0,L_1}$ for $s \ll 0$. If $e$ contains the other end, $h_{e,s} = h_{L_0,L_1}$ for $s \gg 0$. In the exceptional case $I_e = {\mathbb{R}}$, we ask that $h_{e,s} = h_{L_0,L_1}$ for all $s$.
Given such a perturbation datum and a choice of critical points $x_0,x_1$ of $h_{L_0,L_1}$, one considers continuous maps $\bar{u}: \bar{S}_T \rightarrow M$ with $\bar{u}(\bar{z}_k) = x_k$, which satisfy the following equations. For any vertex $v$, the restriction of $\bar{u}$ to $S_v = {\mathbb{R}}\times [0,1]$ yields a smooth map $u_v$ which solves . On the other hand, restriction to an interval $I_e$ yields $$\label{eq:u1-equation}
\left\{
\begin{aligned}
& u_e: I_e \longrightarrow L_0 \cap L_1, \\
& du_e/ds + \nabla h_{e,s}(u_e) = 0.
\end{aligned}
\right.$$ Two maps which are related by a translation of the $u_v$ components are considered to be the same. Similarly, in the exceptional case $I_e = {\mathbb{R}}$, we divide out by translation acting on $u_e$ (which is then also required to be non-constant).
To define the differential on $\mathit{CF}^*(L_0,L_1)$, one has to choose a perturbation datum on every pearly chain, depending smoothly on the lengths. There are additional consistency conditions beyond those in Definition \[th:chain-perturbation\], which appear in the limit when the length of some edge goes to infinity. We will not formulate these in detail (but see [@seidel04] for the general idea, and [@sheridan11] for a case closer to the one discussed here). One then considers the moduli space of solutions $\bar{u}$ of the equations above, varying over all pearly chains. For generic choice of perturbation data, a count of points in the zero-dimensional strata, with appropriate signs and energies, defines the coefficient of $(x_0,x_1)$ in the differential. Assumption \[th:adiscic\] allows us to avoid bubbling of holomorphic discs.
One technical point deserves mention. Given connected components $C_0,C_1 \subset L_0 \cap L_1$, write ${\mathscr{M}}(C_0,C_1)$ for the space of solutions of with limits in those components. Standard transversality theory shows that for generic choice of $J_{L_0,L_1}$, this is smooth of dimension $\mathrm{deg}(C_0) - \mathrm{deg}(C_1) + \mathrm{dim}(C_0) - 1$. In particular, for $C_0 = C_1$ the dimension is $\mathrm{dim}(C_0) - 1$, and by a further application of transversality theory one can achieve that the asymptotic evaluation map ${\mathscr{M}}(C_0,C_0) \rightarrow C_0^2$ avoids the diagonal. This and similar arguments show that for generic choice of almost complex structures, the fibre products ${\mathscr{M}}(C_0,C_1) \times_{C_1} {\mathscr{M}}(C_1,C_2)$ are smooth of the expected dimension. In particular, for $C_0 = C_2$ that dimension is $\mathrm{dim}(C_0) - 2$, and one can again arrange that the asymptotic evaluation map ${\mathscr{M}}(C_0,C_1) \times_{C_1} {\mathscr{M}}(C_1,C_0) \rightarrow C_0^2$ avoids the diagonal. One can iterate that idea to higher fibre products. This is important since those products appear in our moduli spaces when the length of the intervals becomes zero (transversality for positive lengths is much simpler, since one can choose the families of functions $h_{e}$ essentially freely). Interested readers may want to consult [@biran-cornea09c Section 3.1.1], where an argument in the same spirit is used to address the same problem for monotone Lagrangian submanifolds.
\[th:graph\] Take a graded symplectic automorphism $f$ of $M$. Suppose that we have chosen a family $J_f$ as in . Write $M^-$ for the same manifold but with the sign of the symplectic form reversed. The diagonal $\Delta \subset M^- \times M$ is Lagrangian, and admits a distinguished grading. Set $J_\Delta = (-J_{f,1/2}) \times J_{f,1/2}$. Then, holomorphic discs $(D,\partial D) \rightarrow (M^- \times M,\Delta)$ correspond bijectively to holomorphic spheres ${\mathbb{C}}P^1 \rightarrow M$. Hence, the space of points of $\Delta$ lying on a non-constant disc is generically of dimension $\leq 2n-4$ (and the same holds for holomorphic spheres, for even more trivial reasons). Similarly, the graph $\Gamma = \{y = f(x)\}$ is a graded Lagrangian submanifold, which we can equip with $J_{\Gamma} = (-J_{f,1}) \times f_*(J_{f,1}) = (-J_{f,1}) \times J_{f,0}$. A holomorphic strip $(u_x,u_y): {\mathbb{R}}\times [0,1] \longrightarrow M^- \times M$ which satisfies with respect to the family of almost complex structures $$J_{\Gamma,\Delta,t} = (-J_{f,1-t/2}) \times J_{f,t/2}$$ gives rise to a solution of . Namely, consider first $$\label{eq:roll-up}
\begin{aligned}
& u: {\mathbb{R}}\times [0,1] \longrightarrow K, \\
& u(s,t) = \begin{cases} u_y(2s,2t) & t \leq 1/2, \\ u_x(2s,2-2t) & t \geq 1/2. \end{cases}
\end{aligned}$$ This satisfies $\partial_s u + J_{f,t}(u)\partial_t u = 0$, and has the boundary periodicity condition $u(s,0) = f(u(s,1))$. By the removable singularity theorem, we can therefore extend to a solution of . The same machinery runs (a little more easily) in reverse, producing $(u_x,u_y)$ from $u$. Assuming that $M$ is [*Spin*]{} so as to make the Lagrangian submanifolds fit into our framework (one can avoid this assumption by being more careful about the role of relative [*Spin*]{} structures in Fukaya categories, see [@wehrheim-woodward11]), and taking $f$ to have nondegenerate fixed points for simplicity, it is then easy to show that $\mathit{HF}^*(\Gamma,\Delta) {\cong}\mathit{HF}^*(f)$.
The $A_\infty$-structure\[subsec:fukaya\]
-----------------------------------------
We will now carry out the corresponding construction of the $A_\infty$-structure on Lagrangian Floer cochains. Fix, once and for all, a set of Lagrangian submanifolds in $M$. Each of these should satisfy Assumption \[th:adiscic\], and additionally any two should have clean intersection. All Lagrangian submanifolds appearing in the following discussion are assumed to be taken from this set. We suppose that for any two $(L_0,L_1)$, the Floer complex $\mathit{CF}^*(L_0,L_1)$ with its differential $\mu^1_{\mathit{Fuk}(M)}$ has already been defined, which in particular means that functions $h_{L_0,L_1}$ and almost complex structures $J_{L_0,L_1}$ have been chosen. Again, we refer to the previously quoted literature, in particular [@sheridan11], and additionally to [@fukaya-oh-ohta-ono08; @seidel08b].
Fix some $d \geq 1$. A [*pearly tree*]{} with $d$ leaves is a decorated graph of the following kind. The underlying graph $T$ is a ribbon tree with $(d+1)$ ends, one of which is singled out (and called the root, the others being the leaves). Moreover, it is assumed that all vertices $v$ of $T$ have valence $|v| \geq 2$. We orient the tree from the root to the leaves. Each edge $e$ is decorated with a closed interval $I_e \subset {\mathbb{R}}$, with the same properties as in Definition \[th:pearly-chain\]. Each vertex $|v|$ is decorated with a Riemann surface $S_v = D \setminus \{\bar{z}_{v,0},\dots,\bar{z}_{v,|v|-1}\}$, where $D$ is the closed unit disc and the $\bar{z}_{v,i}$ are cyclically ordered distinct boundary points.
For $d = 1$ this reduces to a pearly chain, up to the irrelevant issue of choosing identifications between a two-punctured disc and ${\mathbb{R}}\times [0,1]$. Note that for each vertex $v$, there is a preferred correspondence between ends of $S_v$ and edges adjacent to $v$, which is compatible with the cyclic ordering and assigns the point at infinity $\bar{z}_{v,0}$ to the edge oriented towards $v$ (we call this end of $S_v$ its negative end, and the others positive ends). Using that, we can construct a compact topological space $\bar{S}_T$, obtained by compactifying each $S_v$ to $\bar{S}_v = D$, identifying the endpoints of $I_e$ with the $\bar{z}_{v,k}$, and then adding points at infinity to the noncompact intervals. We denote by $\bar{z}_0,\dots,\bar{z}_d \in \bar{S}_T$ the points added in the last step, starting with the root and proceeding in the ordering given by a planar embedding of $T$.
Let’s clear up few more book-keeping matters. Again supposing that $T$ is embedded properly in ${\mathbb{R}}^2$, we say that it is a [*labeled pearly tree*]{} if each component of ${\mathbb{R}}^2 \setminus T$ comes with a Lagrangian submanifold (taken from our collection). One can in fact number these components by $\{0,\dots,d\}$, compatibly with the cyclic ordering and in such a way that the root separates the first and last component. Therefore, a labeling of $T$ just corresponds to a choice of Lagrangian submanifolds $(L_0,\dots,L_d)$. Suppose from now on that such a labeling has been fixed. For any edge $e$ we then have a pair $(L_{i_{e,0}},L_{i_{e,1}})$, corresponding to the components of ${\mathbb{R}}^2 \setminus T$ lying to the left ($i_{e,1}$) and right ($i_{e,0}$) with respect to the orientation of $e$. By definition, $0 \leq i_{e,0} < i_{e,1} \leq d$. Similarly, for any vertex $v$, there is a canonical correspondence between the boundary components of $S_v$ and connected components of ${\mathbb{R}}^2 \setminus T$ adjacent to $v$. If we label the boundary components by $\partial_0 S_v,\dots, \partial_{|v|-1} S_v$, so that the negative end separates the first and last one, then this leads to having associated Lagrangian submanifolds $(L_{i_{v,0}},\dots,L_{i_{v,|v|-1}})$ for $0 \leq i_{v,0} < \cdots < i_{v,|v|-1} \leq d$.
A [*perturbation datum*]{} on a labeled pearly tree consists of the following data. For each edge $e$, we want to have a family of functions $h_e = (h_{e,s})$ on $L_{i_{e,0}} \cap L_{i_{e,1}}$ parametrized by $s \in I_e$. If $I_e$ is noncompact, we ask that outside a compact subset, $h_{e,s} = h_{L_{i_{e,0}},L_{i_{e,1}}}$ is one of the previously chosen functions (and in the exceptional case $I_e = {\mathbb{R}}$, we impose the same additional condition as in Definition \[th:chain-perturbation\]).
Next, take a vertex of valence $|v| \geq 3$. We then want to choose strip-like ends on $S_v$, which means proper holomorphic embeddings $$\begin{aligned}
& \epsilon_{v,0}: (-\infty,0] \times [0,1] \longrightarrow S_v, \\
& \epsilon_{v,1},\dots,\epsilon_{v,|v|-1}: [0,\infty) \times [0,1] \longrightarrow S_v
\end{aligned}$$ giving preferred coordinates on its ends. Given those, we want to have a family $(J_{v,z})$ of compatible almost complex structures parametrized by $z \in S_v$. If $z \in \partial_k S_v$, then $J_{v,z} = J_{L_{i_{v,k}}}$ should be one of the structures that come from Assumption \[th:adiscic\]. Moreover, on the strip-like ends $$J_{v,\epsilon_{v,k}(s,t)} = \begin{cases} J_{L_{i_{v,0}},L_{i_{v,k}},t} & k = 0 \text{ and } s \ll 0, \\
J_{L_{i_{v,k-1}},L_{i_{v,k}},t} & k > 0 \text{ and } s \gg 0.
\end{cases}$$ Additionally, we want to have a one-form $K_v$ on $S_v$ with values in the space $C^\infty(M,{\mathbb{R}})$ (which means a section of the pullback bundle $T^*S_v \rightarrow S_v \times M$), supported in a compact subset of the interior of $S_v$.
Given such a perturbation datum and critical points $x_0 \in \mathit{Crit}(h_{L_0,L_d})$, $x_k \in \mathit{Crit}(h_{L_{k-1},L_k})$ ($1 \leq k \leq d$), we can define an associated moduli space. Points are (isomorphism classes of) continuous maps $\bar{u}: \bar{S}_T \rightarrow M$, with $\bar{u}(\bar{z}_k) = x_k$, which satisfy the following equations. Let $S_v$ be the Riemann surface associated to a vertex with $|v| \geq 3$. Restriction of $\bar{u}$ to that surface yields a smooth map $$\label{eq:generalized-u2-equation}
\left\{
\begin{aligned}
& u_v: S_v \longrightarrow M, \\
& u_v(\partial_k S_v) \subset L_{i_{v,k}} \text{ for $k = 0,\dots,|v|-1$}, \\
& (du_v - X_{v,z}(u_v)) \circ i = J_{v,z}(u_v) \circ (du_v - X_{v,z}(u_v)).
\end{aligned}
\right.$$ Here, we have $K_{v,z}: TS_z \rightarrow C^\infty(M,{\mathbb{R}})$ and consider the associated Hamiltonian vector field, $X_{v,z}: TS_z \rightarrow C^\infty(M,TM)$, then evaluate that at the point $u(z)$. We extend that to $|v| = 2$ by identifying $S_v {\cong}{\mathbb{R}}\times [0,1]$, and equipping that with $J_{v,s,t} = J_{L_{i_{v,0}},L_{i_{v,1}},t}$ as well as $K_v = 0$, which of course results in being an equation of the form (in that case we again exclude constant solutions). Next, let $I_e$ be the interval associated to an edge. Restriction of $\bar{u}$ to it yields a map $$\label{eq:generalized-u1-equation}
\left\{
\begin{aligned}
& u_e: I_e \longrightarrow L_{i_{e,0}} \cap L_{i_{e,1}}, \\
& du_e/ds + \nabla h_{e,s}(u_e) = 0.
\end{aligned}
\right.$$ To define $\mu^d_{\mathit{Fuk}(M)}$, one has to choose perturbation data for all decorated pearly trees, depending smoothly on the moduli and lengths, and related by other consistency conditions.
\[th:constant-regularity\] The addition of an inhomogeneous term $X_v$ to is necessary to achieve transversality in general. Concretely, the problem with setting $K_v = 0$ is that then, constant maps at points of $L_{v,i_{v,0}} \cap \cdots \cap L_{v,i_{v,|v|-1}}$ would always be solutions, irrespective of the choice of $J_v$. The dimension of the moduli space of constant maps is $\mathrm{dim}(L_{i_{v,0}} \cap \cdots \cap L_{i_{v,|v|-1}}) + (|v|-3)$, whereas its expected dimension is $\mathrm{dim}(L_{i_{v,0}} \cap L_{i_{v,|v|-1}}) + \mathrm{deg}(L_{i_{v,0}} \cap L_{i_{v,|v|-1}}) - \mathrm{deg}(L_{i_{v,0}} \cap L_{i_{v,1}}) - \cdots - \mathrm{deg}(L_{i_{v,|v|-2}} \cap L_{i_{v,|v|-1}}) + (|v|-3)$. Here, the dimensions and degrees really refer to the connected components to which our constant map belongs. One can show that the moduli space is regular if and only if those two numbers agree.
\[th:single-l\] Consider a single $L$, and assume that there are no nonconstant $J_L$-holomorphic discs with boundary on $L$, and no non-constant $J_L$-holomorphic spheres intersecting $L$. To define the $A_\infty$-structure on $\mathit{CF}^*(L,L)$, one can take all the almost complex structures to $J_L$, and all inhomogeneous terms to be zero. The only solutions of are constant maps at points of $L$, which are regular (Remark \[th:constant-regularity\]). The only contribution to $0$-dimensional moduli spaces comes from trees $T$ with only trivalent vertices. Transversality can be achieved by varying the functions, which recovers a version of the picture in [@fukaya-oh98]. In particular, if $\xi$ is the local coefficient system on $L$, we have an isomorphism of rings $$\label{eq:hf-non-deformed}
\mathit{HF}^*(L,L) {\cong}H^*(L;\mathit{Hom}(\xi,\xi)).$$ One can realize this more canonically by an open string analogue of the Piunikhin-Salamon-Schwarz map [@albers08].
The open-closed string map\[subsec:define-open-closed\]
-------------------------------------------------------
We now want to give a similar description of the open-closed string map $$\label{eq:open-closed}
\mathit{QH}^*(M) \longrightarrow \mathit{HH}^*(\mathit{Fuk}(M),\mathit{Fuk}(M)),$$ or at least the part that lands in the subcategory consisting of Lagrangian submanifolds in our fixed collection. Even though it would be possible (and maybe more in tune with our general developments) to represent $\mathit{QH}^*(M) = H^*(M;R)$ Morse-theoretically, we prefer the simpler picture that comes from thinking of cohomology classes as cycles. More specifically, fix a co-oriented submanifold $G \subset M$, and consider the Poincaré dual class $[G] \in H^*(M;{\mathbb{Z}})$.
A [*pointed pearly tree*]{} with $d \geq 0$ leaves is a decorated graph of the following kind. The underlying graph $T$ is a ribbon tree with $(d+1)$ ends, again with a distinguished root. Each edge $e$ is decorated with a closed interval $I_e \subset {\mathbb{R}}$, and each vertex $|v|$ with a Riemann surface $S_v = D \setminus \{\bar{z}_{v,0},\dots,\bar{z}_{v,|v|-1}\}$, as before. The new ingredient is that for exactly one vertex $v_*$, the surface $S_{v_*}$ carries an additional interior marked point $z_*$. Moreover, this particular vertex can be univalent, whereas for all others the condition $|v| \geq 2$ still applies.
We define labelings, and other book-keeping devices, as before.
A [*perturbation datum*]{} on a labeled pointed pearly tree consists of the following data. For each edge $e$, we want to have a family of functions $h_e = (h_{e,s})$ on $L_{i_{e,0}} \cap L_{i_{e,1}}$ as usual. Next, take a vertex, which either satisfies $|v| \geq 3$ or is equal to $v_*$. We then want to choose strip-like ends, a family $J_v$ of almost complex structures, and a one-form $K_v$ as before.
Given this, we can define an associated moduli space which combines gradient flow lines and perturbed pseudo-holomorphic maps, where the component $u_{v_*}$ additionally satisfies $u_{v_*}(z_*) \in G$. Counting solutions of this moduli problem yields a Hochschild cocycle $g$ which represents the image of $[G]$ under the open-closed string map.
Take a single Lagrangian submanifold $L$ as in Example \[th:single-l\]. Assume that $G$ is transverse to $L$, and consider only the part of $g$ involving only $L$, which is an element of the Hochschild complex of the $A_\infty$-algebra $\mathit{CF}^*(L,L)$. Again, one can take all almost complex structure equal to $J_L$, and all inhomogeneous terms to be zero, so $g$ can be expressed in purely Morse-theoretic terms. In particular, the linear part $g^0_L \in \mathit{CF}^*(L,L) = \mathit{CM}^*(h_{L,L})$ is just given by counting gradient half-lines which start in $B \cap L$. This is the Morse-theoretic representative for $[B \cap L] \in H^*(L;R)$. As another consequence of the same observation, if $L \cap G = \emptyset$, all the components $g^d_{L,\dots,L}$ vanish.
\[th:totally-disjoint\] We can generalize the last-mentioned observation as follows. Suppose that $G$ has (real) codimension $1$. Suppose also that every Lagrangian submanifold $L$ in our collection satisfies the condition from Example \[th:single-l\], and additionally is disjoint from $G$. Then, for a suitable choice, $g$ is identically zero. Namely, given a pointed pearly tree with $d>0$ ends, one can forget $z_*$ and then collapse components if necessary, so as to obtain an ordinary pearly tree. This allows one to lift the perturbation data used to define the $A_\infty$-structures to pointed pearly trees, giving a picture whereby $g$ is obtained by cutting down moduli spaces by asking that the holomorphic discs should go through $B$. However, because of the codimension and intersection assumptions, this can never reduce the dimension to zero unless the moduli space is empty.
Abelian coverings\[subsec:pushdown\]
------------------------------------
The following material is not new (compare [@seidel03b]) or difficult, but we will need the statements in a specific form for later reference. The geometric situation is that we have a finite covering of symplectic manifolds $$z: \tilde{M} \longrightarrow M,$$ with abelian covering group $\Gamma$. Both manifolds are supposed to come with trivializations of their anticanonical bundles, related in the obvious way. We consider Lagrangian submanifolds $\tilde{L} \subset \tilde{M}$, equipped with additional structures which turn them into objects of the Fukaya category, and also subject to the following conditions: $$\label{eq:covered-lagrangians}
\parbox{35em}{
$\kappa|\tilde{L}$ is itself a covering (for some subgroup of $\Gamma$) of a Lagrangian submanifold $L = z(\tilde{L}) \subset M$. The grading of $\tilde{L}$ is then automatically lifted from a grading of $L$. We impose the additional requirements that the {\em Spin} structure on $\tilde{L}$ should be the lift of one on $L$, and the same for the almost complex structure $J_{\tilde{L}}$.
}$$ As part of the data, $\tilde{L}$ carries a local system $\tilde{\xi}$, but we do not impose any additional conditions on that. Let $\tilde{F} \subset \mathit{Fuk}(\tilde{M})$ be the full $A_\infty$-subcategory whose objects are Lagrangian submanifolds satisfying . When defining the $A_\infty$-structure, one can similarly lift all choices of Morse functions, almost complex structures, and inhomogeneous terms from $M$. The result is that $\tilde{F}$ comes with a strict action of $\Gamma$ by covering transformations, as well as with a pushforward functor $Z: \tilde{F} \rightarrow \mathit{Fuk}(M)$, which takes $\tilde{L}$ to $Z(\tilde{L}) = L$ with the local coefficient system $\xi = z_*(\tilde{\xi})$. The behaviour of this functor can be fully described in terms of the $\Gamma$-action. We have $$\label{eq:pushdown-cf}
\mathit{CF}^*(Z(\tilde{L}_0),Z(\tilde{L}_1)) = \bigoplus_{\gamma \in \Gamma} \mathit{CF}^*(\tilde{L}_0,\gamma(\tilde{L}_1)),$$ and the $A_\infty$-structure maps $\mu^d_{\mathit{Fuk}(M)}$ are direct sums of $$\begin{aligned}
& \mathit{CF}^*(\tilde{L}_{d-1},\gamma_d(\tilde{L}_d)) \otimes \cdots \otimes \mathit{CF}^*(\tilde{L}_1,\gamma_2(\tilde{L}_2)) \otimes \mathit{CF}^*(\tilde{L}_0,\gamma_1(\tilde{L}_1)) \\ & {\cong}\mathit{CF}^*(\gamma_1 \dots \gamma_{d-1}(\tilde{L}_{d-1}),\gamma_1 \dots \gamma_d(\tilde{L}_d)) \otimes
\cdots \otimes \mathit{CF}^*(\gamma_1(\tilde{L}_1), \gamma_1 \gamma_2(\tilde{L}_2))
\otimes \mathit{CF}^*(\tilde{L}_0, \gamma_1(\tilde{L}_1)) \\ &
\xrightarrow{\mu^d_{\tilde{F}}} \mathit{CF}^*(\tilde{L}_0, \gamma_1\cdots \gamma_d(\tilde{L}_d)).
\end{aligned}$$
Suppose that $\tilde{L} \rightarrow L$ is a covering with group $G \subset \Gamma$. There is an obvious isomorphism of local systems $\mathit{Hom}(z_*\tilde{\xi},z_*\tilde{\xi}) {\cong}z_*\tilde{\xi} \otimes_{\mathbb{Z}}{\mathbb{Z}}[G]$, hence $\mathit{HF}^*(Z(\tilde{L}),Z(\tilde{L})) {\cong}H^*(L;\mathit{Hom}(z_*\tilde{\xi},z_*\tilde{\xi})) {\cong}H^*(\tilde{L}; \tilde{\xi}) \otimes_{\mathbb{Z}}{\mathbb{Z}}[G]$. This describes the decomposition induced by (the summands for elements of $\Gamma \setminus G$ are zero).
This has implications for the open-closed string map as well. Let $\tilde{F}$ be the diagonal bimodule. We can twist it by applying $\gamma \in \Gamma$ to the left (but not the right) actions, thereby obtaining another bimodule $\tilde{F}^\gamma$, which can be thought of as the graph of $\gamma^{-1}$. On the other hand, take the diagonal bimodule of $\mathit{Fuk}(M)$ and pull it back by $Z$ (on both sides) to get a bimodule over $\tilde{F}$. As a consequence of the observations above, we have an isomorphism $$Z^*\mathit{Fuk}(M) {\cong}\bigoplus_{\gamma \in \Gamma} \tilde{F}^\gamma,$$ and the canonical bimodule map $\tilde{F} \rightarrow Z^*\mathit{Fuk}(M)$ is just the inclusion of the $\gamma = e$ summand. The open-closed string maps for $M$ and $\tilde{M}$ (the latter restricted to the category $\tilde{F}$) and the maps associated to the functor $Z$ fit into a commutative diagram $$\label{eq:open-closed-pushforward}
\xymatrix{
\mathit{QH}^*(\tilde{M}) \ar[rr] && \mathit{HH}^*(\tilde{F},\tilde{F}) \ar[d] \\
&& \mathit{HH}^*(\tilde{F},Z^*\mathit{Fuk}(M)) \\
\mathit{QH}^*(M) \ar[rr] \ar[uu]^-{z^*} && \mathit{HH}^*(\mathit{Fuk}(M),\mathit{Fuk}(M)). \ar[u]
}$$
\[th:pullback-functor\] There is also a functor in opposite direction, which is better-behaved since it is defined on the whole Fukaya category, $\mathit{Fuk}(M) \rightarrow \mathit{Fuk}(\tilde{M})$. It maps any object $L$ to its entire primage. As for morphisms, the Floer cochain complex $\mathit{CF}^*(z^{-1}(L_0),z^{-1}(L_1))$ comes with a natural action of $\Gamma$, and the invariant part is the image of $\mathit{CF}^*(L_0,L_1)$ under pullback.
Split-generators\[subsec:split-generate\]
-----------------------------------------
Let $O \subset A^{\mathit{perf}}$ be a full subcategory. One says that the objects of $O$ [*split-generate*]{} $A^{\mathit{perf}}$ if the following holds: any object of $A^{\mathit{perf}}$ up to quasi-isomorphism can be constructed by starting with objects of $O$ and applying the following operations: shifts; mapping cones; and taking direct summands with respect to idempotent endomorphisms. We will quote two abstract split-generation criteria for Fukaya categories from the literature, the second stronger than the first. Both involve Hochschild cohomology and the open-closed string map. More precisely, given $O \subset \mathit{Fuk}(M)$, we consider $$\label{eq:open-closed-restrict}
\xymatrix{
\mathit{QH}^*(M) \ar[r] & \mathit{HH}^*(\mathit{Fuk}(M),\mathit{Fuk}(M)) \ar[r] & \mathit{HH}^*(O,O).
}$$
\[th:split-generation-1\] Suppose that $O$ is smooth [@kontsevich-soibelman00 Definition 8.1.12], and that is an isomorphism. Then the objects in $O$ split-generate $\mathit{Fuk}(M)^{\mathit{perf}}$.
\[th:split-generation-2\] Suppose that there is a linear map $\mathit{HH}^{2n}(O,O) \rightarrow R$ whose composition with yields the integration map $\mathit{QH}^{2n}(M) \rightarrow R$. Then the objects in $O$ split-generate $\mathit{Fuk}(M)^{\mathit{perf}}$.
Because $\mathit{QH}^{2n}(M)$ is one-dimensional, the condition in Theorem \[th:split-generation-2\] is just that should be nonzero in degree $2n$, but we prefer the formulation above, which extends to non-Calabi-Yau cases and more accurately reflects the core of the argument. Note that for a general $M$, there is no reason to suppose that the conditions of either theorem above would apply (no actual counterexamples are known, but there are suggestions coming from mirror symmetry for non-algebraic varieties). However, if they do hold for one set of split-generating objects, then the same is true for any other such set.
Take for instance the two-torus $T$. As discussed in [@abouzaid-smith09], two curves intersecting in a point satisfy the criterion of Theorem \[th:split-generation-1\], hence split-generate the Fukaya category. As already pointed out in Remark \[th:whole-fukaya-torus\], it then follows that the same holds for the two curves from Figure \[fig:abcd1\].
\[th:quartic\] Let $K \subset {\mathbb{C}}P^3$ be a smooth quartic surface, equipped with the restriction of the Fubini-Study form. In this case, homological mirror symmetry [@seidel03b] says that $\mathit{Fuk}(K)^{\mathit{perf}}$ is quasi-equivalent to $D^b\mathit{Coh}(X)$, where the mirror $X$ is a smooth $K3$ surface over $R$. We have $\mathit{HH}^4(X,X) {\cong}R$, and the product $\mathit{HH}^2(X,X)^{\otimes 2} \rightarrow \mathit{HH}^4(X,X)$ is a nondegenerate quadratic form. In particular, any isotropic subspace is at most of dimension ${{\textstyle\frac{1}{2}}}\mathrm{dim} \mathit{HH}^2(X, X) = 11$. The same must then hold for $\mathit{Fuk}(K)^{\mathit{perf}}$. A standard Picard-Lefschetz theory argument shows that the entire orthogonal complement of $[\omega_K]$ in $H_2(K;{\mathbb{Q}})$ is generated by homology classes of Lagrangian spheres. Choose Lagrangian spheres $(L_1,\dots,L_{21})$ whose homology classes form a basis for that subspace, and consider $$\mathit{QH}^2(K) \rightarrow \mathit{HH}^2(\mathit{Fuk}(K),\mathit{Fuk}(K)) \rightarrow \bigoplus_i \mathit{HF}^2(L_i,L_i).$$ This is surjective, which shows that the open-closed string map in degree $2$ is of rank $21$ or $22$. Since the image of that map can’t be an isotropic subspace, it follows that $\mathit{QH}^4(K) \rightarrow \mathit{HH}^4(\mathit{Fuk}(K),\mathit{Fuk}(K))$ must be an isomorphism. This implies that for any set of split-generating objects, the requirement of Theorem \[th:split-generation-2\] is satisfied. In fact, our previous argument shows that the kernel of open-closed string map in degree $2$ is either zero or else spanned by $[\omega_K]$. But the second case is impossible since $[\omega_K]^2$ is nontrivial, so the open-closed string map must be an isomorphism. Since $D^b\mathit{Coh}(X)$, it follows that the assumption of Theorem \[th:split-generation-1\] holds as well.
\[th:product-mirror\] Again following [@abouzaid-smith09], we point out that this strategy extends well to products. For instance, take $M = T \times K$ to be the product of the two-torus and the quartic surface. Let $O$ be the subcategory consisting of Lagrangians which are themselves products. This is quasi-isomorphic to the $A_\infty$-tensor product $\mathit{Fuk}(T) \otimes \mathit{Fuk}(K)$. Therefore, it is again smooth, and the associated open-closed string map is an isomorphism, which shows that $O$ split-generates. This can be used to prove homological mirror symmetry for the product.
Products of Dehn twists
-----------------------
We will now concentrate on constructing specific examples of automorphisms $f$ where the ${\mathbb{Z}}/m$-action on $\mathit{HF}^*(f^m)$ is nontrivial. The symplectic manifold $M$ should still satisfy Assumption \[th:calabi-yau\]. For technical simplicity, we assume that it is a four-manifold ($n = 2$; higher-dimensional generalizations would have to use more advanced methods, as in [@oh10]). Take Lagrangian spheres $L_1,\dots,L_r \subset M$, equipped with an arbitrary choice of grading, the unique [*Spin*]{} structure, and the trivial local coefficient system. Because of our dimensional restriction, generically chosen almost complex structures then satisfy Assumption \[th:calabi-yau\]. Consider the composition of Dehn twists $$f = \tau_{L_r} \cdots \tau_{L_2}\tau_{L_1},$$ which is naturally a graded symplectic automorphism of $M$.
There is a spectral sequence converging to $\mathit{HF}^*(f)$, whose starting page is $$\label{eq:e1fixed}
E_1^{p*} = \begin{cases}
\mathit{QH}^*(M) {\cong}H^*(M;R) & p = 0, \\
\bigoplus_i \mathit{HF}^*(L_i,L_i) {\cong}\bigoplus_i H^*(L_i;R) & p = 1, \\
\bigoplus_{i_1 > \dots > i_p} \mathit{HF}^*(L_{i_p},L_{i_1}) \otimes \mathit{HF}^*(L_{i_{p-1}},L_{i_p}) \otimes \cdots \\ \qquad \qquad \qquad \qquad \qquad \cdots \otimes \mathit{HF}^*(L_{i_1},L_{i_2})[n(p-1)] & 1 < p \leq r, \\
0 & \text{otherwise.}
\end{cases}$$
In fact, Perutz’s work [@perutz10] yields an explicit chain complex which computes $\mathit{HF}^*(f)$ in terms of the Fukaya $A_\infty$-structure and open-closed string map. An appropriate filtration of that chain complex then gives rise to the spectral sequence (for which an alternative approach is due to Ma’u [@mau11]). One can rewrite the nontrivial columns $E_1^{p*}$, $p > 0$, in a way that highlights the cyclic symmetry: $$\label{eq:cyclic-e1}
E_1^{p*} = \Big( \bigoplus_{i_1, \dots, i_p} \mathit{HF}^*(L_{i_p},L_{i_1}) \otimes \mathit{HF}^*(L_{i_{p-1}},L_{i_p}) \otimes \cdots \otimes \mathit{HF}^*(L_{i_1},L_{i_2}) \Big)^{{\mathbb{Z}}/p} [n(p-1)],$$ where the direct sum is over cyclically decreasingly ordered $p$-tuples, and ${\mathbb{Z}}/p$ acts by cyclically permuting these $p$-tuples (with additional signs). This point of view is particularly convenient is when considering conjugation invariance. Namely, the spectral sequences computing $\mathit{HF}^*(f)$ and $\mathit{HF}^*(\tau_{L_1} f \tau_{L_1}^{-1})$ are related by an automorphism, which converges to $C_{\tau_{L_1},f}$. On the $E_1$ level, that automorphism is just the obvious relation between the expressions .
The class of examples of interest to us is where $r = 2m$ and $L_1 = L_3 = \cdots = L_{2m-1}$, $L_2 = L_4 = \cdots = L_{2m}$, so that $f = (\tau_{L_1}\tau_{L_2})^m$. Additionally, we ask that:
\[th:spread\] $\mathit{HF}^*(L_1,L_2)$ is concentrated in degrees $[n/2-k,n/2+k]$ for some $k \geq n/2$. Moreover, both $\mathit{HF}^{n/2-k}(L_1,L_2)$ and $\mathit{HF}^{n/2+k}(L_1,L_2)$ are nonzero, and at least one of those spaces has dimension $>1$.
The argument outlined above shows how the conjugation isomorphisms $$\begin{aligned}
& C_{\tau_{L_2},f}: \mathit{HF}^*(f) \longrightarrow \mathit{HF}^*(\tau_{L_2} f \tau_{L_2}^{-1}), \\
& C_{\tau_{L_1},\tau_{L_2} f \tau_{L_2}^{-1}}: \mathit{HF}^*(\tau_{L_2} f \tau_{L_2}^{-1}) \longrightarrow \mathit{HF}^*(\tau_{L_1} \tau_{L_2} f \tau_{L_2}^{-1} \tau_{L_1}^{-1}) = \mathit{HF}^*(f)
\end{aligned}$$ act on the $E_1$ pages of the respective spectral sequences. The composition of these two isomorphisms defines the standard ${\mathbb{Z}}/m$-action. In particular, this acts on the last column $$\begin{aligned}
E_1^{2m,*} & = \mathit{HF}^*(L_1,L_{2m}) \otimes \mathit{HF}^*(L_2,L_1) \otimes \cdots \otimes \mathit{HF}^*(L_{2m},L_{2m-1})[n(2m-1)] \\
& {\cong}\big(\mathit{HF}^*(L_1,L_2) \otimes \mathit{HF}^*(L_1,L_2)\big)^{\otimes m}[n(2m-1)]
\end{aligned}$$ by cyclically permuting the $m$ tensor factors, up to a degree-dependent sign. That sign could in principle be determined by a more careful argument, but it turns out to be irrelevant for our purpose.
\[th:regular-representation\] $\mathit{HF}^{2m(1-k-n/2)+n}((\tau_{L_1}\tau_{L_2})^m)$ contains a copy of the regular representation of ${\mathbb{Z}}/m$, for any $m \geq 1$.
The $E_1^{0*}$ column contributes only in total degrees $\geq 0$, the $E_1^{1*}$ column in total degrees $\geq 1$, and the $E_1^{p*}$ ($1 < p < 2m$) columns in total degrees $\geq p(1-k-n/2) + n$. This means that the following piece, whose total degree is $2m(1-k-n/2) + n$, survives to $E_\infty$: $$\label{eq:m-tensor}
(\mathit{HF}^{n/2-k}(L_1,L_2) \otimes \mathit{HF}^{n/2-k}(L_2,L_1))^{\otimes m} \subset E_1^{2m,2m(-k-n/2) + n}.$$ By assumption, $\mathit{HF}^{n/2-k}(L_1,L_2) \otimes \mathit{HF}^{n/2-k}(L_2,L_1)$ is at least two-dimensional. If $a_1,a_2$ are linearly independent elements in it, then $a_1 \otimes a_2^{\otimes m-1}$ and its images under the ${\mathbb{Z}}/m$-action are all linearly independent elements of , which proves the claim.
In general, Lemma \[th:regular-representation\] fails without Assumption \[th:spread\]. Namely, suppose that $L_1 = L_2 = L$. In that case, we can find a more compact spectral sequence which also converges to $\mathit{HF}^*((\tau_{L_1}\tau_{L_2})^m) = \mathit{HF}^*(\tau_L^{2m})$ (an algebraic analogue of a Morse-Bott spectral sequence), and whose starting page is $$E_1^{p*} = \begin{cases} H^*(M,L;R) & p = 0, \\
H^*({\mathbb{R}}P^3;R)[p-1] & 1 \leq p \leq m-1, \\
H^*(S^2;R)[2-m] & p = m.
\end{cases}$$ Moreover, the ${\mathbb{Z}}/m$-action is trivial already on this page (but it embeds into a larger ${\mathbb{Z}}/2m$-action, which may be nontrivial on the last column).
The quartic surface
-------------------
Before continuing on to our concrete example, we have to agree on criteria for a symplectic automorphism to be trivial from a topological viewpoint.
\[th:interesting\] Let $f$ be a symplectic automorphism of a closed symplectic manifold $M$. We say that $f$ is [*undistinguishable from the identity by topological means*]{} if there is an isotopy from the identity to $f$ inside the diffeomorphism group, with the following additional properties. First, by starting with $Df$ and deforming it along the isotopy, we get an automorphism of the symplectic vector bundle $TM$ (taking each fibre to itself), and we ask that this should be homotopic to the identity in the group of such automorphisms. Secondly, integrating $\omega_M$ along the isotopy yields a flux-type class in $H^1(M;{\mathbb{R}})$, and we also require that this should vanish.
For the rest of this discussion, we concentrate on the case of smooth quartic surface, as in Example \[th:quartic\]. There are quartic surfaces with a rational singularity of type $(A_3)$ By smoothing out such a singularity and using Moser’s theorem, we see that $K$ contains an $(A_3)$ chain of Lagrangian spheres, which we denote by $(V_1,V_2,V_3)$. Consider the spheres $$\begin{aligned}
& L_1 = \tau_{V_1}\tau_{V_3}(V_2) = \tau_{V_3}\tau_{V_1}(V_2), \\
& L_2 = \tau_{V_1}^{-1}\tau_{V_3}^{-1}(V_2) = \tau_{V_3}^{-1}\tau_{V_1}^{-1}(V_2).
\end{aligned}$$ Figure \[fig:quiver1\] shows a schematic picture of these Lagrangian submanifolds, in the style of [@khovanov-seidel98].
(0,0)![\[fig:quiver1\]](quiver1.pstex "fig:")
\#1\#2\#3\#4\#5[ @font ]{}
(1896,1611)(1753,-4175) (3001,-3286)[(0,0)\[lb\]]{} (3151,-2761)[(0,0)\[lb\]]{} (2551,-2761)[(0,0)\[lb\]]{} (1951,-2761)[(0,0)\[lb\]]{} (3001,-4111)[(0,0)\[lb\]]{}
\[th:2-twists\] $\tau_{L_1}\tau_{L_2}$ is indistinguishable from the identity by topological means.
For any Lagrangian two-sphere, the Dehn twist and its inverse are isotopic as diffeomorphisms, hence so are $\tau_{L_1} = \tau_{V_1}\tau_{V_3}\tau_{V_2}\tau_{V_3}^{-1}\tau_{V_1}^{-1}$ and $\tau_{L_2}^{-1} = \tau_{V_1}^{-1}\tau_{V_3}^{-1}\tau_{V_2}^{-1}\tau_{V_3}\tau_{V_1}$. From an analysis of the simultaneous resolution of the $(A_3)$ singularity as in [@khovanov-seidel98; @seidel04b], one obtains the following stronger [*fragility*]{} statement. There is a deformation $(\omega_{K,r})$ ($0 \leq r < \epsilon$) of the given symplectic form $\omega_K = \omega_{K,0}$, and a smooth family $f_r$ of diffeomorphisms with $f_0 = \tau_{L_1}\tau_{L_2}$ and $f_r^*\omega_{K,r} = \omega_{K,r}$, such that for all $r>0$, $f_r$ is isotopic to the identity within the group of symplectic automorphisms associated to the form $\omega_{K,r}$. Using such an isotopy immediately shows that the second part of Definition \[th:interesting\] holds, and the third part is trivial since $H^1(K;{\mathbb{R}}) = 0$.
A straightforward computation using the relation between Dehn twists and algebraic twists [@seidel04 Corollary 17.17] shows that (for suitable choices of gradings) $$\mathit{HF}^*(L_1,L_2) = \begin{cases} R^2 & * = 0, \\ R & * = 1,2, \\ 0 & \text{otherwise.} \end{cases}$$ This yields a concrete example where Lemma \[th:regular-representation\] applies. In particular, one sees that $\tau_{L_1}\tau_{L_2}$ is not symplectically isotopic to the identity (a known result, compare [@seidel-thomas99]).
Symplectic mapping tori\[sec:mapping-tori\]
===========================================
The symplectic mapping torus construction provides a way of obtaining interesting examples of symplectic manifolds from automorphisms. A complete description of the Fukaya categories of symplectic mapping tori is beyond the aim of this paper (but see Section \[subsec:algebraic-model\] for some conjectural discussion). Instead, we focus on a particular class of mapping tori, and consider only the most obvious Lagrangian submanifolds, which are fibered over circles in the (two-torus) base. For those submanifolds, ad hoc methods parallel to those in Section \[subsec:elliptic\] are sufficient to carry out the necessary Floer cohomology computations. Under suitable additional assumptions, this will allow us to show that the Lagrangian isotopy obtained by moving the circle around the base can be encoded into a perfect family.
Concretely, the starting point for our considerations will always be a symplectic $K3$ surface $K$, by which we mean a closed symplectic four-manifold diffeomorphic to a $K3$ surface. Recall that this is simply connected, admits a perfect Morse function (one without critical points of index $1$ or $3$) [@harer-kas-kirby], and is [*Spin*]{}. The symplectic structure necessarily has $c_1(K) = 0$ [@taubes94], and we choose a trivialization of the anticanonical line bundle in the unique homotopy class. We also suppose that a symplectic automorphism $f \in \mathit{Aut}(K)$ is given which has nondegenerate fixed points as well as nondegenerate $2$-periodic points, and which is indistinguishable from the identity by topological means (Definition \[th:interesting\]). As part of the last-mentioned condition, $f$ is isotopic to the identity in $\mathit{Diff}(K)$, and we fix such an isotopy, as well as a grading of $f$.
Basic geometry\[subsec:basic-mapping-torus\]
--------------------------------------------
Consider $K^- \times K$, where the sign of the symplectic form is reversed on the first factor, as in Example \[th:graph\]. The symplectic mapping torus of $f \times f \in \mathit{Aut}(K^- \times K)$, which we denote by $E = E_f$, is $$\label{eq:symplectic-mapping-torus}
\begin{aligned}
& E = {\mathbb{R}}\times {\mathbb{R}}\times K \times K\;\; / \;\; (p,q,x,y) \sim (p,q-1,x,y) \sim (p-1,q,f(x),f(y)), \\
& \omega_E = dp \wedge dq - \omega_K(x) + \omega_K(y).
\end{aligned}$$ By definition, projection $\pi: E \rightarrow T = {\mathbb{R}}^2/{\mathbb{Z}}^2$ is a locally trivial Hamiltonian fibration, with monodromy $f \times f$ in $p$-direction, and trivial monodromy in $q$-direction. Our assumptions on $f$ ensure that $E$ is diffeomorphic to $T \times K \times K$, in a way which is compatible with the homotopy classes of almost complex structures, and which maps $[\omega_E]$ to $[dp \wedge dq] \times 1 + 1 \times [\omega_{K^- \times K}] \in H^2(T \times K \times K;{\mathbb{R}})$. Moreover, the grading of $f$ yields a trivialization of the anticanonical line bundle of $E$.
As a symplectic fibration over a surface, $E$ is an object of the TCFT with target space $K^- \times K$ discussed in Remark \[th:tcft\]. Hence, there is an associated numerical invariant (a priori an element of $R$, but which will actually turn out to be an integer) counting its pseudo-holomorphic sections. This can be computed in two different ways. On one hand, in terms of , $E$ is obtained by gluing together the two boundary components of $Z_{f \times f} \times [0,1]$ in the trivial way, which means that the numerical invariant is the Euler characteristic of $\mathit{HF}^*(f \times f)$. On the other hand, one can think it as $[0,1] \times S^1 \times K^- \times K$ with both ends glued together using a twist by $f \times f$, in which case the numerical invariant is the supertrace of the action of $f \times f$ on $H^*(K^- \times K;R)$. Both ways of course yield the same result, namely the square of the Lefschetz fixed point number of $f$.
Consider the following Lagrangian submanifolds in $E$: $$\label{eq:lagrangians}
\begin{aligned}
& \Delta_1 = \{q = 0, \;\; y = x\}, \\
& \Delta_2 = \{q = -2p, \;\; y = x\}, \\
& \Delta_{3,u} = \{p = m_0, \;\; y = x \}. \\
\end{aligned}$$ In the last line, the parameter $u \in R^\times$ is written as $u = \hbar^{m_0}a$ with $a \in \mathit{GL}_0(1,R)$. All Lagrangian submanifolds in fibre over loops in $T$, with fibre $K$. These fibrations are actually trivial (tautologically so for $\Delta_{3,u}$, and by using the isotopy $f {\simeq}\mathit{id}$ in the other cases). In particular, we can choose the product of the trivial [*Spin*]{} structures on the underlying loop and the unique [*Spin*]{} structure on $K$. Moreover, our Lagrangian submanifolds also admit gradings (we make a particular choice of gradings, which will become clear in the Floer cohomology formulas below). On $\Delta_{3,u}$ we use the local system $\xi_u$, pulled back from the underlying loop $\{m_0\} \times S^1 \subset T$, which has fibre $R$ and holonomy $a$ in positive $q$-direction. The other two submanifolds carry trivial local systems. The last ingredient needed in order to turn them into objects of the Fukaya category is a choice of almost complex structures as in Assumption \[th:adiscic\]. Choose a one-parameter family $(J_{f,t})$ of almost complex structures on $K$ as in , with the additional (generic, for dimension reasons) property that there are no non-constant $J_{f,t}$-holomorphic spheres for any $t$. We will use the same almost complex structure on $E$ for all three Lagrangian submanifolds : $$\label{eq:j-delta}
(J_\Delta)_{p,q,x,y} = i \times (-J_{f,p+1/2,x}) \times J_{f,p+1/2,y}.$$ Projection to $T$ is $(J_\Delta,i)$-holomorphic, hence there are no non-constant $J_\Delta$-holomorphic spheres. Similarly,
\[th:project\] There are no non-constant $J_\Delta$-holomorphic discs with boundary on any one of the submanifolds .
By projecting to $T$, one sees that any disc must be contained in a fibre. There, it is a map $(D,\partial D) \rightarrow K \times K$ which is holomorphic for $(-J_{f,p+1/2}) \times J_{f,p+1/2}$, and has boundary on the diagonal. By the doubling trick already mentioned in Example \[th:graph\], such discs correspond to $J_{f,p+1/2}$-holomorphic spheres in $K$.
As a consequence, the Floer cohomology of each of our submanifolds with itself is canonically isomorphic to its ordinary cohomology. The other Floer cohomology groups are: $$\begin{aligned}
& \mathit{HF}^*(\Delta_1,\Delta_2) {\cong}H^*(K;R) \oplus H^*(K;R), \label{eq:two-points} \\
& \mathit{HF}^*(\Delta_2,\Delta_{3,u}) {\cong}(\xi_u)_{(m_0,-2m_0)} \otimes H^*(K;R), \\
& \mathit{HF}^*(\Delta_1,\Delta_{3,u}) {\cong}(\xi_u)_{(m_0,0)} \otimes H^*(K;R). \label{eq:one-point-hf}\end{aligned}$$ Dually, one can write $$\begin{aligned}
& \mathit{HF}^*(\Delta_2,\Delta_1) {\cong}H^*(K;R)[-1] \oplus H^*(K;R)[-1], \label{eq:dual-two-point} \\
& \mathit{HF}^*(\Delta_{3,u},\Delta_2) {\cong}(\xi_u)_{(m_0,-2m_0)}^\vee \otimes H^*(K;R)[-1], \\
& \mathit{HF}^*(\Delta_{3,u},\Delta_1) {\cong}(\xi_u)_{(m_0,0)}^\vee \otimes H^*(K;R)[-1]. \label{eq:dual-point}\end{aligned}$$ The proofs of these isomorphisms are straightforward in the Morse-Bott formalism from Section \[subsec:clean\]. For instance, consider : one takes $J_{\Delta_1,\Delta_2,t} = J_\Delta$ to be the constant family, in which case the same argument as in Lemma \[th:project\] shows that there are no non-constant holomorphic strips, immediately reducing the situation to ordinary Morse theory on $\Delta_1 \cap \Delta_2$ (to be precise, one has to check that the local coefficient system $o_C$ on each component $C \subset \Delta_1 \cap \Delta_2$ is trivial; but that is clear since locally near $C$, the geometry splits as a product of base and fibre). The remaining isomorphisms are proved in exactly the same way.
Some of the products on Floer cohomology are also elementary, meaning that they involve no actual count of nontrivial holomorphic maps. For instance, consider $$\label{eq:trivial-products}
\begin{aligned}
& \mathit{HF}^*(\Delta_2,\Delta_2) \otimes \mathit{HF}^*(\Delta_1,\Delta_2) \longrightarrow \mathit{HF}^*(\Delta_1,\Delta_2), \\
& \mathit{HF}^*(\Delta_1,\Delta_1) \otimes \mathit{HF}^*(\Delta_1,\Delta_2) \longrightarrow \mathit{HF}^*(\Delta_1,\Delta_2).
\end{aligned}$$ Each of these turns out to be the action of $H^*(\Delta_k;R)$ on $H^*(\Delta_1 \cap \Delta_2;R)$ by restriction and cup-product. The proof again uses a constant family of almost complex structures equal to $J_\Delta$, a suitable choice of Morse functions on $\Delta_k$, and the arguments from Lemma \[th:project\]; we omit the details. Note that by the cyclic symmetry of the product, this also determines $$\begin{aligned}
& \mathit{HF}^*(\Delta_1,\Delta_1) \otimes \mathit{HF}^*(\Delta_2,\Delta_1) \longrightarrow \mathit{HF}^*(\Delta_2,\Delta_1), \\
& \mathit{HF}^*(\Delta_2,\Delta_1) \otimes \mathit{HF}^*(\Delta_2,\Delta_2) \longrightarrow \mathit{HF}^*(\Delta_2,\Delta_1), \\
& \mathit{HF}^*(\Delta_2,\Delta_1) \otimes \mathit{HF}^*(\Delta_1,\Delta_2) \longrightarrow \mathit{HF}^*(\Delta_1,\Delta_1), \\
& \mathit{HF}^*(\Delta_1,\Delta_2) \otimes \mathit{HF}^*(\Delta_2,\Delta_1) \longrightarrow \mathit{HF}^*(\Delta_2,\Delta_2).\end{aligned}$$
We conclude this preliminary discussion by introducing low-degree generators analogous to those in Section \[subsec:twotorus\], namely $$\begin{aligned}
& \mathit{HF}^0(\Delta_1,\Delta_2) = R \cdot [w_1] \oplus R \cdot [w_2], \\
& \mathit{HF}^1(\Delta_2,\Delta_1) = R \cdot [w_3] \oplus R \cdot [w_4], \\
& \mathit{HF}^0(\Delta_1,\Delta_{3,u}) = R \cdot [z_{1,u}], \\
& \mathit{HF}^1(\Delta_{3,u},\Delta_1) = R \cdot [y_{1,u}], \\
& \mathit{HF}^0(\Delta_2,\Delta_{3,u}) = R \cdot [z_{2,u}], \\
& \mathit{HF}^1(\Delta_{3,u},\Delta_2) = R \cdot [y_{2,u}]
\end{aligned}$$ (the notation $[w_1]$ indicates the cohomology class for some underlying choice of cocycle $w_1$, and we’ve inserted the dots to avoid confusion with polynomial rings). As before, $[w_1]$ and $[-w_2]$ correspond to the class $1 \in H^0(K;R)$ under , for the components lying over $({{\textstyle\frac{1}{2}}},0)$ and $(0,0)$, respectively; $[z_{1,u}]$ and $[z_{2,u}]$ are defined using ; and the other generators are fixed in such a way that the products $$\label{eq:delta-dual-generators}
\begin{aligned}
& [w_4] \cdot [w_1] \in \mathit{HF}^1(\Delta_1,\Delta_1), \\
& [w_1] \cdot [w_4] \in \mathit{HF}^1(\Delta_2,\Delta_2), \\
& -[w_2] \cdot [w_3] \in \mathit{HF}^1(\Delta_2,\Delta_2), \\
& -[w_3] \cdot [w_2] \in \mathit{HF}^1(\Delta_1,\Delta_1), \\
\end{aligned}
\qquad
\begin{aligned}
& [y_{1,u}] \cdot [z_{1,u}] \in \mathit{HF}^1(\Delta_1,\Delta_1), \\
& [z_{1,u}] \cdot [y_{1,u}] \in \mathit{HF}^1(\Delta_{3,u},\Delta_{3,u}), \\
& [y_{2,u}] \cdot [z_{2,u}] \in \mathit{HF}^1(\Delta_2,\Delta_2), \\
& [z_{2,u}] \cdot [y_{2,u}] \in \mathit{HF}^1(\Delta_{3,u},\Delta_{3,u})
\end{aligned}$$ all yield the generator of $H^1(S^1 \times K;R) {\cong}H^1(S^1;R)$ obtained by orienting the underlying loops in $T^2$ as in Section \[subsec:twotorus\]. Moreover, from our computation of it follows that $[w_1] \cdot [w_3]$, $[w_3] \cdot [w_1]$, $[w_2] \cdot [w_4]$ and $[w_4] \cdot [w_2]$ vanish. Hence,
\[th:q-cohomologically\] The subspace of $\bigoplus_{i,j=1}^2 \mathit{HF}^*(\Delta_i,\Delta_j)$ consisting of elements of degree $\leq 1$ is a subalgebra, and in fact isomorphic to the algebra $Q$ from Definition \[th:q-algebra\].
Counting triangles\[subsec:flat-triangles\]
-------------------------------------------
In parallel with our original discussion of the two-torus, we will also need to determine parts of the $A_\infty$-structure which do involve counting holomorphic curves. First of all, we need the counterpart of , which computes the product $$\label{eq:delta-product}
\begin{aligned}
& \mathit{HF}^0(\Delta_2, \Delta_{3,u}) \otimes \mathit{HF}^0(\Delta_1,\Delta_2) \longrightarrow \mathit{HF}^0(\Delta_1, \Delta_{3,u}), \\
& [z_{2,u}] \cdot [w_1] = \vartheta_{2,1}(u) [z_{1,u}], \\
& [z_{2,u}] \cdot [w_2] = -\vartheta_{2,2}(u)[z_{1,u}].
\end{aligned}$$ One can use associativity and to derive two more products from this, namely $$\label{eq:delta-product-2}
\begin{aligned}
& \mathit{HF}^0(\Delta_1,\Delta_2) \otimes \mathit{HF}^1(\Delta_{3,u},\Delta_1) \longrightarrow \mathit{HF}^1(\Delta_{3,u},\Delta_2), \\
& [w_1] \cdot [y_{1,u}] = \vartheta_{2,1}(u) [y_{2,u}], \\
& [w_2] \cdot [y_{1,u}] = -\vartheta_{2,2}(u) [y_{2,u}],
\end{aligned}$$ and $$\label{eq:delta-product-3}
\begin{aligned}
& \mathit{HF}^1(\Delta_{3,u},\Delta_1) \otimes \mathit{HF}^0(\Delta_2,\Delta_{3,u}) \longrightarrow \mathit{HF}^1(\Delta_2,\Delta_1), \\
& [y_{1,u}] \cdot [z_{2,u}] = \vartheta_{2,2}(u)[w_3] + \vartheta_{2,1}(u)[w_4],
\end{aligned}$$ which are the analogues of and , respectively. Next consider $$x = \vartheta_{2,2}(u)[w_1] + \vartheta_{2,1}(u)[w_2] \in \mathit{HF}^0(\Delta_1,\Delta_2).$$ We know from that $[z_{2,u}] \cdot x = 0$, and from that $x \cdot [y_{1,u}] = 0$. Therefore one can form the Massey product $\langle [z_{2,u}], x, [y_{1,u}] \rangle$ (see [@seidel04 Remark 1.2] for the sign conventions in effect here). Generally speaking, such a product takes values in the quotient of $\mathit{HF}^0(\Delta_{3,u},\Delta_{3,u})$ by the two subspaces $[z_{2,u}] \cdot \mathit{HF}^0(\Delta_{3,u},\Delta_2)$ and $\mathit{HF}^{-1}(\Delta_1,\Delta_{3,u}) \cdot [y_{1,u}]$, but both vanish in our case, leading to a strictly well-defined Massey product, which we will show to be the following multiple of the identity class $[e_{3,u}]$: $$\label{eq:delta-massey}
\langle [z_{2,u}], x, [y_{1,u}] \rangle = u(\vartheta_{2,2}'(u)\vartheta_{2,1}(u) - \vartheta_{2,1}'(u)\vartheta_{2,2}(u)) [e_{3,u}].$$
One important aspect of and is that both expressions live on the cohomology level, and are therefore independent of auxiliary choices (of almost complex structures and such) made in the definition of the Fukaya category. Another relevant point is that different homotopy classes of holomorphic triangles can be counted separately. In order the make the latter idea visible on a formal level, let’s introduce a new formal variable $\epsilon$, and temporarily enlarge the ring from by allowing the coefficients $c_k$ to lie in ${\mathbb{C}}[\epsilon]$. At the same time, we mark a point $z \in T^2$ such that the fibre $\pi^{-1}(z)$ is disjoint from the Lagrangian submanifolds under consideration. We can then construct a version of the Fukaya category relative to that fibre, where the $\epsilon^r$ term in $\mu^d$ comes from pseudo-holomorphic maps which have intersection number $r$ with $\pi^{-1}(z)$ (in order for this to be always $\geq 0$, the almost complex structures have to be such that $\pi^{-1}(z)$ is an almost complex submanifold, and the inhomogeneous terms should vanish quadratically near that fibre; both assumptions are unproblematic as far as transversality is concerned). All the Floer cohomology computations from Section \[subsec:basic-mapping-torus\] remain valid in this context.
![\[fig:abcd-more\]](abcd-more.eps)
Suppose for concreteness that $-1/2 < m_0 < 1/2$ (the remaining case can be dealt with in a similar way), and choose $z = (1/2,1/2)$. Consider the $\epsilon^0$ term of the second product in . Suppose that we use only almost complex structures such that projection to $T$ is pseudo-holomorphic, and zero inhomogeneous terms. Then, all contributions to this term come from maps which cover the shaded triangle in Figure \[fig:abcd-more\]. Note that the same holds for any other choice of almost complex structure for which the projection to $T$ is pseudo-holomorphic. Within the given class, we can find an almost complex structure which, on the preimage of a neighbourhood of that triangle, is the product $i \times (-J_{f,t}) \times J_{f,t}$ for some fixed $t$. In that case, pseudo-holomorphic lifts of that triangle are given by $(-J_{f,t}) \times J_{f,t}$-holomorphic maps from a three-punctured disc to $K \times K$, with boundary on the diagonal. But by the same argument as in Lemma \[th:project\], such maps are necessarily constant. This determines the moduli space, and simultaneously shows that it is regular, yielding a contribution of $-1 = -\hbar^0\epsilon^0$ to the product (the sign is a consequence of the convention used when defining $w_2$). On the face of it, this argument would seem to fail in general, since it relied on the fact that the triangle in $T$ was embedded in order to construct the desired almost complex structure. However, one can reduce the computation for any power $\epsilon^r$ to the same kind of situation, by passing to a sufficiently large finite cover of $E$ (which depends on $r$) and using the additional freedom to change the almost complex structure there, breaking its symmetry under the covering group. Formally, this means we are using an $\epsilon$-enhanced version of the pullback from Addendum \[th:pullback-functor\], but remaining on the cohomology level. This allows us to easily compute the following products in the $\epsilon$-enhanced framework: $$\label{eq:epsilon-product}
\begin{aligned}
& [z_{2,u}] \cdot [w_1] = \epsilon^{-1/4} \vartheta_{2,1}(u)_{\hbar \mapsto \hbar\epsilon} \, [z_{1,u}], \\
& [z_{2,u}] \cdot [w_2] = -\vartheta_{2,2}(u)_{\hbar \mapsto \hbar\epsilon}\, [z_{1,u}],
\end{aligned}$$ where $\hbar \mapsto \hbar\epsilon$ is a substitution of variables, and dividing by $\epsilon^{1/4}$ in the first line keeps the powers of $\epsilon$ integral. On the other hand, setting $\epsilon=1$ recovers our previous situation by definition, which concludes our proof of . The strategy for is similar. One first passes to the $\epsilon$-analogue $\langle [z_{2,u}], \vartheta_{2,2}(u)_{\hbar \mapsto \hbar\epsilon} [w_1] + \epsilon^{-1/4}\vartheta_{2,1}(u)_{\hbar \mapsto \hbar\epsilon}[w_2], [y_{1,u}] \rangle$, which is well-defined as a consequence of . For any power $\epsilon^r$, only finitely many homotopy classes can contribute, and after passing to a suitable finite cover one again has only constant maps in fibre direction, which means that the answer is the same as for the two-torus itself, where one can derive it from .
Suppose now that $u \notin \{\pm \hbar^{k/2} \, : \, k \in {\mathbb{Z}}\}$. The computations above imply that the following is an exact triangle in $H^0(\mathit{Fuk}(E)^{\mathit{tw}})$: $$\label{eq:m-triangle}
\xymatrix{\Delta_1 \ar[rrrr]^-{\frac{\vartheta_{2,2}(u)[w_1] + \vartheta_{2,1}(u)[w_2]}{\vartheta_{4,3}'(1)(\vartheta_{4,1}(u) - \vartheta_{4,3}(u))}} &&&& \Delta_2 \ar[dll]^{\;\;\;\;\;\;\; ([z_{2,u}], [z_{2,u^{-1}}])} \\
&& \Delta_{3,u} \oplus \Delta_{3,u^{-1}} \ar[ull]_{[1]}^{([y_{1,u}], -[y_{1,u^{-1}}])\;\;\;\;\;\;\;} &&
}$$ To spell this out a little more, one first proceeds as in Lemma \[th:exact-triangle\] to show that $\Delta_{3,u} \oplus \Delta_{3,u^{-1}}$ is a direct summand of the mapping cone of the horizontal map in . Temporarily denote that cone by $\tilde{C}$. There is a spectral sequence converging to $H^*(\mathit{hom}_{\mathit{Fuk}(E)^{\mathit{tw}}}(\tilde{C},\tilde{C}))$, whose starting page is $$\label{eq:cone-spectral-sequence}
E_1^{pq} = \begin{cases} \mathit{HF}^q(\Delta_2,\Delta_1) & p = -1, \\
\mathit{HF}^q(\Delta_1,\Delta_1) \oplus \mathit{HF}^q(\Delta_2,\Delta_2) & p = 0, \\
\mathit{HF}^q(\Delta_1,\Delta_2) & p = 1, \\
0 & \text{otherwise.}
\end{cases}$$ The differential $d_1: E_1^{pq} \rightarrow E_1^{p+1,q}$ is given by multiplying by the morphism used to form the cone, with suitable signs. In particular, the subspace of elements in of total degree $p+q = 0$ is four-dimensional; and the differentials $E_1^{-1,1} \rightarrow E_1^{0,1}$, $E_1^{0,0} \rightarrow E_1^{1,0}$ are both nonzero. This shows that $H^0(\mathit{hom}_{\mathit{Fuk}(E)^{\mathit{tw}}}(\tilde{C},\tilde{C}))$ is of dimension $\leq 2$, which implies that $\tilde{C}$ must be quasi-isomorphic to $\Delta_{3,u} \oplus \Delta_{3,u^{-1}}$.
\[th:degree1\] For a suitable choice of auxiliary data, the differential on $\bigoplus_{i,j=1}^2 \mathit{CF}^*(\Delta_i,\Delta_j)$ vanishes, and moreover, the subspace of elements of degree $\leq 1$ is an $A_\infty$-subalgebra.
Choose a perfect Morse function $h_K$ on $K$. Choose also a perfect Morse function $h_{S^1}$ on the circle $S^1 = {\mathbb{R}}/{\mathbb{Z}}$, with minimum at $p_0 = 1/4$ and maximum at $p_1 = 3/4$. When defining $\mathit{CF}^*(\Delta_1,\Delta_1)$ in the Morse-Bott formalism from Section \[subsec:clean\], take the Morse function on $\Delta_1 {\cong}S^1 \times K$ given by $$h_{\Delta_1,\Delta_1}(p,x) = h_{S^1}(p) + \mathrm{constant} \cdot h_K(x),$$ where the constant is small and positive (and correspondingly, we choose the product Riemannian metric). The differential is obviously trivial. Moreover, we have that $$\label{eq:barrier}
\begin{aligned}
& \partial_p h_{\Delta_1,\Delta_1} > 0 && \text{along $\{1/2\} \times K$,} \\
& \partial_p h_{\Delta_1,\Delta_1} < 0 && \text{along $\{0\} \times K$.}
\end{aligned}$$ When defining the higher order $A_\infty$-structure on $\mathit{CF}^*(\Delta_1,\Delta_1) = \mathit{CM}^*(h_{\Delta_1,\Delta_1})$, we proceed as in Remark \[th:single-l\], but take care that all the auxiliary families of Morse functions appearing in the process still satisfy (this is an open condition, hence does not stand in the way of transversality arguments). As a result, if $a_1,\dots,a_d$ are generators corresponding to critical points, at least one of which lies in $\{3/4\} \times K$, then $\mu^d_{\mathit{Fuk}(E)}(a_d,\dots,a_1)$ must be a linear combination of critical points also lying in $\{3/4\} \times K$. In particular, take $a$ to be the unique generator corresponding to a critical point of index $1$, which is the minimum of $h_K$ placed in $\{3/4\} \times K$. Then $\mu^d_{\mathit{Fuk}(E)}(a,\dots,a)$ is a linear combination of critical points of index $2$, which moreover lie in $\{3/4\} \times K$. But there are no such points, hence we have proved that these particular $A_\infty$-products must vanish. Of course, all these considerations can be applied to $\Delta_2$ as well.
Similarly, when defining $\mathit{CF}^*(\Delta_1,\Delta_2)$ and $\mathit{CF}^*(\Delta_2,\Delta_1)$, we choose minimal Morse functions, so that $\mu^1 = 0$. Because of this minimality property, the product $\mu^2$ is determined entirely by Lemma \[th:q-cohomologically\], which shows in particular that the product of any two elements of degree $1$ vanishes. Consider a higher product of elements of degree $\leq 1$, $$\label{eq:01-product}
\mu^d_{\mathit{Fuk}(E)}: \mathit{CF}^{\leq 1}(\Delta_{i_{d-1}},\Delta_{i_d}) \otimes \cdots
\mathit{CF}^{\leq 1}(\Delta_{i_0},\Delta_{i_1}) \longrightarrow \mathit{CF}^{\leq 2}(\Delta_{i_0},\Delta_{i_d})$$ for some $d > 2$ and $i_0,\dots,i_d \in \{1,2\}$, and where the $i_k$ are not all equal (since that case has been dealt with before). If one of the inputs has degree $0$, the output automatically has degree $\leq 1$. The only remaining case is when all the inputs have degree $1$, which forces $(i_0,\dots,i_d) = (2,\dots,2,1,\dots,1)$, but then the output would lie in $\mathit{CF}^2(\Delta_2,\Delta_1)$, which vanishes by minimality and .
We have now obtained an embedding $G^1: Q \rightarrow \mathit{Fuk}(E)$ which is compatible with the multiplication $\mu^2$, and which therefore extends to an $A_\infty$-functor $G: Q_{\tilde{p}} \rightarrow \mathit{Fuk}(E)$ for some a priori unknown polynomial $\tilde{p}$.
$\tilde{p} = p$ is the unit torus polynomial.
Fix some $u \notin \{\pm h^{k/2} \, : \, k \in {\mathbb{Z}}\}$, and consider the morphism $v \in e_2 Q e_1 = \mathit{hom}_{Q_{\tilde{p}}}(X_1,X_2)$ given by the same formula as the horizontal arrow in . Denote by $C_v$ the cone of that morphism in $Q_{\tilde{p}}^{\mathit{tw}}$. Because $A_\infty$-functors preserve exact triangles, we have a commutative diagram in the category $H^0(\mathit{Fuk}(E)^{\mathit{tw}})$, $$\label{fukaya-m-triangle}
\xymatrix{
\cdots \ar[r] & G(X_2) \ar@{=}[d] \ar[r] & G^{\mathit{tw}}(C_v) \ar[r] \ar@{-->}[d]^{{\cong}} & G(X_1)[1] \ar@{=}[d] \ar[r] & \cdots \\ \cdots \ar[r] & \Delta_2 \ar[r] & \Delta_{3,u} \oplus \Delta_{3,u^{-1}} \ar[r] & \Delta_1[1] \ar[r] & \cdots
}$$ where the top row is the obvious exact triangle, the bottom one is , and the dotted isomorphism is the only new ingredient. It follows from our previous analysis of that $G$ induces an isomorphism $$\label{eq:g-isomorphism}
H^0(\mathit{hom}_{Q_{\tilde{p}}^{\mathit{tw}}}(C_v,C_v)) \longrightarrow \mathit{HF}^0(\Delta_{3,u},\Delta_{3,u}) \oplus \mathit{HF}^0(\Delta_{3,u^{-1}},\Delta_{3,u^{-1}}) = R^2.$$ Consider the endomorphism $\tilde{t}$ of $\Delta_{3,u} \oplus \Delta_{3,u^{-1}}$ given by the same linear combination of identity elements as in . Using one sees that this satisfies the analogue of . Hence, its preimage under satisfies the criteria from Lemma \[th:exactly-complete-triangle\], which shows that $\tilde{p}(v_1,v_2) = p(v_1,v_2)$, exactly as in Lemma \[th:fukaya-torus\].
We conclude this discussion by looking at the maps on Hochschild cohomology induced by $G$, and how they relate to the open-closed string map.
\[th:g-push\] The map $$\label{eq:g-push}
H(G_*): \mathit{HH}^*(Q_p,Q_p) \longrightarrow \mathit{HH}^*(Q_p,G^*\mathit{Fuk}(E))$$ is an isomorphism in degrees $\leq 1$.
It is convenient to replace $Q_p$ by its quasi-isomorphic image in $\mathit{Fuk}(E)$ described in Lemma \[th:degree1\]. Denoting that by $\tilde{Q}$, what we then have to look at is the effect of the inclusion $\tilde{Q} \hookrightarrow \mathit{Fuk}(E)$. Any Hochschild cochain $g \in \mathit{CC}^{\leq 1}(\tilde{Q},\mathit{Fuk}(E))$ necessarily takes values in the subspace of morphisms of degree $\leq 1$, which is precisely $\tilde{Q}$, so we have $\mathit{CC}^{\leq 1}(\tilde{Q},\tilde{Q}) = \mathit{CC}^{\leq 1}(\tilde{Q},\mathit{Fuk}(E))$. This fails in degree $2$, but at least we have an injective map of cochains there, which is precisely what’s needed to prove the desired statement.
Recall that $\mathit{HH}^1(Q_p,Q_p)$ is two-dimensional, with generators $[g_1]$, $[g_2]$ which were (partially) described in Addendum \[th:quasi-generators\].
\[th:g-pull\] The composition $$\label{eq:g-pull}
\mathit{QH}^*(E) \longrightarrow \mathit{HH}^*(\mathit{Fuk}(E),\mathit{Fuk}(E))
\xrightarrow{H(G^*)} \mathit{HH}^*(Q_p,G^*\mathit{Fuk}(M))$$ sends $[dp] \in H^1(E;{\mathbb{R}})$ to $H(G_*)([g_1] + [g_2])$, and $[dq]$ to $H(G_*)(-2[g_2])$.
By construction, we have a commutative diagram $$\label{eq:open-closed-g}
\xymatrix{
\mathit{HH}^*(Q_p,Q_p) \ar[r]^{G_*} \ar[d] & \ar[d] \mathit{HH}^*(Q_p,\mathit{Fuk}(E)) & \ar[l]_{G^*}
\mathit{HH}^*(\mathit{Fuk}(E),\mathit{Fuk}(E)) \ar[dl] \\
e_1Q_pe_1 \oplus
e_2Q_pe_2
\ar[r]^-{H(G)} &
\mathit{HF}^*(\Delta_1,\Delta_1)
\oplus
\mathit{HF}^*(\Delta_2,\Delta_2)
&
}$$ The vertical arrow on the left is an isomorphism in degree $1$ (by Addendum \[th:quasi-generators\]), and hence so is the one in the middle (using Lemma \[th:g-push\]). On the other hand, composition of the open-closed string map with the diagonal arrow in just yields the ordinary restriction map $\mathit{QH}^*(E) \rightarrow H^*(\Delta_1;R) \oplus H^*(\Delta_2;R)$. The rest is diagram-chasing.
We now consider the analogue of Corollary \[th:t2-family\]. Let ${\mathscr{S}}= \mathit{Spec}({\mathscr{R}})$ be the affine curve associated to the unit torus polynomial $p$, and $\theta$ its standard one-form. The image of $$\label{eq:explicit-current-2}
\theta \otimes [dq] \in H^0({\mathscr{S}},\Omega^1_{\mathscr{S}}) \otimes \mathit{QH}^1(E)$$ under the open-closed string map is a deformation field, which we denote by $[\gamma]$, for the constant family ${\mathscr{Fuk}}(E)$ of Fukaya categories over ${\mathscr{S}}$.
\[th:m-family\] There is a perfect family of modules ${\mathscr{D}}_3$ which follows $[\gamma]$, and whose fibre at a point $(s_1,s_2) \in {\mathscr{S}}$ is isomorphic to $\Delta_{3,u}$, where $u \in R^\times/\hbar^{{\mathbb{Z}}}$ satisfies .
Take the family from Corollary \[th:t2-family\] and map it to $\mathit{Fuk}(E)$ using $G$. As a consequence of the general discussion of functoriality in Section \[subsec:functor\], the image family image indeed follows (the equality of Hochschild cohomology classes required in Assumption \[th:matching-deformations\] comes from Lemma \[th:g-pull\]). By construction, the object of the family at any point is a direct summand of a mapping cone. The triangle identifies that mapping cone with $\Delta_{3,u} \oplus \Delta_{3,u^{-1}}$, and one can follow the same computation as in the case of the two-torus to show that the summand picked out by the projection is indeed $\Delta_{3,u}$.
More Lagrangian submanifolds
----------------------------
$E$ admits a (graded) symplectic automorphism $F$ which is trivial on the base $T$, and equals $\mathit{id} \times f$ in each fibre (this makes sense since it commutes with $f \times f$). By applying that automorphism to our given Lagrangian submanifolds, we get another collection $$\label{eq:lagrangians-2}
\begin{aligned}
& \Gamma_1 = F(\Delta_1) = \{q = 0, \;\; y = f(x)\}, \\
& \Gamma_2 = F(\Delta_2) = \{q = -2p, \;\; y = f(x)\}, \\
& \Gamma_{3,u} = F(\Delta_{3,u}) = \{p = m_0, \;\; y = f(x)\}, \\
\end{aligned}$$ These come with induced gradings and [*Spin*]{} structures. We adopt the almost complex structure $J_{\Gamma,p,q,x,y} = i \times (-J_{f,p+1,x}) \times J_{f,p,y}$, which is the image under $F$ of $i \times (-J_{f,p+1,x}) \times J_{f,p+1,y}$, hence the previous computations apply to as well. We will also need to know how and interact. Unsurprisingly, the answers involve the fixed point Floer cohomology of $f$: $$\begin{aligned}
& \mathit{HF}^*(\Gamma_1,\Delta_1) {\cong}H^*(S^1;R) \otimes \mathit{HF}^*(f), \label{eq:circle-gd} \\
& \mathit{HF}^*(\Gamma_1,\Delta_2) {\cong}\mathit{HF}^*(f) \oplus \mathit{HF}^*(f), \\
& \mathit{HF}^*(\Gamma_1,\Delta_{3,u}) {\cong}(\xi_u)_{(m_0,0)} \otimes \mathit{HF}^*(f), \label{eq:point-gd} \\
& \mathit{HF}^*(\Gamma_2,\Delta_1) {\cong}\mathit{HF}^*(f)[-1] \oplus \mathit{HF}^*(f)[-1], \\
& \mathit{HF}^*(\Gamma_2,\Delta_2) {\cong}H^*(S^1;R) \otimes \mathit{HF}^*(f), \label{eq:circle-gd-2} \\
& \mathit{HF}^*(\Gamma_2,\Delta_3) {\cong}(\xi_u)_{(m_0,-2m_0)} \otimes \mathit{HF}^*(f), \\
& \mathit{HF}^*(\Gamma_{3,u},\Delta_2) {\cong}(\xi_u)_{(m_0,-2m_0)}^\vee \otimes \mathit{HF}^*(f)[-1], \\
& \mathit{HF}^*(\Gamma_{3,u},\Delta_1) {\cong}(\xi_u)_{(m_0,0)}^\vee \otimes \mathit{HF}^*(f)[-1], \\
& \mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u}) {\cong}H^*(S^1;R) \otimes \mathit{HF}^*(f). \label{eq:infty-infty}\end{aligned}$$ For each of these, we adopt a variant of the strategy in Example \[th:graph\], which means that we use the family of almost complex structures on $E$ given by $$\label{eq:j-gamma-delta-2}
(J_{\Gamma,\Delta,t})_{p,q,x,y} = i \times (-J_{f,p+1-t/2,x}) \times J_{f,p+t/2,y}.$$ In all cases listed above, pseudo-holomorphic strips must be contained in a fibre. We take such a strip $(u_x,u_y): {\mathbb{R}}\times [0,1] \rightarrow \pi^{-1}(p,q) {\cong}K \times K$ and transform it to a map $u$ as in , which then satisfies $u(s,t-1) = f(u(s,t))$ and $\partial_s u + J_{f,p+t}(u)\partial_t u = 0$. This makes the isomorphisms above obvious, with and perhaps requiring a little thought. Let’s consider briefly the first of the two. The intersection $\Gamma_1 \cap \Delta_1$ consists of a circle $C_x$ for each fixed point $x$ of $f$. The Maslov index of $C_x$ equals the Conley-Zehnder index of $x$. The local coefficient system $o_{C_x}$ has fibre $o_x$, and its monodromy is given by the natural action of $Df_x$ on $o_x$. It is a nontrivial observation, but one which is well-known as part of the mechanism underlying , that this action is trivial. If we then choose $h_{\Gamma_1,\Delta_1}$ to be the same Morse function $h_{S^1}$ on each circle, we get an isomorphism of graded vector spaces $$\label{eq:cone-product}
\mathit{CF}^*(\Gamma_1,\Delta_1) {\cong}\mathit{CM}^*(h_{S^1}) \otimes \mathit{CF}^*(f).$$ Using the previous observation about pseudo-holomorphic strips, it is not hard to see that this is compatible with the Floer differential. The other case is parallel.
Another way to see where the potential difficulty in lies is to consider for a moment a more general family of Lagrangian submanifolds fibred over the same base circle, namely $$\Gamma_1^m = \{q = 0, \;\; y = f^m(x)\}$$ for some $m \geq 1$. The intersection points of $\Gamma_1^m \cap \Delta_1$ in each fibre $\pi^{-1}(p,0)$ correspond to fixed points of $f^m$ (which we assume to be nondegenerate). However, this correspondence depends on $p \in {\mathbb{R}}$, rather than only on its image in ${\mathbb{R}}/{\mathbb{Z}}$: as we move around the circle, there is nontrivial monodromy which acts by $f$ on the set of these points. Moreover, even for points which are fixed by $f^p$ for some $p|m$, the induced action on $o_x$ (where $x$ is considered as an $m$-periodic point) can be nontrivial; this is the same phenomenon as the “bad orbits” in Symplectic Field Theory. Finally, while the moduli spaces of holomorphic strips fibre over $S^1 \times \{0\}$, that fibration can also be nontrivial. In fact, what one gets is a chain homotopy $$\mathit{CF}^*(\Gamma_1^m,\Delta_1) {\simeq}\mathit{Cone}(\mathit{id} - c_{f,f^m}: \mathit{CF}^*(f^m) \longrightarrow \mathit{CF}^*(f^m)).$$
We will also need to know two related products, namely $$\begin{aligned}
& \mathit{HF}^*(\Delta_1,\Delta_{3,u}) \otimes \mathit{HF}^*(\Gamma_1,\Delta_1) \longrightarrow \mathit{HF}^*(\Gamma_1,\Delta_{3,u}), \label{eq:d-product-1} \\
& \mathit{HF}^*(\Gamma_1,\Delta_{3,u}) \otimes \mathit{HF}^*(\Gamma_{3,u},\Gamma_1) \longrightarrow \mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u}). \label{eq:d-product-2}\end{aligned}$$ After using , the analogue of for the $\Gamma$ Lagrangian submanifolds, as well as , , , and cancelling the $\xi_u$ factors, these maps can be written as $$\begin{aligned}
& H^*(K;R) \otimes H^*(S^1;R) \otimes \mathit{HF}^*(f) \longrightarrow \mathit{HF}^*(f), \label{eq:s1-product-1} \\
& \mathit{HF}^*(f) \otimes H^*(K;R)[-1] \longrightarrow H^*(S^1;R) \otimes \mathit{HF}^*(f). \label{eq:s1-product-2}\end{aligned}$$
\[th:2-products\] The first map vanishes on the $H^1(S^1;R)$ summand, and on the $H^0(S^1;R)$ summand it reproduces the quantum cap module structure of $\mathit{HF}^*(f)$. The second map takes values in the $H^1(S^1;R)$ summand, and again reproduces the quantum cap structure.
To keep the notation simple, we consider only the first product and the case $u = 1$ (so $m_0 = 0$ and $\xi_u$ is trivial). Define $\mathit{CF}^*(\Gamma_1,\Delta_1)$ as in , taking care that $h_{S^1}$ has a single minimum at $(0,0)$ (and a single maximum elsewhere), and using the family of almost complex structures $J_{\Gamma,\Delta}$ from . Next, the intersection $\Delta_1 \cap \Delta_{3,u} {\cong}K$ is the diagonal in the fibre at $(0,0)$. We choose a Morse function $h_{\Delta_1,\Delta_{3,u}} = h_K$ on $K$ in order to define $\mathit{CF}^*(\Delta_1,\Delta_{3,u})$, and use the constant family of almost complex structures $J_\Delta$. Finally, the intersection $\Gamma_1 \cap \Delta_{3,u}$ is transverse, and we use the same family $J_{\Gamma,\Delta}$ for it as before. To form the quantum cap product , we use the same Morse function $h_K$, as well as a two-parameter family of almost complex structures of the form $$J_{\mathrm{cap},s,t} = J_{f,\psi(s,t)},$$ where $\psi: {\mathbb{R}}^2 \rightarrow {\mathbb{R}}$ is a function satisfying $\psi(s,t+1) = \psi(s,t)+1$ for all $(s,t)$, $\psi(s,t) = t$ for $|s| \gg 0$, and $\psi(s,t) = 0$ for $(s,t)$ close to $(0,1/2)$. This leaves enough freedom to achieve the required transversality properties.
The pearly trees that can, in principle, contribute to the product are shown in Figure \[fig:pearly-sequence\]. Consider for a moment the simplest such tree, which has just one trivalent vertex. The Riemann surface associated to that vertex can be written as $S_v = ({\mathbb{R}}\times [0,1]) \setminus \{0,1\}$. We choose the perturbation datum on $S_v$ to have trivial inhomogeneous term, and the following family $J_v$ of almost complex structures: $$(J_{v,s,t})_{p,q,x,y} = i \times (-J_{f,p+\psi(s/2,1-t/2),x}) \times J_{f,p+\psi(s/2,t/2),y}.$$ Importantly, near $(s,t) = (0,1)$ this reduces to $J_\Delta$. Solutions of the associated equation are all contained in the fibre over $(0,0)$. Moreover, in analogy with , they correspond bijectively to maps $u_v: {\mathbb{R}}^2 \rightarrow K$ solving the pseudo-holomorphic part of . Choose the families of Morse functions on the two semi-infinite edges of our pearly tree to be constant equal to $h_{\Gamma_1,\Delta_1}$ and $h_{\Delta_1,\Delta_{3,u}}$, respectively. The associated gradient (half-)flow line of $h_{\Gamma_1,\Delta_1}$ must necessarily be constant, whereas the other one yields the Morse-theoretic part of . Regularity is easy to check.
It remains to exclude contributions from more complicated pearly trees. The maps associated to the two-valent vertices in the upper branch of Figure \[fig:pearly-sequence\] are Floer differentials for the pair $(\Delta_1,\Delta_{3,u})$, but as we have seen before there are none, since they would correspond to non-constant $J_{f,1/2}$-holomorphic spheres in $K$. On the lower branch we equip all the finite length edges with the same constant family of Morse functions. But then, all the associated gradient flow lines are necessarily constant, which means that the length of the edge is a free parameter. After a necessary but easy regularity consideration, it follows that this cannot occur in zero-dimensional moduli spaces.
(0,0)![\[fig:pearly-sequence\]](product1.pstex "fig:")
\#1\#2\#3\#4\#5[ @font ]{}
(3252,2311)(1111,-6050) (1966,-5161)[(0,0)\[lb\]]{} (3376,-4136)[(0,0)\[lb\]]{} (1576,-5986)[(0,0)\[lb\]]{} (3376,-5986)[(0,0)\[lb\]]{} (2326,-5311)[(0,0)\[lb\]]{} (3976,-5311)[(0,0)\[lb\]]{} (2776,-3886)[(0,0)\[lb\]]{} (2776,-4786)[(0,0)\[lb\]]{} (3376,-5086)[(0,0)\[lb\]]{} (1126,-5161)[(0,0)\[lb\]]{} (1201,-4486)[(0,0)\[lb\]]{}
More families\[subsec:overview\]
--------------------------------
We now return to the situation from Corollary \[th:m-family\]. By exactly the same argument (or otherwise by using the functoriality under $F$), one has a family ${\mathscr{G}}_3$ with fibres $\Gamma_{3,u}$, and which otherwise has the same properties as ${\mathscr{D}}_3$.
\[th:constant-families\] The constant families ${\mathscr{D}}_1 = {\mathscr{R}}\otimes_R \Delta_1$, ${\mathscr{G}}_1 = {\mathscr{R}}\otimes_R \Gamma_1$ also follow $[\gamma]$. Moreover, one can choose relative connections on them in such a way that the induced connection on $$H^*(\mathit{hom}_{{\mathscr{Fuk}}(E)}({\mathscr{G}}_1,{\mathscr{D}}_1)) {\cong}{\mathscr{R}}\otimes_R \mathit{HF}^*(\Gamma_1,\Delta_1)$$ is trivial.
The class $[dq]$ is dual to the hypersurface $\{q = 1/2\}$, which is disjoint from both $\Delta_1$ and $\Gamma_1$. As an instance of Example \[th:totally-disjoint\], it follows that $[g]$ vanishes on the subcategory with these two objects. If we then choose trivial relative connections, the result is obviously true.
Without changing the notation, we will now apply the Yoneda embedding and consider ${\mathscr{D}}_1$ and ${\mathscr{G}}_1$ as objects of ${\mathscr{Fuk}}(E)^{\mathit{perf}}$. Lemma \[th:constant-families\] still holds in this context.
\[th:construct-family\] Fix a one-dimensional subspace $B_0 \subset \mathit{HF}^{d-1}(f)$. Then, for suitable choices of relative connections, there is a line bundle ${\mathscr{B}}\subset H^0(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{G}}_3,{\mathscr{D}}_3[d]))$ invariant under the induced connection, whose restriction to any fibre agrees with $B = H^1(S^1;R) \otimes B_0$ under the isomorphism .
Consider the product $$\label{eq:left-right-product}
\begin{aligned}
& H^0(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{D}}_1,{\mathscr{D}}_3)) \otimes H^*(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{G}}_1,{\mathscr{D}}_1)) \otimes H^1(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{G}}_3,{\mathscr{G}}_1)) \\ & \qquad \qquad \longrightarrow H^*(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{G}}_3,{\mathscr{D}}_3))[1].
\end{aligned}$$ We choose relative connections on ${\mathscr{G}}_1$ and ${\mathscr{D}}_1$ as in Lemma \[th:constant-families\]. Consider the subbundle ${\mathscr{R}}\otimes H^0(S^1;R) \otimes B_0$ of $H^*(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{G}}_1,{\mathscr{D}}_1)) {\cong}{\mathscr{R}}\otimes H^*(S^1) \otimes \mathit{HF}^*(f)$, which is of course preserved by the connection. The left and right factors in are line bundles. Hence, the image of our subbundle under yields a subbundle ${\mathscr{B}}\subset H^d(\mathit{hom}_{{\mathscr{Fuk}}(E)^{\mathit{perf}}}({\mathscr{G}}_3,{\mathscr{D}}_3))$ which, because of the compatibility of the product with the connections, is itself preserved by the connection. At any point of ${\mathscr{S}}$, can be written as a map $$H^*(S^1;R) \otimes \mathit{HF}^*(f) \longrightarrow H^*(S^1;R) \otimes \mathit{HF}^*(f)[1].$$ From our computation of and , we know that this is the identity on $\mathit{HF}^*(f)$ times the cup-product with a nonzero class in $H^1(S^1;R)$. This shows that ${\mathscr{B}}$ has the desired property.
This allows one to apply parallel transport at least to a certain part of $\mathit{HF}^d(\Gamma_{3,u},\Delta_{3,u})$ (probably, the same holds for the entire Floer group, but we will not consider this point here). The next issue is uniqueness, which can be dealt with by using Proposition \[th:uniqueness-2\] based on the following observation:
\[th:assumption-satisfied\] Any one-dimensional subspace $B \subset \mathit{HF}^0(\Gamma_{3,u}, \Delta_{3,u}[d])$ satisfies Assumption \[th:augmented-plus\].
Assumption \[th:augmented\] for each object is obvious, since $\mathit{HF}^*(\Delta_{3,u},\Delta_{3,u}) {\cong}H^*(\Delta_{3,u};R)$ as a ring, and the same for $\Gamma_{3,u}$. The products in are part of the Floer product structure $$\label{eq:2-products-2}
\begin{aligned}
& \mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u}) \otimes \mathit{HF}^*(\Gamma_{3,u},\Gamma_{3,u}) \longrightarrow
\mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u}), \\
& \mathit{HF}^*(\Delta_{3,u},\Delta_{3,u}) \otimes \mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u}) \longrightarrow
\mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u}).
\end{aligned}$$ One can determine this explicitly in the manner of Lemma \[th:2-products\], but we prefer to take a shortcut which bypasses computation. Namely, as part of the open-closed string map we have a map $\mathit{QH}^*(E) \rightarrow \mathit{HF}^*(\Delta_{3,u},\Delta_{3,u})$, which in this case agrees with the ordinary restriction map (in particular is surjective), and the same for $\Gamma_{3,u}$. In fact, by assumption on $f$ our two Lagrangian submanifolds are diffeomorphic, and the restriction maps are the same. One combines this restriction map with to yield both a left and a right action of $\mathit{QH}^*(E)$ on $\mathit{HF}^*(\Gamma_{3,u},\Delta_{3,u})$. One can prove geometrically (or alternatively, derived from the fact that this is part of a map landing in Hochschild cohomology) that these two actions coincide up to Koszul signs. This leads directly to the required property.
\[th:sneaky\] The reader may have noticed that, in view of Assumption \[th:augmented-plus\] as originally stated, we only needed to prove the required properties for the degree $0$ parts of $\mathit{HF}^*(\Delta_{3,u},\Delta_{3,u})$ and $\mathit{HF}^*(\Gamma_{3,u},\Gamma_{3,u})$, which is much easier. The real point of the argument above, which will become relevant only later, is that it still yields the desired result if we reduce the grading of the Fukaya category to ${\mathbb{Z}}/2$.
A double covering trick\[subsec:covering-trick\]
------------------------------------------------
Let $z: \tilde{E} \rightarrow E$ be the double cover associated to $(1,0) \in H^1(T;{\mathbb{Z}}/2) {\cong}H^1(E;{\mathbb{Z}}/2)$. Concretely, $$\label{eq:symplectic-mapping-torus-2}
\tilde{E} = {\mathbb{R}}\times {\mathbb{R}}\times K \times K\;\; / \;\; (p,q,x,y) \sim (p,q-1,x,y) \sim (p-2,q,f^2(x),f^2(y)), \\$$ with the symplectic form $\omega_{\tilde{E}}$ pulled back from $E$. This is the mapping torus of $f^2 \times f^2$, except that the area of the base $T$ has been multiplied by $2$. Fukaya category computations for $\tilde{E}$ largely follow those for $E$, so we will only summarize the results. We have Lagrangian submanifolds $\tilde\Delta_1,\tilde\Delta_2,\tilde\Delta_{3,u}$ (fibrewise equal to the diagonal) and $\tilde\Gamma_1,\tilde\Gamma_2,\tilde\Gamma_{3,u}$ (fibrewise equal to the graph of $f^2$) defined analogously to , . As in there are canonical isomorphisms $$\label{eq:infty-infty-lifted}
\mathit{HF}^*(\tilde\Gamma_{3,u},\tilde\Delta_{3,u}) {\cong}H^*(S^1;R) \otimes \mathit{HF}^*(f^2).$$ Recall from Section \[subsec:pushdown\] that $z$ gives rise to a functor $Z$, defined on a full subcategory $\tilde{F} \subset \mathit{Fuk}(\tilde{E})$ (that contains all the Lagrangian submanifolds occurring in our discussion), and which lands in $\mathit{Fuk}(E)$. In particular, $Z(\tilde{\Delta}_{3,u}) = \Delta_{3,u}$, whereas $Z(\tilde{\Gamma}_{3,u})$ is the analogue of $\Gamma_{3,u}$ defined using the graph of $f^2$ in each fibre. Our functor gives an isomorphism $$\label{eq:pushforward-iso}
\mathit{HF}^*(Z(\tilde{\Gamma}_{3,u}), Z(\tilde\Delta_{3,u})) {\cong}\mathit{HF}^*(\tilde{\Gamma}_{3,u},\tilde{\Delta}_{3,u}) {\cong}H^*(S^1;R) \otimes \mathit{HF}^*(f^2).$$ Note that $Z(\tilde{\Gamma}_{3,u})$ and $Z(\tilde\Delta_{3,u})$ only depend on the class of $u$ in ${\mathbb{R}}^\times/\hbar^{\mathbb{Z}}$. However, diagram-chasing shows that going from $u$ to $\hbar u$ changes the isomorphism by the involution $C_{f,f^2}$.
To take into account the difference in the areas of the base $T$, we take the square unit polynomial $p$ and make a substitution $\hbar \mapsto \hbar^2$. This yields a new polynomial $\tilde{p}$ and associated algebraic curve $\tilde{{\mathscr{S}}} = \mathit{Spec}(\tilde{{\mathscr{R}}})$, with its one-form $\tilde{\theta}$. In fact, we had already considered these in Addendum \[th:double-cover\], where it was pointed out that (after removing finitely many points) $\tilde{{\mathscr{S}}}$ is an étale double cover of ${\mathscr{S}}$, and $\tilde{\theta}$ the pullback of $\theta$. We consider the parametrization of the set of points of $\tilde{{\mathscr{S}}}$ by $u \in R^\times/\hbar^{2{\mathbb{Z}}}$ which under the covering map induces . With these slight modifications, the previous argument goes through, yielding perfect families $\tilde{{\mathscr{D}}}_3$ and $\tilde{{\mathscr{G}}}_3$ over $\tilde{F} \subset \mathit{Fuk}(\tilde{E})$ which follow the image of $\tilde{\theta} \otimes [dq]$ under the open-closed string map, and whose fibres at any point $u$ are isomorphic to $\tilde\Delta_{3,u}$ and $\tilde{\Gamma}_{3,u}$, respectively. We now use , as well as the discussion of functoriality from Section \[subsec:functor\], to push these families down to $E$. The outcome are perfect families $Z(\tilde{\mathscr{D}}_3)$ and $Z(\tilde{{\mathscr{G}}}_3)$ over $\mathit{Fuk}(E)$ which follow $\theta \otimes [dq]$, and whose fibres at $u$ are isomorphic to $Z(\tilde{\Delta}_{3,u})$ and $Z(\tilde\Gamma_{3,u})$.
\[th:both-signs\] For some $d \in {\mathbb{Z}}$, $C_{f,f^2}: \mathit{HF}^{d-1}(f^2) \rightarrow \mathit{HF}^{d-1}(f^2)$ is not $\pm \mathit{Id}$.
Supposing from now on that this is the case, we can choose a one-dimensional subspace $B_0 \subset \mathit{HF}^{d-1}(f^2)$ which is not preserved by $C_{f,f^2}$. Let’s temporarily go back to $\tilde{E}$. The analogue of Lemma \[th:construct-family\] says that there is a line bundle $\tilde{\mathscr{B}}\subset H^0(\mathit{hom}_{{\mathscr{Fuk}}(\tilde{E})^{\mathit{perf}}}(\tilde{\mathscr{G}}_3,\tilde{\mathscr{D}}_3[d]))$ invariant under the induced connection on that space, whose restriction to any fibre agrees with $B = H^1(S^1;R) \otimes B_0$ under the isomorphism . Applying $Z$ to this, and using the compatibility of induced connections with functors shown in , we find that the image line bundle $Z(\tilde{\mathscr{B}})$ is still invariant under the induced connection.
\[th:non-iso\] Take two points $\tilde{s}_\pm \in \tilde{{\mathscr{S}}}$ corresponding to $u$ and $\hbar u$ (for any $u$ such that both make sense, which means excluding the finitely many branch points). Then, the triples $$(Z(\tilde{\mathscr{G}}_3)_{\tilde{s}_\pm}, Z(\tilde{\mathscr{D}}_3[d])_{\tilde{s}_\pm}, Z(\tilde{\mathscr{B}})_{\tilde{s}_\pm})$$ are not mutually isomorphic in $H^0(\mathit{Fuk}(E)^{\mathit{perf}})$.
At $\tilde{s}_+$, the relevant triple consists of the objects $z(\tilde{\Gamma}_{3,u})$ and $z(\tilde{\Delta}_{3,u})[d]$ together with the subspace of $\mathit{HF}^d(z(\tilde{\Gamma}_{3,u}),z(\tilde{\Delta}_{3,u}))$ corresponding to $B$ under the isomorphism . The same holds at $\tilde{s}_-$ but where the isomorphism is twisted by $C_{f,f^2}$. Hence, our statement reduces to the following:
There do not exist invertible elements $$\begin{aligned}
& \gamma \in \mathit{HF}^0(z(\tilde{\Gamma}_{3,u}),z(\tilde{\Gamma}_{3,u})) {\cong}H^*(z(\tilde{\Gamma}_{3,u});R) {\cong}H^*(S^1 \times K;R), \\
& \delta \in \mathit{HF}^0(z(\tilde{\Delta}_{3,u}),z(\tilde{\Delta}_{3,u})) {\cong}H^*(z(\tilde{\Delta}_{3,u});R) {\cong}H^*(S^1 \times K;R),
\end{aligned}$$ satisfying $\delta B \gamma = C_{f,f^2}(B)$.
But that is obvious because both $\mathit{HF}^0$ groups only contain multiples of the identity.
\[th:not-iso-even-without-grading\] The Claim above, and therefore Lemma \[th:non-iso\], continues to hold even if we allow $\delta$ and $\gamma$ to have additional terms of higher even degree. This is because the subspace $B$ itself is concentrated in a single degree.
Both points $\tilde{s}_\pm \in \tilde{{\mathscr{S}}}$ map to the same point $s \in {\mathscr{S}}$. This, together with the analogue of Lemma \[th:assumption-satisfied\], triggers Lemma \[th:contradiction\], which shows that:
\[th:not-periodic\] If Assumption \[th:both-signs\] is satisfied, the image of $[dq]$ in $\mathit{HH}^1(\mathit{Fuk}(E),\mathit{Fuk}(E))$ is not a periodic element (for the elliptic curve with one-form obtained as the closure of ${\mathscr{S}}$ and $\theta$).
Let’s have a brief “straight man” conterpart of the previous discussion, concerning the case where the symplectic automorphism is the identity, giving rise to the trivial mapping torus $E^{\mathit{triv}} = T \times K^- \times K$. Arguing as in Example \[th:product-mirror\], one finds that there are quasi-equivalences $$\begin{aligned}
\mathit{Fuk}(E^{\mathit{triv}})^{\mathit{perf}} & {\cong}(\mathit{Fuk}(T) \otimes \mathit{Fuk}(K^-) \otimes \mathit{Fuk}(K))^{\mathit{perf}} \\
& {\cong}(D^b\mathit{Coh}(Y_p) \otimes D^b\mathit{Coh}(X) \otimes D^b\mathit{Coh}(X))^{\mathit{perf}} \\
& {\cong}D^b\mathit{Coh}(Y_p \times X \times X),
\end{aligned}$$ where $Y_p$ and $X$ are the mirrors of $T$ and $K$, respectively (the fact that one of the copies of $K$ has reversed sign of the symplectic form does not affect the statement, since $\omega_K$ and $-\omega_K$ are related by an involution, as one can see by taking $K$ a real quartic). One can construct a family of bimodules exactly as in Section \[subsec:universal-bimodule\], and use that to derive the following analogue of Corollary \[th:periodic-lattice\]: $$\label{eq:lattice-2}
m_1[g_1] + m_2[g_2] \in \mathit{Per}(D^b\mathit{Coh}(Y_p \times X \times X),\bar{{\mathscr{S}}},\bar\theta) \text{ for $m_1 \in {\mathbb{Z}}$, $m_2 \in m_1 + 2{\mathbb{Z}}$},$$ where $[g_1]$, $[g_2]$ are the classes pulled back from $\mathit{HH}^*(Y_p,Y_p) {\cong}\mathit{HH}^*(Q_p,Q_p)$. Under mirror symmetry, the generators $[g_1]+[g_2]$ and $2[g_2]$ of the lattice in correspond to $[dp]$ and $[-dq]$ (compare Lemma \[th:g-pull\]), hence:
\[th:yes-periodic\] Any element in the image of $H^1(E^{\mathit{triv}};{\mathbb{Z}}) {\cong}{\mathbb{Z}}^2 \rightarrow \mathit{HH}^1(\mathit{Fuk}(E^{\mathit{triv}}),\mathit{Fuk}(E^{\mathit{triv}}))$ is periodic (for the same elliptic curve as in Corollary \[th:not-periodic\]).
As a consequence, we see that if $f$ satisfies Assumption \[th:both-signs\], then $E$ is not symplectically isomorphic to $E^{\mathit{triv}}$. Of course, this is by no means the most direct argument available (see the Introduction), but it has the advantage of belonging to the general framework of Fukaya categories.
An algebraic viewpoint\[subsec:algebraic-model\]
------------------------------------------------
Let $A$ be a proper $A_\infty$-category over $R$, together with a functor $G: A \rightarrow A$. The naive [*mapping torus category*]{} $A^{\mathit{torus}}$ is defined as follows. Objects are of the form $X(d)$, where $X$ is an object of $A$ strictly fixed by $G$, meaning that $G(X) = X$, and $d \in {\mathbb{Z}}$ an integer. The definition of the morphism space comes from : $$\label{eq:t-category}
\mathit{hom}_{A^{\mathit{torus}}}(X_0(d_0),X_1(d_1)) = \mathit{hom}_A(X_0,X_1) \otimes F \oplus
\mathit{hom}_A(X_0,X_1) \otimes F[-1],$$ where $F$ is as in , and the tensor product is over $R$. It may be more intuitive to (arbitrarily) choose a basis and write $\mathit{hom}_A(X_0,X_1) = C(X_0,X_1) \otimes_{\mathbb{C}}R$. Then, elements of $\mathit{hom}_A(X_0,X_1) \otimes F$ can be thought of as series $a(t) = c_0 \hbar^{m_0} t^{n_0} + \cdots$, with the same convergence condition as in , but coefficients $c_k \in C(X_0,X_1)$.
Because of the definition as a tensor product, we have the additional condition that for any $a(t)$, the coefficients $c_k$ which occur may span only a finite-dimensional subspace of $C(X_0,X_1)$. This is somewhat unnatural in terms of the topological nature of the ring $F$. However, if $A$ is strictly proper (has finite-dimensional morphism spaces), this point is obviously irrelevant, and of course any proper $A_\infty$-category is quasi-isomorphic to a strictly proper one.
Elements of can be written as pairs $(a(t),b(t))$, where $|b(t)| = |a(t)| - 1$. The differential is $$\label{eq:mu1-torus}
\mu^1_{A^{\mathit{torus}}}(a(t), b(t)) = \big(\mu^1_A(a(t)), \mu^1_A(b(t)) + (-1)^{|a|-1} a(t) + (-1)^{|a|} t^{d_1-d_0} G^1(a(\hbar t)) \big).$$
\[th:multicones\] Consider a single object $X$ fixed by $G$. There is an obvious long exact sequence $$\label{eq:t-sequence}
\cdots H(\mathit{hom}_{A^{\mathit{torus}}}(X(d),X(d))) \rightarrow H(\mathit{hom}_A(X,X)) \otimes F \xrightarrow{\mathit{id} - H(G^1) \otimes T} H(\mathit{hom}_A(X,X)) \otimes F \cdots$$ where $T$ is as in . If we restrict the second map in to series in $t$ with vanishing constant term (in $t$), it is actually an isomorphism. Hence, we have the simpler long exact sequence $$\cdots H(\mathit{hom}_{A^{\mathit{torus}}}(X(d),X(d))) \rightarrow H(\mathit{hom}_A(X,X)) \xrightarrow{\mathit{id}-H(G^1)} H(\mathit{hom}_A(X,X)) \cdots$$
The composition of $(a_k(t),b_k(t)) \in \mathit{hom}_{A^{\mathit{torus}}}(X_{k-1}(d_{k-1}),X_k(d_k))$ ($k = 1,2$) is given by $$\begin{aligned}
& \mu^2_{A^{\mathit{torus}}}((a_2(t),b_2(t)), (a_1(t),b_1(t))) = \big(\mu^2_A(a_2(t),a_1(t)),
(-1)^{|a_2|-1} \mu^2_A(a_2(t),b_1(t)) \\ & \qquad + \mu^2_A(b_2(t),t^{d_1-d_0} G^1(a_1(\hbar t))) +
(-1)^{|a_2|+|a_1|} t^{d_2-d_0} G^2(a_2(\hbar t),a_1(\hbar t)) \big); \\
\end{aligned}$$ and similarly for the higher order structure maps.
\[th:t-model\] Suppose that $X \in \mathit{Ob}\, A$ is fixed by $G$. Let $Q_p$ be the $A_\infty$-category associated to the unit torus polynomial. Then there is an $A_\infty$-functor $Q_p \rightarrow A^{\mathit{torus}}$ which maps the two objects of $Q_p$ to $X(0)$ and $X(2)$, respectively.
After replacing our original category and functor by quasi-isomorphic ones, one can assume that both are strictly unital (while still acting in the same way on objects). Think of $R$ itself as an $A_\infty$-category with a single object $Z$. The embedding $R \rightarrow A$ mapping $Z$ to $X$ induces one $R^{\mathit{torus}} \rightarrow A^{\mathit{torus}}$. Consider the full subcategory of $R^{\mathit{torus}}$ with objects $Z(0)$, $Z(2)$. This is a dg model (actually the one mentioned in Section \[subsec:nonarchimedean\]) for the full subcategory of the derived category of modules over $F \rtimes {\mathbb{Z}}$ with objects $F(0)$, $F(2)$. Lemma \[th:nonarch\] then completes the proof.
The proposed correspondence with geometry goes as follows. If $A$ is the Fukaya category of some compact manifold, and $G$ is given by the action of a (graded) symplectic automorphism, then $A^{\mathit{torus}}$ should conjecturally be quasi-isomorphic to a full subcategory of the Fukaya category of the associated symplectic mapping torus. Objects $X(d)$ correspond to Lagrangian submanifolds in the mapping torus obtained by taking an invariant Lagrangian submanifold in the fibre and moving it along a line in the base which goes through $(0,0)$ and has slope $-d$. Looking back to our previous discussion, $\Delta_1$ and $\Delta_2$ from as well as $\Gamma_1$ and $\Gamma_2$ from are of this type. Example \[th:multicones\] shows that the algebraic model correctly computes the self-Floer cohomology. Lemma \[th:t-model\] would be the algebraic counterpart of the Fukaya category computations in Section \[subsec:flat-triangles\].
The framework above is technically naive, since it asks for strictly fixed objects. In the application to Fukaya categories, this becomes an issue because of the auxiliary choices of almost complex structures. It is probably best to remedy this by making the algebraic formalism more flexible, which could mean considering $A_\infty$-modules over $A \otimes F$ which are equivariant with respect to $G \otimes T$ (this would also allow one to include objects corresponding to Lagrangian submanifolds such as $\Delta_{3,u}$ and $\Gamma_{3,u}$). A less satisfactory but simpler approach is to reduce more general situations to the given framework (which means that given an object $X$ quasi-isomorphic to $G(X)$, one changes $G$ to make $X$ strictly fixed; this can be done for several objects at the same time).
Blowing up\[sec:blowup\]
========================
The topic of this section, namely the behaviour of Fukaya categories under blowups, is of interest from many perspectives, among which our intended application plays only a minor role. Interested readers are referred to [@smith10], from which we have stolen as much as we could (concretely, Sections \[subsec:toy\]–\[subsec:correspondence\] follow [@smith10 Section 4.5] closely). On a technical level, we will freely use and combine a wide range of results, notably: the full-fledged construction of Fukaya categories [@fooo], and split-generation results in that context [@abouzaid-fukaya-oh-ohta-ono11]; degeneration techniques [@ionel-parker04; @li-ruan98]; Lagrangian correspondences [@wehrheim-woodward06; @mww]; and the $h$-principle [@gromov86]. Necessarily, the exposition can’t be self-contained to any extent.
Generally speaking, the passage to ${\mathbb{Z}}/2$-graded Fukaya categories and the introduction of bounding cochains allows us to include many more objects than in the approach from Section \[sec:automorphisms\] (see Remark \[th:compare\] for a precise statement of the relationship). There will be a temporary departure from this framework in Section \[subsec:toy\], when we consider the toy model of blowing up a point in flat space (which happens to be monotone, allowing us to retreat to a simpler version of Floer theory).
Fukaya categories
-----------------
Fix a closed symplectic manifold $M$. Let $R_{\geq 0} \subset R$ be the subalgebra of formal series involving only nonnegative powers of $\hbar$. This comes with a homomorphism $R_{\geq 0} \rightarrow {\mathbb{C}}$ extracting the constant term, and we write $R_{>0}$ for its kernel. For any $\lambda \in R_{>0}$ there is an associated Fukaya category $\mathit{Fuk}(M)_{\lambda}$, which is a proper ${\mathbb{Z}}/2$-graded $A_\infty$-category. We will give an impressionistic sketch of the construction, which is due to [@fooo] (see [@fukaya-oh06; @fukaya11] for more thorough expository accounts).
One first associates to $M$ a [*filtered curved $A_\infty$-category*]{} $\mathit{FO}(M)$. Objects of $\mathit{FO}(M)$ are Lagrangian submanifolds $L \subset M$ equipped with a [*Spin*]{} structure and a local coefficient system $\xi$ with structure group $\mathit{GL}_r({\mathbb{C}})$, for some $r$. The morphism space between any two objects is a finitely generated free ${\mathbb{Z}}/2$-graded module over $R_{\geq 0}$. If we consider a single object, then $$\label{eq:fuk-endo}
\mathit{hom}_{\mathit{FO}(M)}(L,L) \otimes_{R_{\geq 0}} {\mathbb{C}}$$ is an $A_\infty$-algebra (the curvature vanishes since it has no $\hbar^0$ term) quasi-isomorphic to the standard one underlying the cohomology with local coefficients $H^*(L; \mathit{Hom}(\xi,\xi))$. For simplicity we will assume that $\mathit{FO}(M)$ is strictly unital.
The finite-dimensionality of morphism spaces is convenient for expository reasons, since it allows one to worry less about convergence and completeness (in the $\hbar$-adic topology). A Morse theory model as in Section \[subsec:clean\] naturally yields finite-dimensional morphism spaces. On the other hand, one can start with an infinite-dimensional space of cochains (like the singular cochains used in [@fooo]) and then obtain finite-dimensional models a posteriori by applying a version of the Homological Perturbation Lemma [@fooo Theorem W]. Strict units are not an a priori feature of either approach, but can be added by first introducing a homotopy unit through additional moduli spaces [@fooo Section 7.3], and then constructing a strict unit from that [@fooo Section 3.3]. We should point out that from a more abstract viewpoint, unitality is not really the crucial ingredient (see Remark \[th:general-hh\] below).
Objects of $\mathit{Fuk}(M)_{\lambda}$ are [*weakly unobstructed*]{} Lagrangian submanifolds. By this we mean objects of $\mathit{FO}(M)$ together with a [*bounding cochain*]{} $\alpha \in \mathit{hom}_{\mathit{FO}(M)}^1(L,L)$, which vanishes if we tensor with ${\mathbb{C}}$, and which satisfies the following inhomogeneous Maurer-Cartan equation: $$\label{eq:inhomogeneous-mc}
\mu^0_{\mathit{FO}(M)} + \mu^1_{\mathit{FO}(M)}(\alpha) + \mu^2_{\mathit{FO(M)}}(\alpha,\alpha) + \cdots = \lambda e_L \in \mathit{hom}^0_{\mathit{FO}(M)}(L,L)$$ (in the terminology of [@fooo Section 3.6], this would be a “weak bounding cochain”). The morphism spaces, also called Floer cochain groups following the traditional terminology, are defined by $$\label{eq:cf-corrected}
\mathit{CF}^*(L_0,L_1) = \mathit{hom}_{\mathit{Fuk}(M)_\lambda}(L_0,L_1) = \mathit{hom}_{\mathit{FO}(M)}(L_0,L_1) \otimes_{R_{\geq 0}} R.$$ The $A_\infty$-structure is obtained by deforming that on $\mathit{FO}(M)$, as in the construction of twisted complexes. For instance, the differential on is $$\label{eq:deformed-mu1}
\begin{aligned}
\mu^1_{\mathit{Fuk}(M)_{\lambda}}(a) & = \mu^1_{\mathit{FO}(M)}(a) + \mu^2_{\mathit{FO}(M)}(\alpha_1,a) + \mu^2_{\mathit{FO}(M)}(a,\alpha_0) + \mu^3_{\mathit{FO}(M)}(\alpha_1,\alpha_1,a) \\ & \qquad + \mu^3_{\mathit{FO}(M)}(\alpha_1,a,\alpha_0) + \mu^3_{\mathit{FO}(M)}(a,\alpha_0,\alpha_0) + \cdots
\end{aligned}$$
\[th:compare\] Consider the situation where $\lambda = 0$, and let $L$ be some object of $\mathit{FO}(M)$ which is undeformed, meaning that the $A_\infty$-structure on $\mathit{hom}_{\mathit{FO}(M)}(L,L)$ is a trivial deformation of that on . This can be arranged whenever there is a $J_L$ as in Assumption \[th:adiscic\] (the technical details are subtle, but our applications only really involve the simpler case when $J_L$ has no holomorphic spheres or discs). Then reduces to a version of the Maurer-Cartan equation governing the deformation theory of the local coefficient system $\xi$ [@goldman-millson88]. In particular, any deformation of $\xi$ to a local coefficient system with structure group gives rise to a solution (unique up to gauge equivalence), hence produces an object of $\mathit{Fuk}(M)_0$. These observations have the following noteworthy consequence. Suppose that $c_1(M) = 0$, and let $\mathit{Fuk}(M)$ be the Fukaya category according to the more restricted definition used in Section \[sec:automorphisms\]. Then, after reducing the grading of $\mathit{Fuk}(M)$ to ${\mathbb{Z}}/2$, there is a cohomologically full and faithful functor $\mathit{Fuk}(M) \longrightarrow \mathit{Fuk}(M)_0$.
Even though $\mathit{FO}(M)$ has curvature, its Hochschild cohomology is still well-defined as usual, and we have a canonical open-closed string map $$\label{eq:pre-open-closed}
H^*(M;R_{\geq 0}) \longrightarrow \mathit{HH}^*(\mathit{FO}(M),\mathit{FO}(M)),$$ From this one derives, in a way similar to , maps $$\label{eq:lambda-open-closed}
\mathit{QH}^*(M) \longrightarrow \mathit{HH}^*(\mathit{Fuk}(M)_{\lambda}, \mathit{Fuk}(M)_{\lambda}).$$ Concretely, fix some class in $H^*(M;R_{\geq 0})$, and let $g_{\mathit{FO}} \in \mathit{CC}^*(\mathit{FO}(M),\mathit{FO}(M))$ be a Hochschild cocycle representing its image under (this cochain is defined by a generalization of the procedure sketched in Section \[subsec:define-open-closed\]). Then, the image of the same class under is represented by a cochain $g$ whose constant term is $$g^0 = g^0_{\mathit{FO}} + g^1_{\mathit{FO}}(\alpha) + g^2_{\mathit{FO}}(\alpha,\alpha) + \cdots \in CF^*(L,L).$$
\[th:general-hh\] For to square to zero, we do not really need the fact that the left hand side of is a multiple of the unit, but only that it is strictly central. More generally, one can associate a Fukaya category to any $\lambda \in H^{\mathit{even}}(M;R^{>0})$ (called “bulk deformations” in [@fooo Section 3.8]; the special case of $H^0$ corresponds to the previously discussed construction). We will not pursue this further.
There are partial results about what kinds of Lagrangian submanifolds can occur for different values of $\lambda$. If we restrict to monotone symplectic manifolds (and monotone Lagrangian submanifolds $L$, with trivial bounding cochains $\alpha = 0$), Auroux, Kontsevich and the author [@auroux07 Theorem 6.1] showed that the image of $c_1(M) \in \mathit{QH}^*(M)$ in $\mathit{HF}^*(L,L)$ is $\lambda$ times the identity. Therefore, the associated Fukaya category is zero unless $\lambda$ is an eigenvalue of quantum multiplication with $c_1(M)$. It seems that this argument does not generalize \[Abouzaid, private communication\]. There is a related result which does hold without monotonicity assumptions but involves different Gromov-Witten invariants [@fooo Theorem 3.8.11]. Namely, suppose that $L$ can be made into an object of $\mathit{Fuk}(M)_{\lambda}$. Then its homology class must satisfy $$\label{eq:2-point-invariant}
\sum_{A \in H_2(M;{\mathbb{Z}})} \hbar^{\omega_M(A)}\big( \langle \mathit{PD}([L]), x \rangle_A^M - \langle \mathit{PD}([L]), \lambda, x \rangle_A^M \big) = 0$$ for all $x \in \mathit{QH}^*(M)$. Here, $\langle x_1,\dots,x_n \rangle_A^M$ is the notation for the Gromov-Witten invariant counting genus $0$ (no higher genus invariants will occur in this paper) pseudo-holomorphic curves in homology class $A$, whose $n$ marked points lie on the Poincar[é]{} duals of the cohomology classes $x_i$. In the second term of , $\lambda$ is thought of as a multiple of the identity; hence, that term vanishes unless $A = 0$, and the whole equation reduces to the fact that $[L]$ is an eigenvector with eigenvalue $\lambda$ for the endomorphism induced by the two-point Gromov-Witten invariant (the reason why we chose the formulation is that it accomodates “bulk deformed” generalizations [@fooo Theorem 3.8.88]). Unfortunately, we do not yet have a broader theorem that would allow us to decide for which values of $\lambda$ the category $\mathit{Fuk}(M)_{\lambda}$ can be nonzero. Instead, we’ll allow arbitrary $\lambda$ and then formally cut down the resulting category.
Projections
-----------
We make a short detour into homological algebra. Let $A$ be an $A_\infty$-category (${\mathbb{Z}}/2$-graded and cohomologically unital), and $Q_+ \in \mathit{HH}^0(A,A)$ a Hochschild cohomology class which is idempotent with respect to the natural ring structure. We want to [*project*]{} the category $A$ accordingly, which will give rise to a new category $A_+$. The simplest way to go about that is as follows. $Q_+$ determines, for any $X \in \mathit{Ob}\,A$, an idempotent endomorphism $Q_+^0 \in H^0(\mathit{hom}_A(X,X))$. After lifting that to a homotopy idempotent $p_+$ (compare Section \[subsec:splittings\]), one gets a (perfect) module $(X,p_+)^{\mathit{yon}}$. Choose one such $p_+$ for each $X$, and define $A_+$ to be the full $A_\infty$-subcategory of the module category of $A$ with objects $(X,p_+)^{\mathit{yon}}$.
Here is another approach, which turns out to be equivalent but yields additional properties. Let $Q_-$ be the complementary idempotent. Fix modules $(X,p_+)^{\mathit{yon}}$ and their complementary counterparts $(X,p_-)^{\mathit{yon}}$, and consider those as well as the standard Yoneda images $X^{\mathit{yon}}$. Let $A_{\pm}$ be the full subcategory of the module category of $A$ containing all those objects, and then pass to the quotient category $A_{\pm}/A_-$ in the sense of [@drinfeld02], in which all the objects $(X,p_-)^{\mathit{yon}}$ become quasi-isomorphic to zero. This comes with a canonical quotient functor $A_{\pm} \rightarrow A_{\pm}/A_-$.
\[th:projection-as-quotient\] The quotient functor restricts to a quasi-equivalence $$\label{eq:ppm}
A_+ \longrightarrow A_{\pm}/A_-.$$
Since $X^{\mathit{yon}} {\cong}(X,p_+)^{\mathit{yon}} \oplus (X,p_-)^{\mathit{yon}}$ by construction, the objects $X^{\mathit{yon}}$ and $X^{\mathit{yon}}_+$ become quasi-isomorphic in the quotient, so is essentially onto. Note that for any two objects $X_0,X_1$ we have $$\begin{aligned}
& H(\mathit{hom}_{A_\pm}((X_0,p_{0,+})^{\mathit{yon}},(X_1,p_{1,-})^{\mathit{yon}})) = 0, \\
& H(\mathit{hom}_{A_\pm}((X_0,p_{0,-})^{\mathit{yon}},(X_1,p_{1,+})^{\mathit{yon}})) = 0.
\end{aligned}$$ By general nonsense [@drinfeld02 Theorem 1.6.2(ii)], this implies that is cohomologically full and faithful.
So far, it may not have been evident why we have included the Yoneda images themselves in $A_\pm$. The point is that we can combine the Yoneda embedding and the quotient functor to get a canonical functor $A \rightarrow A_\pm/A_-$. From the proof of Lemma \[th:projection-as-quotient\] it follows that on the cohomological level, this is projection to the part of the morphism space singled out by $Q_+$: $$H(\mathit{hom}_{A_{\pm}/A_-}(X_0^{\mathit{yon}},X_1^{\mathit{yon}})) {\cong}Q_+^0 H(\mathit{hom}_A(X_0,X_1)) =
H(\mathit{hom}_A(X_0,X_1)) Q_+^0.$$ Here is an even simpler way to describe the resulting situation. Let $\tilde{A}$ be the category with the same objects as $A$, but where $$\mathit{hom}_{\tilde{A}}(X_0,X_1) = \mathit{hom}_{A_\pm/A_-}(X_0^{\mathit{yon}},X_1^{\mathit{yon}}) \oplus \mathit{hom}_{A_\pm/A_+}(X_0^{\mathit{yon}},X_1^{\mathit{yon}}),$$ with the obvious choice of $A_\infty$-structure. Then, the functor $A \rightarrow A_\pm/A_-$ and its analogue for the complementary quotient combine to yield a quasi-isomorphism $A \rightarrow \tilde{A}$. If we allow ourselves to replace $A$ by $\tilde{A}$, the situation is that we have a category whose morphism spaces are split into two parts compatibly with all compositions, and the projection just throws away one of those parts. In particular, there are canonical isomorphism $$\mathit{HH}^*(A,A) {\cong}\mathit{HH}^*(\tilde{A},\tilde{A}) {\cong}\mathit{HH}^*(A_+,A_+) \oplus \mathit{HH}^*(A_-,A_-).$$
Inspection of our argument shows that construction does not really involve the Hochschild cohomology class for $A$, but only on its image in $\mathit{HH}^0(H(A),H(A))$ (which is the collection of idempotents $[p_+^0] \in H^0(\mathit{hom}_A(X,X))$ with their compatibility property). In fact, as a byproduct we have shown that every idempotent element in $\mathit{HH}^0(H(A),H(A))$ can be lifted to an idempotent in $\mathit{HH}^0(A,A)$, and one can also see that this lift is unique.
We will now apply this to Fukaya categories, with modified notation. Take an idempotent $q \in \mathit{QH}^0(M)$. Its image under is an idempotent in Hochschild cohomology, and we denote the outcome of the resulting projection by $\mathit{Fuk}(M)_{\lambda,q}$. The following result is a modified version of Theorem \[th:split-generation-2\], and can be proved by adapting arguments from [@abouzaid-fukaya-oh-ohta-ono11].
\[th:split-generation-3\] Let $O \subset \mathit{Fuk}(M)_{\lambda,q}$ be a full $A_\infty$-subcategory. Suppose that there is a linear map $\mathit{HH}^0(O,O) \rightarrow R$ such that the following diagram commutes: $$\xymatrix{\mathit{QH}^0(M)q \ar[rr] \ar[dr]_{\int_M} && \mathit{HH}^0(O,O) \ar[dl] \\
& R &}$$ Then the objects in $O$ split-generate $\mathit{Fuk}(M)^{\mathit{perf}}_{\lambda,q}$.
A toy model\[subsec:toy\]
-------------------------
Consider ${\mathbb{C}}P^r$ blown up at a point as a toric symplectic (and K[ä]{}hler) manifold. We denote this by $\bar{B}^{\mathit{toy}}$. It is characterized symplectically by its moment polytope, which is $$\big\{ t \in {\mathbb{R}}^r \,:\, t_1, \dots, t_r \geq 0, \; \delta \leq t_1 + \cdots + t_r \leq \epsilon \big\} \label{eq:polytope}$$ for some $0 < \delta < \epsilon$. The preimage of $\{t_1 + \cdots + t_r = \epsilon\}$ under the moment map is the hyperplane at infinity, denoted by $H^{\mathit{toy}}$. The preimage of $\{t_1 + \cdots + t_r = \delta\}$ is the exceptional divisor, denoted by $D^{\mathit{toy}}$. In particular, $$\label{eq:c1-toy-compactified}
c_1(\bar{B}^{\mathit{toy}}) = -(r-1)\mathit{PD}([D^{\mathit{toy}}]) + (r+1)\mathit{PD}([H^{\mathit{toy}}]).$$ Write $B^{\mathit{toy}} = \bar{B}^{\mathit{toy}} \setminus H^{\mathit{toy}}$. This is non-compact, but because the divisor we remove has positive normal bundle, doing pseudo-holomorphic curve theory in its complement is unproblematic. We denote by $Z \in H_2(B^{\mathit{toy}})$ the class of a line lying in $D^{\mathit{toy}}$, and by $u = -\mathit{PD}([D^{\mathit{toy}}]) \in H^2_{\mathit{cpt}}(B^{\mathit{toy}})$ the negative Poincar[é]{} dual of the exceptional divisor. With this sign convention, $$\begin{aligned}
& u(Z) = 1, \label{eq:negative-pd} \\
& c_1(B_{\mathit{toy}}) = (r-1) u, \label{eq:c1-toy} \\
& [\omega_{B^{\mathit{toy}}}] = 2\pi\delta \, u, \label{eq:omega-toy} \\
& u^r = -\mathit{PD}([\mathit{point}]), \label{eq:toy-gitler}\end{aligned}$$ where in and we’ve mapped $u$ to $H^2(B^{\mathit{toy}})$. It follows that $B^{\mathit{toy}}$ is monotone: $$\label{eq:monotonicity}
c_1(B^{\mathit{toy}}) = \textstyle\frac{(r-1)}{2\pi\delta} [\omega_{B^{\mathit{toy}}}] \in H^2(B^{\mathit{toy}};{\mathbb{R}}).$$ We will take advantage of the resulting possible simplification and omit the formal parameter $\hbar$, which means working with quantum cohomology and Floer theory over ${\mathbb{C}}$.
Let’s first look at the small quantum product. $B^{\mathit{toy}}$ is a line bundle over $D^{\mathit{toy}}$, and the fibres are Poincaré dual to $u^{r-1} \in H^{2r-2}(B^{\mathit{toy}})$, as one can see from . The contribution of lines lying inside the exceptional divisor is $$\begin{aligned}
\textstyle \int_{B^{\mathit{toy}}} (u^{r-1} \ast_Z u) \, u^{r-1} & = \langle u^{r-1}, u, u^{r-1} \rangle_{Z}^{B^{\mathit{toy}}} \\ & = \langle u^{r-1}, u^{r-1} \rangle_{Z}^{B^{\mathit{toy}}} = 1.
\end{aligned}$$ Comparison with yields $$\label{eq:toy-quantum-gitler}
u^{r-1} \ast_Z u = - u.$$ In principle, the multiples $dZ$, $d>1$, could also contribute to the quantum product. But the virtual dimension of the associated moduli space of three-pointed holomorphic spheres is $2r + (2r-2) d = (2+2d)r - 2d$, while the image of the evaluation map is contained in a subspace $({\mathbb{C}}P^{r-1})^3$ of smaller dimension $6r-6$. Hence the virtual fundamental cycle maps to zero under evaluation.
Suppose from now on that $\epsilon > \frac{r}{r-1}\delta$, and consider the Lagrangian torus $C^{\mathit{toy}} \subset B^{\mathit{toy}}$ which is the fibre of the moment map over the point $(\delta/(r-1),\dots,\delta/(r-1))$ in . For another description, recall that $B^{\mathit{toy}} \setminus D^{\mathit{toy}}$ is $U(1)^r$-equivariantly symplectically isomorphic to the open subset $$\label{eq:annulus}
\{ z \in {\mathbb{C}}^r \,:\, \delta < {{\textstyle\frac{1}{2}}}\| z\|^2 < \epsilon \}$$ with the standard (constant) symplectic form. In this isomorphism, $C^{\mathit{toy}}$ corresponds to the Clifford torus with all radii equal to $\sqrt{2\delta/(r-1)}$. We can use areas of discs in to show that $C^{\mathit{toy}}$ is monotone as well, meaning that the analogue of holds in $H^2(B^{\mathit{toy}},C^{\mathit{toy}};{\mathbb{R}})$.
Following [@cho-oh02] (see [@hori-vafa00] for physics motivation, [@auroux07 §4] for an exposition, and [@fukaya-oh-ohta-ono10] for generalizations), it is convenient to formulate results about the Floer cohomology of $C^{\mathit{toy}}$ in terms of the superpotential $$\begin{aligned}
& W: H^1(C^{\mathit{toy}};{\mathbb{C}}^*) \longrightarrow {\mathbb{C}}, \\
& z \longmapsto \sum_A n_A \, z^{\partial A}.
\end{aligned}$$ Here, the sum is over $A \in H_2(B^{\mathit{toy}}, C^{\mathit{toy}})$, and $n_A \in {\mathbb{Z}}$ is the number of pseudo-holomorphic discs in class $A$ going through a generic point of $C^{\mathit{toy}}$. To fix the signs, we equip $C^{\mathit{toy}}$ with the trivial [*Spin*]{} structure (the one compatible with the rotation-invariant framing; or equivalently, the unique one which is invariant under the action of $\mathit{SL}_r({\mathbb{Z}})$). The domain of $W$ should be thought of as the moduli space of flat ${\mathbb{C}}^*$-bundles on $C^{\mathit{toy}}$, and the superpotential is obtained by counting discs weighted with their boundary holonomy. Equip $C^{\mathit{toy}}$ with the bundle $\xi$ corresponding to some point $z$, and consider the spectral sequence [@oh96] going from $H^*(C^{\mathit{toy}};{\mathbb{C}})$ to $\mathit{HF}^*(C^\mathit{toy}, C^\mathit{toy})$. Part of the differential of this sequence is the map $$\label{eq:d1-oh}
\begin{aligned}
& H^1(C^{\mathit{toy}};{\mathbb{C}}) \longrightarrow H^0(C^{\mathit{toy}};{\mathbb{C}}) = {\mathbb{C}}, \\
& w \longmapsto dW_z(zw) = \sum_A n_A \, w(\partial A) z^{\partial A}.
\end{aligned}$$ Because the spectral sequence is multiplicative [@buhovsky10], there are only two possible outcomes. Either is nonzero, in which case the Floer cohomology vanishes. Or else it is zero, in which case the spectral sequence degenerates. From now on let’s focus exclusively on the second case, which happens exactly when $z$ is a critical point of $W$. In that situation, we have an isomorphism $H^*(C^{\mathit{toy}};{\mathbb{C}}) {\cong}\mathit{HF}^*(C^{\mathit{toy}},C^{\mathit{toy}})$, which is canonical up to composition with automorphisms of $H^*(C^{\mathit{toy}};{\mathbb{C}})$ of the form $\mathit{Id} + R$, where $R$ decreases degrees by at least $2$. In particular, the degree $0$ and $1$ parts $$\begin{aligned}
\label{eq:h0-inclusion}
& H^0(C^{\mathit{toy}};{\mathbb{C}}) {\cong}\mathit{HF}^0(C^{\mathit{toy}},C^{\mathit{toy}}), \\
\label{eq:h1-inclusion}
& H^1(C^{\mathit{toy}};{\mathbb{C}}) {\cong}\mathit{HF}^1(C^{\mathit{toy}},C^{\mathit{toy}}),\end{aligned}$$ are strictly canonical. The multiplicative nature of the spectral sequence ensures that yields the unit $e$ in Floer cohomology, and generates Floer cohomology as a ring. The relations between these generators are determined by the Hessian of $W$ at $z$: $$\label{eq:mult-relation}
w \cdot w = (D^2W)_z(zw,zw)\, e = \sum_A n_A w(\partial A)^2 z^{\partial A}\, e.$$ Being a toric fibre, $C^{\mathit{toy}}$ is contractible in $B^{\mathit{toy}}$, hence the restriction map in ordinary cohomology vanishes. However, the specialization of the open-closed string map $\mathit{H}^*(B^{\mathit{toy}};{\mathbb{C}}) \longrightarrow \mathit{HF}^*(C^{\mathit{toy}},C^{\mathit{toy}})$ has quantum corrections [@fukaya-oh-ohta-ono11]. In degree $2$, the outcome is as follows: $$\label{eq:toy-open-closed}
\begin{aligned}
& H^2(B^{\mathit{toy}};{\mathbb{C}}) \longrightarrow \mathit{HF}^0(C^{\mathit{toy}},C^{\mathit{toy}}), \\
& v \longmapsto \sum_A n_A \, v(A) z^{\partial A} e,
\end{aligned}$$ where we have chosen an arbitrary lift of $v$ to $H^2(B^{\mathit{toy}},C^{\mathit{toy}})$ to define the pairings $v(A)$ (vanishing of ensures that the choice of lift doesn’t matter).
It remains to spell out the data in our particular case, again following [@cho-oh02]. Thinking of $C^{\mathit{toy}}$ as a torus orbit yields an identification $H^1(C^{\mathit{toy}}) {\cong}{\mathbb{Z}}^r$. The numbers $n_A$ are $+1$ for unit vectors $A = (0,\dots,1,\dots,0)$ as well as for $A = (1,\dots,1)$, and vanish otherwise, yielding the Hori-Vafa mirror superpotential $$W(z_1,\dots,z_r) = z_1 + \cdots + z_r + z_1 \cdots z_r.$$ There are $(r-1)$ critical points $(z_1,\dots,z_r) = (\lambda,\dots,\lambda)$, where $\lambda^{r-1} = -1$. Each of them is nondegenerate, which in view of means that the Floer cohomology rings are complex Clifford algebras. More precisely, using the generators one gets a canonical isomorphism $$\label{eq:toy-clifford}
{\mathbb{C}}\langle w_1,\dots,w_r \rangle/ \{w_i^2 \text{ for all $i$},\, w_iw_j+w_jw_i+2\lambda \text{ for all $i \neq j$}\} \longrightarrow \mathit{HF}^*(C^{\mathit{toy}},C^{\mathit{toy}}),$$ where ${\mathbb{C}}\langle w_1,\dots,w_r \rangle$ is the free ${\mathbb{Z}}/2$-graded algebra with the $w_i$ as odd generators. Finally, the open-closed string map is given by $$u \longmapsto - z_1 \cdots z_r = \lambda \, e.$$ One checks that this is compatible with the ring structure . It also agrees with the general statement from [@auroux07 Theorem 6.1], which says that the critical value of the superpotential, in our case $(r-1)\lambda$, must be an eigenvalue of quantum multiplication with the first Chern class.
Suppose that $r$ is even. Then the Clifford algebra can be (non-canonically) thought of as the total endomorphism algebra of a ${\mathbb{Z}}/2$-graded category with $2^{r/2}$ objects, any two of which are isomorphic up to a shift. It is convenient to first diagonalize the underlying quadratic form by using the modified basis elements $$\tilde{w}_i = w_i + \textstyle(-\frac{1}{r} + \frac{1}{r\sqrt{1-r}})(w_1 + \cdots + w_r),$$ which satisfy $\tilde{w}_i^2 = \lambda$ and $\tilde{w}_i\tilde{w}_j + \tilde{w}_j\tilde{w}_i = 0$ ($i \neq j$). Then the identity endomorphisms of our objects can be taken to be the minimal idempotents $$\label{eq:minimal-idempotent}
p = \textstyle {{\textstyle\frac{1}{2}}}(1 \pm \frac{\sqrt{-1}}{\lambda} \tilde{w}_1 \tilde{w}_2) \cdots {{\textstyle\frac{1}{2}}}(1 \pm \frac{\sqrt{-1}}{\lambda} \tilde{w}_{r-1} \tilde{w}_r).$$
The blowup
----------
Let $M$ be a symplectic manifold, and $i: N \hookrightarrow M$ a symplectic submanifold (as usual, both are assumed to be connected) of codimension $2r$. Let $B$ be the result of blowing up that submanifold, with size $\delta > 0$. We will always suppose that $\delta$ is sufficiently small, making the existence and uniqueness of $B$ unproblematic. Write $D \subset B$ for the exceptional divisor, and $u = -\mathit{PD}([D]) \in H^2(B)$ for its negative Poincar[é]{} dual. The pushforward $H^*(D) \rightarrow H^{*+2}(B)$ and the pullback $H^*(M) \rightarrow H^*(B)$ are both injective, their images are disjoint, and together they span $H^*(B)$. Using Leray-Hirsch one then writes $$\label{eq:blowup-cohomology}
H^*(B) {\cong}H^*(M) \oplus H^*(D)[-2] {\cong}H^*(M) \oplus u H^*(N) \oplus \cdots \oplus u^{r-1} H^*(N).$$ With respect to this decomposition, $$\begin{aligned}
& [\omega_B] = [\omega_M] + 2\pi \delta\, u, \\
& c_1(B) = c_1(M) + (r-1)u. \label{eq:c1}\end{aligned}$$
At this point, we will introduce additional conditions which drastically simplify the blowup construction and its effects.
\[th:few-spheres\] Both $M$ and $N$ have zero first Chern class, and $N$ has trivial (as a symplectic vector bundle) normal bundle. The (real) dimensions satisfy $$\begin{aligned}
\label{eq:dimension-ineq}
& \mathrm{dim}(M) \geq 2\, \mathrm{dim}(N), \\
& \mathrm{dim}(M) - \mathrm{dim}(N) \equiv 0 \, \mathrm{mod} \, 4.\end{aligned}$$ $N$ admits a compatible almost complex structure for which there are no non-constant pseudo-holomorphic spheres. For $B$ we require a weaker kind of condition, namely that the Gromov-Witten invariants $\langle v_1,\dots,v_n \rangle_{A}^B$ for genus $0$ curves in any homology class $A \in H_2(M)$ with $u(A) = -A \cdot [D] \leq 0$ vanish.
In particular, $D$ is diffeomorphic to ${\mathbb{C}}P^{r-1} \times N$, and the local structure near it is described by $B^{\mathit{toy}} \times N$. This first of all simplifies the structure of the cohomology ring of $B$ somewhat. Take the given ring structures on $H^*(M)$ and $H^*(N)$, and the restriction map $i^*: H^*(M) \rightarrow H^*(N)$. These define a ring structure on $H^*(M) \oplus u H^*(N)[u]$. To get the ring structure on one imposes the relations $$\label{eq:gitler}
u^r v = -i_!(v) \in H^{2r+|v|}(M)$$ for $v \in H^*(N)$. This is easily seen by arguing Poincaré dually in terms of intersections. Note that since $i^*i_!$ vanishes, multiplying by $u$ yields $u^{r+1} = 0$ (for a discussion of cohomology rings of blowups going beyond the case of trivial normal bundle, see [@gitler92; @lambrechts-stanley08]).
\[th:qh-blowup\] In terms of , the (small) quantum product $x \ast y$ can be described as follows. If $x$ or $y$ lie in $H^*(M;R)$, the product agrees with the classical cup product. The same is true if $x \in u^j H^*(N;R)$, $y \in u^k H^*(N;R)$ with $j+k < r$. Finally, we have a modified version of : $$\label{eq:quantum-gitler}
u \ast u^{r-1}v = - i_!(v) - \hbar^{2\pi \delta} uv.$$
Take an almost complex structure which, near $D$, is the product of the standard (toric K[ä]{}hler) structure on $B^{\mathit{toy}}$ and a compatible almost complex structure on $N$ with no pseudo-holomorphic spheres in it. Then, any pseudo-holomorphic curve in $B$ either intersects $D$ nonnegatively or is contained in a fibre of the projection $D \rightarrow N$. Our assumption says that the first kind of curves contribute trivially to the quantum product, and the contribution of the second kind is as in the previously discussed toy model case , with the formal parameter $\hbar^{\omega_B(Z)}$ re-inserted.
Multiplying by $u$, but this time with respect to the quantum product, yields the counterpart of : $$\overbrace{u \ast \cdots \ast u}^{r+1} = u \ast (u \ast u^{r-1}) = - \hbar^{2\pi \delta} u^2 = - \hbar^{ 2\pi \delta} u \ast u.$$ Hence, the operator of quantum multiplication with $u$ has eigenvalues $0$ as well as $\lambda$, where $$\label{eq:eigenvalues}
\lambda^{r-1} = -\hbar^{2\pi \delta}.$$ The generalized eigenspaces are $$\begin{aligned}
\label{eq:eigenspace0}
E_0 & = \mathit{ker}(u \ast u \ast \cdot) = \big\{ v + \hbar^{-2\pi \delta} u^{r-1} i^*(v) \, : \, v \in \mathit{H}^*(M;R)\big\}, \\
\label{eq:eigenspace-lambda} E_{\lambda} & = \mathit{ker}(\lambda - u \ast \cdot) \\
\notag
& = \big\{ -i_!(v) + \lambda^{r-1} u v + \lambda^{r-2} u^2 v + \cdots + \lambda u^{r-1} v
\, : \, v \in H^*(N;R) \big\}.\end{aligned}$$ Each of these is a subalgebra. In particular, is isomorphic to $H^*(N;R)$ as a ring, with the unit element being the idempotent $$\label{eq:lambda-idempotent}
q = \frac{1}{(1-r)\lambda \hbar^{2\pi \delta}}(-i_!(1) + \lambda^{r-1} u + \lambda^{r-2} u^2 + \cdots + \lambda u^{r-1}).$$
We also need to consider the (genus zero, with no descendants) relative Gromov-Witten invariants of the pair $(B,D)$. In general, such invariants have the form $$\label{eq:relative-gw}
\langle (\mu_1,w_1),\dots,(\mu_m,w_m),v_1,\dots,v_n \rangle_{A}^{(B,D)} \in {\mathbb{Q}}$$ where $m > 0$, $\mu_1,\dots,\mu_m > 0$ are the multiplicities of tangency with $D$ at the marked points, $w_1,\dots,w_m \in H^*(D;{\mathbb{Q}})$, $v_1,\dots,v_n \in H^*(B;{\mathbb{Q}})$, and $A \in H_2(B)$ is a class such that $A \cdot D = \mu_1 + \cdots + \mu_m$. The virtual dimension formula constrains $$\sum_i (|v_i| - 2) + \sum_j (|w_j| + 2\mu_j - 2) = \mathrm{dim}(B) + 2c_1(A) - 6.$$ The divisor axiom shows that this vanishes if $|v_i| < 2$ for any $i$, so we exclude this case from now on. By and Assumption \[th:few-spheres\] we have $\mathrm{dim}(B) + 2c_1(A) - 6 = (\mathrm{dim}(N) - 2) + (2r-2)(1 - A \cdot D) - 2 \leq (2r-2)(2 - A \cdot D) - 2$. Hence, the only possibly nontrivial invariants have $A \cdot D = 1$, hence $m = 1$ and $\mu_1 = 1$.
\[th:relative-vanish\] $\sum_A \hbar^{\omega_B(A)} \langle (1,w) \rangle_{A}^{(B,D)} = 0$.
The proof follows the strategy of [@maulik-pandharipande06] (see also [@hu-li-ruan08] for similar arguments in a symplectic setting), which is to compare the relative invariants of $(B,D)$ with the absolute invariants of $B$ through degeneration to the normal cone and deformation. A symplectic cut [@lerman95] allows us to write $B$ as the symplectic sum of two pieces $(B^{\mathit{left}},D^{\mathit{left}})$ and $(B^{\mathit{right}},D^{\mathit{right}})$. The first of these is symplectically deformation equivalent to $(B,D)$ itself, but with the cohomology class of the symplectic form changed to $$\label{eq:moving-symplectic-form}
[\omega_{B^{\mathit{left}}}] = [\omega_B] + 2\pi (\epsilon - \delta) u.$$ The other piece is $B^{\mathit{right}} = \bar{B}^{\mathit{toy}} \times N$ containing $D^{\mathit{right}} = H^{\mathit{toy}} \times N$. It is unproblematic to assume that $\epsilon = (r+1)/(r-1) \delta$, which slightly simplifies the situation since $\bar{B}^{\mathit{toy}}$ becomes monotone.
To keep the notation compact, we will mostly omit the homology classes of pseudo-holomorphic curves. In that case, the convention is that we sum over all classes with $\hbar$ weights given by symplectic areas. Choose a basis $\{w_i\}$ of $H^*(D^{\mathit{right}};{\mathbb{Q}})$ and the corresponding Poincaré dual basis $\{w_i^*\}$ , so that $\sum_i w_i \otimes w_i^*$ represents the Poincaré of the diagonal. In view of the restrictions on relative Gromov-Witten invariants observed above, the symplectic sum formula [@ionel-parker04; @li-ruan98] takes on the form $$\label{eq:symplectic-sum}
\begin{aligned}
& \langle uw \rangle^B - \sum_{A^{\mathit{right}} \cdot D^{\mathit{right}} = 0} \hbar^{\omega_{B^{\mathit{right}}}(A^{\mathit{right}})} \langle uw \rangle_{A^{\mathit{right}}}^{B^{\mathit{right}}} = \\ &
\sum_i \langle (1,w_i) \rangle^{(B^{\mathit{left}},D^{\mathit{left}})} \langle (1,w_i^*), uw \rangle^{(B^{\mathit{right}},D^{\mathit{right}})} \\
& + \sum_{i_1,i_2} \langle (1,w_{i_1}) \rangle^{(B^{\mathit{left}},D^{\mathit{left}})} \langle (1,w_{i_2}) \rangle^{(B^{\mathit{left}},D^{\mathit{left}})}
\langle (1,w_{i_1}^*), (1,w_{i_2}^*), uw \rangle^{(B^{\mathit{right}},D^{\mathit{right}})} \\
& + \cdots
\end{aligned}$$ Here, we think of $uw$ as being represented by (minus) the Poincaré dual of $w$ inside $D$. Degeneration moves such a cycle into $B^{\mathit{right}}$, and therefore the simplest term corresponds to curves in $B^{\mathit{right}}$ which do not intersect $D^{\mathit{right}}$. The other terms measure configurations consisting of a curve in $B^{\mathit{right}}$ with an arbitrary number of “tails” in $B^{\mathit{left}}$ (see Figure \[fig:sum\]). Assumption \[th:few-spheres\] ensures that the left hand side of this equation vanishes, since it implies that only curves lying inside $D$ contribute to $\langle uw \rangle^B$, and the contribution there is the same as to $\langle uw \rangle^{B^{\mathit{right}}}$. The lowest energy contribution to the relative invariants of $(B^{\mathit{right}},D^{\mathit{right}})$ comes from rational curves lying in the $\bar{B}^{\mathit{toy}}$ fibres, and in fact those which represents proper transforms of lines through the origin in ${\mathbb{C}}P^r$. Using that one rewrites as $$0 = \langle (1,w) \rangle^{(B^{\mathit{left}},D^{\mathit{left}})} + \left\{\parbox{20em}{series whose terms are monomials of degree $>0$ in the $\langle (1,w_i) \rangle^{(B^{\mathit{left}},D^{\mathit{left}})}$, with coefficients containing only $>0$ powers of $\hbar$}\right\}.$$ If we assume that $(1,\cdot)^{(B^{\mathit{left}},D^{\mathit{left}})}$ is nonzero, this immediately leads to a contradiction by looking at the lowest possible power of $\hbar$ which occurs. Deformation invariance shows that the relative Gromov-Witten invariants of $(B,D)$ and $(B^{\mathit{left}},D^{\mathit{left}})$ agree. Of course, the symplectic areas change, but because of they change in the same way for all the homology classes of curves involved in $\langle (1,w) \rangle^{(B,D)}$ (in fact, the last step is not even needed for our applications, since the relative invariants of $(B^{\mathit{left}},D^{\mathit{left}})$ is what will really be relevant).
(0,0)![\[fig:sum\]](sum.pstex "fig:")
\#1\#2\#3\#4\#5[ @font ]{}
(3426,3115)(1014,-2014) (3451,-1036)[(0,0)\[lb\]]{} (2701,389)[(0,0)\[lb\]]{}
From now on, we will assume for technical convenience that $N$ is [*Spin*]{} (the reason is the same as in Example \[th:graph\]). Using the local model $B^{\mathit{toy}} \times N \subset B$, we now introduce a Lagrangian correspondence $$C = C^{\mathit{toy}} \times \Delta_N \subset B^{\mathit{toy}} \times N \times N^- \subset B \times N^-.$$
\[th:hf-correspondence\] Take $\lambda$ as in . By a suitable choice of local coefficient system and bounding cocycle, $C$ can be made into an object of $\mathit{Fuk}(B \times N^-)_{(r-1)\lambda}$, in such a way that the following holds. We have a canonical ring isomorphism $$\label{eq:c-ring}
\begin{aligned}
& \mathit{HF}^*(C,C) {\cong}\mathit{HF}^*(C^{\mathit{toy}},C^{\mathit{toy}}) \otimes H^*(N;R) \\
& \quad {\cong}R\langle w_1,\dots,w_r \rangle/ \{w_i^2 \text{ for all $i$},\, w_iw_j+w_jw_i+2\lambda \text{ for all $i \neq j$}\} \otimes H^*(N;R).
\end{aligned}$$ Moreover, specializing the open-closed string map yields a ring homomorphism $$\label{eq:double-open-closed}
\mathit{QH}^*(B) \otimes \mathit{QH}^*(N) \longrightarrow \mathit{HF}^*(C,C),$$ which can be partially described as follows. On the factor $\mathit{QH}^*(N) = H^*(N;R)$ we just have the obvious inclusion; and elements of the form $uv \in u H^*(N;R) \subset \mathit{QH}^*(B)$ are mapped to $\lambda v$.
Recall that, up to a suitable notion of homotopy [@fooo Chapter 4], the curved $A_\infty$-algebra $$\label{eq:endo-curved}
\mathit{hom}_{\mathit{FO}(B \times N^-)}(C,C)$$ is independent of auxiliary choices (like almost complex structures) made in defining $\mathit{FO}(B \times N^-)$. This is proved by using parametrized moduli spaces [@fooo Section 4.6], and the same argument allows one to degenerate $B \times N^-$ to $(B^{\mathit{left}} \times N^-) \cup (B^{\mathit{right}} \times N^-)$, where $C$ goes to $C^{\mathit{toy}} \times \Delta_N \subset (B^{\mathit{right}} \times N^-) = \bar{B}^{\mathit{toy}} \times N \times N^-$. In parallel with , The resulting curved $A_\infty$-structure consists of that inherited from $C^{\mathit{toy}} \subset B^{\mathit{toy}}$ and correction terms involving relative Gromov-Witten invariants of $(B^{\mathit{left}},D^{\mathit{left}})$. To be precise, the $A_\infty$-structure involves actual cycles representing the invariants from Lemma \[th:relative-vanish\]. However, again up to homotopy, the specific choice of cycles is irrelevant, so one can take them to be empty. The structure of up to homotopy determines the possible bounding cycles, as well as the endomorphism algebras of the resulting objects of the actual (unobstructed) Fukaya category [@fooo Theorem 4.1.3]. With that in mind, follows from the computations in the model case (one uses the same flat ${\mathbb{C}}^*$-bundles).
The same argument applies to the first order infinitesimal “bulk” deformations induced by cycles in $B$, as long as those cycles can be moved entirely into $B^{\mathit{right}} \times N$ when degenerating, which is true for all the elements considered in the statement of the Lemma. The resulting computation is simple, and we leave it to the reader.
\[th:idempotent-goes-to-unit\] Consider again just the part of concerning $\mathit{QH}^*(B)$. Using the ring structure we deduce that $$i_!(1) = -\overbrace{u \ast \cdots \ast u}^r - \hbar^{2\pi \delta} u \longmapsto - \lambda^r - \hbar^{2\pi \delta} \lambda = 0.$$ Furthermore, the image of the idempotent $q$ from is the identity element in $\mathit{HF}^0(C,C)$.
A correspondence functor\[subsec:correspondence\]
-------------------------------------------------
Pick some $\lambda$ as in , and suppose that $C$ has been made into an object of $\mathit{Fuk}(B \times N^-)_{(r-1)\lambda}$ following Lemma \[th:hf-correspondence\]. Take the idempotent $q \in \mathit{QH}^0(B)$ which defines the projection to the corresponding eigenspace , and let $p \in \mathit{HF}^*(C,C)$ be one of the idempotents from .
\[th:ww\] The formal direct summand of $C$ associated to $p$ gives rise to a cohomologically full and faitful functor $$\label{eq:c-functor}
\mathit{Fuk}(N)_0^{\mathit{perf}} \longrightarrow \mathit{Fuk}(B)_{(r-1)\lambda,q}^{\mathit{perf}}.$$ For brevity, denote these two categories by $A_N$ and $A_B$. The induced maps on Hochschild cohomology and open-closed string maps fit into a commutative diagram $$\label{eq:open-closed-correspondence}
\xymatrix{
\mathit{QH}^*(N) \ar[rr] \ar[d] && p\mathit{HF}^*(C,C)p \ar[d] && \ar[ll] \ar[d] \mathit{QH}^*(B)q \\
\mathit{HH}^*(A_N,A_N) \ar[rr] && \mathit{HH}^*(A_N,A_B) && \ar[ll] \mathit{HH}^*(A_B,A_B)
}$$ where the maps in the top row come from .
Parts of this statement come from the general theory of Fukaya categories and Lagrangian correspondences, and do not need specific proofs. Given that $q$ maps to the identity in $\mathit{HF}^*(C,C)$, the correspondence $C$ gives rise to a functor taking values in a module category: $$\label{eq:formal-functor}
\mathit{Fuk}(N)_0 \longrightarrow \mathit{Fuk}(B)_{\lambda,q}^{\mathit{mod}},$$ and this fits into a commutative diagram analogous to . As pointed out in [@smith10], any decomposition of $C$ into formal direct summands, such as the one provided by $p$, yields a corresponding decomposition of this functor. The two additional facts that need to be proved are first of all that the functor takes values in the subcategory $\mathit{Fuk}(B)_{\lambda,q}^{\mathit{perf}}$, and that it is full and faithful. There are general results which ensure that the first property holds under suitable assumptions [@wehrheim-woodward09; @lekili-lipyanskiy10], and which also make it easy to determine the action of the functor on Floer cohomology groups. However, they do not apply exactly to the situation under discussion, and we will instead argue as in Lemma \[th:hf-correspondence\].
It is convenient to introduce a quilted version $\mathit{Fuk}^\sharp(B)_{(r-1)\lambda}$ of the Fukaya category, as in [@mww] but tailored to our specific application. As objects, this admits objects of $\mathit{Fuk}(B)_{(r-1)\lambda}$ as well as generalized Lagrangian submanifolds of a specific kind, namely pairs consisting of our fixed Lagrangian correspondence $C$ and an object of $\mathit{Fuk}(N)_0$. Morphisms are defined by (chain complexes underlying) quilted Floer cohomology [@ww]. We can again use $q$ to project to a piece of the quilted category, and denote the result by $\mathit{Fuk}^\sharp(B)_{(r-1)\lambda,q}$. By construction, can be factored as follows: $$\label{eq:quilted-functor}
\mathit{Fuk}(N)_0 \longrightarrow \mathit{Fuk}^\sharp(B)_{(r-1)\lambda,q} \longrightarrow \mathit{Fuk}(B)_{(r-1)\lambda,q}^{\mathit{mod}}.$$ The second arrow is a Yoneda-type functor, which is full and faithful when restricted to the subcategory $\mathit{Fuk}(B)_{(r-1)\lambda,q} \subset \mathit{Fuk}^\sharp(B)_{(r-1)\lambda,q}$. What we want to show is that the first arrow takes any Lagrangian submanifold to an object that’s quasi-isomorphic to one in $\mathit{Fuk}(B)_{(r-1)\lambda,q}$.
Let’s temporarily turn to a simpler geometric situation, which is $C^{\mathit{toy}} \times \Delta_N \subset B^{\mathit{toy}} \times N \times N^-$. We have an analogue of , and the first part of it maps any object $L_0$ of $\mathit{Fuk}(N)_0$ to the generalized Lagrangian submanifold $(C^{\mathit{toy}} \times \Delta_N, L_0)$ in $\mathit{Fuk}^\sharp(B^{\mathit{toy}} \times N)_{(r-1)\lambda,q}$. In particular, if $L_1$ is any object of $\mathit{Fuk}(B^{\mathit{toy}} \times N)_{(r-1)\lambda}$, then the quilted Floer cohomology can be expressed in terms of ordinary Floer cohomology as $$H(\mathit{hom}_{\mathit{Fuk}^\sharp(B^{\mathit{toy}} \times N)_{(r-1)\lambda}}(L_1, (C^{\mathit{toy}} \times \Delta_N, L_0))) {\cong}\mathit{HF}^*(L_1, C^{\mathit{toy}} \times L_0).
$$ This is fairly straightforward, involving only re-folding strips and changing their widths, as indicated in Figure \[fig:re-quilt\] (but not the deeper analytic degeneration arguments from [@wehrheim-woodward09]). The same thing holds for morphisms in the other direction, and these isomorphisms are compatible with products. From this, one easily concludes that $(C^{\mathit{toy}} \times \Delta_N,L_0)$ is quasi-isomorphic to $C^{\mathit{toy}} \times L_0$.
If one now looks at the original picture and applies the degeneration argument from Lemma \[th:hf-correspondence\], the outcome is that the computations carried out inside $B^{\mathit{toy}} \times N$ could in principle be deformed by contributions lying in $B^{\mathit{left}}$, but that these in fact vanish, ensuring that the argument above remains valid. If instead of $C$ one now uses its direct summand defined by $p$, the resulting functor no longer lands in the actual Fukaya category, but in its idempotent completion, which then allows a formal extension as in . The proof that the resulting functor is cohomologically full and faithful uses the same kind of argument, the concrete input being that $p\mathit{HF}^*(C^{\mathit{toy}},C^{\mathit{toy}})p {\cong}R$, and we leave it to the reader.
(0,0)![\[fig:re-quilt\]](quilt.pstex "fig:")
\#1\#2\#3\#4\#5[ @font ]{}
(6912,7911)(61,-6664) (1726,-1186)[(0,0)\[lb\]]{} (1201,-1186)[(0,0)\[lb\]]{} (451,-136)[(0,0)\[lb\]]{} (2701,-136)[(0,0)\[lb\]]{} (5701,-6586)[(0,0)\[lb\]]{} ( 76,-2836)[(0,0)\[lb\]]{} (826,-3886)[(0,0)\[lb\]]{} (3226,-2836)[(0,0)\[lb\]]{} (4251,-3211)[(0,0)\[lb\]]{} (4251,-61)[(0,0)\[lb\]]{} (2401,-3886)[(0,0)\[lb\]]{} (1126,-1636)[(0,0)\[lb\]]{} (2176,-1636)[(0,0)\[lb\]]{} (1651,-3886)[(0,0)\[lb\]]{} (4551,-3476)[(0,0)\[lb\]]{} (4576,-5461)[(0,0)\[lb\]]{} (6601,-5461)[(0,0)\[lb\]]{} (4051,-6136)[(0,0)\[lb\]]{} (2701,-5461)[(0,0)\[lb\]]{} (1276,1064)[(0,0)\[lb\]]{} (1876,-6586)[(0,0)\[lb\]]{} (1201,-6586)[(0,0)\[lb\]]{} (451,-5461)[(0,0)\[lb\]]{} (1576,-4336)[(0,0)\[lb\]]{}
\[th:blowup-generate\] Suppose that we have a full $A_\infty$-subcategory $O_N \subset \mathit{Fuk}(N)_0$ which satisfies the assumptions of Theorem \[th:split-generation-3\], meaning that there is a linear map $$\label{eq:n-int}
\mathit{HH}^*(O_N,O_N) \longrightarrow R$$ whose composition with the open-closed string map yields $\int_N$. Let $O_B \subset A_B$ be its image under the functor induced by $(C,p)$. By restricting we get a commutative diagram $$\label{eq:o-diagram}
\xymatrix{
\mathit{QH}^*(N) \ar[rr] \ar[d] && p\mathit{HF}^*(C,C)p \ar[d] && \ar[ll] \ar[d] \mathit{QH}^*(B)q \\
\mathit{HH}^*(O_N,O_N) \ar[rr]^{{\cong}} && \mathit{HH}^*(O_N,O_B) && \ar[ll]_{{\cong}} \mathit{HH}^*(O_B,O_B).
}$$ Take $v \in \mathit{QH}^*(N;R)$, and associate to it an element $x$ of $\mathit{QH}^*(B)q = E_\lambda$, as in but additionally dividing by the nonzero constant $(1-r)\lambda^r \hbar^{2\pi\delta}$. Crucially, $x$ has a nonzero top degree component (with respect to the ordinary grading of cohomology) if and only if $v$ does. Lemma \[th:hf-correspondence\] (see also Addendum \[th:idempotent-goes-to-unit\]) shows that $v$ and $x$ have the same image in $\mathit{HF}^*(C,C)$. Hence, if we combine with the maps in to define a map $$\label{eq:b-int}
\mathit{HH}^*(O_B,O_B) \longrightarrow R,$$ the assumptions of Theorem \[th:split-generation-3\] are satisfied. This shows that the image of our functor split-generates, which means that it is a quasi-equivalence $A_N {\cong}A_B$.
The examples\[subsec:final\]
----------------------------
Let $K \subset {\mathbb{C}}P^3$ be a quartic surface, containing a smooth elliptic curve $T \subset K$. We equip $K$ with the Fubini-Study form rescaled in such a way that $T$ has area $1$. Take a symplectic automorphism $f$ of $K$ as in Lemma \[th:2-twists\], and form the symplectic mapping torus of $f \times f$ as in , calling the result $E$. We will compare this to the same process applied to the identity map, which gives $E^{\mathit{triv}} {\cong}T \times K^2$ (the sign of the symplectic form does not matter, see the discussion at the end of Section \[subsec:covering-trick\]).
\[th:deform-torus\] There is a diffeomorphism $g: E \rightarrow E^{\mathit{triv}}$ which is compatible with the cohomology classes of symplectic forms, and with the homotopy classes of almost complex structures. One can even construct $g$ in such a way that $g^*\omega_{E^{\mathit{triv}}}$ is deformation equivalent to $\omega_E$, which means that there is a family of symplectic forms $(\omega_{E,r})$ interpolating between the two. Finally, one can arrange that the associated family $[\omega_{E,r}]$ of cohomology classes is arbitrarily close to a constant one.
The first sentence follows directly from Lemma \[th:2-twists\], while the rest requires inspection of the proof of that Lemma. As discussed there, one can find a deformation $(\omega_{K,r})$ of the symplectic form $\omega_{K,0} = \omega_K$, as well as an isotopy $(f_r)_{0 \leq r \leq 1}$ of diffeomorphisms starting with $f_0 = f$, such that $f_r^*\omega_{K,r} = \omega_{K,r}$, and $f_r$ is isotopic to the identity in $\mathit{Aut}(K,\omega_{K,r})$ for all $r>0$. The concatenation of $(f_r)$ (going from $0$ to some small value $r = \rho$), the symplectic isotopy between $f_\rho$ and the identity, and deformation back to $\omega_{K,0}$, yields a family of symplectic mapping tori, interpolating between $E$ and $E^{\mathit{triv}}$. Ordinary (differentiable) parallel transport in that family yields the desired diffeomorphism.
Define $N = T \times E$ and $N^{\mathit{triv}} = T \times E^{\mathit{triv}}$, with the product symplectic structures as usual. We embed $N^{\mathit{triv}}$ symplectically into $M = K^7$ by identifying it with $T^2 \times K^2 \times \{\mathit{point}\}^3$. Denote the embedding by $i^{\mathit{triv}}$.
\[th:gromov\] There is a symplectic embedding $i: N \rightarrow M$ which is deformation equivalent to $i^{\mathit{triv}} \circ g$ (for $g$ as in Lemma \[th:deform-torus\]) provided that we allow deformations of the symplectic form on $M$. More precisely, this means the following: there is a family of symplectic forms $\omega_{M,r}$ ($0 \leq r \leq 1$) on $M$ which starts and ends at the given one, $\omega_{M,0} = \omega_{M,1} = \omega_M$, and a family $i_r: N^{\mathit{triv}} \rightarrow M$ of embeddings such that $i_0 = i^{\mathit{triv}}$, $i_r^*\omega_{M,r}$ is symplectic for all $r$, and $g^*i_1^*\omega_{M,1} = \omega_N$. One then sets $i = i_1 \circ g$.
Since $T \subset K$ represents a nonzero homology class, the composition $$H^2(M;{\mathbb{R}}) \xrightarrow{(i^{\mathit{triv}})^*} H^2(N^{\mathit{triv}};{\mathbb{R}}) \longrightarrow H^2(E^{\mathit{triv}};{\mathbb{R}})$$ is onto. Take $g$ as in Lemma \[th:deform-torus\], and a family $\omega_{E,r}$ of symplectic forms interpolating between $g^*\omega_{E^{\mathit{triv}}}$ and $\omega_E$. One can then find a family of closed two-forms $\omega_{M,r}$ such that $\omega_{M,0} = \omega_{M,1} = \omega_M$ and $$\label{eq:coh-restriction}
g^*(i^{\mathit{triv}})^*[\omega_{M,r}] = [\omega_T \times 1] + [1 \times \omega_{E,r}].$$ Moreover, since the cohomology classes on the right hand side of can be chosen arbitrarily close to constant ones, one can arrange that the family $\omega_{M,r}$ itself is close to a constant one, hence that all the $\omega_{M,r}$ are symplectic.
At this point, we apply a parametrized version of the [*h*]{}-principle for symplectic immersions. The original result is [@gromov86 (3.4.2.A)], see also [@eliashberg-mishachev (16.4.3)], and the parametrized version should be parallel. The statement is that there is a family of immersions $i_r$ starting with $i_0 = i^{\mathit{triv}}$, and such that $g^*i_r^*\omega_{M,r} = \omega_T \times 1 + 1 \times \omega_{E,r}$. Finally, since $\mathrm{dim}(M) > 2\,\mathrm{dim}(N) + 1$, a $C^1$-small perturbation will turn these immersions into embeddings (alternatively, one could try to appeal to a stronger result of Gromov that produces embeddings directly, see [@gromov86 (3.4.2.B)] or [@eliashberg-mishachev (12.1.1)], which would result in an argument closer to [@mcduff87 Section 3]).
Let $B$ be the result of blowing up $M$ along $N$ (embedded through the map $i$ we have just constructed) with small parameter $\delta>0$, and $B^{\mathit{triv}}$ the same with $N^{\mathit{triv}}$. Lemma \[th:gromov\] implies that $B$ is symplectically deformation equivalent to $B^{\mathit{triv}}$.
Both blowups $B$ and $B^{\mathit{triv}}$ satisfy Assumption \[th:few-spheres\].
Let $J_K^{\mathit{int}}$ be the given (integrable) complex structure on $K$, for which $T \subset K$ is a holomorphic curve. By standard transversality methods, we can find another compatible almost complex structure $J_K$ which agrees with $J_K^{\mathit{int}}$ in a neighbourhood of $T$, and which has no non-constant $J_K$-holomorphic spheres. Equip $K^5$ with the product structure induced by $J_K$. Then $T^2 \times \{\mathit{point}\}^3$ is a complex submanifold, and if we choose the point to lie on $T$ as well, then $J_K$ is integrable near that submanifold. We can therefore carry out the blowup process following the local algebro-geometric model. Having done that, take the product with two more copies of $K$ equipped with $J_K$. The outcome is that we get an almost complex structure $J_{B^{\mathit{triv}}}$ on the blowup and a pseudo-holomorphic blowdown map $(B^{\mathit{triv}},J_{B^{\mathit{triv}}}) \rightarrow (K,J_K)^7$. Hence, all the pseudo-holomorphic spheres must be contained in the fibres of this map, which means that they lie in multiples of the homology class $Z$ of a line in the exceptional divisor. Since $u(Z) = 1$, this implies the vanishing of Gromov-Witten invariants as required in Assumption \[th:few-spheres\]. The rest of the requirements for $B^{\mathit{triv}}$ are obvious.
Because of the deformation linking it to $i^{\mathit{triv}}$, $i$ inherits the property that the normal bundle is trivial. Since $B$ is deformation equivalent to $B^{\mathit{triv}}$, it has the same Gromov-Witten invariants. The absence of pseudo-holomorphic spheres in $N$ is easy to arrange, see Section \[subsec:basic-mapping-torus\].
Choose some $\lambda$ as in . The subspace $E_\lambda \subset \mathit{QH}^*(B)$ has an intrinsic characterization as the eigenspace of $(r-1)\lambda$ for quantum multiplication by $c_1(B)$. It is also an algebra with identity element $q$ as in . Take $\lambda^{-1} H^3(B;{\mathbb{Z}}) = \lambda^{-1} u H^1(N;{\mathbb{Z}})$, project it to $E_\lambda$ by taking its quantum multiplication with $q$, and denote the outcome by $\Gamma_\lambda \subset \mathit{QH}^1(B)$. Consider the associated Fukaya category $\mathit{Fuk}(B)_{(r-1)\lambda,q}$. Let ${\mathscr{S}}= \mathit{Spec}({\mathscr{R}})$ be the curve associated to the unit torus polynomial, with its standard one-form $\theta$. We also have their closures $\bar{{\mathscr{S}}}$ and $\bar{\theta}$.
\[th:1\] The image of $\Gamma_{\lambda}$ under the open-closed string map is not contained in the set $\mathit{Per}(\mathit{Fuk}(B)_{(r-1)\lambda,q},\bar{{\mathscr{S}}},\bar{\theta})$ of periodic elements.
Corollary \[th:not-periodic\] implies that there is a class in $H^1(N;{\mathbb{Z}})$ whose image is not contained in $\mathit{Per}(\mathit{Fuk}(N)_0,\bar{{\mathscr{S}}},\bar{\theta})$. More precisely, this is not quite the result as originally stated, but has the following minor differences. First of all, the grading of the Fukaya category has been reduced to ${\mathbb{Z}}/2$, which means that we have to check whether the uniqueness results from Section \[subsec:unique\] are applicable. However, that was already taken care of in Addenda \[th:sneaky\] and \[th:not-iso-even-without-grading\]. The second difference is that we are considering the product of the symplectic mapping torus with an additional copy of $T$, which correspondingly means that we have to take the product of the Lagrangian submanifolds under consideration with a fixed circle in $T$. This requires the same kind of check, but the argument from Lemma \[th:assumption-satisfied\] goes through as before, and similarly for Addendum \[th:not-iso-even-without-grading\]. The third difference is that the Fukaya category $\mathit{Fuk}(N)_0$ contains more objects than our original version $\mathit{Fuk}(N)$ (see Remark \[th:compare\]), but that is clearly irrelevant for this argument.
We can now use the full and faithful functor from Lemma \[th:ww\] to transfer this result to $B$. By our computation of , any class $v \in H^1(N;{\mathbb{Z}}) \subset \mathit{QH}^1(N)$ and its counterpart $\lambda^{-1} uv \ast q \in \Gamma_{\lambda}$ have the same image in $\mathit{HH}^*(\mathit{Fuk}(N)_0,\mathit{Fuk}(B)_{(r-1)\lambda,q})$. As explained in Section \[subsec:functor\], this allows one to map families that follow a given deformation field. The rest of the argument carries over without any changes.
We now consider the analogous construction for $B^{\mathit{triv}}$, defining the idempotent $q^{\mathit{triv}} \in \mathit{QH}^0(B^{\mathit{triv}})$ and subgroup $\Gamma^{\mathit{triv}}_{\lambda} \subset \mathit{QH}^1(B^{\mathit{triv}})$ as before.
The image of $\Gamma^{\mathit{triv}}_{\lambda}$ under the open-closed string map is contained in the set $\mathit{Per}(\mathit{Fuk}(B^{\mathit{triv}})_{(r-1)\lambda,q^{\mathit{triv}}},\bar{{\mathscr{S}}},\bar\theta)$.
As before, the first step is to establish a version of Corollary \[th:yes-periodic\] for $\mathit{Fuk}(N^{\mathit{triv}})_0$, the key additional consideration being that we are now working with a larger Fukaya category than before. However, one can use Theorem \[th:split-generation-3\] and the argument from Example \[th:product-mirror\] to show that $\mathit{Fuk}(N^{\mathit{triv}})_0$ is split-generated by the subcategory $\mathit{Fuk}(N^{\mathit{triv}})$ (with its grading reduced to ${\mathbb{Z}}/2$). One takes the family of bimodules used in the proof of Corollary \[th:yes-periodic\], reduces its grading to ${\mathbb{Z}}/2$ as well, and then extends it to a family of bimodules over $\mathit{Fuk}(N^{\mathit{triv}})_0$. When carrying over the results to $B^{\mathit{triv}}$, one uses Addendum \[th:blowup-generate\] for split-generation, and the same computation as in Proposition \[th:1\].
Comparing the two Propositions above shows that, as announced in the Introduction,
$B$ and $B^{\mathit{triv}}$ are not symplectically isomorphic.
[100]{}
M. Abouzaid, *A geometric criterion for generating the [F]{}ukaya category*, Publ. Math. IHES, To appear.
M. Abouzaid, K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono, In preparation.
M. Abouzaid and I. Smith, *Homological mirror symmetry for the four-torus*, Duke Math. J. **152** (2010), 373–-440.
P. Albers, *A [L]{}agrangian [P]{}iunikhin-[S]{}alamon-[S]{}chwarz morphism and two comparison homomorphisms in [F]{}loer homology*, Int. Math. Res. Not. (2008), Art. ID 134, 56 pages.
M. Atiyah, *Complex analytic connections in fibre bundles*, Trans. Amer. Math. Soc. **85** (1957), 181–207.
D. Auroux, *Mirror symmetry and [T]{}-duality in the complement of an anticanonical divisor*, J. Gökova Geom. Topol. **1** (2007), 51–91.
A. Beilinson, V. Ginsburg, and V. Schechtman, *Koszul duality*, J. Geom. Physics **5** (1988), 317–350.
A. Beilinson, V. Ginzburg, and W. Soergel, *Koszul duality patterns in representation theory*, Journal of the Amer. Math. Soc. **9** (1996), 473–527.
J. Bernstein, *Algebraic theory of [$D$]{}-modules*, Unpublished lecture notes.
P. Biran and O. Cornea, *A [L]{}agrangian quantum homology*, New perspectives and challenges in symplectic field theory, CRM Proc. Lecture Notes, vol. 49, Amer. Math. Soc., 2009, pp. 1–44.
[to3em]{}, *Rigidity and uniruling for [L]{}agrangian submanifolds*, Geom. Topol. **13** (2009), 2881–2989.
A. Bondal and M. Van den Bergh, *Generators and representability of functors in commutative and noncommutative geometry*, Moscow Math. J. **3** (2003), 1–36.
A. Bondal and M. Kapranov, *Enhanced triangulated categories*, Math. USSR Sbornik **70** (1991), 93–107.
F. Bourgeois, *A [M]{}orse-[B]{}ott approach to contact homology*, Ph.D. thesis, Stanford University, 2002.
L. Buhovsky, *The [M]{}aslov class of [L]{}agrangian tori and quantum products in [F]{}loer cohomology*, J. Topol. Anal. **2** (2010), 57–75.
H. Cartan and S. Eilenberg, *Homological [A]{}lgebra*, Princeton Univ. Press, 1956.
C.-H. Cho and Y.-G. Oh, *Floer cohomology and disc instantons of [L]{}agrangian torus fibers in toric [F]{}ano manifolds*, Asian J. Math. **10** (2006), 773–814.
B. Conrad, *Several approaches to non-[A]{}rchimedean geometry*, [$P$]{}-adic geometry, Univ. Lecture Ser., vol. 45, Amer. Math. Soc., 2008, pp. 9–63.
P. Deligne, P. Griffiths, J. Morgan, and D. Sullivan, *Real homotopy theory of [K]{}[ä]{}hler manifolds*, Invent. Math. **29** (1975), 245–274.
A. Dold, *Zur [H]{}omotopietheorie der [K]{}ettenkomplexe*, Math. Ann. **140** (1960), 278––298.
[to3em]{}, *Lectures on [A]{}lgebraic [T]{}opology (2nd ed.)*, Springer, 1980.
S. Dostoglou and D. Salamon, *Self dual instantons and holomorphic curves*, Annals of Math. **139** (1994), 581–640.
V. Drinfeld, *D[G]{} quotients of [DG]{} categories*, J. Algebra **272** (2004), 643–691.
Ya. Eliashberg and N. Mishachev, *Introduction to the [$h$]{}-principle*, Amer. Math. Soc., 2002.
A. Floer, *Symplectic fixed points and holomorphic spheres*, Commun. Math. Phys. **120** (1989), 575–611.
R. Fr[ö]{}berg, *Koszul algebras*, Advances in commutative ring theory ([F]{}ez, 1997), Lecture Notes in Pure and Appl. Math., vol. 205, Dekker, 1999, pp. 337–350.
K. Fukaya, *Floer homology and mirror symmetry. [II]{}*, Minimal surfaces, geometric analysis and symplectic geometry (Baltimore, MD, 1999), Math. Soc. Japan, 2002, pp. 31–127.
K. Fukaya, *Floer homology for families—a progress report*, Integrable systems, topology, and physics ([T]{}okyo, 2000), Amer. Math. Soc., 2002, pp. 33–68.
[to3em]{}, *Mirror symmetry of abelian varieties and multi-theta functions*, J. Algebraic Geom. **11** (2002), 393–512.
K. Fukaya, *Cyclic symmetry and adic convergence in [L]{}agrangian [F]{}loer theory*, Preprint arXiv:0907.4219, 2009.
[to3em]{}, *Floer homology of [L]{}agrangian submanifolds*, Preprint arXiv:1106.4882, 2011.
K. Fukaya and Y.-G. Oh, *Zero-loop open strings in the cotangent bundle and [M]{}orse homotopy*, Asian J. Math. **1** (1998), 96–180.
K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono, *Canonical models of filtered [$A_\infty$]{}-algebras and [M]{}orse complexes*, New perspectives and challenges in symplectic field theory, CRM Proc. Lecture Notes, vol. 49, Amer. Math. Soc., 2009, pp. 201–227.
K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono, *Lagrangian [F]{}loer theory on compact toric manifolds. [I]{}*, Duke Math. J. **151** (2010), 23–174.
K. Fukaya, Y.-G. Oh, H. Ohta, and K. Ono, *Lagrangian intersection [F]{}loer theory - anomaly and obstruction*, Amer. Math. Soc., 2010.
[to3em]{}, *Lagrangian [F]{}loer theory on compact toric manifolds. [II]{}: [B]{}ulk deformations*, Selecta Math. (2011), In press.
S. Gitler, *The cohomology of blow ups*, Bol. Soc. Mat. Mex. **37** (1992), 167–175.
W. Goldman and J. Millson, *The deformation theory of the fundamental group of compact [K]{}[ä]{}hler manifolds*, IHES Publ. Math. **67** (1988), 43–96.
I. Gordon, *Symplectic reflection algebras*, Trends in representation theory of algebras and related topics, Eur. Math. Soc., 2008, pp. 285–347.
E. Green, G. Hartman, E. Marcos, and [Ø]{}. Solberg, *Resolutions over [K]{}oszul algebras*, Arch. Math. **85** (2005), 118–127.
M. Gromov, *Partial differential relations*, Springer, 1986.
J. Harer, A. Kas, and R. Kirby, *Handlebody decompositions of complex surfaces*, Mem. Amer. Math. Soc., vol. 62, Amer. Math. Soc., 1986.
H. Hofer and D. Salamon, *Floer homology and [N]{}ovikov rings*, The [F]{}loer memorial volume (H. Hofer, C. Taubes, A. Weinstein, and E. Zehnder, eds.), Progress in Mathematics, vol. 133, Birkh[ä]{}user, 1995, pp. 483–524.
K. Hori and C. Vafa, *Mirror symmetry*, Preprint hep-th/0002222, 2000.
J. Hu, T.-J. Li, and Y. Ruan, *Birational cobordism invariance of uniruled symplectic manifolds*, Invent. Math. **172** (2008), 231–275.
D. Huybrechts and R. Thomas, *Deformation-obstruction theory for complexes via [A]{}tiyah and [K]{}odaira-[S]{}pencer classes*, Math. Ann. **346** (2010), 545–569.
E. Ionel and T. Parker, *The symplectic sum formula for [G]{}romov-[W]{}itten invariants*, Ann. of Math. (2) **159** (2004), 935–1025.
J. Johns, *Complexifications of [M]{}orse functions and the directed [D]{}onaldson-[F]{}ukaya category*, J. Symplectic Geom. **8** (2010), 403–500.
T. Kadeishvili, *The structure of the [$A_\infty$]{}-algebra, and the [H]{}ochschild and [H]{}arrison cohomologies*, Trudy Tbiliss. Mat. Inst. Razmadze Akad. Nauk Gruzin. SSR **91** (1988), 19–27 (Russian).
B. Keller, *Introduction to [$A$]{}-infinity algebras and modules*, Homology Homotopy Appl. (electronic) **3** (2001), 1–35.
B. Keller, *Hochschild cohomology and derived [P]{}icard groups*, J. Pure Appl. Algebra **190** (2004), 177–196.
[to3em]{}, *On differential graded categories*, International [C]{}ongress of [M]{}athematicians. Vol. [I]{}I, Eur. Math. Soc., 2006, pp. 151–190.
M. Khovanov and P. Seidel, *Quivers, [F]{}loer cohomology, and braid group actions*, J. Amer. Math. Soc. **15** (2002), 203–271.
M. Kontsevich, *Homological algebra of mirror symmetry*, Proceedings of the International Congress of Mathematicians (Z[ü]{}rich, 1994), Birkh[ä]{}user, 1995, pp. 120–139.
M. Kontsevich and Y. Soibelman, *Homological mirror symmetry and torus fibrations*, Symplectic geometry and mirror symmetry, World Scientific, 2001, pp. 203–263.
H. Krause, *The stable derived category of a [N]{}oetherian scheme*, Compos. Math. **141** (2005), 1128–1162.
F. Lalonde, D. McDuff, and L. Polterovich, *Topological rigidity of [H]{}amiltonian loops and quantum homology*, Invent. Math. **135** (1999), 369–385.
P. Lambrechts and D. Stanley, *The rational homotopy type of a blow-up in the stable case*, Geom. Topol. **12** (2008), 1921–1993.
K. Lefevre, *Sur les [$A_\infty$]{}-catégories*, Ph.D. thesis, Université Paris 7, 2002.
Y. Lekili and M. Lipyanskiy, *Geometric composition in quilted [F]{}loer theory*, Preprint arXiv:1003.4493, 2010.
Y. Lekili and T. Perutz, *Fukaya categories of the torus and [D]{}ehn surgery*, Proc. Nat. Acad. Sci., to appear, 2011.
E. Lerman, *Symplectic cuts*, Math. Res. Lett. **2** (1995), 247–258.
A.-M. Li and Y. Ruan, *Symplectic surgery and [G]{}romov-[W]{}itten invariants of [C]{}alabi-[Y]{}au 3-folds*, Invent. Math. **145** (2001), 151–218.
N. Markarian, *The [A]{}tiyah class, [H]{}ochschild cohomology and the [R]{}iemann–-[R]{}och theorem*, J. London Math. Soc. **79** (2009), 129––143.
S. Ma’u, *Quilted [F]{}loer modules*, Conference talk, recording available at <http://media.scgp.stonybrook.edu/video/video.php?f=20110518_3_qtp.mp4>, 2011.
S. Ma’u, K. Wehrheim, and C. Woodward, *A-infinity functors for [L]{}agrangian correspondences*, Manuscript available at <http://math.mit.edu/~katrin/papers/ainfty.pdf>.
D. Maulik and R. Pandharipande, *A topological view of [G]{}romov-[W]{}itten theory*, Topology **45** (2006), 887–918.
D. McDuff, *Examples of symplectic structures*, Invent. Math. **89** (1987), 13–36.
S. Mukai, *Duality between [$D(X)$]{} and [$D(\hat{X})$]{} with its application to [P]{}icard sheaves*, Nagoya J. Math. **81** (1981), 153–175.
D. Mumford, *Tata lectures on theta. [I]{}*, Progress in Mathematics, vol. 28, Birkhäuser, 1983.
Y.-G. Oh, *Floer cohomology, spectral sequences, and the [M]{}aslov class of [L]{}agrangian embeddings*, Int. Math. Res. Notices (1996), 305–346.
[to3em]{}, *Seidel’s long exact sequence on [C]{}alabi-[Y]{}au manifolds*, Preprint arXiv:1002.1648, 2010.
Y.-G. Oh and K. Fukaya, *Floer homology in symplectic geometry and in mirror symmetry*, International [C]{}ongress of [M]{}athematicians. [V]{}ol. [II]{}, Eur. Math. Soc., Zürich, 2006, pp. 879–905.
K. Ono, *Floer-[N]{}ovikov cohomology and the flux conjecture*, Geom. Funct. Anal. **16** (2006), 981–1020.
M. Papikian, *Rigid-analytic geometry and the uniformization of abelian varieties*, Snowbird lectures in algebraic geometry, Contemp. Math., vol. 388, Amer. Math. Soc., 2005, pp. 145–160.
T. Perutz, Talk at the [AIM]{} workshop [*Cyclic homology and symplectic topology*]{}, 2009.
S. Piunikhin, D. Salamon, and M. Schwarz, *Symplectic [F]{}loer-[D]{}onaldson theory and quantum cohomology*, Contact and symplectic geometry (C. B. Thomas, ed.), Cambridge Univ. Press, 1996, pp. 171–200.
A. Polishchuk, *[A]{}-infinity algebra of an elliptic curve and [E]{}isenstein series*, Preprint arXiv:0911.2814, 2009.
A. Polishchuk and E. Zaslow, *Categorical mirror symmetry: the elliptic curve*, Adv. Theor. Math. Phys. **2** (1998), 443–470.
M. Po[ź]{}niak, *Floer homology, [N]{}ovikov rings and clean intersections*, Northern California Symplectic Geometry Seminar, Amer. Math. Soc., 1999, pp. 119–181.
S. Priddy, *Koszul resolutions*, Transactions of the Amer. Math. Soc. **152** (1970), 39–60.
M. Schwarz, *A quantum cup-length estimate for symplectic fixed points*, Invent. Math. **133** (1998), 353–397.
P. Seidel, *Floer homology and the symplectic isotopy problem*, Ph.D. thesis, Oxford Univ., 1997.
[to3em]{}, *Graded [L]{}agrangian submanifolds*, Bull. Soc. Math. France **128** (2000), 103–146.
[to3em]{}, *Homological mirror symmetry for the quartic surface*, Preprint math.SG/0310414, 2003.
[to3em]{}, *[F]{}ukaya categories and [P]{}icard-[L]{}efschetz theory*, European Math. Soc., 2008.
[to3em]{}, *Lectures on four-dimensional [D]{}ehn twists*, Symplectic 4-manifolds and algebraic surfaces, Lecture Notes in Math., vol. 1938, Springer, 2008, pp. 231–267.
[to3em]{}, *Homological mirror symmetry for the genus two curve*, J. Algebraic Geom. **20** (2011), 727–769.
[to3em]{}, *Braids and symplectic four-manifolds with abelian fundamental group*, Turkish J. Math. **26** (Gököva Geometry and Topology Conference Proceedings Special Volume, 2002), 93–100.
P. Seidel and R. Thomas, *Braid group actions on derived categories of coherent sheaves*, Duke Math. J. **108** (2001), 37–108.
N. Sheridan, *On the homological mirror symmetry conjecture for pairs of pants*, Journal of Diff. Geom., to appear.
I. Smith, *Floer cohomology and pencils of quadrics*, Preprint arXiv:1006.1099, 2010.
J. Solomon and P. Seidel, *Symplectic cohomology and $q$-intersection numbers*, Preprint arXiv:1005.5156, 2010.
N. Spaltenstein, *Resolutions of unbounded complexes*, Compositio Math. **65** (1988), 121–154.
C. Taubes, *The [S]{}eiberg-[W]{}itten invariants and symplectic forms*, Math. Research Letters **1** (1994), 809–822.
B. To[ë]{}n and M. Vaqui[é]{}, *Moduli of objects in dg-categories*, Ann. Sci. École Norm. Sup. **40** (2007), 387–444.
K. Wehrheim and C. Woodward, *Orientations for pseudoholomorphic quilts*, Manuscript, available from <http://www-math.mit.edu/~katrin/orient.pdf>.
[to3em]{}, *Floer cohomology and geometric composition of [L]{}agrangian correspondences*, Preprint arXiv:0905.1368, 2009.
[to3em]{}, *Quilted [F]{}loer cohomology*, Geom. Topol. **14** (2010), 833–902.
[to3em]{}, *Functoriality for [L]{}agrangian correspondences in [F]{}loer theory*, Quantum Topol. **1** (2010), 129–170.
E. Whittaker and G. Watson, *A course of modern analysis*, 4th ed., Cambridge Univ. Press, 1927.
A. Yekutieli, *Dualizing complexes, [M]{}orita equivalence and the derived [P]{}icard group of a ring*, J. London Math. Soc. (2) **60** (1999), 723–746.
[to3em]{}, *The derived [P]{}icard group is a locally algebraic group*, Algebr. Represent. Theory **7** (2004), 53–57.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $A=\oplus_{i\in \mathbb{Z}}A_i$ be a finite dimensional graded symmetric cellular algebra with a homogeneous symmetrizing trace of degree $d$. We prove that $A_{-d}$ contains the Higman ideal $H(A)$ of the center of $A$ and $\dim H(A)\leq \dim A_{0}$ if $d\neq 0$, and provide a semisimplicity criterion of $A$ in terms of the centralizer of $A_0$, which is a graded version of [@L Theorem 3.2].'
address:
- 'Li: School of Mathematics and Statistics, Northeastern University at Qinhuangdao, Qinhuangdao, 066004, China'
- 'Zhao: School of Applied Mathematics, Beijing Normal University at Zhuhai, Zhuhai, 519087, China'
author:
- Yanbo Li and Deke Zhao
title: Notes on graded symmetric cellular algebras
---
[^1]
Introduction
============
Cellular algebras were introduced by Graham and Lehrer [@GL] in 1996 motivated by the work of Kazhdan and Lusztig [@KL]. It provides a systematic framework for studying the representation theory of many interesting and important algebras coming from mathematics and physics, such as Schur algebras, Temperley-Lieb algebras, Brauer algebras [@GL], partition algebras [@Xi1], Birman-Wenzl algebras [@Xi2], Hecke algebras of finite types [@G], and so on.
The ($\mathbb{Z}$-)gradings is a subtle structure on a finite dimensional algebra, which plays an important role in Lie theory and the representation theory (ref. [@BGS; @CPS; @F-Zimmer; @R]). Motivated by the works of Brundan, Kleshchev (and Wang) [@BK; @BK2; @BKW], Hu and Mathas [@HM] introduced graded cellular algebras, which include the Khovanov diagram algebras and their quasi-hereditary covers [@BS; @BS2], the level two degenerate cyclotomic Hecke algebras [@AST], graded walled Brauer algebras [@BS3], Temperley-Lieb algebras of type $A$ and $B$ [@PR], etc (see references in [@HM2; @HM3]).
The Auslander-Reiten conjecture [@ARS] claims that if the stable categories of two Artin algebras are equivalent then they have the same number of non-projective simple modules up to isomorphism. Recently, Liu, Zhou and Zimmermann’s work [@LZZ] indicates that the projective center is the main obstruction to attack the Auslander-Reiten conjecture, which is exactly the Higman ideal for a symmetric algebra. It is natural and interesting to investigate the Higman ideal of the center of a symmetric algebra.
The aim of this note is to study the Higman ideal of a graded symmetric cellular algebra by applying the dual basis method which has been used in [@L; @L1; @L2; @L3; @LX; @LZ]. More precisely, assume that $A=\oplus_{i\in \mathbb{Z}}A_i$ is a finite dimensional graded symmetric cellular algebra over a field $K$ with a homogeneous symmetrizing trace $\tau$ of degree $d$. Our first main result claims that the Higman ideal of the center $\mathcal{Z}(A)$ of $A$ is contained in $A_{-d}$, and $\dim H(A)\leq \dim A_{0}$ whenever $d\neq 0$ (see Theorem \[3.4\]). Secondly, we provide a semisimplicity criterion for finite dimensional graded symmetric cellular algebras in terms of the centralizer of $A_0$ (see Theorem \[Them:semisimple\]), which is a graded version of [@L Theorem 3.2].
The note is organized as follows. In Section 2, we review quickly some known results on symmetric algebras, graded algebras and cellular algebras. In Section 3, we study the Higman ideal of finite dimensional graded symmetric cellular algebras and prove the first main result. Section 4 devotes to investigate the centralizer of $A_0$ and gives a graded version of [@L Thoerem 3.2]. The semisimplicity criterion for finite dimensional graded symmetric cellular algebras is provided in the last section.
Preliminaries
=============
In this section, we recall some facts on symmetric algebras, graded algebras and cellular algebras, and introduce the graded symmetric algebras.
Symmetric algebras
------------------
Let $K$ be a field and let $A$ be a finite dimensional $K$-algebra. Recall that a bilinear form $f: A\times A\rightarrow K$ is [*non-degenerate*]{} if the determinant of the matrix $(f(x_{i},\,x_{j}))_{x_{i},\,\, x_{j}\in \mathscr{B}}$ is invertible for some basis $\mathscr{B}=\{x_1, \cdots, x_n\}$ of $A$ and $f$ is [*associative*]{} if $f(ab,c)=f(a,bc)$ for all $a, b, c\in A$. We say that $A$ is a [*symmetric algebra*]{} if there is a non-degenerate associative bilinear form $f$ on $A$ such that $f(x,y)=f(y,x)$ for all $x, y\in A$. Note that if $A$ is a symmetric algebra, then we can define a linear map $\tau$ by $$\tau: A\rightarrow K\quad\quad a\mapsto
f(a,1),$$ which is called the [*symmetrizing trace*]{} of $A$ induced by $f$.
Now let $A$ be a finite dimensional symmetric algebra with a basis $\mathscr{B}=\{x_1, \cdots, x_n\}$ and denote by $\mathscr{D}=\{y_1, \cdots, y_n\}$ the [*dual basis*]{} of $\mathscr{B}$, that is, $\mathscr{D}$ is a basis of $A$ satisfying $\tau(x_iy_j)=\delta_{ij}$ for all $i, j=1,\ldots,n$. Then the [*Higman ideal*]{} $H(A)$ of the center $\mathcal{Z}(A)$ of $A$ is $$H(A):=\biggl\{\sum_ix_iay_i\biggl| \,a\in A\biggr.\biggr\},$$ which is independent of the choice of $\tau$ and the basis of $A$.
For arbitrary $1\leq i,j \leq n$, write $x_{i}x_{j}=\sum_{k}r_{ijk}x_{k},$ where $r_{ijk}\in K$. The first named author proved the following lemma.
\[2.0\] Let $A$ be a symmetric $K$-algebra. Then the following hold: $$x_{i}y_{j}=\sum_{k}r_{kij}y_{k};\,\,\,\,\,y_{i}x_{j}=\sum_{k}r_{jki}y_{k}.$$
Graded symmetric algebras
-------------------------
By a [*graded space*]{} we mean a $\mathbb{Z}$-graded $K$-space $V$, namely a $K$-space with a decomposition into subspaces $V=\bigoplus_{i\in \mathbb{Z}}V_i$. A nonzero element $v$ of $V_i$ is said to be a [*homogeneous element*]{} of degree $i$ and denoted by ${\rm deg}(v)=i$. We will view the field $K$ as a graded space concentrated in degree $0$. Given two graded spaces $V$ and $W$, the $K$-space $\mathrm{Hom}_K(V,W)$ of all $K$-linear maps from $V$ to $W$ is a graded space with $\mathrm{Hom}_K(V,W)_i$ consisting of all the $K$-linear maps $\alpha: V\rightarrow W$ such that $\alpha(V_j)\subseteq W_{j+i}$ for all $i, j\in \mathbb{Z}$. Nonzero element of $\mathrm{Hom}_K(V,W)_i$ will be called a *homogeneous map* of degree $i$.
By a *graded algebra* $A$ we always mean a finite dimensional $\mathbb{Z}$-graded associative $K$-algebra with identity, that is, $A$ is a graded space $A=\bigoplus_{i\in \mathbb{Z}}A_i$ such that $A_iA_j\subseteq A_{i+j}$ for all $i, j\in \mathbb{Z}$.
\[2.1\] A graded algebra $A$ is said to be a **graded symmetric algebra** if there is a homogeneous symmetrizing trace $\tau: A\rightarrow K$ of degree $d$ for some $d\in \mathbb{Z}$.
Cellular algebras
-----------------
Now we recall the definitions of cellular algebras, Gram matrices and cell modules.
\[2.2\] Let $R$ be a commutative ring with identity. An associative unital $R$-algebra is called a **cellular algebra** with cell datum $(\Lambda, M, C, \ast)$ if the following conditions are satisfied:
1. The finite set $\Lambda$ is a poset. Associated with each ${\lambda}\in\Lambda$, there is a finite set $M({\lambda})$. The algebra $A$ has an $R$-basis $\{C_{S,T}^{\lambda}\mid S,T\in
M({\lambda}),{\lambda}\in\Lambda\}$.
2. The map $\ast$ is an $R$-linear anti-automorphism of $A$ such that $(C_{S,T}^{\lambda})^{\ast}= C_{T,S}^{\lambda}$ for all ${\lambda}\in\Lambda$ and $S,T\in M({\lambda})$.
3. If ${\lambda}\in\Lambda$ and $S,T\in M({\lambda})$, then for any element $a\in A$, we have $$aC_{S,T}^{\lambda}\equiv\sum_{S^{'}\in
M({\lambda})}r_{a}(S',S)C_{S^{'},T}^{{\lambda}} \,\,\,\,(\rm {mod}\,\,\,
A(<{\lambda})),$$ where $r_{a}(S^{'},S)\in R$ is independent of $T$ and $A(<{\lambda})$ is the $R$-submodule of $A$ generated by $\{C_{U,V}^\mu \mid U,\,\,V\in M(\mu),\mu<{\lambda}\}$.
Let ${\lambda}\in\Lambda$. For arbitrary elements $S,T,U,V\in M({\lambda})$, Definition \[2.2\] implies that $$C_{S,T}^{\lambda}C_{U,V}^{\lambda}\equiv\Phi(T,U)C_{S,V}^{\lambda}\,\,\,\, (\rm mod\,\,\, A(<{\lambda})),$$ where $\Phi(T,U)\in R$ depends only on $T$ and $U$. It is easy to check that $\Phi(T,U)=\Phi(U,T)$ for arbitrary $T,U\in M({\lambda})$.
For each ${\lambda}\in \Lambda$, fix an order on $M({\lambda})$. The associated [*Gram matrix*]{} $G({\lambda})$ is the following symmetric matrix $$G({\lambda})=(\Phi(S, T))_{S,T\in M({\lambda})}.$$ Note that the determinant of $G({\lambda})$ is independent of the choice of the order on $M({\lambda})$.
Given a cellular algebra $A$, we note that $A$ has a family of modules defined by its cellular structure.
\[2.3\] Let $A$ be a cellular algebra with cell datum $(\Lambda, M, C,
\ast)$. For each ${\lambda}\in\Lambda$, the *cell module* $W({\lambda})$ is a free $R$-module with basis $\{C_{S}\mid S\in M({\lambda})\}$ and the left $A$-action defined by $$aC_{S}=\sum_{S^{'}\in M({\lambda})}r_{a}(S^{'},S)C_{S^{'}}
\,\,\,\,(a\in A,\,\,S\in M({\lambda})),$$ where $r_{a}(S^{'},S)$ is the element of $R$ defined in Definition \[2.2\](GC3).
Symmetric cellular algebras
---------------------------
Let $A$ be a symmetric cellular algebra with cell datum $(\Lambda, M, C, \ast)$. Fix a symmetrizing trace $\tau$ and denote the dual basis by $$D=\{D_{S,T}^{\lambda}\mid S,T\in M({\lambda}),{\lambda}\in\Lambda\},$$ which is determined by $$\tau(C_{S,T}^{{\lambda}}D_{U,V}^{\mu})=\delta_{{\lambda}\mu}\delta_{SV}\delta_{TU}.$$
Set $e_{{\lambda}}=\sum_{S\in M({\lambda})}C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}}$. The first named author [@L] introduced the following ideal $L(A)$ of $\mathcal{Z}(A)$ $$\begin{aligned}
L(A)&:=& \biggl\{\sum_{{\lambda}\in\Lambda} r_{{\lambda}}e_{{\lambda}}\biggl|\, r_{{\lambda}}\in K\biggr.\biggr\},\end{aligned}$$ and proved that $H(A)\subseteq L(A)$.
For any ${\lambda}, \mu\in \Lambda$, $S,T\in M({\lambda})$, $U,V\in M(\mu)$, write $$C_{S,T}^{{\lambda}}C_{U,V}^{\mu}=\sum\limits_{\epsilon\in\Lambda,X,Y\in M(\epsilon)}
r_{(S,T,{\lambda}),(U,V,\mu),(X,Y,\epsilon)}C_{X,Y}^{\epsilon}.$$ The following lemma is important to our purpose.
\[2.4\] Let $A$ be a symmetric cellular algebra with a cell datum $(\Lambda,
M, C, \ast)$ and $\tau$ a given symmetrizing trace. For arbitrary ${\lambda},\mu\in\Lambda$ and $S,T,P,Q\in M({\lambda})$, $U,V\in M(\mu)$, the following hold:
1. $D_{U,V}^{\mu}C_{S,T}^{{\lambda}}=\sum\limits_{\epsilon\in
\Lambda, X,Y\in
M(\epsilon)}r_{(S,T,{\lambda}),(Y,X,\epsilon),(V,U,\mu)}D_{X,Y}^{\epsilon}.$
2. $C_{S,T}^{{\lambda}}D_{U,V}^{\mu}=\sum\limits_{\epsilon\in
\Lambda, X,Y\in
M(\epsilon)}r_{(Y,X,\epsilon),(S,T,{\lambda}),(V,U,\mu)}D_{X,Y}^{\epsilon}.$
3. $C_{S,T}^{{\lambda}}D_{T,Q}^{{\lambda}}=C_{S,P}^{{\lambda}}D_{P,Q}^{{\lambda}}.$
4. $D_{T,S}^{{\lambda}}C_{S,Q}^{{\lambda}}=D_{T,P}^{{\lambda}}C_{P,Q}^{{\lambda}}.$
5. $C_{S,T}^{{\lambda}}D_{P,Q}^{{\lambda}}=0 \text{ if } T\neq P.$
6. $D_{P,Q}^{{\lambda}}C_{S,T}^{{\lambda}}=0 \text{ if } Q\neq S.$
7. $C_{S,T}^{{\lambda}}D_{U,V}^{\mu}=0 \text{ if } \mu\nleq {\lambda}.$
8. $D_{U,V}^{\mu}C_{S,T}^{{\lambda}}=0 \text{ if } \mu\nleq{\lambda}.$
Denote by $G'({\lambda})$ the Gram matrices defined by the dual basis. The first named author [@L2 Lemma 3.6] showed $G({\lambda})G'({\lambda})=k_{{\lambda}}E$ for some $k_{{\lambda}}\in K$, where $E$ is the identity matrix.
The following facts on the constants $k_{{\lambda}}$, ${\lambda}\in \Lambda$ will be used later.
\[2.6\] Let $A$ be a symmetric cellular algebra. Then $(C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}})^2=k_{{\lambda}}C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}}$ for arbitrary ${\lambda}\in\Lambda$ and $S\in M({\lambda})$.
\[2.5\] Let $A$ be a symmetric cellular algebra with a cell datum $(\Lambda, M, C, \ast)$. Then for any ${\lambda}\in\Lambda$, the cell module $W({\lambda})$ is projective if and only if $k_{\lambda}\neq
0$.
\[2.8\] Let $A$ be a finite dimensional symmetric cellular algebra. Then the following statements are equivalent.
1. $A$ is semisimple.
2. $k_{{\lambda}}\neq 0$ for all ${\lambda}\in\Lambda$.
3. $\{C_{S,T}^{{\lambda}}D_{T,T}^{{\lambda}}\mid{\lambda}\in\Lambda, S,T\in M({\lambda})\}$ is a basis of $A$.
Homogeneous symmetrizing trace and Higman ideal
===============================================
Let $A$ be a cellular algebra. Following Hu and Mathas [@HM], we say that $A$ is a [*graded cellular algebra*]{} if $A$ is in addition a $\mathbb{Z}$-graded algebra satisfying the following condition:
1. Let ${\rm deg}:
\coprod_{{\lambda}\in\Lambda}M({\lambda})\rightarrow \mathbb{Z}$ be a function. Each basis element $C_{S,T}^{\lambda}$ is homogeneous of degree ${\rm deg}(C_{S,T}^{\lambda})= {\rm deg}(S)+{\rm deg}(T)$.
A graded cellular algebra $A$ is called a [*graded symmetric cellular algebra*]{} if $A$ is in addition a graded symmetric algebra. Note that a finite dimensional semi-simple algebra is a graded symmetric cellular algebra with a non-trivial grading (see Section 5 for details), which is a generalization of [@HM Example 2.2].
The following is an example of non-semisimple graded symmetric cellular algebras.
\[3.1\] Let $K$ be a field and $A=K\oplus Kx$ with ${\rm deg}(x)=2$ and $x^2=0$. Then $A$ is a graded symmetric algebra with a non-degenerate trace form $\tau(1)=0$ and $\tau(x)=1$. It is a homogeneous trace form of degree $-2$. It is easy to check that the dual basis of $\{1,
x\}$ is $\{x, 1\}$. Thus the dual basis is homogeneous too.
The following easy verified fact is useful, which implies that the dual basis of the homogeneous basis of graded symmetric algebras with respect to a homogeneous symmetrizing trace must be homogeneous.
\[3.2\] Let $A$ be a graded symmetric algebra with homogeneous symmetrizing trace $\tau$ of degree $d$ and let $\mathscr{B}=\{x_i\mid i\in 1,\cdots, n\}$ be a homogeneous basis of $A$. Then the dual basis $\mathscr{D}=\{y_i\mid \tau(x_iy_j)=\delta_{ij},\,\, i, j\in 1,\cdots, n\}$ of $A$ is a homogeneous basis with $${\rm deg}(x_i)+{\rm deg}(y_i)=-d.$$
Combining Lemma \[3.2\] with [@L1 Theorem 2.5], we get the following
\[3.3\] Let $A$ be a graded symmetric cellular algebra with a homogeneous symmetrizing trace $\tau$ of degree $d$. Then the dual basis is a graded cellular basis if and only if $d$ is even and $\tau(a)=\tau(a^{\ast})$ for all $a\in A$.
$``\Rightarrow"$ It follows from [@L1 Theorem 2.5] that if the dual basis $\{D_{S,T}^{{\lambda}}\mid{\lambda}\in\Lambda, S,T\in M({\lambda})\}$ is cellular, then $\tau(a)=\tau(a^{\ast})$ for all $a\in A$. Since the dual basis is graded cellular, the condition ($GC_d$) implies that there is a function ${\operatorname{codeg}}:
\coprod_{{\lambda}\in\Lambda}M({\lambda})\rightarrow \mathbb{Z}$ such that $\deg (D_{S,S}^{{\lambda}})=2\operatorname{codeg}(S)$. Thanks to Lemma \[3.2\], $d=-2(\deg(S)+\operatorname{codeg}(S))$ is even.
$``\Leftarrow"$ Suppose $\tau(a)=\tau(a^{\ast})$ for all $a\in A$. Then [@G2] or [@L1 Theorem 2.5], and Lemma \[3.2\] imply that the dual basis is cellular and homogeneous. Now we define $${\operatorname{codeg}}: \coprod_{{\lambda}\in\Lambda}M({\lambda})\rightarrow \mathbb{Z},\qquad\quad\operatorname{codeg}(S)=-\deg(S)-\frac{d}{2}.$$ Applying Lemma \[3.2\], $\deg(D_{S,T}^{{\lambda}})=-d-\deg(S)-\deg(T)=\operatorname{codeg}(S)+\operatorname{codeg}(T)$. This completes the proof.
Now we are in a place to give the main result of this section.
\[3.4\] Let $A$ be a finite dimensional graded symmetric cellular algebra with a homogeneous symmetrizing trace $\tau$ of degree $d\neq 0$. Then
1. If $\rho$ is a homogeneous symmetrizing trace of degree $d'$, then $d=d'$.
2. All cell modules are non-projective.
3. $H(A)\subseteq L(A)\subseteq A_{-d}$ and $\dim H(A)\leq \dim A_{0}$.
\(1) Let $\{D_{S,T}^{{\lambda}}\mid {\lambda}\in\Lambda,\,\, S, T\in M({\lambda})\}$ and $\{d_{S,T}^{{\lambda}}\mid {\lambda}\in\Lambda,\,\, S, T\in M({\lambda})\}$ be dual bases determined by $\tau$ and $\rho$, respectively. It follows from Lemma \[3.2\] that $\deg(C_{S, S}^{{\lambda}}D_{S, S}^{{\lambda}})=-d$ and $\deg(C_{S, S}^{{\lambda}}d_{S, S}^{{\lambda}})=-d'$ for arbitrary $S\in M({\lambda})$. According to [@L2 Lemma 2.3], $$d_{S, S}^{{\lambda}}=\sum\limits_{\varepsilon\in\Lambda, X, Y\in M(\varepsilon)}\tau(C_{X,Y}^{\varepsilon}d_{S, S}^{{\lambda}})D_{Y,X}^{\varepsilon}.$$ Therefore $$\begin{aligned}
C_{S, S}^{{\lambda}}d_{S, S}^{{\lambda}} &=& \sum\limits_{\varepsilon\in\Lambda, X, Y\in M(\varepsilon)}\tau(C_{X,Y}^{\varepsilon}d_{S, S}^{{\lambda}})C_{S, S}^{{\lambda}}D_{Y,X}^{\varepsilon} \\
&=& \sum\limits_{ X\in M({\lambda})}\tau(C_{X,S}^{{\lambda}}d_{S, S}^{{\lambda}})C_{S, S}^{{\lambda}}D_{S,X}^{{\lambda}}\\
&=& \sum\limits_{ X\in M({\lambda})}\tau(d_{S, S}^{{\lambda}}C_{X,S}^{{\lambda}})C_{S, S}^{{\lambda}}D_{S,X}^{{\lambda}}\\
&=&\tau(C_{S,S}^{{\lambda}}d_{S, S}^{{\lambda}})C_{S, S}^{{\lambda}}D_{S,S}^{{\lambda}},\end{aligned}$$ where the second and the last equality follow by applying Lemma \[2.4\]. Clearly, $\tau(C_{S,S}^{{\lambda}}d_{S, S}^{{\lambda}})\neq 0$ and this forces $d=d'$.
\(2) Suppose that $W({\lambda})$ is a projective cell module. Then Lemma \[2.5\] implies $k_{{\lambda}}\neq 0$. Thus $k_{{\lambda}}^{-1}C_{S,S}^{\lambda}D_{S,S}^{\lambda}$ is an idempotent of $A$ for arbitrary $S\in M({\lambda})$ due to Lemma \[2.6\]. This forces ${\deg}(C_{S,S}^{\lambda})+{\deg}(D_{S,S}^{\lambda})=0$, while Lemma \[3.2\] shows ${\deg}(C_{S,S}^{\lambda})+{\deg}(D_{S,S}^{\lambda})=-d\neq 0$. It is a contradiction and we complete the proof.
\(3) According to Lemma \[3.2\], we have $L(A)\subseteq A_{-d}$. Note that Li [@L] proved that $H(A)\subseteq L(A)$. So $H(A)\subseteq A_{-d}$ and we only need to prove $\dim H(A)\leq \dim A_{0}$. In fact, For each $C_{X,Y}^\varepsilon$ with $\deg(C_{X,Y}^\varepsilon)\neq 0$, it follows from $H(A)\subseteq A_{-d}$ and Lemma \[3.2\] that $$\sum_{{\lambda}\in\Lambda, \,\,S,T\in M({\lambda})} C_{S, T}^{\lambda}C_{X,Y}^\varepsilon D_{T, S}^{\lambda}=0.$$ This implies that $H(A)$ is a $K$-span of $$\biggl\{\sum_{{\lambda}\in\Lambda, \,\,S,T\in M({\lambda})} C_{S, T}^{\lambda}C_{X,Y}^\varepsilon D_{T, S}^{\lambda}\mid C_{X,Y}^\varepsilon\in A_0\biggr\}$$ and consequently, $\dim H(A)\leq \dim A_{0}$.
\[3.5\] Hu and Mathas [@HM] proved that the blocks $\mathscr{H}_{\beta}^{\Lambda}$ of the cyclotomic Hecke algebras of type $G(m, 1, n)$ are graded symmetric cellular algebras with homogeneous trace form of degree $-2{\rm def}\beta$. Now Theorem \[3.4\] implies that the degree $-2{\rm def}\beta$ is the only one which makes $\mathscr{H}_{\beta}^{\Lambda}$ to be graded symmetric cellular and the (ungraded) cell modules of $\mathscr{H}_{\beta}^{\Lambda}$ are non-projective when ${\rm def}\beta\neq 0$.
Let us remark that the equality in Theorem \[3.4\](3) may be held. The following is an example given in [@L], which is a graded symmetric cellular algebra.
\[3.6\] Let $K$ be a field with ${\rm Char}K\nmid n+1$ and let $Q$ be the following quiver $$\xymatrix@C=13mm{
\bullet \ar@<2.5pt>[r]^{\alpha_1} & \bullet \ar@<2.5pt>[r]^(0.4){\alpha_2}
\ar@<2.5pt>@[r][l]^(1){1}^{\alpha_1'}^(0){2}
&\bullet\ar@<2.5pt>@[r][l]^(0.20){3}^(0.6){\alpha_2'}\cdots
\bullet\ar@<2.5pt>[r]^(0.6){\alpha_{n-1}} & \bullet\ar@<2.5pt>@[r][l]^(0.35){\alpha_{n-1}'}^(0.75){n-1}^(0){n}\\
}$$ with relation $\rho$ given as follows:
1. all paths of length $\geq 3$;
2. $\alpha'_{i}\alpha_{i}-\alpha_{i+1}\alpha'_{i+1}$, $i=1,\cdots, n-2$;
3. $\alpha_{i}\alpha_{i+1}$, $\alpha'_{i+1}\alpha'_{i}$, $i=1,\cdots, n-2$.
Then $A=K(Q,\rho)$ is a graded algebra in a natural way. Now we define the homogeneous symmetrizing trace $\tau$ of $A$ by
1. $\tau(e_{1})=\cdots=\tau(e_{n})=0$;
2. $\tau(\alpha_{i})=\tau(\alpha'_{i})=0$, $i=1,\cdots, n-1$;
3. $\tau(\alpha_{i}\alpha'_{i})=\tau(\alpha'_{i}\alpha_{i})=1$, $i=1,\cdots, n-1$.
Then the degree of $\tau$ is $-2$, $\tau(a)=\tau(a^\ast)$ for all $a\in A$, and $A_0=\{\sum k_ie_i\mid k_i\in K\}$. Furthermore, $A$ is a graded symmetric cellular algebra with a homogeneous cellular basis $\{C_{i,j}^k\mid 1\leq k\leq n+1, 1\leq i,j\leq 2\}$ as follows: $$\begin{matrix}
\begin{matrix} e_{1}\end{matrix} ;&
\begin{matrix} \alpha_{1}\alpha_{1}^{'} & \alpha_{1}\\ \alpha_{1}^{'} &
e_{2}\end{matrix} ; &
\begin{matrix} \alpha_{2}\alpha_{2}^{'} & \alpha_{2}\\ \alpha_{2}^{'} &
e_{3}\end{matrix} ;& \cdots ;&
\begin{matrix} \alpha_{n-1}\alpha_{n-1}^{'} & \alpha_{n-1}\\ \alpha_{n-1}^{'} &
e_{n}\end{matrix} ;&
\begin{matrix} \alpha_{n-1}^{'}\alpha_{n-1}\end{matrix},
\end{matrix}$$ and the Higman ideal $H(A)$ is generated by $$\begin{aligned}
&\left\{2\alpha_{1}\alpha_{1}^{'}+\alpha_{2}\alpha_{2}^{'},
\alpha_{1}\alpha_{1}^{'}+2\alpha_{2}\alpha_{2}^{'}+\alpha_{3}\alpha_{3}^{'},
\alpha_{2}\alpha_{2}^{'}+2\alpha_{3}\alpha_{3}^{'}+\alpha_{4}\alpha_{4}^{'},
\cdots,\right.\\
&\left.\qquad\alpha_{n-3}\alpha_{n-3}^{'}+2\alpha_{n-2}\alpha_{n-2}^{'}+\alpha_{n-1}\alpha_{n-1}^{'},
\alpha_{n-2}\alpha_{n-2}^{'}+2\alpha_{n-1}\alpha_{n-1}^{'}\right\}.\end{aligned}$$ It is easy to check that $H(A)\subset A_2$ and $\dim H(A)=\dim A_0=n$. Furthermore, a direct computation yields that $C_{i,i}^kD_{i,i}^k\in\mathcal {Z}_A(A_0)$ for $i=1,2$ and $k=1, \ldots, n+1$. However, these elements can not form the whole $\mathcal {Z}_A(A_0)$.
Centralizer of $A_0$
====================
This section devotes to give a graded version of [@L Theorem 3.2]. Throughout this section $A$ is a finite dimensional graded symmetric cellular $K$-algebra with a homogeneous symmetrizing trace $\tau$ of degree $d$.
For any integer $c$, it is interesting to consider the following subset $H_c(A)$ of the Higman ideal $H(A)$ of $A$ $$H_c(A):=\biggl\{\sum\limits_{{\rm deg}(x_i)=c} x_iay_i\mid a\in A\biggr\}.$$
\[4.1\]Let $H_{\rm gr}(A)$ be the $K$-span of all $H_c(A)$ and $\mathcal {Z}_A(A_0)$ the centralizer of $A_0$ in $A$. Then $H_{\rm gr}(A)\subseteq\mathcal {Z}_A(A_0)$.
Clearly, we only need to prove $H_c(A)\subseteq\mathcal {Z}_A(A_0)$ for each integer $c$. Assume that ${\rm deg}(x_j)=0$. Thanks to Lemma \[2.0\], $$\sum\limits_{{\rm deg}(x_i)=c} x_jx_iay_i=\sum\limits_{{\rm deg}(x_i)=c}\sum\limits_kr_{jik}x_kay_i,$$ where $r_{jik}=0$ when ${\rm deg}(x_k)\neq c$. This implies that $$\sum\limits_{{\rm deg}(x_i)=c} x_jx_iay_i=\sum\limits_{\substack{i,\,k\\{\rm deg}(x_i)={\rm deg}(x_k)=c}}r_{jik}x_kay_i. \eqno(*)$$ While Lemma\[2.0\] implies $$\sum\limits_{{\rm deg}(x_i)=c} x_iay_ix_j=\sum\limits_{{\rm deg}(x_i)=c}\sum\limits_k r_{jki}x_iay_k,$$ where $r_{jki}=0$ if ${\rm deg}(x_k)\neq c$. Thus $$\sum\limits_{{\rm deg}(x_i)=c} x_iay_ix_j=\sum\limits_{\substack{i,\,k\\{\rm deg}(x_i)={\rm deg}(x_k)=c}}r_{jki}x_iay_k. \eqno(**)$$ Comparing the equalities ($*$) and ($**$), we complete the proof.
Now we consider the following elements $$e_{{\lambda}, \,c}:=\sum\limits_{\mathrm{deg}(S)=c}C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}}.$$
Applying Lemmas \[2.4\] and \[2.6\], we get the following lemma.
\[4.2’\]Keep notations as above. Then $e_{{\lambda}, \,c}e_{\mu, \,c'}=\delta_{{\lambda}\mu}\delta_{cc'}k_{{\lambda}}e_{{\lambda}, \,c}.$
Define $$L_{\rm gr}(A):=\biggl\{\sum\limits_{{\lambda}\in\Lambda, \,c \,\in\,{\mathbb Z}} r_{{\lambda}, \,c}e_{{\lambda}, \,c}\mid r_{{\lambda}, \,c}\in K\biggr\}.$$ Using the similar argument as [@L Propositition 3.3 (1)], we can show that $L_{\rm gr}(A)$ is independent of the choice of $\tau$.
The following fact will be used later.
\[4.2\]Keep notations as above. Then $L_{\rm gr}(A)\subseteq \mathcal {Z}_A(A_0)$.
Clearly, we only need to prove $e_{{\lambda}, \,c}\in \mathcal {Z}_A(A_0).$ Let $C_{U,V}^{\mu}$ be a basis element of degree 0. Then by Lemma \[2.4\], $$\begin{aligned}
\sum_{{\rm deg}(S)=c}C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}}C_{U,V}^{\mu}& =&\sum_{{\rm deg}(S)=c}\sum_{\substack{\epsilon\in \Lambda,\\ X,Y\in
M(\epsilon)}}r_{(U,V,\mu),(X,Y,\epsilon),(S,S,{\lambda})}C_{S,S}^{{\lambda}}D_{Y,X}^{\epsilon}\\
&=&\sum_{{\rm deg}(S)=c}\sum_{X\in
M({\lambda})}r_{(U,V,\mu),(X,S,{\lambda}),(S,S,{\lambda})}C_{S,S}^{{\lambda}}D_{S,X}^{{\lambda}}\\&=&\sum_{{\rm deg}(S)={\rm
deg}(X)=c}r_{(U,V,\mu),(X,S,{\lambda}),(S,S,{\lambda})}C_{S,S}^{{\lambda}}D_{S,X}^{{\lambda}},\end{aligned}$$ where the second equality follows from Definition \[2.3\] and Lemma \[2.4\](7), the last one follows by comparing the degree of both sides.
On the other hand, we have $$\begin{aligned}
C_{U,V}^{\mu}\sum_{
\mathrm{deg}(S)=c}C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}}&=\sum_{
\mathrm{deg}(S)=c}\sum_{\substack{\epsilon\in \Lambda, \\X,Y\in
M(\epsilon)}}r_{(U,V,\mu),(S,S,{\lambda}),(X,Y,\epsilon)}C_{X,Y}^{\epsilon}D_{S,S}^{{\lambda}}\\
&=\sum_{\mathrm{deg}(S)=c}\sum_{X\in
M({\lambda})}r_{(U,V,\mu),(S,S,{\lambda}),(X,S,{\lambda})}C_{X,S}^{{\lambda}}D_{S,S}^{{\lambda}}\\
&=\sum_{\mathrm{deg}(S)=\mathrm{deg}(X)=c}r_{(U,V,\mu),(S,S,{\lambda}),(X,S,{\lambda})}C_{X,S}^{{\lambda}}D_{S,S}^{{\lambda}},\end{aligned}$$ where the second equality and the third one follow form Lemma \[2.4\](5, 7). Thus $e_{{\lambda},c}\in
\mathcal {Z}_A(A_0)$ as required.
Keep notations as above. If $d\neq 0$ then $L_{\rm gr}(A)\subsetneqq \mathcal {Z}_A(A_0)$.
Note that $L_{\rm gr}(A)\subseteq A_{-d}$. Hence $\mathcal {Z}(A_0)$ is not contained in $L_{\rm gr}(A)$.
The relationship between $H_{\rm gr}(A)$ and $L_{\rm gr}(A)$ is given by the following lemma, which can be proved by the similar argument of [@L Theorem 3.2].
\[4.3\]Keep notations as above. Then $H_{\rm gr}(A)\subseteq L_{\rm gr}(A)$.
Combining Lemmas \[4.2\] and \[4.3\] yields the main result of this section.
\[4.4\]Keep notations as above. Then $H_{\rm gr}(A)\subseteq L_{\rm gr}(A)\subseteq \mathcal {Z}_A(A_0)$.
Semisimple case
===============
In this section, we give a semisimlicity criterion for a graded symmetric cellular algebra, that is, $A$ is semisimple if and only if $L_{\rm gr}(A)=\mathcal {Z}_A(A_0)$.
Now let $S_n$ be the symmetric group on $n$ letters and $A=M_n(K)$. For $\sigma_1, \sigma_2\in S_n$, we set $e_{ij}=C_{\sigma_1(i)\sigma_2(j)}$, $1\leq i, j\leq n$, where $e_{ij}$ is the $n\times n$ matrix with only one non-zero $(i,j)$-entry 1.
\[5.1\] Keep notations as the above and define $\sigma=\sigma_1\sigma_2^{-1}$. Let $``\rm deg"$ be a function from $\{1,2,\cdots, n\}$ to $\mathbb{Z}$. Then the basis $\{C_{i,j}\mid 1\leq i,j\leq n\}$ is graded cellular if and only if
1. $\sigma^2=\mathrm{id}$;
2. ${\rm deg}(i)=-{\rm deg}(\sigma(i))$ for $1\leq i\leq n$.
$``\Longrightarrow"$ For all $1\leq j\leq n$, the cellularity of $C_{i,j}$ shows $$\begin{aligned}
e_{ij}e_{jk}&=&C_{\sigma_1(i)\sigma_2(j)}C_{\sigma_1(j)\sigma_2(k)}\\
&=&C_{\sigma_1(i)\sigma_1(j)}C_{\sigma_2(j)\sigma_2(k)}=e_{i, \,\sigma_2^{-1}\sigma_1(j)}e_{\sigma_1^{-1}\sigma_2(j),\,k},\end{aligned}$$ which implies $\sigma_2^{-1}\sigma_1(j)=\sigma_1^{-1}\sigma_2(j)$ for $1\leq j\leq n$, i.e., $(\sigma_1^{-1}\sigma_2)^2=\mathrm{id}$. While the degree of $\sigma_1^{-1}\sigma_2$ equals to that of $\sigma=\sigma_2\sigma_1^{-1}$. Hence $\sigma^2=id$.
Since $e_{ii}$ is an idempotent of $A$ for all $i$, $C_{\sigma_1(i), \,\sigma_2(i)}$ is also an idempotent. Thus ${\rm deg}(\sigma_1(i))=-{\rm deg}(\sigma_2(i))$ for all $1\leq i\leq n$ and $${\rm deg}(i)={\rm deg}(\sigma_1(\sigma_1^{-1}(i)))=-{\rm deg}(\sigma_2(\sigma_1^{-1}(i)))=-{\rm deg}(\sigma(i)).$$
$``\Longleftarrow"$ Firstly we prove (GC2). We need to check that the linear map $\ast$ sending $C_{ij}$ to $C_{ji}$ is an anti-morphism of $A$. Note that $$e_{ij}^\ast=C_{\sigma_1(i),\,\sigma_2(j)}^\ast=C_{\sigma_2(j),\,\sigma_1(i)}=e_{\sigma_1^{-1}\sigma_2(j), \,\sigma_2^{-1}\sigma_1(i)},$$ which implies $$(e_{ij}e_{kl})^{\ast}=\delta_{jk}e_{il}^\ast=\delta_{jk}e_{\sigma_1^{-1}\sigma_2(l), \,\sigma_2^{-1}\sigma_1(i)}.$$ While $$\begin{aligned}
e_{kl}^{\ast}e_{ij}^{\ast}&=&e_{\sigma_1^{-1}\sigma_2(l), \,\sigma_2^{-1}\sigma_1(k)}e_{\sigma_1^{-1}\sigma_2(j), \,\sigma_2^{-1}\sigma_1(i)}\\&=&\delta_{\sigma_2^{-1}\sigma_1(k), \,\sigma_1^{-1}\sigma_2(j)}e_{\sigma_1^{-1}\sigma_2(l), \,\sigma_2^{-1}\sigma_1(i)}.\end{aligned}$$ Now $\sigma^2=\mathrm{id}=(\sigma_2^{-1}\sigma_1)$ makes $\sigma_2^{-1}\sigma_1=\sigma_1^{-1}\sigma_2$. Therefore $j=k$ if and only if $\sigma_2^{-1}\sigma_1(k)=\sigma_1^{-1}\sigma_2(j)$, that is, $\delta_{jk}=\delta_{\sigma_2^{-1}\sigma_1(k), \,\sigma_1^{-1}\sigma_2(j)}$. As a consequene, $(e_{ij}e_{kl})^{\ast}=e_{kl}^{\ast}e_{ij}^{\ast}$. This completes the proof of (GC2).
Secondly we prove (GC3). According to the definition of $C_{ij}$, easy computations give that $C_{ij}C_{kl}=\delta_{\sigma_2^{-1}(j),\,\sigma_1^{-1}(k)}C_{il}.$
Finally, assume that $C_{ij}C_{kl}\neq 0$. Then $\sigma_2^{-1}(j)=\sigma_1^{-1}(k)$, that is, $\sigma(j)=k$. Then ${\rm deg}(j)=-{\rm deg}(\sigma(j))=-{\rm deg}(k)$ implies ${\rm deg}(C_{ij}C_{kl})={\rm deg}(C_{il})$.
\[5.2\]Keep notations as Proposition \[5.1\]. If $\sigma(i)=i$ then ${\rm deg}(i)=0$.
Using Proposition \[5.1\], we can give a graded symmetric cellular structure of $A=M_n(K)$ as follows: Let $\Lambda=\{\diamond\}$, $M(\diamond)=\{1, 2, \cdots, n\}$, and set $C_{ij}=e_{i,n-j+1}$ for all $1\leq i,j \leq n$. Then $C_{ij}C_{kl}=\delta_{k,n-j+1}C_{il}$ shows $\{C_{ij}\mid 1\leq i, j\leq n\}$ is a cellular basis of $A$.
Now if $n>1$ is odd then we define $${\rm deg}(i)=
\begin{cases}
i,& i\leq \frac{n-1}{2};\\
0,& i=\frac{n+1}{2};\\
i-n-1, & i\geq \frac{n+3}{2}.
\end{cases}$$ If $n$ is even then we define $${\rm deg}(i)=
\begin{cases}
i,& i\leq \frac{n}{2};\\
i-n-1, & i> \frac{n}{2}.
\end{cases}$$ This makes $\{C_{ij}\mid 1\leq i, j\leq n\}$ a graded cellular basis of $A$. Now we define a homogeneous $K$-linear map $\tau$ from $A$ to $K$ by $$\tau(C_{ij})=
\begin{cases}
1,& {\rm deg}(C_{ij})=0;\\
0, & {\rm Otherwise.}
\end{cases}$$ Then $\tau(C_{ij}C_{kl})=\delta_{k,n-j+1}\delta_{i,n-l+1}=\tau(C_{kl}C_{ij})$ implies that $\tau(ab)=\tau(ba)$ for all $a, b\in A$. Clearly $\tau$ is non-degenerate. As a Consequence, $\tau$ is a homogeneous symmetrizing trace of degree 0.
Let us remark that the above argument can be generalized to the semisimple case, that is, we have the following
\[5.3\] Let $A=\bigoplus M_{n_i}(K)$ be a semisimple algebra. If there exists some $n_i\neq 1$, then $A$ is a graded symmetric cellular algebra with a non-trivial grading.
Now let $A$ be a semisimple algebra with a homogeneous symmetrizing trace of degree $d$ and let $\{C_{S,T}^{\lambda}\mid{\lambda}\in\Lambda, S,T\in M({\lambda})\}$ be a homogeneous cellular basis of $A$. Then $k_{{\lambda}}^{-1}C_{S,S}^{\lambda}D_{S,S}^{\lambda}$ is an idempotent of $A$ (Lemma \[2.6\]), of course its degree is 0. While Lemma \[3.2\] gives that the degree of $k_{{\lambda}}^{-1}C_{S,S}^{\lambda}D_{S,S}^{\lambda}$ is $-d$. Hence $d=0$ and we have proved the following lemma.
\[5.4\] The degree of all symmetrizing traces $\tau$ of semisimple algebras is 0.
\[5.5\] Let $A$ be a semisimple $K$-algebra with a homogeneous symmetrizing trace $\tau$. Then $L_{\rm gr}(A)=\mathcal {Z}_A(A_0)$.
It follows from Theorem \[4.4\] that we only need to prove $\mathcal {Z}_A(A_0)\subseteq L_{\rm gr}(A)$. Since $A$ is semisimple, we have from Proposition \[5.3\] that $A$ is graded symmetric cellular. Let $\{C_{S,T}^{{\lambda}}\mid{\lambda}\in\Lambda, S, T\in M({\lambda})\}$ be a homogeneous cellular basis of $A$. Note that Lemma \[2.8\] shows $\{C_{S,S}^{{\lambda}}D_{S,T}^{{\lambda}}\mid S,T\in M({\lambda}),{\lambda}\in\Lambda\}$ is a basis of $A$. Assume that $$a=\sum_{S,T\in M({\lambda}),\, {\lambda}\in\Lambda}r_{S,T,{\lambda}}C_{S,S}^{{\lambda}}D_{S,T}^{{\lambda}}\in \mathcal {Z}_A(A_0).$$ Combining Lemmas \[5.4\] and \[3.2\], $\deg (C_{X,X}^{\epsilon}D_{X,X}^{\epsilon})=0$ and $C_{X,X}^{\epsilon}D_{X,X}^{\epsilon}\in A_0$ for $\epsilon\in \Lambda$, $X\in M(\epsilon)$. Therefore $$\begin{aligned}
\sum\limits_{P\in M(\epsilon)}r_{P,X,\epsilon}k_{\epsilon}C_{P,P}^{\epsilon}D_{P,X}^{\epsilon}
&=&\sum_{\substack{{\lambda}\in\Lambda,\\S,T\in M({\lambda})}} r_{S,T,{\lambda}}C_{S,S}^{{\lambda}}D_{S,T}^{{\lambda}}C_{X,X}^{\epsilon}D_{X,X}^{\epsilon}\\
&=& C_{X,X}^{\epsilon}D_{X,X}^{\epsilon}\sum_{\substack{{\lambda}\in\Lambda,\\S,T\in M({\lambda})}}r_{S,T,{\lambda}}C_{S,S}^{{\lambda}}D_{S,T}^{{\lambda}} \\
&=& \sum\limits_{Q\in M(\epsilon)}r_{X,Q,\epsilon}k_{\epsilon}C_{X,X}^{\epsilon}D_{X,Q}^{\epsilon},
\end{aligned}$$ where the first equality and the last one follow by Lemmas \[2.4\] and \[2.6\].
Since $A$ is semisimple, $k_{\epsilon}\neq 0$ according to Lemma \[2.8\]. Thus $r_{P, X, \epsilon}=0$ if $P\neq X$ for $\epsilon\in \Lambda$, $P, X\in M({\lambda})$, that is, $a=\sum\limits_{{\lambda}\in\Lambda, \,S\in M({\lambda})}r_{S,{\lambda}}C_{S,S}^{{\lambda}}D_{S,S}^{{\lambda}}.$
Now assume that $P, Q\in M(\epsilon)$ and $\deg(P)=\deg(Q)$. Then Lemma \[3.2\] and Lemma \[5.4\] imply $\deg(C_{P,P}^{\epsilon}D_{P,Q}^{\epsilon})=0$. Thus $aC_{P,P}^{\epsilon}D_{P,Q}^{\epsilon}=C_{P,P}^{\epsilon}D_{P,Q}^{\epsilon}a$. By employing the same argument as above, one obtain $r_{P, \epsilon}=r_{Q, \epsilon}$. So $a\in L_{\mathrm{gr}}(A)$, that is, $\mathcal {Z}_A(A_0)\subseteq L_{\rm gr}(A)$.
Let $A$ be a finite dimensional graded symmetric cellular algebra. If $L_{\rm gr}(A)=\mathcal {Z}_A(A_0)$ then $A$ is semisimple.
Since $L_{\rm gr}(A)=\mathcal {Z}_A(A_0)$, we assume that $1=\sum\limits_{\varepsilon\in\Lambda, \,c \,\in\,{\mathbb Z}} r_{\varepsilon, \,c}e_{\varepsilon, \,c}$. For ${\lambda}\in\Lambda$, $S\in M({\lambda})$, Lemma \[4.2’\] implies $e_{{\lambda}, \,c_0}=r_{{\lambda}, \,c_0}k_{{\lambda}}e_{{\lambda}, \,c_0}$, where $c_0=\deg(S)$. Clearly, $e_{{\lambda}, \,c_0}\neq 0$ and consequently $k_{{\lambda}}\neq 0$. Thus $A$ is semisimple owing to Lemma \[2.8\].
Combining Lemmas 5.5 and 5.6, we obtain the main result of this section.
\[Them:semisimple\] Let $A$ be a finite dimensional graded symmetric cellular algebra. Then $A$ is semisimple if and only if $L_{\rm gr}(A)=\mathcal {Z}_A(A_0)$.
Acknowledgements {#acknowledgements .unnumbered}
----------------
The authors thank the Chern Institute of Mathematics in Nankai University for the hospitality during their visits.
[99]{}
M. Auslander, I. Reiten and S.O. Smalo, [*Representation theory of Artin algebras*]{}, Cambridge University Press, 1995.
H.H. Andersen, C. Stroppel and D. Tubbenhauer, [*Cellular structures using $U_q$-tilting modules*]{}, Pacific J. Math. **292** (2018), 21–59.
A. Beilinson, V. Ginzburg and W. Soergel, [*Koszul duality patterns in representation theory*]{}, J. Amer. Math. Soc., [**9**]{} (1996), 473–528.
J. Brundan and A. Kleshchev, [*Blocks of cyclotomic Hekce algebras and Khovanov-Lauda algebras*]{}, Invent. Math., [**178**]{} (2009), 451–484.
J. Brundan and A. Kleshchev, [*Graded deconposition numbers for cyclotomic Hekce algebras*]{}, Adv. Math., [**222**]{} (2009), 1883–1942.
J. Brundan and C. Stroppel, [*Highest weight categories arising from Khovanov’s diagram algebra I: cellularity*]{}, Moscow Math. J., **11** (2011), 685–722.
J. Brundan and C. Stroppel, [*Highest weight categories arising from Khovanov’s diagram algebra III: category O*]{}, Represent. Theory, **15** (2011), 170–243.
J. Brundan and C. Stroppel, [*Gradings on Walled Brauer Algebras and Khovanov’s arc Algebra*]{}, Adv. Math., [**231**]{} (2012), 709–773.
J. Brundan, A. Kleshchev and W. Wang, [*Graded Specht modules*]{}, J. Reine Angew. Math., [**655**]{} (2011), 61–87.
E. Cline, B. Parshall and L. Scott, [*Graded and non-graded Kazhdan-Lusztig theories in algebraic groups and Lie groups*]{}, Aust. Math. Lecture Series, Cambridge University Press, [**9**]{} (1997), 105–125.
K.R. Fuller and B. Zimmermann-Huisgen, [*On the generalized Nakayama conjecture and the Cartan determinant problem*]{}, Trans. Amer. Math. Soc., [**294**]{} (1986), 679-691.
J.J. Graham, [*Modular representations of Hecke algebras and related algebras*]{}, PhD Thesis, Sydney University, 1995.
M. Geck, [*Hecke algebras of finite type are cellular*]{}, Invent. Math., [**169**]{} (2007), 501–517.
J.J. Graham and G.I. Lehrer, [*Cellular algebras*]{}, Invent. Math., [**123**]{} (1996), 1–34.
J. Hu and A. Mathas, [*Graded cellular bases for the cyclotomic Khovanov-Lauda-Rouquier algebras of type $A$*]{}, Adv. Math., [**225**]{} (2010), 598–642.
J. Hu and A. Mathas, [*Cyclotomic quiver Schur algebras for linear quivers*]{}, Proc. London Math. Soc., **110** (2015), 1315–1386.
J. Hu and A. Mathas, [*Seminormal forms and cyclotomic quiver Hecke algebras of type A*]{}, Math. Ann., **364** (2016), 1189–1254.
D. Kazhdan and G. Lusztig, [*Representations of Coxeter groups and Hecke algebras*]{}, Invent. Math., **53** (1979), 165–184.
Y.B. Li, [*Centres of symmetric cellular algebras*]{}, Bull. Aust. Math. Soc., [**82**]{} (2010), 511–522.
Y.B. Li, [*Nakayama automorphisms of Frobenius cellular algebras*]{}, Bull. Aust. Math. Soc., [**86**]{} (2012), 515–524.
Y.B. Li, [*Radicals of symmetric cellular algebras*]{}, Colloq. Math., **133** (2013), 67–83.
Y.B. Li, [*On the radical of the group algebra of a symmetric group*]{}, J. Algebra Appl., **16** (2017) 1750175:1–11.
Y.B. Li and Z.K. Xiao, [*On cell modules of symmetric cellular algebras*]{}, Monatsh. Math., [**168**]{} (2012), 49–64.
Y.B. Li and D.K. Zhao, [*Projective cell modules of Frobenius cellular algebras*]{}, Monatsh. Math., [**175**]{} (2014), 283–291.
Y.M. Liu, G.D. Zhou and A. Zimmermann, [*Higman ideal, stable Hochschild homology and Auslander-Reiten conjecture*]{}, Math. Z., [**270**]{} (2012), 759–781.
David Plaza and Steen Ryom-Hansen, [*Graded cellular bases for Temperley-Lieb algebras of type A and B*]{}, J. Algebra. Comb., **40** (2014), 137–177.
R. Rouquier, [*Derived equivalences and finite dimensional algebras*]{}, Proc. Interna. Congr. Math. (Madrid, 2006), Vol. II, EMS Publishing House, Zurich 2006, 191–221.
C.C. Xi, [*Partition algebras are cellular*]{}, Compos. Math., **119** (1999), 99–109.
C.C. Xi, [*On the quasi-heredity of Birman-Wenzl algebras*]{}, Adv. Math., [**154**]{} (2000), 280–298.
[^1]: The first named author is supported by the Natural Science Foundation of Hebei Province, China (A2017501003) and the Science and Technology support program of Northeastern University at Qinhuangdao (No. XNK201601). The second named author is supported partly by NSFC 11571341, 11671234.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We discuss a mechanical model which mimics the main features of the radiation matter interaction in the black body problem. The pure classical dynamical evolution, with a simple discretization of the action variables, leads to the Stefan- Boltzmann law and to the Planck distribution without any additional statistical assumption.'
address:
- 'International Center for the Study of dynamical Systems, Università degli Studi dell’ Insubria, Via Valleggio,11, 22100 Como, Italy'
- 'Istituto Nazionale di Fisica della Materia, Unità di Milano, Via Celoria 16, 20133 Milano, Italy'
- 'Istituto Nazionale di Fisica Nucleare, Sezione di Milano, Via Celoria 16, 20133 Milano, Italy'
author:
- Giulio Casati
title: Quantum Chaos and the Black Body Radiation
---
[2]{}
The problem of black body radiation occupies a central role in the history of physics. The long debate around its properties stimulated a profound revision of old and firmly established concepts and eventually led to the birth of quantum mechanics. Yet it is somehow unfortunate that such a profound revision of a fundamental theory has been based on a problem of so great complexity. Indeed, from the dynamical viewpoint, one is dealing with a nonlinear system with infinite degrees of freedom. Needless to say, such an infinite set of nonlinear differential equations is not solvable. What is more important, and perhaps less noticed, is that even the statistical description do not rest on solid grounds. As a matter of facts, classical ergodic theory which is now quite well developed, is valid for systems with a [*finite*]{} number $N$ of degrees of freedoms. A main difficulty stems from the fact that the two limits $N \rightarrow \infty$ and $ t\rightarrow\infty$ do not commute. The quantum statistical description is even more complex since it involves the additional difficulty that also the two limits $t\rightarrow \infty$, $\hbar\to 0$ do not commute.
In view of the relevance of the black body problem, it is perhaps worthwhile, after hundred years, a re-examination of the statistical properties with the help of modern computers and at the light of the recent progress in the study of nonlinear classical and quantum dynamical systems. This implies to tackle the problem of the statistical behaviour of infinite systems.
A distinctive feature of the radiation matter interaction in the black body is that each normal mode, or field’s oscillator, interacts with the matter’s degrees of freedom only, and that the strength of this interaction decreases as the mode’s frequency increases. In this paper we introduce a simplified model which however shares the main features of the black body. We consider a system of $N$ oscillators with mass $m_i = c/i^2$ and frequency $\omega_i= \sqrt{k/m_i} = \alpha {i}$ with $\alpha=\sqrt{k/c}$. When at its central position $x=0$ each oscillator collides elastically with a particle of mass $ M>>m_i$. Therefore the whole system is conservative with total energy
$$\begin{aligned}
\label{model}
E= E_0 +\sum_{i=1}^{N} E_i = (1/2)M{V^2}+ \sum_{i=1}^{N} I_i\omega_i + h.c.\end{aligned}$$
where $E_0$ is the energy of the heavy particle and $I_i$ are the actions of the oscillators. The interaction between oscillators and the heavy particle is provided by the hardcore(h.c.) collisions. In our numerical computations we have taken the mass of the heavy particle $M= (\sqrt{5}+1)/2$, $c=0.51$, $k=0.1$ so that $\omega_i=\alpha {i}$ with $\alpha \approx 0.443$. Notice that our system is of a billiard type and therefore the trajectory does not depend on the total energy $E$ which is merely a scale factor. We recall that the one-dimensional system of two hard core point particles with fixed boundary conditions is equivalent to the billiard in a right triangle which is assumed to be ergodic and weakly mixing[@prosen]. In our model the particle exchanges energy with each oscillator and moreover, after each collision, we assign at random, the sign of the velocity of the heavy particle. Our classical model has therefore a high degree of chaoticity. We would like to stress that for the purpose of the present paper we only need a mechanisms which conserves the total energy and allows energy exchange between the different oscillators. Therefore, we are not interested in the detailed motion of the heavy particle. Instead we need a sufficient degree of chaotic behaviour in order to ensure ergodicity of the entire system with $N+1$ degrees of freedom[@nota].
The model system (\[model\]) is a mechanical version of the one dimensional black body problem discussed in [@benenti] and one expects a similar statistical behaviour. Indeed we have numerically computed the time averaged energies of the oscillators and of the particle. As expected, we have observed that for any $N$, and independently of the initial condition, the system approaches the equipartition state in which the time averaged energies of the oscillators and of the particle are all equal (an example is given by the open squares in fig. 1) . Therefore, for any finite value of the total energy $E$, with increasing the number $N$ of oscillators the temperature of the system $T=E/N$ will decrease down to zero as $N\rightarrow\infty$ . The mechanisms through which such state is approached is the one already envisaged by Jeans[@jeans]: energy flows from the matter(our particle) to higher and higher modes of the electromagnetic field(our oscillators) in such a way that the time averaged energy in each mode is zero while the total energy remains constant. Therefore the field continuously absorbs energy from the matter, energy will move endless to higher and higher modes, and the whole system cools down to absolute zero temperature. This would be the behaviour according to classical laws. Fortunately however, this is not the case since our world is governed by quantum mechanics. The latter leads to the celebrated Planck distribution which, in one dimension, reads:
$$\begin{aligned}
\label{planck}
E_i =E(\omega_i) = \hbar{\omega_i}/(\exp{\beta{\hbar}{\omega_i}}-1)\end{aligned}$$
The temperature $T=1/\beta$ is finite and connected to the total energy $E_f$ in the field (in the oscillators in our case) by the relation $E_f=\sigma{T^2}$ which is known as Stefan-Boltzmann law (in the one dimensional case).
Expression(\[planck\]) has been derived by statistical methods. A [*dynamical*]{} derivation based on the quantum theory or on the numerical solution of the Schrodinger equation without any statistical assumption is still lacking. In this sense we are asking, for the quantum case, the same question that was posed about 50 years ago by Fermi, Pasta and Ulam[@fermi] regarding the problem of classical equipartition (which is the limit of (\[planck\]) for $\hbar\to 0$). Equipartition was a very well established consequence of classical statistical mechanics even though it rested on statistical assumptions. Fermi however, wanted to derive this property by a direct solution of Newton equations. The result was unexpected and Fermi considered it as one of the most important of his life. Indeed, as we now know, it is at the root of the modern field of nonlinear classical dynamics and chaos. It is therefore worthwhile to ask now a similar question: can we, by direct numerical integration of the time dependent Schrodinger equation, derive expression(\[planck\])?
A convenient guide in this direction can be provided by recent progress in the so-called field of quantum chaos. One of the main results in the study of this field is the quantum suppression of classically chaotic diffusion[@ford; @review]. This phenomenon is known as quantum dynamical localization since is the dynamical analog of Anderson localization which takes place in disordered solids[@fishman]. Localization represents a strong deviation from quantum ergodicity and only when localization length is larger than the sample size then eigenfunctions are extended, quantum ergodicity takes place and statistical methods can be applied. For example, in such situation, statistical properties of energy levels can be described by random matrix theory. Therefore, if the eigenfunctions are localized, then quantum evolution leads to a stationary state which is typically exponentially localized around the initially excited state and therefore strongly depends on the initial condition. On the other hand, if eigenfunctions are delocalised then the quantum stationary state is close to the corresponding classical one. In both cases it is not clear according to which quantum [*dynamical*]{} mechanisms Planck distribution sets in [@sred] .
In order to provide a clear answer to the above question it would be necessary to compute the quantum evolution for a system with many interacting particles like the system (\[model\]). This is a too difficult task.
In the following we present instead the results of a numerical integration of classical system (\[model\]) in which however we allow for the actions $I_i$ to take only integers values. Namely, after each collision, which obeys the usual classical conservation laws, we substitute the values $I_j=E_j/\omega_j$ with the nearby integers $n_j$. The choice between the upper or the lower nearest integer is made at random after each collision. The roundoff energy of the oscillator is then given to the heavy particle in order to ensure total energy conservation. We will refer to this model as to the discrete model to distinguish it from the usual model(\[model\]) in which classical evolution is computed without any approximation. Our surmise, to be verified, is that such a simple discretization procedure will qualitatively reproduce the main results of an exact integration of Schrodinger equation (with $\hbar=1$). Such possibility was also suggested in [@chirikov; @berman1; @berman2]. More recently we have found that application of such procedure to the standard map leads to exponential quantum localization[@casati]. Clearly the difference between quantum and classical mechanics goes much beyond the discrete nature of phase space. It is however our hope to gain a better understanding of their relationship.
In fig.1 we show the time averaged energies of the particle and of the oscillators, obtained with the above described numerical scheme for $N=64$ and total energy $E=60$. The thin curve is the theoretical Planck law (\[planck\]) with $\hbar=1$ and $\beta$ given by (\[total\])with $E=60$.
$$\begin{aligned}
\label{total}
E = 1/\beta + \sum_{i=1}^{\infty} \omega_i/(\exp{\beta\omega_i}-1)\end{aligned}$$
It is quite remarkable that, [*independently on the initial condition*]{}, the discrete model always reaches the same stationary distribution. We have also checked that the time-averaged values do not change by increasing the integration time. As an example we show in fig. 2 the time-averaged energies for few oscillators as a function of time measured in number of collisions, up to $t=10^7$. These values do not change by further increasing the integration time.
We will not enter here in several questions of detail and simply notice that the overall agreement between the numerical results and the theoretical curve (\[planck\]) (with the correct value of $\hbar =1$)is sufficiently satisfactory. Solution of the classical model starting with the same initial conditions leads to equipartition as is shown by the open squares in fig.1.
In order to analyse the behaviour of the system as a function of the energy, at fixed $N$, we plot in fig.3 the total energy $E$ of the system over the time averaged energy$<E_0>$ of the particle. Clearly in our discret model the classical limit is reached for $E\rightarrow \infty$, since the effects of discretization become less and less important as the total energy is increased. Therefore, for sufficiently large energy, one expects equipartition which is given by the relation $<E>=(N+1)<E_0>$. According to standard statistical mechanics, the temperature is defined as the average kinetic energy $T= <E_0)>$.
As we decrease energy, at fixed particle number $N$, we expect deviation from equipartition. It is really remarkable that integration of our discrete model leads to the Stefan-Bolzmann law which, in our one-dimensional case, implies that the total energy increases quadratically with the temperature. The solid line in fig. 3 is given by $E=\sigma{<E_0>^2}
=\sigma{T^2}$ with $\sigma\approx 6$ (independent on $N$). The four parallel lines are given instead by $E= (N+1)<E_0>$ ($N=8, 16, 32, 64$)which is the exact equipartition law of classical mechanics. According to our data, the transition from Stefan-Boltzmann to equipartition is quite sharp and takes place at $E \approx N^2/\sigma$.
In order to analyse instead the behaviour of the system as a function of the number $N$ of oscillators, at fixed total energy $E$, we plot in fig.4 the quantity $E/<E_0>$ over $N$, for different values of the total energy $E$. The dotted line corresponds to equipartition and it is approached by decreasing $N$, at fixed energy $E$. Instead, by increasing $N$ one approaches a constant value $\propto \sqrt E$. This is again in agreement with the Stefan-Boltzmann law. In summary, the results presented in figs 1-4 agree with the predictions of quantum mechanics. For a given $N$, at low temperature, the Stefan-Boltzmann law and the Planck distribution are obtained; as temperature is increased, one approaches the semiclassical limit in which equipartition takes place.
It may be interesting to examine this transition in relation to the general theory of quantum dynamical localisation. According to [@pichard], the scaling [*Ansatz*]{} is equivalent to postulating the existence of a function $f(x)$ such that
$$\begin{aligned}
\label{loc}
l/N= f(x), x= \xi/N\end{aligned}$$
where $l$ is the actual localization length in the sample of finite size $N$, and $\xi$ is the localization length for the infinite sample. As a measure of localization, following standard procedure, we take the inverse participation ratio:
$$\begin{aligned}
\label{loc}
l=<E_f>^2/\sum_{j=1}^{N}<E_j>^2\end{aligned}$$
In fig.5 we plot the actual localization length $l/N$ versus $\xi/N$ where $\xi$ is computed from the distribution (\[total\]). The localisation length $l$ is obtained from expression(\[loc\]) where $<E_i>$ are the time-averaged energies of the oscillators computed for the discrete model and $<E_f>$ is the total average energy of the oscillators. A quite nice scaling behaviour is observed.
Certainly the results presented here require a better understanding. In particular it is not clear to what extent they are related to the peculiarity of the radiation matter interaction in which the field modes interact only through the matter. Needless to say the conclusions we have drawn here refer to the discrete model (\[model\]). It is our personal opinion that the solution of the time dependent Schrodinger equation will lead to the same qualitative results of figs. 1-5. This however needs a carefull study of true quantum mechanics of system(\[model\]) or of similar models.
[10]{}
G. Casati and T. Prosen, Phys Rev Lett [**83**]{}(1999)4729. One may consider more realistic models than the one studied here. For example one may take the heavy particle moving between two fixed walls in one dimension or in a two-dimensional billiard. However the main results will remain unchanged. The main ingredients are the fact that the system is chaotic, the total energy is conserved and that the oscillators exchange energy only via interaction with the heavy particle. The choice of this particular model is only due to numerical convenience. G. Benenti, G. Casati and I. Guarneri, Europhys. Lett. [**46**]{} 307(1999). J.H.Jeans, [*The dynamical Theory of Gases*]{}, 2nd edition (Cambridge)1916. E. Fermi, J. Pasta, S. Ulam, and M. Tsingou, “Studies of Nonlinear Problems I“, Los Alamos preprint LA-1940, 7 november 1955; See also E. Fermi, collected papers, vol.II (University of Chicago Press, Chicago, 1965), p. 978. G. Casati, B. V. Chirikov, J. Ford and F.M. Izrailev, in Lectures Notes in Physics, Vol 93 (Springer Verlag, Berlin, 1979), 334. G. Casati and B.V. Chirikov, in ”Quantum Chaos", Cambridge University press, 1995. S. Fishman, D.R.Grempel, and R.E. Prange, Phys. Rev.Lett. [**49**]{}, 508(1982) M. Srednicki, Phys. Rev. E 50, 888 (1994);J.Phys.A29, 75 (1996);J.Phys. A 32, 1163 (1999). B.V. Chirikov, F.M.Izrailev and D.L. Shepelyansky, Sov. Sci. Rev. C2, 209 (1981) G.P.Berman,and A.R. Kolovsky, PhysicaD 17, 183(1985). G.P.Berman, A.R. Kolovsky, and F.M. Izrailev, Physica A 152 (1988)273 G. Casati, unpublished J.L. Pichard, J. Phys. C [**19**]{}, 1519 (1986).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Galaxies and the dark matter halos that host them are not spherically symmetric, yet spherical symmetry is a helpful simplifying approximation for idealised calculations and analysis of observational data. The assumption leads to an exact conservation of angular momentum for every particle, making the dynamics unrealistic. But how much does that inaccuracy matter in practice for analyses of stellar distribution functions, collisionless relaxation, or dark matter core-creation?
We provide a general answer to this question for a wide class of aspherical systems; specifically, we consider distribution functions that are “maximally stable”, [*i.e.*]{} that do not evolve at first order when external potentials (which arise from baryons, large scale tidal fields or infalling substructure) are applied. We show that a spherically-symmetric analysis of such systems gives rise to the false conclusion that the density of particles in phase space is ergodic (a function of energy alone).
Using this idea we are able to demonstrate that: (a) observational analyses that falsely assume spherical symmetry are made more accurate by imposing a strong prior preference for near-isotropic velocity dispersions in the centre of spheroids; (b) numerical simulations that use an idealised spherically-symmetric setup can yield misleading results and should be avoided where possible; and (c) triaxial dark matter halos (formed in collisionless cosmological simulations) nearly attain our maximally-stable limit, but their evolution freezes out before reaching it.
author:
- 'Andrew Pontzen$^{1,2,3,4}$, Justin I. Read$^5$, Romain Teyssier$^{6}$, Fabio Governato$^{7}$,'
- |
Alessia Gualandris$^{5}$, Nina Roth$^{1}$, Julien Devriendt$^{2}$\
$^1$ [Department of Physics and Astronomy, University College London, London WC1E 6BT]{}\
$^2$ [Oxford Astrophysics, Denys Wilkinson Building, Keble Road, Oxford, OX1 3RH]{}\
$^3$ [Balliol College, Broad Street, Oxford, OX1 3BJ]{}\
$^4$ [Email: [email protected]]{}\
$^5$ [Department of Physics, University of Surrey, Guildford, GU2 7XH, Surrey, UK]{}\
$^6$ [Institute for Theoretical Physics, University of Zurich, CH-8057 Zürich, Switzerland]{}\
$^7$ [Astronomy Department, University of Washington, Seattle, WA 98195, US]{}\
bibliography:
- '../bibtex/refs.bib'
date: ' Received —; published—. '
title: 'Milking the spherical cow — on aspherical dynamics in spherical coordinates'
---
Introduction
============
Spherical symmetry is a foundational assumption of many dynamical analyses. The primary motivation is simplicity, since few astronomical objects are actually spherical. For example, observations and simulations both suggest that gravitational potential wells generated by dark matter halos are typically triaxial [e.g. @1991ApJ...378..496D; @1996MNRAS.281..716C; @2002ApJ...574..538J; @2005ApJ...629..781K; @2007MNRAS.377...50H; @2012JCAP...05..030S; @Loebman12triaxialMW]. Characterising dark matter halos by spherically-averaged densities and velocities [e.g. @1991ApJ...378..496D; @1996ApJ...462..563N; @TaylorNavarro01NFWPhaseSpace; @2009MNRAS.398L..21S] at best tells only part of the story. At worst, it could be severely misleading.
The question of whether baryonic processes can convert dark matter cusps into cores [@pontzen2014nature] provides one motivation for a detailed study of the relationship between spherical and near-spherical dynamics. To explain why, we need to look forward to some of our results. Later in this paper, we cut a dark matter halo out of a cosmological simulation, then match it to an exactly spherical halo with an identical density and velocity anisotropy profile. This gives us two easy-to-compare equilibrium structures – the first triaxial, the second spherical – to perform a dynamical comparison. We expose each to the same time-varying gravitational potential, mimicking the effects of stellar feedback (there are no actual baryons in these runs). After $1\,{\mathrm{Gyr}}$, the triaxial halo’s averaged density profile flattens into a convincing dark matter core, but the spherical halo maintains its cusp (see Figure 1).
This example, which is fully explored in Section \[sec:cusp-core\], illustrates how it is dangerous to use spherically-symmetric simulations to infer anything about dynamics — even spherically-averaged dynamics — in the real universe. A spherical system does not evolve in the same way as the spherical averages of a triaxial system.
Ignoring asphericity can also lead to observational biases [e.g. @HayashiNavarro06; @Corless07triaxlens]. From a dynamical standpoint, the nature of orbits in triaxial potentials is fundamentally different from those in spherical potentials: although the total angular momentum of any self-gravitating system must always be conserved, it is only in the spherical case that this conservation holds for individual particles. Conversely, a large fraction of dark matter particles near the centre of cosmological halos will be on box orbits which do not conserve their individual angular momenta [@1983ApJ...267..571D; @1996ApJ...471...82M; @2001ApJ...549..862H; @2007ApJ...670.1027A]. One practical consequence is that asphericity may be responsible for filling the loss cones of supermassive black holes at the centre of the corresponding galaxies [@2004MerritLossCone].
Finally, it is known that asphericity plays a fundamental role in setting the equilibrium density profile during gravitational cold collapse [e.g. @1999ApJ...517...64H]. The underlying process is known as the radial orbit instability or ROI ; a related effect was discussed by [@2007ApJ...670.1027A]. The name arises because particles on radial orbits are perturbed onto more circular trajectories. At the same time, the density distribution becomes triaxial. Even in the case of a uniform spherical collapse, this symmetry-breaking process is still triggered, presumably by numerical noise; the tangential component of forces must be unphysically suppressed for the system to remain spherical [@1999ApJ...517...64H; @2006ApJ...653...43M].
Despite all this, assuming spherical symmetry is very tempting because it makes life so much easier. Defining spherically-averaged quantities is a well-defined and sensible procedure even if we actually have the full distribution function in hand (as in simulations): departures from spherical symmetry are sufficiently small that different averaging procedures lead to consistent results [@saharead09]. Additionally, when an aspherical halo is in equilibrium, we have shown numerically that a “sphericalised” version of it is also in equilibrium (see Appendix B of @2013MNRAS.430..121P 2013). This is helpful because it allows one to make a meaningful analysis in spherical coordinates, even when the system is aspherical. But it breaks down when out-of-equilibrium processes are included, as in the stellar-feedback-driven core-creation example above.
![One motivation for studying the relationship between spherical and aspherical dynamics is that the conversion of a dark matter cusp into a core by baryonic processes is qualitatively different in the two cases. Here we show the inner log density slope from a numerical experiment on two halos. One is spherical (dashed line) and the other triaxial (solid line) but their spherically-averaged properties are initially identical. An external potential has been added at the centre during the times indicated by the grey bands, with the fluctuations mimicking stellar feedback. The triaxial halo develops a clear core, whereas the spherical halo almost maintains its central density cusp. A complete description and analysis is given in Section \[sec:cusp-core\]. ](slope_magic.pdf){width="49.00000%"}
The present paper formalizes the idea of spherical analysis performed on aspherical systems and follows through the consequences. We will study equilibrium distribution functions in nearly, but not exactly, spherically-symmetric potentials, and focus on maximally-stable systems (which we define as being stable against all possible external linear perturbations). We will find that, in spherical coordinates, such systems appear to be “ergodic" (meaning that their distribution functions depend on energy alone) because the individual particles move randomly in angular momentum while maintaining a near-constant energy. It is important to emphasise that this describes the appearance of the system when analysed in a spherical coordinate system and the true system need not be chaotic for the result to hold, provided any isolating integrals are not closely related to angular momentum.
The formal statement of this idea is derived in Section \[sec:most-stable-system\]. A brief overview of the required background is given in Appendix \[sec:background\], and a second-order derivation of the evolution is given in Appendix \[sec:f0-evolution\]. Section \[sec:ROI-connection\] outlines the practical consequences, starting by recasting and extending the phenomenology of the radial orbit instability. We describe an immediate implication for observational studies of aspherical systems, namely a new way to break the anisotropy degeneracy. Finally we return to the motivating problem above and explain why triaxial systems can undergo cusp-core transitions more easily than spherical systems. Section \[sec:conclusions\] concludes and points to open questions and future work.
Aspherical dynamics in spherical coordinates {#sec:most-stable-system}
============================================
In this section we consider an aspherical system which is maximally stable against external linear perturbations. We assume that an observer of this system analyses it assuming spherical symmetry. We will show that this observer (falsely) concludes that the system is ergodic, [*i.e.*]{} that the density of particles in phase space is a function of energy alone. The derivation requires the use of action-angle coordinates; a crash course is provided in Appendix \[sec:background\].
Single particles
----------------
Given any near-spherical system, the Hamiltonian in the spherical action-angle variables is $$H({\bm{J}}, {\bm{\Theta}}) = H_0({\bm{J}}) + \delta H({\bm{J}}, {\bm{\Theta}})\textrm{,}\label{eq:H-perturbed}$$ where $H_0$ is the sum of kinetic and potential energies in the spherical background, ${\bm{J}}=(J_r,j,j_z)$ is the vector of spherical actions (see Appendix \[sec:background\]), ${\bm{\Theta}}$ is the vector of spherical angles and $\delta H$ contains the perturbation (which includes the aspherical correction to the potential). The orbit of a particle in exact spherical symmetry, $\delta H=0$, is described by Hamilton’s equations: $$\dot{{\bm{J}}}_0 = -\frac{\partial H_0}{\partial {\bm{\Theta}}} = 0 \textrm{;} \hspace{1cm}
\dot{{\bm{\Theta}}}_0 = \left. \frac{\partial H_0}{\partial {\bm{J}}} \right|_{{\bm{J}}={\bm{J}}_0} \equiv
\Omega_0({\bm{J}}_0)\textrm{,}$$ which defines the constant background orbital frequencies $\Omega_0({\bm{J}}_0)$. The expressions ${\bm{J}}_0$, ${\bm{\Theta}}_0$ and $\Omega_0({\bm{J}}_0)$ will be used throughout to refer to the background ($\delta H=0$) solution. This algebraically simple form of the equations of motion is the reason for using action-angle variables, since it immediately integrates to $${\bm{J}}_0(t) = {\bm{J}}_0 = \textrm{constant; }\hspace{0.5cm} {\bm{\Theta}}_0(t) =
{\bm{\Theta}}_0(0) + {\bm{\Omega}}_0({\bm{J}}_0) t\textrm{,}\label{eq:background-solution}$$ where ${\bm{J}}_0$ and ${\bm{\Theta}}_0(0)$ specify the initial action and angle coordinates of the orbit.
We now consider the effect of the aspherical correction to the potential encoded in $\delta H$, using standard Hamiltonian perturbation theory [e.g. @1992rsm..book.....L; @BinneyTremaine2008]. First, taking advantage of the angle coordinates ${\bm{\Theta}}$ being periodic in $2\pi$, $\delta H$ is expressed as $$\delta H({\bm{J}},{\bm{\Theta}}) = \sum_{{\bm{n}}} \delta H_{{\bm{n}}}({\bm{J}}) e^{i{\bm{n}}\cdot{\bm{\Theta}}}\textrm{.}\label{eq:H-expansion}$$ This equation states that, at any fixed ${\bm{J}}$, one can expand the periodic $\Theta$ dependence in a Fourier series without loss of generality.
We are interested in the evolution of ${\bm{J}}$ at first order in the perturbation. Hamilton’s relevant equation now reads: $$\dot{{\bm{J}}} = - \frac{\partial H}{\partial {\bm{\Theta}}} = - \sum_{{\bm{n}}} i {\bm{n}}
\delta H_{{\bm{n}}}({\bm{J}}) e^{i {\bm{n}} \cdot {\bm{\Theta}}}\textrm{.}\label{eq:hamilton-deltaH}$$ Because $\delta H$ is small, the result to first-order accuracy is given by substituting the zero-order solution into equation and integrating to give $${\bm{J}}(t) = {\bm{J}}_0 - \sum_{{\bm{n}}} \frac{{\bm{n}}}{{\bm{n}} \cdot {\bm{\Omega}}_0} \delta H_{{\bm{n}}}({\bm{J}}_0) e^{i {\bm{n}} \cdot {\bm{\Theta}}_0(t)} + \cdots \textrm{.}\label{eq:J-first-order}$$ Consequently as ${\bm{n}} \cdot {\bm{\Omega}}_0 \to 0$, the linear-order correction to the orbit of a particle can become large even if the aspherical correction to the potential ($\delta H$) is small, an effect known as resonance [@BinneyTremaine2008]. Consider now the evolution of the background energy along the perturbed trajectory, given by $$H_0({\bm{J}}(t)) \simeq H_0({\bm{J}}_0) + \frac{\partial H_0}{\partial {\bm{J}}} \cdot \left({\bm{J}}(t) - {\bm{J}}_0\right) + \cdots\textrm{,}$$ where we have Taylor-expanded to first order around ${\bm{J}}_0$. Substituting equation \[eq:J-first-order\] for $J(t)$, there is a cancellation between numerator and denominator: $$H_0({\bm{J}}(t)) \simeq H_0({\bm{J}}_0) - \sum_{{\bm{n}}}\delta H_{{\bm{n}}}({\bm{J}}_0) e^{i {\bm{n}} \cdot {\bm{\Theta}}_0(t)} +\cdots\label{eq:H0-variation-small}$$ and so the fractional variation in $H_0$ remains small, even if ${\bm{J}}$ changes significantly over time. In other words, according to linear perturbation theory, particles migrate large distances in ${\bm{J}}$ along surfaces of approximately constant background energy $H_0$. One can verify this constrained-migration prediction in numerical simulations of dark matter halos – an explicit demonstration is given in Appendix \[sec:Jr-j-halos\]. This gives us some intuition for the result to come: a population of particles will seem to “randomise" their actions (including angular momentum), but not their energy distribution.
The extent of the migration will depend on the nature of the potential in which a particle orbits. To quantify this requires going beyond linear perturbation theory and is the subject of “KAM theory” after [@kolmogorov1954], [@arnold1963] and [@moser1962]; see e.g. [@BinneyTremaine2008; @goldstein2002classical; @1992rsm..book.....L] for introductions. Colloquially the result is that for any given small perturbation the migration of typical orbits is also small. Arnold diffusion offers the most famous route to more significant diffusion through action space [see @1992rsm..book.....L]; but in our case, there is a more immediate reason why the KAM result does not in fact hold. Specifically, KAM theory relies on the frequencies ${\bm{\Omega}}({\bm{J}})$ being non-degenerate – i.e. that any change in the action leads also to a change in the frequencies, thus shutting off resonant migration. In smooth potentials, $\Omega$ is almost a function of energy alone (see Appendix \[sec:Jr-j-halos\]) and so the migration can be substantial.
Overall we informally expect particles to redistribute themselves randomly within the action shell of fixed background energy until they are evenly spread, implying a distribution function that appears ergodic in a spherical analysis. This does not imply the orbits are chaotic in the traditional sense; it is only because we are analysing an aspherical system in spherical coordinates that the phenomenon arises. With this in mind, we now turn to a more formal demonstration of the result.
The distribution function {#sec:population}
-------------------------
So far we have discussed how a single particle orbiting in a mildly aspherical potential does not conserve its spherical actions (e.g. angular momentum). We informally suggested that a population will appear to ‘randomise’ the spherical actions at fixed energy. We now show more formally that a distribution function of particles subject to aspherical perturbations will be most stable when it is spread evenly on surfaces of constant $H_0$.
We start by decomposing the true distribution function of particles in phase space, $f$, in terms of a spherical background $f_0$ and a perturbation $\delta f$. To make sure the split between spherical background and aspherical perturbation is uniquely defined, we take $f_0$ as the distribution function obtained when we perform a naive analysis averaging out the aspherical contribution: $$f_0({\bm{J}}) \equiv \frac{1}{(2\pi)^3} \int {\mathrm{d}}^3\Theta f({\bm{J}},{\bm{\Theta}})\textrm{.}\label{eq:f0_from_spherical_average}$$ By Jean’s Theorem, $f_0$ is an equilibrium distribution function in the spherical background because it is constructed from spherical invariants ${\bm{J}}$ alone. Analogous to equations and one can write the full distribution function $f$ as $$\begin{aligned}
f({\bm{J}},{\bm{\Theta}}) & = f_0({\bm{J}}) + \delta f({\bm{J}}, {\bm{\Theta}}) \nonumber \\
& = f_0({\bm{J}}) + \sum_{{\bm{n}}} \delta f_{{\bm{n}}}({\bm{J}}) e^{i
{\bm{n}} \cdot {\bm{\Theta}}}\textrm{.}\end{aligned}$$
The whole $f$ is to be in equilibrium in the true system, $\partial f/\partial t=0$. We can turn this into an explicit condition on $f$ using the the collisionless Boltzmann equation, $$0=\frac{\partial f}{\partial t} = [H,f] \equiv \frac{\partial
H}{\partial {\bm{\Theta}}} \cdot \frac{\partial f}{\partial {\bm{J}}}
- \frac{\partial
H}{\partial {\bm{J}}} \cdot \frac{\partial f}{\partial {\bm{\Theta}}}\textrm{.}$$ Expanding to linear order in $H$ and $f$ gives the condition $$\sum_{{\bm{n}}} \left( {\bm{\Omega}}_0({\bm{J}}) \cdot {\bm{n}}
\delta f_{{\bm{n}}} ({\bm{J}}) - \frac{\partial f_0}{\partial {\bm{J}}}
\cdot {\bm{n}} \,\delta H_{{\bm{n}}}({\bm{J}}) \right) e^{i {\bm{n}} \cdot
{\bm{\Theta}}} = 0\label{eq:aspherical-evolution}\textrm{.}$$ The different ${\bm{\Theta}}$ dependence of each term in the sum means that the term in brackets must be zero for each ${\bm{n}}$. In particular, for the resonant terms ${\bm{n}}_{\perp}$ where ${\bm{\Omega}}_0 \cdot {\bm{n}}_{\perp}=0$ one has the condition $$\left(\frac{\partial f_0}{\partial {\bm{J}}}\cdot {\bm{n}}_{\perp}\right) \,\delta H_{{\bm{n}}_{\perp}}({\bm{J}})=0\textrm{.}\label{eq:sphergodicity}$$ A sufficient condition for stability of $\delta f$ is therefore that $f_0$ is a function only of $H_0$, since then $\partial f_0 / \partial {\bm{J}} = {\bm{\Omega}}_0\, {\mathrm{d}}f_0 / {\mathrm{d}}H_0$ and consequently the dot product in equation vanishes.
This is the core result claimed at the start of the section: $f_0 = f_0(H_0)$, [*i.e.*]{} the distribution function implied by a spherical analysis appears to be ergodic. It is not a necessary condition for achieving equilibrium, since for any given aspherical system certain $\delta H_{{\bm{n}}_{\perp}}$ will be zero. Rather, the result should be read as applying to the maximally stable distribution function – a distribution function that does not evolve under any linear perturbation to its potential.
We again emphasise that the distribution function $f_0$ is fictional. There is no sense in which the true distribution function, $f$, is actually ergodic. The statement is about how the system appears to be when it is analysed using spherically averaged quantities, equation . Yet, it establishes a way in which we can understand these spherically-averaged quantities in a systematic way – the system is most stable if it [*appears*]{} ergodic, regardless of what the underlying dynamics is really up to. In the remainder of this paper we will refer to such systems as ‘spherically ergodic’.
Testable predictions {#sec:sphergodic-predictions}
--------------------
We have established that systems which appear to be ergodic in a spherical analysis are maximally stable. Now we need to devise a connection to observable or numerically-measurable quantities.
A distribution function $f(H_0)$ that is truly a function of energy alone has an isotropic velocity distribution [@BinneyTremaine2008]. To test for isotropy, one calculates $\beta(r)$ according to the usual spherically-averaged definition $$\beta(r) = 1 - \frac{\langle v_t^2\rangle(r)}{2 \langle v_r^2 \rangle(r)}$$ where $v_r$ is a particle’s radial velocity, $v_t$ its tangential velocity and the angle-bracket averages are taken in radial shells. For a population on radial orbits, $\beta(r)=1$; conversely for purely circular motion, $\beta(r)=-\infty$. Between these two extremes, an ergodic population has $\beta(r)=0$ [@BinneyTremaine2008].
Intuitively, a spherically ergodic system (in the sense defined in the previous section) should therefore be approximately isotropic. However one has to handle that expectation with a little care because the true population $f$ is [*not*]{} ergodic and the measured velocity dispersions, even in spherical polar coordinates, may inherit information from $f$ that is not present in $f_0$.
Instead we will now construct a more rigorously justifiable, slightly different statement that still connects spherically ergodic populations to velocity isotropy. Measuring the mean of any function of the spherical actions $q({\bm{J}})$, we obtain $$\int {\mathrm{d}}^3 J\, {\mathrm{d}}^3 \Theta\, f({\bm{J}},{\bm{\Theta}}) q({\bm{J}}) = \int {\mathrm{d}}^3 J \, f_0\left(H_0({\bm{J}})\right) q({\bm{J}})\textrm{,}\label{eq:q-average}$$ an exact result. Therefore any statement about averages over spherical actions automatically knows only about $f_0$ – the spherical part of the distribution function. This allows us to derive unambiguous implications of a spherically ergodic population.
The most familiar action is the specific scalar angular momentum $j$. Because it is a scalar for each particle, averages over this quantity do not express anything about a net spin of the halo but rather about the mix of circular and radial orbits, just like the traditional velocity anisotropy. Radially-biased populations have $\langle j\rangle\simeq 0$ whereas populations on circular orbits have $\langle j \rangle= j_c$, where $j_c$ is the maximum angular momentum available at a given energy. So, velocity anisotropy can be conveniently represented in terms of the mean scalar angular momentum.
We can go further and calculate a function, $\langle j\rangle(E)$, where the average is taken only over particles at a particular specific energy. This quantity can be represented in terms of the ratio of two integrals of the form : $$\begin{aligned}
\langle j \rangle(E) &= \frac{\iiint {\mathrm{d}}J_r\, {\mathrm{d}}j\, {\mathrm{d}}j_z\,
f_0(H_0) j \delta(H_0-E)}{\iiint {\mathrm{d}}J_r\, {\mathrm{d}}j\, {\mathrm{d}}j_z\,
f_0(H_0) \delta(H_0-E)}\textrm{.}\end{aligned}$$ The triple integral ranges over the physical phase space coordinates: $0\le J_r < \infty$, $0 \le j < \infty$, $-j \le j_z \le j $. One can immediately perform the $j_z$ integrals; then the $J_r$ integral can be completed by changing variables to $H_0$ (recalling $\partial
H_0/\partial J_r \equiv \Omega_r$) and consuming the $\delta$ function. After this manipulation $j$ can only range between $0$ and $j_c(E)$ where $j_c(E)$ is the specific angular momentum corresponding to a circular orbit with specific energy $E$; there are no physical orbits with more angular momentum at the specified $E$. The final, exact result is: $$\begin{aligned}
\langle j \rangle(E) &= \left.\int_0^{j_c(E)} {\mathrm{d}}j\, \Omega_r(E,j)^{-1}\, j^2\right/\int_0^{j_c(E)} {\mathrm{d}}j\, \Omega_r(E,j)^{-1}\, j\textrm{.} \label{eq:mean-j-expression}\end{aligned}$$ We now have a firm prediction for spherically ergodic populations. Namely, if we bin particles in $E$ and measure $\langle j
\rangle(E)$ in each bin, the results should be predicted by equation \[eq:mean-j-expression\], which is a function only of the potential (through $\Omega_r$). Equation does not exhaust the possible tests for spherical ergodicity, but it is sufficient for our present exploratory purposes.
For smooth potentials, $\Omega_r(E,j)$ varies very little between $j=0$ and $j=j_c$, and one can approximate it very well as a function of $E$ alone (Appendix \[sec:Jr-j-halos\]). In this case, the integrals follow analytically and one has the result $$\langle j \rangle(E) \simeq \frac{2}{3} j_c(E)\textrm{.}\label{eq:mean-j-approx}$$ This is a helpful simplification to set expectations, but throughout this paper when showing the spherically ergodic limit, we will use the exact expression given by equation .
Example applications {#sec:ROI-connection}
====================
So far we have motivated and derived a formal result that aspherical systems are most stable when they appear ergodic in spherical coordinates. We derived one practical consequence for the angular momentum distribution, equation (\[eq:mean-j-expression\]), which in an approximate sense states that the velocity distribution will appear isotropic. We are now in a position to test whether numerical simulations actually tend towards this maximally-stable limit in a variety of situations, beginning with cosmological collisionless dark matter halos.
Cosmological dark matter halos {#sec:cosmo-halos}
------------------------------
Let us re-examine the three high-resolution, dark-matter-only zoom cosmological simulations used in the analysis of [@2013MNRAS.430..121P]. The three each have several million particles in their $z=0$ halos which correspond in turn to a dwarf irregular, $L_{\star}$ galaxy and cluster. The force softening $\epsilon$, virial radius $r_{200}$ (at which the mean density enclosed is 200 times the critical density) and virial masses $M_{200}$ are $65$, $170$, $690\,{\mathrm{pc}}$; $98$, $301$, $1430\,{\mathrm{kpc}}$ and $2.8\times 10^{10}$, $8.0\times 10^{11}$, $8.7 \times 10^{13}\,{\mathrm{M}_{\odot}}$ respectively. For further details see [@2013MNRAS.430..121P].
![The velocity anisotropy of the inner parts of three sample high-resolution cosmological dark matter halos (simulated without baryons), plotted as a function of radius (upper panel) and in energy shells (lower panel). The upper panel shows the classic velocity anisotropy $\beta(r)$ defined in the text, for which a purely radial population has $\beta=1$ and a population on circular orbits $\beta=-\infty$. The lower panel shows our alternative in energy space which can be more precisely related to the theoretical arguments presented in Section \[sec:most-stable-system\]; here $\langle
j \rangle (E)/{j_{\mathrm{circ}}}=0$ for radial orbits and $1$ for circular orbits, and approximately $2/3$ for an isotropic population. Both panels show that the halos have near-isotropic orbits with a slight radial bias. The range of the two plots is roughly comparable, but we caution that the mapping from $r$ to $E$ is not unique (see Figure \[fig:r-to-E\]). []{data-label="fig:cosmo-halos"}](j_of_E.pdf){width="49.00000%"}
The upper panel of Figure \[fig:cosmo-halos\] shows the anisotropy for our cosmological halos. To compare the three directly, we scale the radius by $r_{\mathrm{max}}$ (respectively $27$, $57$ and $340\,{\mathrm{kpc}}$), the radius at which the circular velocity $(GM(<r)/r)^{1/2}$ reaches its maximum, $v_{\mathrm{max}}$ ($=56$, $150$ and $610\,{\mathrm{km\,s^{-1}}}$). We restrict attention to the region well within the virial radius; here, the anisotropy $\beta(r)$ typically lies between the purely radial and isotropic cases [e.g. @2008ApJ...685..739B; @2010MNRAS.402...21N].
We now want to link this relatively familiar velocity anisotropy to the alternative $\langle j \rangle (E)$ statistic that was directly predicted by the spherically ergodic property in Section \[sec:most-stable-system\]. For each particle we calculate the specific energy $E = \dot{{\bm{x}}}^2/2 + \Phi\left(|{\bm{x}}|\right)$, where ${\bm{x}}$ is the vector displacement from the halo centre. To make this quantity agree exactly with $H_0$ in the terminology of Section \[sec:most-stable-system\], we ignore asphericity when calculating the potential energy, defining it as $$\Phi(r) \equiv \int_0^r \frac{G M(<r')}{r'^2} {\mathrm{d}}r'$$ where $M(<r')$ is the mass enclosed inside a sphere of radius $r'$. (The numerical integration is performed by binning particles in shells of fixed width $\epsilon$, chosen to coincide with the force softening in the simulation; within these bins the density is taken to be constant.) The physics is invariant if a constant is added to the potential; we have chosen to fix its scale by setting $\Phi(0)=0$.
![The relationship between $r/{r_{\mathrm{max}}}$ and $E/{v_{\mathrm{max}}}^2$ for circular orbits (dashed line), radial orbits at apocentre (solid line) and particles in the ‘MW’ run (density shows the number of particles with energy $E$ at each radius $r$). The mapping between energy and radius is fuzzy, so that anisotropy at high $E$ can easily contaminate $\beta(r)$ at small $r$.[]{data-label="fig:r-to-E"}](r_E_conversion.pdf){width="49.00000%"}
For each particle we also calculate the specific angular momentum $j=\left| {\bm{x}} \times {\bm{\dot{x}}}
\right|$, and the specific angular momentum of a circular orbit at the same energy, $j_{\mathrm{circ}}(E)$ which is given by simultaneously solving $$\begin{aligned}
E &= \Phi(r)+\frac{{j_{\mathrm{circ}}}^2}{2 r^2} \hspace{0.5cm}\textrm{ and }\hspace{0.5cm} \Phi'(r) = \frac{{j_{\mathrm{circ}}}^2}{r^3}\textrm{,}\end{aligned}$$ to eliminate $r$ in favour of ${j_{\mathrm{circ}}}$.
We plot $\langle j \rangle(E)/{j_{\mathrm{circ}}}(E)$ in bins containing 1000 particles each in the lower panel of Figure \[fig:cosmo-halos\]. To facilitate comparison with the top panel, a population on purely radial orbits would have $\beta=1$ and $\langle j \rangle/{j_{\mathrm{circ}}}=0$, whereas a purely circular distribution function corresponds to $\beta=-\infty$ or $\langle j \rangle/{j_{\mathrm{circ}}}=1$. Isotropic, purely spherical populations have $\beta(r)=0$ and $\langle j \rangle/{j_{\mathrm{circ}}}\simeq 2/3$ as discussed at the end of Section \[sec:sphergodic-predictions\]. When compared against each other in this way, the two panels agree well: the populations are on near-isotropic orbits with a slight radial bias.
Quantitatively, how well do these results agree? As a guideline, we can compare results for models with constant anisotropy. From [@BinneyTremaine2008], a distribution function $f(j,E)=j^{-2\beta} f_1(E)$ generates a constant anisotropy $\beta(r)=\beta$. We can calculate the connection to the new statistic by generalising the reasoning of Section \[sec:sphergodic-predictions\], with the result that $$\langle j \rangle(E) = \frac{\int_0^{{j_{\mathrm{circ}}}} {\mathrm{d}}j \, \Omega_r^{-1} \, j^{2-2\beta}}{\int_0^{{j_{\mathrm{circ}}}} {\mathrm{d}}j \, \Omega_r^{-1} j^{1-2\beta}} \simeq \frac{2-2\beta}{3-2\beta} {j_{\mathrm{circ}}}\textrm{,}\label{eq:beta-to-j}$$ where the first result is exact and the second follows from assuming $\Omega_r$ is independent of $j$ (which is an excellent approximation; see Appendix \[sec:Jr-j-halos\]). Consistent with equation , $\beta=0$ gives $\langle j \rangle(E)/{j_{\mathrm{circ}}}\simeq 2/3$. But as the system becomes more radially biased we can now calculate that, for example, $\beta=0.2$ corresponds to $\langle j \rangle(E)/{j_{\mathrm{circ}}}\simeq 0.62$. Despite the various approximations involved, these values therefore correctly relate the values of $\beta$ in the top panel of Figure \[fig:cosmo-halos\] with the $\langle j \rangle$ results in the lower panel.
That said, a detailed comparison as a function of radius is hard because particles at a given radius $r$ have a wide spread of energies $E$. Figure \[fig:r-to-E\] illustrates this relationship for the ‘MW’ halo. The density shows the probability of a particle at radius $r$ also having specific energy $E$, $p(E|r)$. The minimum $E$ at each $r$ is set by $\Phi(r)$, which gives the energy of a particle at apocentre (solid line). A more typical $E$ is given by the energy of a circular orbit at $r$ (dashed line), and this gives some intuition for mapping results from the top panel of Figure \[fig:cosmo-halos\] onto the bottom panel. However, any $E$ exceeding $\Phi(r)$ is theoretically possible. So $\beta$ at any given radius actually represents an average over particles of many different energies.
The classical radial orbit instability {#sec:classic-roi}
--------------------------------------
![The $\beta(r)$ velocity anisotropy (upper panel) and its equivalent in energy space $\langle j \rangle(E)$ (lower panel, as Figure \[fig:cosmo-halos\]) for the evolution over time of a halo that is initially in equilibrium, but unstable against the radial orbit instability. The initial conditions are represented by the dark blue line, $t=0$; there is a delay of $12 {t_{\mathrm{dyn}}}$ before any significant evolution can be seen (light blue line). Subsequent lines are shown every $4 {t_{\mathrm{dyn}}}$. Once the instability kicks in and generates aspherical perturbations, there is a rapid evolution towards the predicted spherically ergodic limit (lower panel).[]{data-label="fig:j_E_evolution"}](roi.pdf){width="49.00000%"}
Having established that, loosely speaking, $\langle j \rangle(E)$ represents the anisotropy in energy shells in the same way that $\beta(r)$ does in radial shells, we can return to our prediction (Section \[sec:sphergodic-predictions\]) for the former quantity, which is shown by the dashed line in the lower panel of Figure \[fig:cosmo-halos\]. The prediction is almost, but not quite, satisfied in cosmological dark matter halos. Since the condition is only reached in a maximally-stable object, approximate agreement is an acceptable situation.
Because cosmological halos initially form from near-cold collapse, the radial orbit instability [@vanAlbada82ROI; @Barnes86ROI; @1991MNRAS.248..494S; @Weinberg91] is invoked to explain how the radially infalling material gets scattered onto a wider variety of orbits [@2006ApJ...653...43M; @2008ApJ...685..739B; @2009ApJ...704..372B], isotropising the velocity dispersion. Most tellingly, numerical experiments by [@1999ApJ...517...64H; @2006ApJ...653...43M] show that the suppressing the instability (by switching off non-radial forces) results in a qualitatively different density profile as the end-point of collapse. We will now show that the isotropisation of velocity dispersion during the radial orbit instability can be interpreted in terms of an evolution towards stability in the terms of Section \[sec:most-stable-system\].
Consider what happens to a halo that is intentionally designed to be unstable. First, we will show the classic radial orbit instability (ROI) at work by constructing a spherical halo with particles that are on radially-biased orbits. We initialise our particles such that they solve the Boltzmann equation and so are stable in exact spherical symmetry. In practice, however, the strong radial bias means that any slight numerical noise will trigger the ROI. By initialising an unstable equilibrium in this way, we avoid confusion from violent relaxation processes associated with out-of-equilibrium collapse [@1967MNRAS.136..101L].
Specifically, the initial conditions are set up in a similar fashion to @2006MNRAS.367..387R, with particle positions drawn from a generalised Hernquist density profile [@Hernquist1990; @1993MNRAS.265..250D]: $$\rho(r) = \frac{M(3-\gamma)}{4\pi a^3}
\left(\frac{r}{a}\right)^{-\gamma}\left(1+\frac{r}{a}\right)^{\gamma-4}\textrm{,}
\label{eqn:hern}$$ which has a circular velocity reaching a peak at ${r_{\mathrm{max}}}=(2-\gamma)a$ and implies the gravitational potential $$\Phi(r) = \frac{GM}{2-\gamma}\left[\left(1+a/r\right)^{\gamma-2}-1\right]\textrm{.}$$ We choose $\gamma=1$ to roughly mimic an NFW halo [@1997ApJ...490..493N] in the innermost parts of interest. The velocities are sampled (using an accept-reject algorithm) from a numerically-calculated Osipkov-Merritt distribution function [@BinneyTremaine2008]: $$f(Q) = \frac{\sqrt{2}}{4 \pi^2} \frac{{\mathrm{d}}}{{\mathrm{d}}Q} \int_Q^0 \frac{{\mathrm{d}}\Phi}{\sqrt{\Phi - Q}} \frac{{\mathrm{d}}}{{\mathrm{d}}\Phi} \left[\left(
1+r(\Phi)^2/r_a^2\right) \rho(r(\Phi) \right]
\label{eqn:OM}$$ where the parameter $Q$ is defined by $$Q \equiv E + \frac{j^2}{2r_a^2}\textrm{.}$$ The value of $r_a$ is known as the anisotropy radius because the velocity anisotropy is given by $$\beta(r) = \frac{1}{1 + r_a^2/r^2}\textrm{,}$$ showing that $\beta(r) \simeq 0$ (isotropic) for $r \ll r_a$ and $\beta(r) \simeq 1$ (radial) for $r \gg r_a$. We used the minimum value of $r_a$ for which the distribution function is everywhere positive, making the orbits as radially biased as possible; for $\gamma =
1$, this is $r_a \simeq 0.21$ [@1997ApJ...490..136M]. We draw $10^6$ particles and evolve the system using RAMSES [@Teyssier02ramses], with mesh refinement based on the number of particles per cell, resulting in a naturally adaptive force softening reaching a minimum of $\epsilon = 90\,{\mathrm{pc}}$.
Our expectation that numerical noise triggers the ROI is borne out by the numerical experiment. The upper panel of Figure \[fig:j\_E\_evolution\] shows the radial anisotropy $\beta(r)$ over time. We have defined a single dynamical time, ${t_{\mathrm{dyn}}}$, at the peak of the velocity curve so that ${t_{\mathrm{dyn}}}\equiv {r_{\mathrm{max}}}/{v_{\mathrm{max}}}$. The six solid lines show the population at $t=0, 12, 16, 20, 24$ and $28$ dynamical times. At first, $\beta(r)$ appears stable, but suddenly after $16\,{t_{\mathrm{dyn}}}$ it becomes considerably more isotropic. Over the same timescale, asphericity in the potential develops. To demonstrate this, we determine the inertia tensor of the entire density distribution and calculate the ratio of the principal axes; at $t=0$, the ratio is $1.0$ by construction. By $16\,{t_{\mathrm{dyn}}}$ the ratio is $1.8$. It stabilises at around $26\,{t_{\mathrm{dyn}}}$ with a ratio of $4.3$.
This is symptomatic of the classic radial orbit instability in action. We can follow the same process from our energy/angular-momentum standpoint in the bottom panel of Figure \[fig:j\_E\_evolution\]. The initial conditions ($\Delta t = 0$) have a kink in them, with most regions of energy space appearing radially biased but some showing a slight circular preference (around $E \simeq 1.5 {v_{\mathrm{max}}}^2$). We verified that this is an artefact of the Opsikov-Merritt construction and is consistent with the radial $\beta(r)$ being radially biased everywhere; recall Figure \[fig:r-to-E\] shows that the relationship between $r$ and $E$ is non-trivial.
As time progresses, the $\langle j \rangle(E)$ distribution correctly tends towards the spherically ergodic (SE) limit, as predicted. The SE limit is attained to good accuracy for energies $E<{v_{\mathrm{max}}}^2$ by $\Delta
t = 20 \,{t_{\mathrm{dyn}}}$. At larger energies, it is likely difficult to achieve SE because of the long time-scales and weak gravitational fields involved. Like $\langle j \rangle$, $\beta(r)$ at $28\,{t_{\mathrm{dyn}}}$ shows an isotropic distribution at the centre ($r \to 0$); but $\beta(r)>0$ everywhere for $r>0.1\,{r_{\mathrm{max}}}$. We verified that this continuing radial bias throughout the halo is produced by a number of high-energy, loosely-bound particles plunging through, i.e. the high-$E$ particles from the lower panel in Figure \[fig:j\_E\_evolution\] pervade all radii in the upper panel.
Let us briefly recap: what we have so far is a new view of an existing phenomenon. We have recast the ROI and its impact on spherically-averaged quantities as an evolution towards an analytically-derived maximally stable class of distribution functions. Viewed in energy shells instead of radial shells, the distinction between the regions that reach the SE limit and the slowly-evolving, loosely-bound regions that remain radially biased is considerably clearer. Now, because our underlying analytic result is derived without requiring the population to be self-gravitating, we can now go further and consider a different version of the radial orbit instability – and show that it continues to operate even when a system is in a completely stable equilibrium.
The continuous radial orbit instability {#sec:cont-roi}
---------------------------------------
![The mean angular momentum of particles as a function of energy, like the lower panel of Figure \[fig:j\_E\_evolution\] – but now for a subpopulation in a dark matter halo that is globally in equilibrium. At $\Delta t=0$, we select particles on predominantly radial ($j/{j_{\mathrm{circ}}}<0.2$) orbits; at later times, the subpopulation mean evolves back towards the population mean. The lines from bottom to top show the state at selection and after $0.2$, $0.4$, $0.6$, $1.0$ and $4.0\, {t_{\mathrm{dyn}}}$ respectively. []{data-label="fig:subpopulation"}](subpopulation.pdf){width="49.00000%"}
Our next target for investigation is to broaden the conditions: according to the results of Section \[sec:most-stable-system\], a subpopulation of particles should undergo something much like the radial orbit instability even when the [*global*]{} potential is completely stable. A suitable name for this phenomenon would seem to be the “continuous radial orbit instability", since it continues indefinitely after the global potential has stabilised.
We perform another numerical experiment to demonstrate the effect. First, to avoid confusion from cosmological infall and tidal fields, we create a stable, isolated, triaxial cosmological halo by extracting from our cosmological run a region of $3 r_{200}$ around our ‘Dwarf’. We then evolve this isolated region for $2\,{\mathrm{Gyr}}$ to allow any edge effects to die away, and verify that the density profile out to $r_{200}$ is completely stable. As before we define a dynamical time for the system of $t_{\mathrm{dyn}} \equiv
r_{\mathrm{max}}/v_{\mathrm{max}} = 470\,{\mathrm{Myr}}$. After the $2\,{\mathrm{Gyr}}\simeq 4 {t_{\mathrm{dyn}}}$ has elapsed, we select all particles with $j<0.2\,j_{\mathrm{circ}}$. These particles are, at the moment of selection, on preferentially radial orbits. We trace our particles forward through time, measuring $\langle j \rangle(E)$ in each snapshot. The results are shown in Figure \[fig:subpopulation\] for various times between $\Delta t = 0.0$ and $4.0\,t_{\mathrm{dyn}}$ after selection. Over this period, the mean angular momentum significantly increases towards the spherically ergodic limit at every energy. The changes are much faster at low energies where the particles are tightly bound and the local dynamical time is short compared to the globally-defined ${t_{\mathrm{dyn}}}$.
The evolution is rapid until $4\,{t_{\mathrm{dyn}}}$ after which the $\langle j
\rangle(E)$ remains near-static (except at $E>2.5 {v_{\mathrm{max}}}^2$ where it continues to slowly rise). Eventually the subpopulation has $\langle j
\rangle(E) \simeq 0.5 {j_{\mathrm{circ}}}(E)$ independent of $E$. This establishes that particles at low angular momentum within a completely stable, unevolving halo, automatically evolve towards higher angular momentum. After a few dynamical times, their angular momentum becomes comparable to that of a randomly-selected particle from the full population, although with a continuing slight radial bias.
One can understand this incompleteness in a number of equivalent ways. Within our formal picture, it arises from the fact that only certain $\delta H_{{\bm{n}}}$ are non-zero in equation . More intuitively, the continuing process of subpopulation evolution is being driven by particles on box orbits that change their angular momentum at near-fixed energy – but not all particles are on such orbits, and so some memory of the initial selection persists.
Observations of dwarf spheroidal galaxies {#sec:dsph}
-----------------------------------------
![The $\beta(r)$ (upper panel) and $\langle j \rangle(E)$ (lower panel) relations for the stellar populations of dwarf satellites around a Milky-Way-like central. In the upper panel arrows indicate the location of $r_{1/2}$, the radius enclosing half the stellar mass. The profile is plotted exterior to $3\epsilon$ where $\epsilon$ is the softening length. Because of the tidal interaction with the parent galaxy, the outer parts of each object are out of equilibrium and display biases ranging from strongly radial to circular. The lower panel shows the same populations in energy space. At low energies, tightly bound particles are now seen to be close to the spherically ergodic limit. At higher energies, a spread is seen but not as large as would naively be expected from the $\beta(r)$ relation. []{data-label="fig:stellar"}](stellar_j_of_E_2.pdf){width="49.00000%"}
In the previous section, we established that subpopulations on initially radially-biased orbits evolve towards velocity isotropy even if the global potential is stable. We explained this in terms of our earlier calculations, and now turn to how the results might impact on observations.
An important current astrophysical question is how dark matter is distributed within low-mass galaxies: this can discriminate between different particle physics scenarios [@pontzen2014nature]. The smallest known galaxies, dwarf spheroidals surrounding our Milky Way, in principle provide a unique laboratory from this perspective. But determining the distribution of dark matter is a degenerate problem because of the unknown transverse velocities of the stellar component [e.g. @Charbonnier11]. Furthermore the use of spherical analyses seems inappropriate since the underlying systems are known to be triaxial [e.g. @Bonnivard15].
This is exactly the kind of situation we set up in our initial calculations (Section \[sec:most-stable-system\]): an aspherical system being analysed in spherical symmetry. We have shown that the results apply both to tracer and self-gravitating populations. So, we can go ahead and apply it to the stars in observed systems: the stellar population of a dwarf spheroidal system will be maximally stable if it appears ergodic in a spherical analysis. This implies a strong prior on what spherical distribution functions are actually acceptable and therefore, in principle, lessens the anisotropy degeneracy.
This idea warrants exploration in a separate paper; here we will briefly test whether the idea is feasible by inspecting some simulated dwarf spheroidal satellite systems. In particular we use a gas-dynamical simulation of “MW" region (see above) using exactly the same resolution and physics as [@2012arXiv1207.0007Z]; see that work for technical details (although note the actual simulation box is a different realisation).
We analyse all satellites with more than $10^4$ star particles, which gives us objects lying in the ranges $5.4 \times 10^7<M_{\star}<4.9 \times 10^8\,{\mathrm{M}_{\odot}}$, $28<{v_{\mathrm{max}}}<60\,{\mathrm{km\,s^{-1}}}$ and $4.6<{r_{\mathrm{max}}}<9.6\,{\mathrm{kpc}}$. We first verified that these do not host rotationally supported disks; we then calculated $\beta(r)$ profiles between $3 \epsilon \simeq 0.5\,{\mathrm{kpc}}$ and $0.6\,{r_{\mathrm{max}}}$. The stellar half-light radius is $0.15 < r_{1/2}/{r_{\mathrm{max}}}< 0.3$, so our calculations extend into the outer edges of each visible system.
The results are shown in the upper panel of Figure \[fig:stellar\], with arrows indicating $r_{1/2}$. The outer regions display a range of different behaviours from strongly radial to circular. However, in each case the $\beta(r)$ is nearly isotropic as $r\to 0$. This is consistent with a view in which the centres have achieved stability while the outer parts are being harassed by the parent tides and stripping.
The picture is reinforced by considering the $\langle j \rangle(E)$ statistic (lower panel of Figure \[fig:stellar\]). Here, stars on tightly bound orbits (small $E$) are very close to the spherically ergodic limit. In the less-bound regions, the agreement worsens. However a quantitative analysis using the approximate relation shows that the $\langle j \rangle(E)$ relation stays much closer to the spherically ergodic limit than the $\beta(r)$ relation naively implies. This is probably because the shape of $\beta(r)$ is in part determined by out-of-equilibrium processes, coupled to the multivalued relationship between $r$ and $E$ (Figure \[fig:r-to-E\]).
In any case these results suggest that we should adopt priors on spherical Jeans or Schwarzschild analyses that strongly favour near-isotropy in the centre of spheroidal systems – or, more accurately, strongly favour near-ergodicity for tightly bound stars. In future work we will expand on these ideas and apply them to observational data, since they could lead to a substantial weakening of the problematic density-anisotropy degeneracies.
Dark matter cusp – core transitions {#sec:cusp-core}
-----------------------------------
![The velocity anisotropy ($\beta(r)$, upper panel) and angular momentum ($\langle j \rangle(E)$, lower panel) for cusped and cored halos from PG12; note the much-expanded $y$-axis scales relative to previous figures, which are required to highlight the differences. The cored cases (solid lines) are almost perfectly spherically ergodic (lower panel) and hence isotropic (upper panel). This contrasts with the cusped case (dotted lines) which has a slight but significant radial bias (seen as high $\beta$ in the upper panel and low $j$ in the lower panel). []{data-label="fig:jE_cusp_v_core"}](cusp_v_core.pdf){width="49.00000%"}
In @2012MNRAS.421.3464P [henceforth PG12] we established that dark matter can be redistributed when intense, short, repeated bursts of star formation repeatedly clear the central regions of a forming dwarf galaxy of dense gas [see also @2005MNRAS.356..107R; @2006Natur.442..539M]. The resulting time-changing potential imparts net energy to the dark matter in accordance with an impulsive analytic approximation. This type of activity has now been seen or mimicked in a large number of simulations, allowing glimpses of the dependency of the process on galactic mass, feedback type and efficiency [@2012MNRAS.422.1231G; @2012ApJ...759L..42P; @2011arXiv1112.2752M; @teyssier13; @2012arXiv1207.0007Z; @GK13; @2013arXiv1306.0898D; @2015arXiv150202036O].
We based our PG12 analysis on 3D zoom cosmological simulations, but our analytic model assumed exact spherical symmetry. By definition, in the analytic model all particles conserve their angular momentum at all times. Taken at face value, energy gains coupled to constant angular momentum would leave a radially biased population in the centre of the halo.
The top panel of Figure \[fig:jE\_cusp\_v\_core\] shows the measured velocity anisotropy in the inner $5\,{\mathrm{kpc}}$ for the zoom simulations in PG12 at $z=0$; the solid line shows the feedback simulation which has developed a core, whereas the dotted line shows the dark-matter-only simulation which maintains its cusp. The difference between the two cases is the opposite to that naively expected: the centre of the cored halo has a more isotropic velocity dispersion than the centre of the cusped halo. The lower panel shows the equivalent picture in energy space (noting that $\Phi(0)=0$ and $\Phi(5\,{\mathrm{kpc}})=3\,790\,\mathrm{km^2\,s^{-2}}$). For clarity only a small fraction of the $0<j/{j_{\mathrm{circ}}}<1$ interval is now plotted; the differences between the cusped (dotted) and cored (solid) lines are relatively small, but significant. At all energies the cored profile has more angular momentum than the cusped profile; in fact, it lies right on the spherically-ergodic limit (dashed line) whereas the cusped profile is biased by around $0.06$ to radial orbits[^1].
While the PG12 model correctly describes the energy shifts, it misses the angular momentum aspect of core creation, which has been emphasised elsewhere [e.g. @2006ApJ...649..591T]. We will now show that a complete description of cusp–core transitions involves two components: energy gains consistent with PG12, and re-stabilisation of the distribution function consistent with the description of Sections \[sec:population\] and \[sec:cont-roi\]. We again use the RAMSES code to follow isolated halos with particle mass $2\times
10^4\,{\mathrm{M}_{\odot}}$; however we adapted the code to add an external, time-varying potential to the self-gravity.
![The PG12 mechanism is reproduced by adding an external potential (to mimic ‘gas’) to a DM-only simulation for the time intervals indicated by the grey bands. We measure the response of a completely spherical halo (dotted lines) and a realistically triaxial halo (solid lines) as described in the text. In both cases we track the mean energy (top panel) and angular momentum (middle panel) of the $0.1\%$ of particles that start out most bound. We also measure the log slope of the density profile at each time output (bottom panel). The energy shift (top) predicted by PG12 is found to hold in both cases. In the spherical case, the angular momentum remains constant up to discreteness effects; it therefore drops relative to the circular angular momentum (middle panel). Conversely, in the triaxial case, the continuous radial orbit instability causes the angular momentum to rise in proportion to $j_{circ}$. Only when the angular momentum is allowed to rise does a significant core form, meaning that the spherical case unphysically suppresses core formation (lower panel). []{data-label="fig:cuspflattening"}](cusp_flattening.pdf){width="49.00000%"}
We took our stabilised, isolated, triaxial dark matter halo extracted as described in Section \[sec:cont-roi\] and created a sphericalised version of it as follows. For each particle we generate a random rotation matrix following an algorithm given by [@graphicsgems], then multiply the velocity and position vector by this matrix. Finally, we verified that the final particles are distributed evenly in solid angle, and that the spherically-averaged density profile and velocity anisotropy is unchanged. The triaxial and spherical halos are both NFW-like and stable over more than $2\,{\mathrm{Gyr}}$ when no external potential is applied.
For our science runs, we impose an external potential corresponding to $10^8\,{\mathrm{M}_{\odot}}$ gas in a spherical ball $1\,{\mathrm{kpc}}$ in radius, distributed following $\rho\propto r^{-2}$; this implies a potential perturbation of $\Delta \Phi = -700\,{\mathrm{km}}^2\,{\mathrm{s}}^{-2}$ at $500\,{\mathrm{pc}}$, for instance. The potential instantaneously switches off at $100\,{\mathrm{Myr}}$, back on at $200\,{\mathrm{Myr}}$, off at $300\,{\mathrm{Myr}}$ and so forth until it has accomplished four “bursts”. The period and the mass in gas is motivated by Figures 1 and 2 of PG12 and Figure 7 of T+13. We also tried imposing potentials with different regular periods and with random fluctuations, none of which altered the behaviour described below.
The top two panels of Figure \[fig:cuspflattening\] show the time-dependent behaviour of the central, most-bound $0.1\,\%$ of particles. The triaxial and spherical halo results are illustrated by solid and dashed lines respectively. The top panel shows how the coupling of the external potential is similar; in particular, it results in a mean increase in specific energy of $\Delta E \simeq
200\,{\mathrm{km}}^2\,{\mathrm{s}}^{-2}$ for particles in both cases. The final shift in the spherical case is very slightly larger than that in the triaxial case, a difference which is unimportant for what follows (it would tend to create a larger core if anything).
The middle panel displays the mean specific angular momentum $j$ of the same particles as a fraction of the circular angular momentum ${j_{\mathrm{circ}}}(E)$. In the spherical case (dashed line), this quantity drops significantly because $j$ is fixed but ${j_{\mathrm{circ}}}$ is rising over time (since it increases with $E$). By contrast in the triaxial case $j/{j_{\mathrm{circ}}}$ returns to its original value, meaning that $j$ has risen by the same fraction as ${j_{\mathrm{circ}}}(E)$. This is dynamic confirmation of the discussion earlier in this Section: the spherically symmetric approach predicts an increasing bias to radial orbits – whereas in the realistic aspherical case, the stability requirements quickly erase this bias.
The bottom panel of Figure \[fig:cuspflattening\] shows the measured density slope $\alpha = {\mathrm{d}}\ln \rho / {\mathrm{d}}\ln r$ at $500\,{\mathrm{pc}}$ for the two experiments. Both start at $\alpha \simeq -1.5$; the triaxial case correctly develops a core (with $\alpha\simeq -0.1$) whereas the completely spherical case maintains a cusp (with $\alpha \simeq
-0.9$). This discrepancy is causally connected to the angular momentum behaviour: the mean radius of a particle increases with increasing angular momentum $j$, even at fixed energy $E$, so a typical particle migrates further outwards in the triaxial case compared to the spherical.
We can therefore conclude that asphericity is a pre-requisite for efficient cusp–core transitions. However there is a subsidiary issue worth mentioning. After about $3\,{\mathrm{Gyr}}$ we find that the spherical halo autonomously does start increasing $\langle j \rangle / {j_{\mathrm{circ}}}$ and the dark matter density slope becomes shallower. This is because the potential fluctuations have generated a radially biased population which is unstable, and a global radial orbit instability is triggered by numerical noise over a sufficiently long period (as in Section \[sec:classic-roi\]).
In fact, with ‘live’ baryons rather than imposed fluctuations, there are aspherical perturbations which accelerate the re-equilibration process further and renew a global radial orbit instability (Section \[sec:classic-roi\]), encouraging the entire population towards spherical ergodicity. As we saw in Figure \[fig:jE\_cusp\_v\_core\], the cored halo from PG12 has an almost perfectly spherically-ergodic population for $E<3\,000\,{\mathrm{km}}^2\,{\mathrm{s}}^{-2}$ (unlike the cusped case). We verified that this is also true of T+13. Generating a core through potential fluctuations seems to complete a relaxation process that otherwise freezes out at an incomplete stage during collisionless collapse.
Conclusions {#sec:conclusions}
===========
Let us return to the original question: how much do inaccuracies inherent in the spherical approximation really matter in practical situations? The answer is, unfortunately, that “it depends”; but we can now distinguish two cases as a rule of thumb:
1) For dynamical calculations or simulations, the inaccuracy matters a great deal. Neglecting the aspherical part of the potential unphysically freezes out the radial orbit instability and related effects, so can lead to qualitatively incorrect behaviour.
2) For the [*analysis*]{} of observations or simulations in equilibrium, the assumption is far more benign – it is, in fact, extremely powerful when handled with care. The underlying aspherical system and the fictional spherical system both appear to be in equilibrium; the mapping between the two views yields striking insight into (for example) spheroidal stellar distribution functions and dark matter halo equilibria.
These conclusions are based on the fact that, when aspherical systems are analysed in spherical coordinates, there is an attractor solution for the spherically-averaged distribution function $f_0$ – namely, it tends towards being ergodic (i.e. $f_0$ is well-approximated as a function of energy alone). We demonstrated this using Hamiltonian perturbation theory (see Section \[sec:most-stable-system\] and Appendix \[sec:f0-evolution\]), and subsequently used the term “spherically ergodic” (SE) to describe a distribution function $f$ with the property that its spherical average $f_0$ is ergodic in this way.
The result follows because the orbits do not respect spherical integrals-of-motion such as angular momentum. Note, however, that the physical orbits may still possess invariants in a more appropriate set of coordinates. The apparent chaotic behaviour and tendency of particles to spread evenly at each energy is a helpful illusion caused by adopting coordinates that are only partially appropriate.
In Section \[sec:ROI-connection\] we showed that the idea has significant explanatory power. First, we inspected the selection of equilibrium in triaxial dark matter halos (Section \[sec:cosmo-halos\]) which led us to consider the classical radial orbit instability (Section \[sec:classic-roi\]). We demonstrated that the instability naturally terminates very near our SE limit (lower panel, Figure \[fig:j\_E\_evolution\]). Particles at high energies have long dynamical times which causes them to freeze out: they evolve towards, but do not reach, the SE limit. Because these high-energy particles often stray into the innermost regions (Figure \[fig:r-to-E\]), the velocity anisotropy $\beta(r)$ continues to display a significant radial bias at all $r$ after the instability has frozen out. Moreover in the case of self-consistently formed cosmological halos (Figure \[fig:cosmo-halos\]), even at low energies there is a slight radial bias. The bias is only erased when a suitable external potential is applied ([*e.g.*]{} during baryonic cusp-core transformations), forcing the system to stabilise itself against a wider class of perturbations than it can self-consistently generate (Figure \[fig:jE\_cusp\_v\_core\]); we will return to this issue momentarily.
One novel aspect of our analysis compared to previous treatments of the radial orbit instability is that it applies as much to tracer particles as to self-gravitating populations. As a first example, in Section \[sec:cont-roi\] we demonstrated how a subset of dark matter particles chosen to be on radially-biased orbits mix back into the population (Figure \[fig:subpopulation\]). This is the case even for halos that, as a whole, are in stable equilibrium; therefore we referred to the phenomenon as a ‘continuous’ radial orbit instability.
The same argument implies that stars undergo the radial orbit instability as easily as dark matter particles. In particular, without a stable disk to enforce extra invariants, dwarf spheroidal galaxies likely have stellar populations with a near-SE distribution (Section \[sec:dsph\]; Figure \[fig:stellar\]). This provides a footing on which to base spherical Jeans or Schwarzschild analyses of observed systems: it implies an extra prior which can be formulated loosely as stating that $\beta(r) \to 0$ as $r \to 0$. Such a prior could be powerful in breaking the degeneracy between density estimates and anisotropy [e.g. @Charbonnier11], In turn tightening limits on the particle physics of dark matter [@pontzen2014nature].
As a final application, we turned to the question of the baryonic processes that convert a dark matter cusp into a core (Section \[sec:cusp-core\]). Angular momentum is gained by individual particles during the cusp-flattening process [@2006ApJ...649..591T] but our earlier work (especially PG12) has focussed primarily on the energy gains instead. In Figure \[fig:cuspflattening\] we see that, in spherical or in triaxial dark matter halos, time-dependent perturbations (corresponding to the behaviour of gas in the presence of bursty star formation feedback) always lead to a rise in energy. However, in the spherical case the mean angular momentum as a fraction of ${j_{\mathrm{circ}}}$ drops because the angular momentum of each particle is fixed, whereas ${j_{\mathrm{circ}}}$ rises with $E$. Only in the triaxial case does $\langle
j\rangle / {j_{\mathrm{circ}}}$ stay constant, indicating a re-isotropisation of the velocities. We tied the increase in $\langle j
\rangle$ to the continuous radial orbit instability which always pushes a population with low $\langle j \rangle / {j_{\mathrm{circ}}}$ back towards the SE limit (Section \[sec:cont-roi\]). The consequence is that – in realistically triaxial halos – the final dark matter core size will be chiefly determined by the total energy lost from baryons to dark matter, with little sensitivity to details of the coupling mechanism.
Although the PG12 model can be completed neatly in this way, other analytic calculations or simulations based on spherical symmetry will need to be considered on a case-by-case basis. In our exactly-spherical test cases (Figure \[fig:cuspflattening\]), core development is substantially suppressed. This certainly cautions against taking the results of purely spherical analyses at face value. On the other hand, any slight asphericity is normally sufficient to prevent the unphysical angular-momentum lock-up – in particular, we verified that the simulations in [@teyssier13] were unaffected by this issue because their baryons settle into a flattened disk. The analytic result is generic, so the exact shape and strength of the asphericity is a secondary effect in determining the final spherically-averaged distribution function $f_0$.
That said, to understand the way in which $f_0$ actually evolves towards the SE limit (and perhaps freezes out before it gets there) requires going to second order in perturbation theory, as shown in Appendix \[sec:f0-evolution\]. At this point it may also become important to incorporate self-gravity, i.e. the instantaneous connection between $\delta f$ and $\delta H$. This approach has been investigated more fully elsewhere, leading to a different set of insights regarding the onset of the radial orbit instability [e.g. @1991MNRAS.248..494S] as opposed to its end state. The present work actually suggests that the radial orbit instability can be cast largely as a kinematical process, and that the self-gravity is a secondary aspect; it would be interesting to further understand how these two views relate. But of more immediate practical importance is to apply the broader insights about dwarf spheroidal stellar equilibria to observational data, something that we will attempt in the near future.
Acknowledgments {#acknowledgments .unnumbered}
===============
All simulation analysis made use of the pynbody suite [@2013ascl.soft05002P]. AP acknowledges helpful conversations with James Binney, Simon White, Chervin Laporte and Chiara Tonini and support from the Royal Society and, during 2013 when a significant part of this work was undertaken, the Oxford Martin School. JIR acknowledges support from SNF grant PP00P2\_128540/1. FG acknowledges support from HST GO-1125, NSF AST-0908499. NR is supported by STFC and the European Research Council under the European Community’s Seventh Framework Programme (FP7/2007-2013) / ERC grant agreement no 306478-CosmicDawn. JD’s research is supported by Adrian Beecroft and the Oxford Martin School. This research used the DiRAC Facility, jointly funded by STFC and the Large Facilities Capital Fund of BIS.
Background {#sec:background}
==========
{width="90.00000%"}
A brief review of actions
-------------------------
This Appendix contains a very brief review of actions which are necessary for deriving the main result in the paper. For more complete introductions see [@BinneyTremaine2008] or [@goldstein2002classical]. We consider any mechanical problem described by phase-space coordinates ${\bm{q}}$ and momenta ${\bm{p}}$, with Hamiltonian $H({\bm{p}},{\bm{q}})$ so that the equations of motion are $$\dot{{\bm{p}}} = -\frac{\partial H}{\partial {\bm{q}}}\textrm{,}
\hspace{1cm} \dot{{\bm{q}}} = \frac{\partial H}{\partial
{\bm{p}}}\textrm{,}\label{eq:eom}$$ where $\dot{{\bm{q}}}={\mathrm{d}}{\bm{q}}/{\mathrm{d}}t$ and $t$ denotes time.
The actions can be defined starting from any coordinate system in which the motion is separately periodic in each dimension ([*i.e.*]{} for each $i$, $q_i$ and $p_i$ repeat every $\Delta t_i$). The actions $J_i$ are then given by $$J_i \equiv \frac{1}{2\pi}\int_0^{\Delta t_i} p_i \dot{q}_i \,{\mathrm{d}}t\textrm{ (no sum
over $i$)}\textrm{.}\label{eq:define-action}$$ By construction the actions $J_i$ do not change under time evolution and are therefore integrals of motion.
We can complete the set of 6 phase-space coordinates with angles ${\bm{\Theta}}$ in such a way that equations of motion of the form apply. Since $J_i$ is constant, we must then have $$0 = \dot{J}_i=-\frac{\partial H}{\partial \Theta_i}\textrm{,}\hspace{1cm}
\dot{\Theta}_i = \frac{\partial H}{\partial J_i} \equiv \Omega_i({\bm{J}})\textrm{,}\label{eq:action-eom}$$ where the first equation establishes that $H$ can have no $\Theta$ dependence, and the second that consequently the $\Theta_i$ each increase at a constant rate in time specified by the frequency $\Omega_i({\bm{J}})$. The convenience of this set of coordinates is that all the time evolution of a particle trajectory is represented in a very simple way: $$J_i = \textrm{constant,} \hspace{1cm} \Theta_i = \textrm{constant} +
\Omega_i\, t\textrm{.}\label{eq:action-eom-integrated}$$ Furthermore the equations of motion are canonical, which is sufficient to demonstrate that the coordinates are canonical – in other words, the measure appearing in phase space integrals is ${\mathrm{d}}^3 J\,
{\mathrm{d}}^3 \Theta$.
We now specialise to the spherical case, with polar coordinates ${\bm{q}}=(r,\theta,\phi)$ and momenta ${\bm{p}}=(\dot{r},r^2
\dot{\theta}, r^2 \sin^2 \theta \dot{\phi})$. Consider first $J_{\phi}$, $$J_{\phi} = \frac{1}{2\pi} \int_0^{\Delta t_{\phi}} {\mathrm{d}}t \dot{\phi}^2
r^2 \sin^2 \theta =
\frac{1}{2\pi} \int_0^{2\pi} {\mathrm{d}}\phi \dot{\phi} r^2 \sin^2 \theta = j_z$$ where $j_z$ is the $z$ component of the specific angular momentum, $j_z=\dot{\phi} r^2 \sin^2 \theta$ which is a constant of motion. Next consider $J_{\theta}$, $$J_{\theta} = \frac{1}{2\pi} \int_0^{\Delta t_{\theta}} {\mathrm{d}}t
\dot{\theta}^2 r^2 = \frac{1}{\pi} \int_{\theta_a}^{\theta_b} {\mathrm{d}}\theta
\sqrt{j^2 - \frac{j_z^2}{\sin^2 \theta}}
\textrm{,}$$ where $j^2=j_x^2+j_y^2+j_z^2$ is the square of the total specific angular momentum, and $\theta$ varies between $\theta_a$ and $\theta_b$ over the course of an orbit. To evaluate the integral requires a change of variables to relate $\theta_a$ and $\theta_b$ to the inclination of the orbit and so to $j$ and $j_z$ [e.g. @goldstein2002classical eq 10.135]; the final result is that $$J_{\theta} = j-j_z\textrm{.}$$ Because linear combinations of actions are still actions ([*i.e.*]{} they still satisfy equation ) one can take $j_z$ as the first and $j$ as the second action in place of $J_{\phi}$ and $J_{\theta}$.
The last action is the radial action, $$J_r = \frac{1}{2 \pi} \int_0^{\Delta t_r} {\mathrm{d}}t \dot{r}^2 = \frac{1}{
\pi} \int_{r_a}^{r_b} {\mathrm{d}}r \sqrt{E - j^2/2 r^2 - \Phi(r)}\textrm{,}\label{eq:define-Jr}$$ where $\dot{r}$ has been evaluated by energy conservation, and $r$ librates between $r_a$ and $r_b$ over the period of an orbit.
Although the complexity of expression appears to make using the actions cumbersome, the great simplification it brings to the equations of motion in the background (i.e. equation \[eq:action-eom\]) makes the perturbation theory tractable. For that reason we have used the action-angle coordinates in our analytic derivation, Section \[sec:most-stable-system\], but avoided them when discussing results from simulations in Section \[sec:ROI-connection\].
Perturbed trajectories in the harmonic oscillator case {#sec:harmonic-intuition}
------------------------------------------------------
Section \[sec:most-stable-system\] used Hamiltonian perturbation theory to discuss the behaviour of particles in aspherical potentials. To connect this more firmly with the equations of motion, it can be helpful to study orbits in a specific potential and connect the solutions with the more general statements made by the perturbation theory. In this Appendix, we use the harmonic oscillator as such an illustrative example.
Consider the Hamiltonian for a single particle in an anisotropic harmonic oscillator potential, $$H = \frac{1}{2} \left(\pi_x^2 + \pi_y^2 + \pi_z^2\right) +
\frac{\omega_0^2}{2} \left( x^2+(1+\epsilon) y^2+(1-\delta) z^2\right)\textrm{,}\label{eq:anisotropic-oscillator}$$ where $\pi_x$, $\pi_y$, $\pi_z$ denote the momentum in the three cartesian directions and without loss of generality $\epsilon\ge 0$ and $\delta\ge 0$.
We want to compare the motion in this potential to the behaviour in a sphericalised version with Hamiltonian $H_0$; the change in the Hamiltonian between the true and sphericalised cases is given by $$\delta H = H-H_0 = \frac{1}{2} \omega_0^2 \left(\epsilon y^2 - \delta
z^2 \right)\textrm{.}$$ We can immediately read off the first result, which is that the magnitude of $\delta H$ is bounded. The true solution moves around on a fixed $H$ surface, meaning the fractional error takes a maximum value given by $$\frac{\left| \delta H \right |}{H}< \max(\epsilon,\delta)\textrm{.}\label{eq:approx-H0-conserve}$$ This is the equivalent of the general statement that $H_0$ variations are small, given by equation .
On the other hand, the angular momentum does change significantly. We can see this as follows: each of $x$, $y$ and $z$ undergoes oscillation at the frequencies $\omega_0$, $(1+\epsilon)^{1/2}\omega_0$ and $(1-\delta)^{1/2} \omega_0$ respectively. Assuming these new frequencies are not commensurate, the relative phases between the different oscillations slowly shifts until at some point all three separated oscillators reach $x=0$, $y=0$ and $z=0$ at the same moment. At this point, since the velocity remains finite, the angular momentum has become zero. It may take a number of dynamical times before this happens, but (for example) at the centre of dark matter halos the dynamical time is very short compared to the Hubble time so angular momentum conservation is effectively destroyed.
All the above is illustrated in Figure \[fig:orbit-Jr-j\]. The left panel contrasts the orbits for a spherical harmonic oscillator (dashed line) and an aspherical oscillator (solid line) projected in the $(x,y)$ plane. The latter obeys equation with $\epsilon=\delta=0.1$ (and the former with $\epsilon=\delta=0$). The orbits for the spherical case are closed because the frequencies are identical, so the relative phase of the $x$ and $y$ part of the motion remains fixed. Orbits in the aspherical potential are more complex as the relative phase of the cartesian components gradually changes; in fact, a particle will sometimes plunge through the centre of the potential. This is known as a ‘box orbit’. In more general triaxial potentials, a variety of orbit types are possible [e.g. @1996ApJ...471...82M]; the importance of these will be considered momentarily.
The same portion of the orbit is illustrated in the right panel, but now projected into the spherical actions plane $(J_r,j)$. For the spherical case, $J_r$ and $j$ are exactly conserved by construction and the orbit appears as a single point. In the aspherical case (solid line), $J_r$ and $j$ are not even approximately conserved over more than a dynamical time. However, even then, the orbit remains close to the dashed line of constant $H_0$, as required by equation and more generally by equation . From this figure, it is intuitively plausible that the particle is equally likely to be found anywhere along the constant $H_0$ contour, which is the essential result of Section \[sec:population\].
Another way to look at the effect is to consider the relationship between the angular momentum and the harmonic oscillator’s cartesian actions, $J_x$, $J_y$ and $J_z$ which remain constant even in the aspherical case. Specialising for simplicity to the case that $J_z=0$, one can show that $$j = 2 \sqrt{J_x J_y} \sin(\Theta_x-\Theta_y)\textrm{,}$$ where $\Theta_x$ and $\Theta_y$ are the angles conjugate to $J_x$ and $J_y$. Only if $\Theta_x-\Theta_y$ remains constant is $j$ an integral of motion; as soon as the oscillator is aspherical, one has $$j = 2 \sqrt{J_x J_y} \sin( \phi_0 + (\Omega_x - \Omega_y ) t)\label{eq:j-not-conserved}$$ so that $j$ oscillates on the timescale $2\pi (\Omega_x-\Omega_y)^{-1} \simeq \pi \epsilon^{-1}$. The situation in 3D is qualitatively similar.
The harmonic oscillator orbits discussed above are all box orbits. More general triaxial potentials support other orbit types (loop orbits, for example, which are much more tightly constrained; or chaotic orbits, which are even less tightly constrained than box orbits). However the Hamiltonian analysis in the main text is general for all these types of possible orbit. The fraction of different orbit types will determine how fast and how far particles diffuse along the contour. In realistic cosmological dark matter halos, most orbits in the central regions are indeed of the box or chaotic type [@2012MNRAS.423.1955Z]. Even with the baryonic contribution to the gravitational force included (which partially sphericalises the central potential), the large majority of particles remain on the same class of orbit as the dark progenitor [@2010MNRAS.403..525V] so long as feedback is strong enough to prevent long-lived central baryon concentrations developing [@2012MNRAS.422.1863B].
The action space of dark matter halos {#sec:Jr-j-halos}
-------------------------------------
![Orbits in the action space of the equilibrium ‘Dwarf’ cosmological halo. Dotted contours of contant spherical specific energy $H_0$ are spaced at $600\,{\mathrm{km}}^2\,{\mathrm{s}}^{-2}$. (For this halo ${v_{\mathrm{max}}}^2 \simeq 3\,100\,{\mathrm{km}}^2\,{\mathrm{s}}^{-2}$.) The straightness of these contours is part of the reason that orbits can efficiently explore the space, as described in the text. []{data-label="fig:scribble"}](scribble.pdf){width="49.00000%"}
The action-angle space of our equilibrium ‘Dwarf’ dark-matter-only halo is illustrated in Figure \[fig:scribble\], along with some particle orbits (solid lines) over $1.5\,{\mathrm{Gyr}}\simeq 3\,{t_{\mathrm{dyn}}}$. The Hamiltonian is a function of $J_r$ and $j$ only for any spherical potential, and so we have suppressed the third action $j_z$. As expected, the particles explore the space, approximately running along the $H_0$ contours, giving rise to the continuous radial orbit instability described in Section \[sec:cont-roi\]. The contours of $H_0$ give a great deal of dynamical information, because the background frequencies are defined by ${\bm{\Omega}}_0 \equiv \partial
H_0 / \partial {\bm{J}}$. These frequencies are at the core of perturbation theory through equation , but they also determine the higher-order behaviour as follows.
The first step to understanding the behaviour of resonances is to move to secular perturbation theory (e.g. section 2.4 of @1992rsm..book.....L 1992). Secular analysis splits resonant orbits into two classes, known as accidentally- and intrinsically-degenerate. The intrinsic case refers to the situation where the resonance condition applies globally; in other words, that ${\bm{n}} \cdot {\bm{\Omega}}_0$ is near-constant along lines of constant $H_0$. Suppose, conversely, that ${\bm{\Omega}}_0$ did vary along these directions; then, since ${\bm{\Omega}}_0$ is defined by the normal to the $H_0$ contours (${\bm{\Omega}}_0 = \partial H_0/\partial
{\bm{J}}$), this would imply a significant curvature of the dotted lines in Figure \[fig:scribble\]. Since there is no such curvature, we can read off that the frequencies do not change and the dynamics is in the intrinsically-degenerate regime, giving rise to large-scale migrations. (The same property also means that the approximation $\Omega_r = \Omega_r(H_0)$ used in reaching equation will be extremely accurate.) We have established using numerical investigations that this intrinsic degeneracy property is generic to any spherical action space with smooth potentials, rather than being specific to cosmological halos.
The evolution of $f_0$ {#sec:f0-evolution}
======================
In the main text (Section \[sec:most-stable-system\]), we showed that a distribution function $f$ is maximally stable to linear perturbations if its sphericalised part $f_0$ appears ergodic, $f_0 = f_0(H_0)$. However we did not discuss the actual evolution of $f_0$ to see whether this limit is likely to be achieved.
This requires time-dependent perturbation theory. We start by writing the collisionless Boltzmann equation, $$\frac{\partial f}{\partial t} = [H, f] = [H_0, f_0] + [H_0, \delta f] + [\delta H, f_0] + [\delta H, \delta f]\textrm{,}\label{eq:tdep-boltzmann}$$ which is an exact expression. The first term vanishes identically; the second and third terms are linear order, and the final term is second order. The evolution of $f_0$ is given by taking the time derivative of equation and interchanging the derivative and integral: $$\begin{aligned}
\frac{\partial f_0}{\partial t} &=
\frac{1}{(2\pi)^3}\int {\mathrm{d}}^3 \, \Theta \, [H,f] \nonumber \\
& = \frac{1}{(2\pi)^3}\sum_{{\bm{m}},{\bm{n}}} \int {\mathrm{d}}^3 \Theta \, \left[\delta H_{{\bm{m}}}\,e^{i {\bm{m}} \cdot {\bm{\Theta}}} ,\, \delta f_{{\bm{n}}} \,e^{i {\bm{n}} \cdot {\bm{\Theta}}} \right]\textrm{,}\end{aligned}$$ where the two linear-order terms have vanished after integrating over $\Theta$. Expanding the Poisson bracket and integrating the remaining term gives $$\frac{\partial f_0}{\partial t} = - i \sum_{{\bm{n}}} {\bm{n}} \cdot \frac{\partial}{\partial {\bm{J}}} \left( \delta f_{{\bm{n}}} \delta H_{-{\bm{n}}} \right)\textrm{.}\label{nonlinear-f0-evolution}$$ This shows that the evolution of $f_0$ is a fundamentally non-linear phenomenon. To make further progress, we can eliminate $\delta f_{{\bm{n}}}$, showing that $f_0$ evolution depends only on $\delta H_{{\bm{n}}}$. First, Fourier transform the time-dependence of $f_0$ and $\delta f$, so that $$\begin{aligned}
f_0({\bm{J}},t) & = \int {\mathrm{d}}\omega\, e^{i \omega t}\,
\tilde{f}_0({\bm{J}}, \omega)\hspace{0.4cm} \textrm{ and} \\
\delta f({\bm{J}}, {\bm{\Theta}}, t) & = \sum_{{\bm{n}}} \int {\mathrm{d}}\omega \delta\tilde{f}_{{\bm{n}}} ({\bm{J}}, \omega)\,e^{i \omega t + i \,{\bm{\Theta}} \cdot {\bm{n}}}\textrm{.}\end{aligned}$$ For simplicity, we will restrict attention to the case where $\delta H$ is constant in time. Then the evolution of $\delta f$ is given by the linear-order part of equation , Fourier transformed to give: $$\left( \omega + {\bm{\Omega}}_0 \cdot {\bm{n}} \right) \delta \tilde{f}_{{\bm{n}}} = \left(\frac{\partial \tilde{f}_0}{\partial {\bm{J}}} \cdot {\bm{n}}\right) \delta H_{{\bm{n}}}\textrm{.}\label{eq:linear-order-df-dH}$$ This is just the time-dependent, Fourier-transformed version of equation . Note that, for $\delta H$ and $\delta f$ to be real, the Fourier coefficients must satisfy $$\delta \tilde{f}_{{\bm{n}}}(\omega)^* = \delta \tilde{f}_{-{\bm{n}}}(-\omega) \textrm{ and } \delta H_{{\bm{n}}}^* = \delta H_{-{\bm{n}}}\textrm{.}\label{eq:mode-reality}$$ These requirements are consistent with the relation .
We can now substitute the linear-order solution for $\delta f_{{\bm{n}}}$ in equation to get the leading-order evolution equation for $f_0$: $$\omega \tilde{f}_0 = -\sum_{{\bm{n}}} {\bm{n}} \cdot \frac{\partial}{\partial {\bm{J}}} \left[ \frac{\delta H_{{\bm{n}}} \delta H_{{\bm{-n}}} {\bm{n}} \cdot \partial \tilde{f}_0/\partial {\bm{J}}}{\omega + {\bm{\Omega}}_0 \cdot {\bm{n}}} \right]\textrm{.}$$ Using the reality condition, equation , we can pair up negative and positive ${\bm{n}}$ modes, giving an alternative version of the expression that is more explicitly symmetric: $$\tilde f_0 = \sum_{{\bm{n}}} {\bm{n}} \cdot \frac{\partial}{\partial {\bm{J}}} \left[ \frac{\left| \delta H_{{\bm{n}}} \right|^2 {\bm{n}} \cdot \partial \tilde{f}_0/\partial {\bm{J}} }{\left| {\bm{\Omega}}_0 \cdot {\bm{n}} \right|^2 - \omega^2} \right]\textrm{,}\label{eq:f0-fourier}$$ where we have divided both sides by $\omega$ and so the result is technically only applicable for $\omega \ne 0$. Provided the evolution of $f_0$ is smooth, its Fourier transform $\tilde{f}_0$ is continuous, so this is not a problem in practice.
Equation is enough to give some insight into the relaxation process. For simplicity, consider a perturbation consisting of a single, resonant ${\bm{n}}_{\perp}$-mode $\delta
H_{{\bm{n}}_{\perp}}$ with ${\bm{n}}_{\perp} \cdot
{\bm{\Omega}}_0=0$. Then we can explicitly invert the Fourier transform to yield $$\frac{\partial^2 f_0}{\partial t^2} = {\bm{n}}_{\perp} \cdot \frac{\partial}{\partial {\bm{J}}} \left[ \left| \delta H_{{\bm{n}}_{\perp}} \right|^2 {\bm{n}}_{\perp} \cdot \partial f_0/\partial {\bm{J}} \right]\textrm{.}\label{eq:almost-wave-eqn}$$ This is a wave equation in the ${\bm{n}}_{\perp}$ direction with varying speed-of-sound proportional to $|\delta
H_{{\bm{n}}_{\perp}}|$. Any unevenness in the resonant directions will flow away at a speed proportional to the strength of the asphericity, showing explicitly that $\tilde{f}_0$ evolves towards the spherically ergodic limit. We hope to investigate the full behaviour of equation for a variety of regimes in future work.
[^1]: The dashed line for the isotropic $j/{j_{\mathrm{circ}}}$ in the lower panel is calculated using the cored potential. However the dependence on potential is very weak, as discussed in Section \[sec:ROI-connection\]; calculating it instead with the cusped potential leads to differences of less than one percent at every energy.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
This paper describes the resource- and system-building efforts of an eight-week Johns Hopkins University Human Language Technology Center of Excellence [*Summer Camp for Applied Language Exploration*]{} (SCALE-2009) on Semantically-Informed Machine Translation (SIMT). We describe a new modality/negation (MN) annotation scheme, the creation of a (publicly available) MN lexicon, and two automated MN taggers that we built using the annotation scheme and lexicon. Our annotation scheme isolates three components of modality and negation: a trigger (a word that conveys modality or negation), a target (an action associated with modality or negation) and a holder (an experiencer of modality). We describe how our MN lexicon was semi-automatically produced and we demonstrate that a structure-based MN tagger results in precision around 86% (depending on genre) for tagging of a standard LDC data set.
We apply our MN annotation scheme to statistical machine translation using a syntactic framework that supports the inclusion of semantic annotations. Syntactic tags enriched with semantic annotations are assigned to parse trees in the target-language training texts through a process of tree grafting. While the focus of our work is modality and negation, the tree grafting procedure is general and supports other types of semantic information. We exploit this capability by including named entities, produced by a pre-existing tagger, in addition to the MN elements produced by the taggers described in this paper. The resulting system significantly outperformed a linguistically naïve baseline model (Hiero), and reached the highest scores yet reported on the NIST 2009 Urdu-English test set. This finding supports the hypothesis that both syntactic and semantic information can improve translation quality.
author:
- Kathryn Baker
- |
Michael Bloodgood and\
Bonnie J. Dorr
- 'Chris Callison-Burch and Nathaniel W. Filardo and Christine Piatko'
- Lori Levin
- Scott Miller
bibliography:
- 'paper.bib'
title: 'Use of Modality and Negation in Semantically-Informed Syntactic MT'
---
=1
Introduction
============
This paper describes the resource- and system-building efforts of an eight-week Johns Hopkins Human Language Technology Center of Excellence [ *Summer Camp for Applied Language Exploration*]{} (SCALE-2009) on Semantically-Informed Machine Translation (SIMT) [@Baker:2010a; @Baker:2010d; @Baker:2010b; @Baker:2010c]. Specifically, we describe our modality/negation (MN) annotation scheme, a (publicly available) MN lexicon, and two automated MN taggers that were built using the lexicon and annotation scheme.
Our annotation scheme isolates three components of modality and negation: a trigger (a word that conveys modality or negation), a target (an action associated with modality or negation) and a holder (an experiencer of modality). Two examples of MN tagging are shown in Figure \[modality-example\].
[lp[5in]{}]{} (1)&[**Input:**]{} Americans should know that we can not hand over Dr. Khan to them.\
&[**Output:**]{} Americans `<`TrigRequire should`>` `<`TargRequire know`>` that we `<`TrigAble can`>` `<`TrigNegation not`>` `<`TargNOTAble hand`>` over Dr. Khan to them.\
\
(2)&[**Input:**]{} He managed to hold general elections in the year 2002, but he can not be ignorant of the fact that the world at large did not accept these elections.\
&[**Output:**]{} He `<`TrigSucceed managed`>` to `<`TargSucceed hold`>` general elections in the year 2002, but he `<`TrigAble can`>` `<`TrigNegation not`>` `<`TargNOTAble be`>` ignorant of the fact that the world at large did `<`TrigNegation not`>` `<`TrigBelief accept`>` these `<`TargNOTBelief elections`>`.
Note that modality and negation are unified into single MN tags (e.g., the “Able” modality tag is combined with “NOT” to form the “NOTAble” tag) and also that MN tags occur in pairs of triggers (e.g., TrigAble and TrigNegation) and targets (e.g., TargNOTAble).
We apply our modality and negation mechanism to the problem of Urdu-English machine translation using a technique that we call [*tree grafting*]{}. This technique incorporates syntactic labels and semantic annotations in a unified and coherent framework for implementing semantically-informed machine translation. Our framework is not limited to the semantic annotations produced by the MN taggers that are the subject of this paper and we exploit this capability to additionally include named-entity annotations produced by a pre-existing tagger. By augmenting hierarchical phrase-based translation rules with syntactic labels that were extracted from a parsed parallel corpus, and further augmenting the parse trees with markers for modality, negation, and entities (through the tree grafting process), we produced a better model for translating Urdu and English. The resulting system significantly outperformed the linguistically naïve baseline Hiero model, and reached the highest scores yet reported on the NIST 2009 Urdu-English translation task.
We note that while our largest gains were from syntactic enrichments to the model, smaller (but significant) gains were achieved by injecting semantic knowledge into the syntactic paradigm. Verbal semantics (modality and negation) contributed slightly more gains than nominal semantics (named entities) while their combined gains were the sum of their individual contributions.
Of course, the limited semantic types we explored (modality, negation and entities) are only a small piece of the much larger semantic space, but demonstrating success on these semantic aspects of language, the combination of which has been unexplored by the statistical machine translation community, bodes well for (larger) improvements based on the incorporation of other semantic aspects (e.g., relations and temporal knowledge). Moreover, we believe this syntactic framework to be well suited for further exploration of the impact of many different types of semantics on the quality of machine-translation (MT) output. Indeed, it would not have been possible to initiate the current study without the foundational work that gave rise to a syntactic paradigm that could support these semantic enrichments.
In the SIMT paradigm, semantic elements (e.g., modality/negation) are identified in the English portion of a parallel training corpus and projected to the source language (in our case, Urdu) during a process of syntactic alignment. These semantic elements are subsequently used in the translation rules that are extracted from the parallel corpus. The goal of adding them to the translation rules is to constrain the space of possible translations to more grammatical and more semantically coherent output. We explored whether including such semantic elements could improve translation output in the face of sparse training data and few source language annotations. Results were encouraging. Translation quality, as measured by the Bleu metric [@Papineni:2002], improved when the training process for the Joshua machine translation system [@Li:2009] used in the SCALE workshop included MN annotation.
We were particularly interested in identifying modalities and negation because they can be used to characterize events in a variety of automated analytic processes. Modalities and negation can distinguish realized events from unrealized events, beliefs from certainties, and can distinguish positive and negative instances of entities and events. For example, the correct identification and retention of negation in a particular language—such as a single instance of the word “not”—is very important for a correct representation of events and likewise for translation.
The next two sections examine related work and the motivation behind the SIMT approach. Section \[modality-section\] defines the theoretical framework for our MN lexicon and automatic MN taggers. Section \[modality-annotation-scheme\] presents the MN annotation scheme used by our human annotators and describes the creation of a MN lexicon based on this scheme. Section \[Automatic-Modality-Annotation\] presents two types of MN taggers—one that is string-based and one that is structure-based—and evaluates the effectiveness of the structure-based tagger. Section \[joshua\] then presents implementation details of the semantically-informed syntactic system and describes the results of its application. Finally, Section \[conclusions\] presents conclusions and future work.
Related Work
============
The development of annotation schemes has become an area of computational linguistics development in its own right, often separate from machine learning applications. Some projects began as strictly linguistic projects that were later adapted for computational linguistics. When an annotation scheme is consistent and well developed, its subsequent application to NLP systems is most effective. For example, the syntactic annotation of parse trees in the Penn Treebank [@Marcus:93] had a tremendous effect on parsing and on Natural Language Processing in general.
In the case of semantic annotations, each tends to have its unique area of focus. While the labeling conventions may differ, a layer of modality annotation over verb role annotation, for example, can have a complementary effect of providing more information, rather than being viewed as a competing scheme. We review some of the major semantic annotation efforts below.
Propbank [@Palmer:2005] is a set of annotations of predicate-argument structure over parse trees. First annotated as an overlay to the Penn Treebank, Propbank annotation now exists for other corpora. Propbank annotation aims to answer the question [*Who did what to whom?*]{} for individual predicates. It is tightly coupled with the behavior of individual verbs. FrameNet [@Baker:1998], a frame-based lexical database that associates each word in the database with a semantic frame and semantic roles, is also associated with annotations at the lexical level. WordNet [@fellbaum98wordnet] is a very widely used online lexical taxonomy which has been developed in numerous languages. WordNet nouns, verbs, adjectives and adverbs are organized into synonym sets. PropBank, FrameNet, and WordNet cover the word senses and argument-taking properties of many modal predicates.
The Prague Dependency Treebank [@PDT10; @trmanEn2005] (PDT) is a multi-level system of annotation for texts in Czech and other languages, with its roots in the Prague school of linguistics. Besides a morphological layer and an analytical layer, there is a Tectogrammatical layer. The Tectogrammatical layer includes functional relationships, dependency relations and co-reference. The PDT also integrates propositional and extra-propositional meanings in a single annotation framework.
The Penn Discourse Treebank (PDTB) [@Webber:2003; @PRASAD08.754] annotates discourse connectives and their arguments over a portion of the Penn Treebank. Within this framework, senses are annotated for the discourse connectives in a hierarchical scheme. Relevant to the current work, one type of tag in the scheme is the Conditional tag, which includes hypothetical, general, unreal present, unreal past, factual present, and factual past arguments.
The PDTB work is related to that of Wiebe, Wilson, and Cardie for establishing the importance of attributing a belief or assertion expressed in text to its agent (equivalent to the notion of [*holder*]{} in our scheme). The annotation scheme is designed to capture the expression of opinions and emotions. In the PDTB, each discourse relation and its two arguments are annotated for attribution. The attribute features are the Source or agent, the Type (assertion propositions, belief propositions, facts and eventualities), scopal polarity, and determinacy. Scopal polarity is annotated on relations and their arguments to identify cases when verbs of attribution are negated on the surface but the negation takes scope over the embedded clause. An example is the sentence “Having the dividend increases is a supportive element in the market outlook [*but I don’t think it’s a main consideration*]{}.” Here, the second argument (the clause following [*but*]{}) is annotated with a “Neg” marker, meaning “I think it’s not a main consideration.”
Wilson, Wiebe, and Hoffman describe the importance of correctly interpreting polarity in the context of sentiment analysis, which is the task of identifying positive and negative opinions, emotions and evaluations. The authors have established a set of features to distinguish between positive and negative polarity and discuss the importance of correctly analyzing the scope of the negation and the modality (e.g., whether the proposition is asserted to be real or not real).
A major annotation effort for temporal and event expressions is the TimeML specification language, which has been developed in the context of reasoning for question answering [@SauriVP06]. TimeML, which includes modality annotation on events, is the basis for creating the TimeBank and FactBank corpora [@Pustejovsky06; @Sauri09]. In FactBank, event mentions are marked with their degree of factuality.
Recent work incorporating modality annotation includes work on detecting certainty and uncertainty. Rubin describes a scheme for five levels of certainty, referred to as Epistemic modality, in news texts. Annotators identify explicit certainty markers and also take into account Perspective, Focus, and Time. Focus separates certainty into facts and opinions, to include attitudes. In our scheme, focus would be covered by [*want*]{} and [*belief*]{} modality. Also, separating focus and uncertainty can allow the annotation of both on one trigger word. Prabhakaran et al. describe a scheme for automatic committed belief tagging. Committed belief indicates the writer believes the proposition. The authors use a previously annotated corpus of committed belief, non-committed belief, and not applicable [@Diab:2009], and derive features for machine learning from parse trees. The authors desire to combine their work with FactBank annotation.
The CoNLL-2010 shared task [@Farkas:2010] was about the detection of cues for uncertainty and their scope. The task was described as “hedge detection,” that is, finding statements which do not or cannot be backed up with facts. Auxiliary verbs such as [*may*]{}, [*might*]{}, [*can*]{}, etc. are one type of hedge cue. The training data for the shared task included the BioScope corpus [@Szarvas:2008], which is manually annotated with negation and speculation cues and their scope, and paragraphs from Wikipedia possibly containing hedge information. Our scheme also identifies cues in the form of triggers, but our desired outcome is to cover the full range of modalities and not just certainty and uncertainty. To identify scope, we use syntactic parse trees, as was allowed in the CoNLL task.
The textual entailment literature includes modality annotation schemes. Identifying modalities is important to determine whether a text entails a hypothesis. Bar-Haim et al. include polarity based rules and negation and modality annotation rules. The polarity rules are based on an independent polarity lexicon [@Nairn:2006]. The annotation rules for negation and modality of predicates are based on identifying modal verbs, as well as conditional sentences and modal adverbials. The authors read the modality off parse trees directly using simple structural rules for modifiers.
Earlier work describing the difficulty of correctly translating modality using machine translation includes [@Sigurd:1994] and [@Murata:2005]. Sigurd et al. write about rule based frameworks and how using alternate grammatical contructions such as the passive can improve the rendering of the modal in the target language. Murata et al. analyze the translation of Japanese into English by several systems, showing they often render the present incorrectly as the progressive. The authors trained a support vector machine to specifically handle modal contructions, while our modal annotation approach is a part of a full translation system.
We now consider other literature, relating to tree-grafting and machine translation. Our tree-grafting approach builds on a technique used for tree augmentation in Miller et al. , where parse-tree nodes are augmented with semantic categories. In this earlier work, tree nodes were augmented with relations, while we augmented tree nodes with modality and negation. The parser is subsequently retrained for both semantic and syntactic processing. The semantic annotations were done manually by students who were provided a set of guidelines and then merged with the syntactic trees automatically. In our work we tagged our corpus with entities, modality, and negation automatically and then grafted them onto the syntactic trees automatically, for the purpose of training a statistical machine translation system. An added benefit of the extracted translation rules is that they are capable of producing semantically-tagged Urdu parses, despite the fact that the training data were processed by only an English parser and tagger.
Related work in syntax-based MT includes that of Huang and Knight , where a series of syntax rules are applied to a source language string to produce a target language phrase structure tree. The Penn English Treebank [@Marcus:93] is used as the source for the syntactic labels and syntax trees are relabeled to improve translation quality. In this work, node-internal and node-external information is used to relabel nodes, similar to earlier work where structural context was used to relabel nodes in the parsing domain [@KleinManning03]. Klein and Manning’s methods include lexicalizing determiners and percent markers, making more fine-grained VP categories, and marking the properties of sister nodes on nodes. All of these labels are derivable from the trees themselves and not from an auxiliary source. Wang et al. employ this type of node splitting in machine translation and report a small increase in BLEU score.
We use the methods described in [@ZollmannVenugopal:2006; @Venugopal:2007] to induce synchronous grammar rules, a process which requires phrase alignments and syntactic parse trees. Venugopal et al. use generic non-terminal category symbols, as in [@Chiang:2005], as well as grammatical categories from the Stanford parser [@KleinManning03]. Their method for rule induction generalizes to any set of non-terminals. We further refine this process by adding semantic notations onto the syntactic non-terminals produced by a Penn Treebank trained parser, thus making the categories more informative.
In the parsing domain, the work of Petrov and Klein is related to the current work. In this work, rule splitting and rule merging are applied to refine parse trees during machine learning. Hierarchical splitting leads to the creation of learned categories that have linguistic relevance, such as a breakdown of a determiner category into two subcategories of determiners by number, i.e., [*[this]{}*]{} and [*[that]{}*]{} group together as do [*[some]{}*]{} and [*[these]{}*]{}. We augment parse trees by category insertion in cases where a semantic category is inserted as a node in a parse tree, after the English side of the corpus has been parsed by a statistical parser.
SIMT Motivation {#motivation}
===============
As in many of the frameworks described above, the aim of the SIMT effort was to provide a generalized framework for representing structured semantic information, such as modality and negation. Unlike many of the previous semantic annotation efforts (where the emphasis tends to be on English), however, our approach is designed to be directly integrated into a translation engine, with the goal of translating highly divergent language pairs, such as Urdu and English. As such, our choice of annotation scheme—illustrated in the trigger-target example shown in Figure \[modality-example\]—was based on a simplified structural representation that is general enough to accommodate divergent modality/negation phenomena, easy for language experts to follow, and straightforward to integrate into a tree-grafting mechanism for MT. Our objective is to investigate whether incorporating this sort of information into machine translation systems could produce better translations, particularly in settings where only small parallel corpora are available.
---------- ------- -------- ------- -------- -------
set lines tokens types tokens types
training 202k 1.7M 56k 1.7M 51k
dev 981 21k 4k 19k 4k
devtest 883 22k 4k 19-20k 4k
test 1792 42k 6k 38-41k 5k
---------- ------- -------- ------- -------- -------
: The size of the various data sets used for the experiments in this paper including the training, development (dev), incremental test set (devtest) and blind test set (test). The dev/devtest was a split of the NIST08 Urdu-English test set, and the blind test set was NIST09.[]{data-label="data-set-sizes"}
![An example of Urdu-English translation. Shown are an Urdu source document, a reference translation produced by a professional human translator, and machine translation output from a phrase-based model (Moses) without linguistic information, which is representative of state-of-the-art MT quality before the SIMT effort. []{data-label="Urdu-example"}](graphics/intro-example-revised-2){width="\linewidth"}
It is informative to look at an example translation to understand the challenges of translating important semantic elements when working with a low-resource language pair. Figure \[Urdu-example\] shows an example taken from the 2008 NIST Urdu-English translation task, and illustrates the translation quality of a state-of-the-art Urdu-English system (prior to the SIMT effort). The small amount of training data for this language pair (see Table \[data-set-sizes\]) results in significantly degraded translation quality compared, e.g., to an Arabic-English system that has more than 100 times the amount of training data.
The output in Figure \[Urdu-example\] was produced using Moses [@Moses], a state-of-the-art phrase-based MT system that by default does not incorporate any linguistic information (e.g., syntax or morphology or transliteration knowledge). As a result, words that were not directly observed in the bilingual training data were untranslatable. Names, in particular, are problematic. For example, the lack of translation for [*Nagaland*]{} and [*Nagas*]{} induces multiple omissions throughout the translated text, thus producing several instances where the [*holder*]{} of a claim (or [ *belief*]{}) is missing. This is because out of vocabulary words are deleted from the Moses output.
![The evolution of a semantically informed approach to our synchronous context free grammars (SCFGs). At the start of summer the decoder used translation rules with a single generic non-terminal symbol, later syntactic categories were used, and by the end of the summer the translation rules included semantic elements such as modalities and negation, as well as named entities.[]{data-label="Evolution-to-SIMT-Joshua"}](graphics/Evolution-to-SIMT-Joshua.pdf){height="3in"}
We use syntactic and semantic tags as higher-order symbols inside the translation rules used by the translation models. Generic symbols in translation rules (i.e., the non-terminal symbol “X”) were replaced with structured information at multiple levels of abstraction, using a tree-grafting approach that we describe below. Figure \[Evolution-to-SIMT-Joshua\] illustrates the evolution of the translation rules that we used, first replacing “X” with grammatical categories and then with categories corresponding to semantic units.
The semantic units that we examined in this effort were modalities and negation (indications that a statement represents something that has/hasn’t taken place or is/isn’t a belief or an intention) and named entities (such as people or organizations). Other semantic units such as relations between entities and events, were not part of this effort, but we believe they could be similarly incorporated into the framework. We chose to examine semantic units that canonically exhibit two different syntactic types: verbal, in the case of modality and negation, and nominal, in the case of named entities.
Although used in this effort, named entities were not the focus of our research efforts in SIMT. Rather, we focused on the development of an annotation scheme for modality and negation and its use in MT, while relying on a pre-existing HMM-based tagger derived from Identifinder [@Bikel:1999] to produce entity tags. Thus, the remainder of this paper will focus on our modality/negation annotation scheme, two MN taggers produced by the effort, and on the integration of semantics in the SIMT paradigm.
Modality and Negation {#modality-section}
=====================
Modality is an extra-propositional component of meaning. In [*John may go to NY*]{}, the basic proposition is [*John go to NY*]{} and the word [*may*]{} indicates modality and is called the [ *trigger*]{} in our work. Van der Auwera and Amman define core cases of modality: [*John must go to NY*]{} (epistemic necessity), [*John might go to NY*]{} (epistemic possibility), [*John has to leave NY now*]{} (deontic necessity) and [*John may leave NY now*]{} (deontic possibility). Larreya defines the core cases slightly differently as [*root*]{} and [*epistemic*]{}. Root modality in Larreya’s taxonomy includes physical modality ([*He had to stop. The road was blocked*]{}) and deontic modality ([*You have to stop*]{}). Epistemic modality includes problematic modality ([*You must be tired*]{}) and implicative modality ([*You have to be mad to do that*]{}). Many semanticists [@Kratzer; @VonFintelIatridou] define modality as quantification over possible worlds. [*John might leave NY*]{} means that there exist some possible worlds in which John leaves NY. Another view of modality relates more to a speaker’s attitude toward a proposition [@NirenburgMcShane; @McShaneEtAl:2004].
We incorporate negation as an inextricably intertwined component of modality, using the term “modality/negation (MN)” to refer to our resources (lexicons) and processes (taggers). We adopt the view that modality includes several types of attitudes that a speaker might have (or not have) toward an event or state. From the point of view of the reader or listener, modality might indicate factivity, evidentiality, or sentiment. Factivity is related to whether an event, state, or proposition happened or didn’t happen. It distinguishes things that happened from things that are desired, planned, or probable. Evidentiality deals with the source of information and may provide clues to the reliability of the information. Did the speaker have first hand knowledge of what he or she is reporting, or was it hearsay or inferred from indirect evidence? Sentiment deals with a speaker’s positive or negative feelings toward an event, state, or proposition.
Our project was limited to modal words and phrases—and their negations—that are related to factivity. However, beyond the core cases of modality we include some aspects of speaker attitude such as intent and desire. We included these because they are often not separable from the core cases of modality. For example, [*He had to go*]{} may include the ideas that someone wanted him to go, that he might not have wanted to go, that at some point after coercion he intended to go, and that at some point he was able to go [@Larreya:2009].
Our focus was on the eight modalities in Figure \[8-modalities\], where P is a proposition (the [*target*]{} of the [*triggering*]{} modality) and H is the holder (experiencer or cognizer of the modality). Some of the eight factivity-related modalities may overlap with sentiment or evidentiality. For example, [*want*]{} indicates that the proposition it scopes over may not be a fact (it may just be desired), but it also expresses positive sentiment toward the proposition it scopes over. We assume that sentiment and evidentiality are covered under separate coding schemes, and that words like [*want*]{} would have two tags, one for sentiment and one for factivity.
The Modality/Negation Annotation Scheme {#modality-annotation-scheme}
=======================================
The challenge of creating a MN annotation scheme was to deal with the complex scoping of modalities with each other and with negation, while at the same time creating a simplified operational procedure that could be followed by language experts without special training. Below we describe our MN annotation framework, including a set of linguistic simplifications, and then we present our methodology for creation of a publicly available MN lexicon. The modality annotation scheme is fully documented in a set of guidelines that were written with English example sentences [@Baker:2010b]. The guidelines can be used to derive hand tagged evaluation data for English and they also include a section that contains a set of Urdu trigger-word examples.
During the SCALE workshop, some Urdu speakers used the guidelines to annotate a small corpus of Urdu by hand, which we reserved for future work. The Urdu corpus could be useful as an evaluation corpus for automatically tagged Urdu, such as one derived from rule projection in the Urdu-English machine translation system, a method we describe further in section \[joshua\]. Also, although we did not annotate a very large Urdu corpus, more data could be manually annotated to train an automatic Urdu tagger in the future.
Anatomy of Modality/Negation in Sentences
-----------------------------------------
In sentences that express modality, we identify three components: a trigger, a target, and a holder. The trigger is the word or string of words that expresses modality or negation. The target is the event, state, or relation over which the modality scopes. The holder is the experiencer or cognizer of the modality. The trigger can be a word such as [*should*]{}, [*try*]{}, [*able*]{}, [*likely*]{}, or [*want*]{}. It can also be a negative element such as [*not*]{} or [*n’t*]{}. Often, modality or negation is expressed without a lexical trigger. For a typical declarative sentence (e.g., [*John went to NY*]{}), the default modality is strong belief when no lexical trigger is present. Modality can also be expressed constructionally. For example, Requirement can be expressed in Urdu with a dative subject and infinitive verb followed by a verb that means to happen or befall.
- [**Requirement:**]{} does H require P?
- [**Permissive:**]{} does H allow P?
- [**Success:**]{} does H succeed in P?
- [**Effort:**]{} does H try to do P?
- [**Intention:**]{} does H intend P?
- [**Ability:**]{} can H do P?
- [**Want:**]{} does H want P?
- [**Belief**]{}: with what strength does H believe P?
Linguistic Simplifications for Efficient Operationalization {#linguistic-simplifications}
-----------------------------------------------------------
Six linguistic simplifications were made for the sake of efficient operationalization of the annotation task. The first linguistic simplification deals with the scope of modality and negation. The first sentence below indicates scope of modality over negation. The second indicates scope of negation over modality:
- He tried not to criticize the president.
- He didn’t try to criticize the president.
The interaction of modality with negation is complex, but was operationalized easily in the menu of thirteen choices shown in Figure \[13-menu-choices-for-modality\]. First consider the case where negation scopes over modality. Four of the thirteen choices are composites of negation scoping over modality. For example, the annotators can choose [*try*]{} or [*not try*]{} as two separate modalities. Five modalities (Require, Permit, Want, Firmly Believe, and Believe) do not have a negated form. For three of these modalities (Want, Firmly Believe, and Believe), this is because they are often transparent to negation. For example, [*I do not believe that he left NY*]{} sometimes means the same as [*I believe he didn’t leave NY*]{}. Merging the two is obviously a simplification, but it saves the annotators from having to make a difficult decision.
The second linguistic simplification is related to a duality in meaning between [*require*]{} and [*permit*]{}. Not requiring P to be true is similar in meaning to permitting P to be false. Thus, annotators were instructed to label [*not require P to be true*]{} as [*Permit P to be false*]{}. Conversely, [*not Permit P to be true*]{} was labeled as [*Require P to be false*]{}.
After the annotator chooses the modality, the scoping of modality over negation takes place as a second decision. For example, for the sentence [*John tried not to go to NY*]{}, the annotator first identifies [*go*]{} as the target of a modality and then chooses [*try*]{} as the modality. Finally, the annotator chooses [*false*]{} as the polarity of the target.
- H requires \[P to be true/false\]
- H permits \[P to be true/false\]
- H succeeds in \[making P true/false\]
- H does not succeed in \[making P true/false\]
- H is trying \[to make P true/false\]
- H is not trying \[to make P true/false\]
- H intends \[to make P true/false\]
- H does not intend \[to make P true/false\]
- H is able \[to make P true/false\]
- H is not able \[to make P true/false\]
- H wants \[P to be true/false\]
- H firmly believes \[P is true/false\]
- H believes \[P may be true/false\]
The third simplification relates to entailments between modalities. Many words have complex meanings that include components of more than one modality. For example, if one managed to do something, one tried to do it and one probably wanted to do it. Thus, annotators were provided a specificity-ordered modality list in Figure \[13-menu-choices-for-modality\], and were asked to choose the first applicable modality. We note that this list corresponds to two independent “entailment groupings,” ordered by specificity:
- {[*requires*]{} $\rightarrow$ [*permits*]{}}
- {[ *succeeds*]{} $\rightarrow$ [*tries*]{} $\rightarrow$ [ *intends*]{} $\rightarrow$ [*is able*]{} $\rightarrow$ [ *wants*]{}}
Inside the entailment groupings, the ordering corresponds to an entailment relation, e.g., [*succeeds*]{} can only occur if [*tries*]{} has occurred. Also, the {[*requires $\rightarrow$ $\ldots$* ]{}} entailment grouping is taken to be more specific than (ordered before) the {[*succeeds $\rightarrow$ $\ldots$* ]{}} entailment grouping. Moreover, both entailment groupings are taken to be more specific than [*believes*]{}, which is not in an entailment relation with any of the other modalities.
The fourth simplification, already mentioned above, is that sentences without an overt trigger word are tagged as [*firmly believes*]{}. This heuristic works reasonably well for the types of documents we were working with, although one could imagine genres such as fiction in which many sentences take place in an alternate possible world (imagined, conditional, or counterfactual) without explicit marking.
The fifth linguistic simplification is that we did not require annotators to mark nested modalities. For a sentence like [*He might be able to go to NY*]{} the target word [*go*]{} is marked as ability, but [*might*]{} is not annotated for Belief modality. This decision was based on time limits on the annotation task; there was not enough time for annotators to deal with syntactic scoping of modalities over other modalities.
Finally, we did not mark the holder H because of the short time frame for workshop preparation. We felt that identifying the triggers and targets would be most beneficial in the context of machine translation.
The English Modality/Negation Lexicon {#English-Modality-Lexicon}
-------------------------------------
Using the framework described above, we created a MN lexicon that was incorporated into a MN tagging scheme to be described below in Section \[Automatic-Modality-Annotation\]. Entries in the MN lexicon consist of: (1) A string of one or more words: for example, [ *should*]{} or [*have need of*]{}. (2) A part of speech for each word: the part of speech helps us avoid irrelevant homophones such as the noun [*can*]{}. (3) A MN designator: one of the thirteen modality/negation cases described above. (4) A head word (or [ *trigger*]{}): the primary phrasal constituent to cover cases where an entry is a multi-word unit, e.g., the word [*hope*]{} in [*hope for*]{}. (5) One or more subcategorization codes.
We produced the full English MN lexicon semi-automatically. First, we gathered a small seed list of MN trigger words and phrases from our modality annotation manual [@Baker:2010b]. Then, we expanded this small list of MN trigger words by running an online search for each of the words, specifically targeting free online thesauri (e.g., thesaurus.com), to find both synonymous and antonymous words. From these we manually selected the words we thought triggered modality (or their corresponding negative variants) and filtered out words that we thought didn’t trigger modality. The resulting list of MN trigger words and phrases contained about 150 lemmas.
We note that most intransitive LDOCE codes were not applicable to modality/negation constructions. For example, [*hunger*]{} (in the [*Want*]{} modality class) has a modal reading of “desire” when combined with the preposition [*for*]{} (as in [*she hungered for a promotion*]{}), but we do not consider it to be modal when it is used in the somewhat archaic sentence [*He hungered*]{}, meaning that he did not have enough to eat. Thus the LDOCE code `I` associated with the verb [*hunger*]{} was hand-changed to `I-FOR`. There were 43 such cases. Once the LDOCE codes were hand-verified (and modified accordingly), the mapping to subcategorization codes was applied.
We note that most intransitive LDOCE codes were not applicable to modality/negation constructions. For example, [*hunger*]{} (in the [*Want*]{} modality class) has a modal reading of “desire” when combined with the preposition [*for*]{} (as in [*she hungered for a promotion*]{}), but not in its pure intransitive form (e.g., [*he hungered all night*]{}). Thus the LDOCE code `I` associated with the verb [*hunger*]{} was hand-changed to `I-FOR`. There were 43 such cases. Once the LDOCE codes were hand-verified (and modified accordingly), the mapping to subcategorization codes was applied.
The MN lexicon is publicly available at <http://www.umiacs.umd.edu/~bonnie/ModalityLexicon.txt>. An example of an entry is given in Figure \[need-entry\], for the verb [*need*]{}.
------------------- ----------------------------------------------------------------------------------------------------------------------------------
[**String:**]{} Need
[**Pos:**]{} VB
[**Modality:**]{} Require
[**Trigger:**]{} Need
[**Subcat:**]{} [**V3-passive-basic**]{} – More citizens are needed to vote.
[**Subcat:**]{} [**V3-I3-basic**]{} – The government will need to work continuously for at least a year. We will need them to work continuously.
[**Subcat:**]{} [**T1-monotransitive-for-V3-verbs**]{} – We need a Sir Sayyed again to maintain this sentiment.
[**Subcat:**]{} [**T1-passive-for-V3-verb**]{} – Tents are needed.
[**Subcat:**]{} [**Modal-auxiliary-basic**]{} – He need not go.
------------------- ----------------------------------------------------------------------------------------------------------------------------------
Automatic Modality/Negation Annotation {#Automatic-Modality-Annotation}
======================================
A MN tagger produces text or structured text in which modality or negation triggers and/or targets are identified. Automatic identification of the holders of modalities was beyond the scope of our project because the holder is often not explicitly stated in the sentence in which the trigger and target occur. This section describes two types of MN taggers—one that is string-based and one that is structure-based.
The String-based English Modality/Negation Tagger {#string-based-english-modality-tagger}
-------------------------------------------------
The string-based tagger operates on text that has been tagged with parts of speech by a Collins-style statistical parser [@Miller:1998]. The tagger marks spans of words/phrases that exactly match MN trigger words in the MN lexicon described above, and that exactly match the same parts of speech. This tagger identifies the target of each modality/negation using the heuristic of tagging the next non-auxiliary verb to the right of the trigger. Spans of words can be tagged multiple times with different types of triggers and targets.
We found the string-based modality/negation tagger to produce output that matched about 80% of the sentence-level tags produced by our structure-based tagger, the results of which are described next. While string-based tagging is fast and reasonably accurate in practice, we opted to focus on the in-depth analysis of modality/negation of our SIMT results using the more accurate structure-based tagger.
The Structure-based English Modality/Negation Tagger {#structured-based-english-modality-tagger}
----------------------------------------------------
The structure-based MN tagger operates on text that has been parsed [@Miller:1998]. We used a version of the parser that produces flattened trees. In particular, the flattener deletes VP nodes that are immediately dominated by VP or S and NP nodes that are immediately dominated by PP or NP. The parsed sentences are processed by TSurgeon rules. Each TSurgeon rule consists of a pattern and an action. The pattern matches part of a parse tree and the action alters the parse tree. More specifically, the pattern finds a MN trigger word and its target and the action inserts tags such as [TrigRequire]{} and [TargRequire]{} for triggers and targets for the modality Require. Figure \[structure-based-tagger-output\] shows output from the structure-based MN tagger. (Note that the sentence is disfluent: [*Pakistan which could not reach semi-final, in a match against South African team for the fifth position Pakistan defeated South Africa by 41 runs.*]{}) The example shows that [*could*]{} is a trigger for the Ability modality and [ *not*]{} is a trigger for negation. [*Reach*]{} is a target for both Ability and Negation, which means that it is in the category of “H is not able \[to make P true/false\]” in our coding scheme. [ *Reach*]{} is also a trigger for the Succeed modality and [ *semi-final*]{} is its target.
(TOP
(S
(NP
(NNP Pakistan)
(SBAR (WDT which)
(S (MD TrigAble could)
(RB TrigNegation not)
(VB B TargAble TrigSucceed
TargNegation reach)
(ADJP
(JJ TargSucceed semi-final))
(, ,)
(PP (IN in) (DT a)
(NN match) (PP (IN against)
(ADJP (JJ South) (JJ African))
(NN team))
(PP (IN for) (DT the)
(JJ fifth) (NN position))
(NP (NNP Pakistan))))))
(VB D defeated)
(NP (NNP South) (NNP Africa))
(PP (IN by) (CD 41) (NNS runs)) (. .)))
The TSurgeon patterns are automatically generated from the verb class codes in the MN lexicon along with a set of fifteen templates. Each template covers one situation such as the following: the target is the subject of the trigger; the target is the direct object of the trigger; the target heads an infinitival complement of the trigger; the target is a noun modified by an adjectival trigger, etc. The verb class codes indicate which templates are applicable for each trigger word. For example, a trigger verb in the transitive class may use two target templates, one in which the trigger is in active voice and the target is a direct object ([*need tents*]{}) and one in which the trigger is in passive voice and the target is a subject ([*tents are needed*]{}).
In developing the TSurgeon rules, we first conducted a corpus analysis for forty of the most common trigger words in order to identify and debug the most broadly applicable templates. We then used LDOCE to assign verb classes to the remaining verbal triggers in the MN lexicon, and we associated one or more debugged templates with each verb class. In this way, the initial corpus work on a limited number of trigger words was generalized to a longer list of trigger words. Because the TSurgeon patterns are tailored to the flattened structures produced by our parser, it is not easily ported to new parser outputs. However, the MN lexicon itself is portable. Switching parsers would entail writing new TSurgeon templates, but the trigger words in the MN lexicon would still be automatically assigned to templates based on their verb classes.
The following example shows an example of a TSurgeon pattern-action pair for a sentence like [*They were required to provide tents*]{}. The pattern-action pair is intended to be used after a pre-processing stage in which labels such as “VoicePassive” and “AUX” have been assigned. “VoicePassive” is inserted by a pre-processing TSurgeon pattern because, in some cases, the target of a passive modality trigger word is in a different location from the target of the corresponding active modality trigger word. “AUX” is inserted during pre-processing to distinguish auxiliary uses of [*have*]{} and [ *be*]{} from their uses as main verbs. The pattern portion of the pattern-action pair matches a node with label VB that is not already tagged as a trigger and that is passive and dominates the string “required”. The VB node is also a sister to an S node, and the S node dominates a VB that is not an auxiliary ([*provide*]{} in this case). The action portion of the pattern-action pair inserts the string “TargReq” as the second daughter of the second VB and inserts the string “TrigReq” as the second daughter of the first VB.
VB=trigger !< /^Trig/ < VoicePassive < required $..
(S < (VB=target !< AUX))
insert (TargReq) >2 target
insert (TrigReq) >2 trigger
Verb-specific patterns such as this one were generalized in order to gain coverage of the whole modality lexicon. The specific lexical item, [*required*]{}, was replaced with a variable, as were the labels “TrigReq” and “TargReq.” The pattern was then given a name, V3-passive-basic, where V3 is a verb class tag from LDOCE (described above in Section \[English-Modality-Lexicon\]) for verbs that take infinitive complements. We then looked up the LDOCE verb class labels for all of the verbs in the modality lexicon. Using this information, we could then generate a set of new, verb-specific patterns for each V3 verb in the modality lexicon.
Evaluating the Effectiveness of Structure-based MN Tagging {#modality-tagging-evaluation}
----------------------------------------------------------
We performed a manual inspection of the structure-based tagging output. We calculated precision by examining 229 instances of modality triggers that were tagged by our tagger from the English side of the NIST 09 MTEval training sentences. We analyzed precision in two steps, first checking for the correct syntactic position of the target and then checking the semantic correctness of the trigger and target. For 192 of the 229 triggers (around 84%), the targets were tagged in the correct syntactic location.
For example, for the sentence [*A solution must be found to this problem*]{} shown in Figure \[embedded-targ-example\], the word [ *must*]{} is a modality trigger word, and the correct target is the first non-auxiliary verb heading a verb phrase that is contained in the syntactic complement of [*must*]{}. The syntactic complement of [*must*]{} is the verb phrase [*be found to this problem*]{}. The syntactic head of that verb phrase, [*be*]{}, is skipped because it is an auxiliary verb. The correct (embedded) target [*found*]{} is the head of the syntactic complement of [*be*]{}.
(S (NP (DT A) (NN solution))
(VP (MD-TrigBelief must)
(VP (VB be)
(VP (VBN-TargBelief found)
(PP (TO to) (NP (DT this) (NN problem)))))))
The 192 modality instances with structurally correct targets do not all have semantically correct tags. In the example above, [*must*]{} is tagged as [TrigBelief]{}, where the correct tag would be [TrigRequire]{}. Also, because the MN lexicon was used without respect to word sense, words were sometimes erroneously identified as triggers. This includes non-modal uses of [*work*]{} (work with refugees), [*reach*]{} (reach a destination), and [*attack*]{} (attack a physical object), in constrast to modal uses of these words: [*work for peace*]{} (effort), [*reach a goal*]{} (succeed), and [*attack a problem*]{} (effort). Fully correct tagging of modality would need to include word sense disambiguation.
For 37 of the 229 triggers we examined, a target was not tagged in the correct syntactic position. In 12 of 37 incorrectly tagged instances the targets are inside compound nouns or coordinate structures (NP or VP), which are not yet handled by the modality tagger. The remaining 25 of the 37 incorrectly tagged instances had targets that were lost because the tagger does not yet handle all cases of nested modalities. Nested modalities occur in sentences like [*They did not want to succeed in winning*]{} where the target words [*want*]{} and [*succeed*]{} are also modality trigger words. Proper treatment of nested modalities requires consideration of scope and compositional semantics.
Nesting was treated in two steps. First, the modality tagger marked each word as a trigger and/or target. In [*They did not want to succeed in winning*]{}, [*not*]{} is marked as a trigger for negation, [*want*]{} is marked as a target of negation and a trigger of wanting, [*succeed*]{} is marked as a trigger of succeeding and a target of wanting, and [*win*]{} is marked as a target of succeeding. The second step in the treatment of nested modalities occurs during tree grafting, where the meanings of the nested modalities are composed. The tree grafting program correctly composes some cases of nested modalities. For example, the tag [TrigAble]{} composed with [TrigNegation]{} results in the target tag [TargNOTAble]{}, as shown in Figure \[tree-grafting-composition\]. In other cases, where compositional semantics are not yet accommodated, the tree grafting program removed target labels from the trees, and those cases were counted as incorrect for the purpose of this evaluation.
(S
(NP (EX there))
(VP (VBZ is)
(NP (NP (DT no) (NN difficulty))
(SBAR (WHNP (WDT which))
(S (VP (MD-TrigAble can)
(RB-TrigNegation not)
(VP (VB be)
(VP-TargNOTAble (VBN-TargNOTAble solved)))))))))
In the 229 instances that we examined, there were 14 in which a light verb or noun was the correct syntactic target, but not the correct semantic target. [*Decision*]{} would be a better target than [*taken*]{} in [*The decision [**should**]{} be [**taken**]{} on delayed cases on the basis of merit.*]{} We counted sentences with semantically light targets as correct in our evaluation because our goal was to identify the synactic head of the target. The semantics of the target is a general issue, and we often find lexico-syntactic fluff between the trigger and the most semantically salient target in sentences like [*We succeeded in our goal of winning the war*]{} where “success in war” is the salient meaning.
With respect to recall, the tagger primarily missed special forms of negation in noun phrases and prepositional phrases: [*There was [**no**]{} place to seek shelter.*]{}; [*The buildings should be reconstructed, [**not**]{} with RCC, but with the wood and steel sheets.*]{} More complex constructional and phrasal triggers were also missed: [*President Pervaiz Musharraf has said that he will [**not rest unless**]{} the process of rehabilitation is completed.*]{} Finally, we discovered some omissions from our MN lexicon: [*It is not [**possible**]{} in the middle of winter to re-open the roads.*]{} Further annotation experiments are planned, which will be analyzed to close such gaps and update the lexicon as appropriate.
Providing a quantitative measure of recall was beyond the scope of this project. At best we could count instances of sentences containing trigger words that were not tagged. We are also aware of many cases of modality that were not covered such as the modal uses of the future tense auxiliary [*will*]{} as in [*That’ll be John*]{} (conjecture), [*I’ll do the dishes*]{} (volition), [*He won’t do it*]{} (non-volition), and [*It will accommodate five*]{} (ability) [@Larreya:2009]. However, because of the complexity and subtlety of modality and negation, it would be impractical to count every clause (such as the [*not rest unless*]{} clause above) that had a nuance of non-factivity.
Semantically-Informed Syntactic MT {#joshua}
==================================
This section describes the incorporation of our structured-based MN tagging into an Urdu-English machine-translation system using [*tree grafting*]{} for combining syntactic symbols with semantic categories (e.g. modality/negation). We note that a [*de facto*]{} Urdu MN tagger resulted from identifying the English MN trigger and target words in a parallel English-Urdu corpus, and then projecting the trigger and target labels to the corresponding words in Urdu syntax trees.
Refinement of Translation Grammars with Semantic Categories
-----------------------------------------------------------
We used synchronous context free grammars (SCFGs) as the underlying formalism for our statistical models of translation. SCFGs provide a convenient and theoretically grounded way of incorporating linguistic information into statistical models of translation, by specifying grammar rules with syntactic non-terminals in the source and target languages. We refine the set of non-terminal symbols so that they not only include syntactic categories, but also semantic categories.
Chiang re-popularized the use of SCFGs for machine translation, with the introduction of his hierarchical phrase-based machine translation system, Hiero. Hiero uses grammars with a single non-terminal symbol “X” rather than using linguistically informed non-terminal symbols. When moving to linguistic grammars, we use Syntax Augmented Machine Translation (SAMT) developed by Venugopal et al. . In SAMT the “X” symbols in translation grammars are replaced with nonterminal categories derived from parse trees that label the English side of the Urdu-English parallel corpus.[^1] We refine the syntactic categories by combining them with semantic categories. Recall this progression was illustrated in Figure \[Evolution-to-SIMT-Joshua\] on page .
We extracted SCFG grammar rules containing modality, negation and named entities using an extraction procedure that requires parse trees for one side of the parallel corpus. While it is assumed that these trees are labeled and bracketed in a syntactically motivated fashion, the framework places no specific requirement on the label inventory. We take advantage of this characteristic by providing the rule extraction algorithm with augmented parse trees containing syntactic labels that have semantic annotations grafted onto them so that they additionally express semantic information.
![A sentence on the English side of the bilingual parallel training corpus is parsed with a syntactic parser, and also tagged with our modality tagger. The tags are then [*grafted*]{} onto the syntactic parse tree to form new categories like VP-TargNOTAble and VP-TargRequire. Grafting happens prior to extracting translation rules, which happens normally except for the use of the augmented trees.[]{data-label="Parse-Tree-for-Sample-Sentence"}](graphics/tree-grafting-figure-new){width="\linewidth"}
Our strategy for producing semantically-grafted parse trees involves three steps:
1. The English sentences in the parallel training data are parsed with a syntactic parser. In our work, we used the lexicalized probabilistic context free grammar parser provided by Basis Technology Corporation.
2. The English sentences are MN-tagged by the system described above and named-entity-tagged by the Phoenix tagger [@richman-schone:2008:ACLMain].
3. The modality/negation and entity markers are grafted onto the syntactic parse trees using a tree-grafting procedure. The grafting procedure was implemented as a part of the SIMT effort. Details are further spelled out in Section \[algorithm\].
Figure \[Parse-Tree-for-Sample-Sentence\] illustrates how modality tags are grafted onto a parse tree. Note that while we focus the discussion here on the modality and negation, our framework is general and we were able to incorporate other semantic elements (specifically, named entities) into the SIMT effort.
Once the semantically-grafted trees have been produced for the parallel corpus, the trees are presented, along with word alignments (produced by the Berkeley aligner), to the rule extraction software to extract synchronous grammar rules that are both syntactically and semantically informed. These grammar rules are used by the decoder to produce translations. In our experiments, we used the Joshua decoder [@Li:2009], the SAMT grammar extraction software [@Venugopal2009], and special purpose-built tree-grafting software.
Figure \[derivation-with-modalities\] shows example semantic rules that are used by the decoder. The verb phrase rules are augmented with modality and negation, taken from the semantic categories listed in Table \[Modalitytable\]. Because these get marked on the Urdu source as well as the English translation, semantically enriched grammars also act as very simple named entity or MN taggers for Urdu. However, only entities, modality, and negation that occurred in the parallel training corpus are marked in the output.
![Example translation rules with tags for modality, negation, and entities combined with syntactic categories.[]{data-label="derivation-with-modalities"}](graphics/example-semantic-derivations-modality){width="3.5in"}
Tree-Grafting Algorithm {#algorithm}
-----------------------
The overall scheme of our tree-grafting algorithm is to match semantic tags to syntactic categories. There are two inputs to the process. Each is derived from a common text file of sentences. The first input is a list of standoff annotations for the semantically tagged word sequences in the input sentences, indexed by sentence number. The second is a list of parse trees for the sentences in Penn Treebank format, indexed by sentence number.
Table \[Modalitytable\] lists the modality/negation types that were produced by the MN tagger. For example, the sentence [*The students are able to swim*]{} is tagged as [*The students are $\langle$TrigAble able$\rangle$ to $\langle$TargAble swim$\rangle$*]{}. The distinction between “Negation” and “NOT” corresponds to the difference between negation that is inherently expressed in the triggering lexical item and negation that is expressed explicitly as a separate lexical item. Thus, [*I achieved my goal*]{} is tagged “Succeed” and [*I did not achieve my goal*]{} is tagged as “NOTSucceed,” but [*I failed to win*]{} is tagged as “SucceedNegation,” and [*I did not fail to win*]{} is tagged as “NOTSucceedNegation.”
---------------------- -------------------------
Require NOTRequire
Permit NOTPermit
Succeed NOTSucceed
SucceedNegation NOTSucceedNegation
Effort NOTEffort
EffortNegation NOTEffortNegation
Intend NOTIntend
IntendNegation NOTIntendNegation
Able NOTAble
AbleNegation NOTAbleNegation
Want NOTWant
WantNegation NOTWantNegation
Belief NOTBelief
BeliefNegation NOTBeliefNegation
Firm\_Belief NOTFirm\_Belief
Firm\_BeliefNegation NOTFirm\_BeliefNegation
Negation
---------------------- -------------------------
: Modality Tags with their Negated Versions. Note that [*Require*]{} and [*Permit*]{} are in a dual relation, and thus RequireNegation is represented as NOTPermit and PermitNegation is represented as NOTRequire.[]{data-label="Modalitytable"}
The tree-grafting algorithm proceeds as follows. For each tagged sentence, we iterate over the list of semantic tags. For each semantic tag, there is an associated word or sequence of words. For example, the modality tag TargAble may tag the word [*swim*]{}.
For each semantically tagged word, we find the parent node in the corresponding syntactic parse tree that dominates that word. For a word sequence, we find and compare the parent nodes for all of the words. Each node in the syntax tree has a category label. The following tests are then made and tree grafts applied:
- If there is a single node in the parse tree that dominates all and only the words with the semantic tag, graft the name of the semantic tag onto the highest corresponding syntactic constituent in the tree. For example, in Figure \[Parse-Tree-for-Sample-Sentence\], which shows the grafting process for modality tagging, the semantic tag TargNOTAble that “hand over” receives is grafted onto the VB node that dominates all and only the words “hand over.” Then the semantic tag TargNOTAble is passed up the tree to the VP node, which is the highest corresponding syntactic constituent.
- If the semantic tag corresponds to words that are adjacent daughters in a syntactic constituent, but less than the full constituent, insert a node dominating those words into the parse tree, as a daughter of the original syntactic constituent. The name of the semantic tag is grafted onto the new node and becomes its category label. This is a case of tree augmentation by node insertion.
- If a syntactic constituent selected for grafting has already been labeled with a semantic tag, overlay the previous tag with the current tag. We chose to tag in this manner simply because our system was not set up to handle the grafting of multiple tags onto a single constituent. An example of this occurs in the sentence “The Muslims had obtained Pakistan.” If the NP node dominating [ *Pakistan*]{} is grafted with a named entity tag such as NP-GPE, we overlay this with the NP-TargSucceed tag in a modality tagging scheme.
- In the case of a word sequence, if the words covered by the semantic tag fall across two different syntactic constituents, do nothing. This is a case of crossing brackets.
![Results for a range of experiments conducted during the SIMT effort show the score for our top-performing baseline systems derived from a hierarchical phrase-based model (Hiero). Substantial improvements obtained when syntax was introduced along with feature functions (FFs) and further improvements resulted from the addition of semantic elements. The scores are lowercased Bleu calculated on the held-out devtest set.[]{data-label="joshua-experiments"}](graphics/joshua-experiments-scores-revised){width="3.5in"}
Our tree-grafting procedure was simplified to accept a single semantic tag per syntactic tree node as the final result. The algorithm keeps the last tag seen as the tag of precedence. In practice, we established a precedence ordering for modality/negation tags over named entity tags by grafting named entity tags first and modality/negation second. Our intuition was that, in case of a tie, finer-grained verbal categories would be more helpful to parsing than finer-grained nominal categories.[^2] In cases where a word was tagged both as a MN target and a MN trigger, we gave precedence to the target tag. This is because, while MN targets vary, MN triggers are generally identifiable with lexical items. Finally, we used the simplified specificity ordering of MN tags described in Section \[linguistic-simplifications\] to ensure precedence of more specific tags over more general ones. Table \[Modalitytable\] lists the modality/negation types from highest (Require modality) to lowest (Negation) precedence.[^3]
SIMT Results {#results}
------------
We evaluated our tree grafting approach by performing a series of translation experiments. Each version of our translation system was trained on the same bilingual training data. The bilingual parallel corpus that we used was distributed as part of the 2008 NIST Open Machine Translation Evaluation Workshop.[^4] The training set contained 88,108 Urdu-English sentence pairs, and a bilingual dictionary with 113,911 entries. For our development and test sets, we split the NIST MT-08 test set into two portions (with each document going into either test or dev, and preserving the genre split). Our test set contained 883 Urdu sentences, each with four translations into English, and our dev set contained 981 Urdu sentences, each with four reference translations. To extract a syntactically informed translation model, we parsed the English side of the training corpus using a Penn Treebank trained parser [@Miller:1998]. For the experiments that involved grafting named entities onto the parse trees, we tagged the English side of the training corpus with the Phoenix tagger [@richman-schone:2008:ACLMain]. We word-aligned the parallel corpus with the the Berkeley aligner. All models used a 5-gram language model trained on the English Gigaword corpus (v5) using the SRILM toolkit with modified KN smoothing. The Hiero translation grammar was extracted using the Joshua toolkit [@Li:2009]. The other translation grammars were extracted using the SAMT toolkit [@Venugopal2009].
Figure \[joshua-experiments\] gives the results for a number of experiments conducted during the SIMT effort.[^5] The experiments are broken into three groups: baselines, syntax, and semantics. To contextualize our results we experimented with a number of different baselines that were composed from two different approaches to statistical machine translation—phrase-based and hierarchical phrase-based SMT—along with different combinations of language model sizes and word aligners. Our best-performing baseline was a Hiero model. The Bleu score for this baseline on the development set was 22.9 Bleu points.
![An example of the improvements to Urdu-English translation before and after the SIMT effort. Output is from the baseline Hiero model, which does not use linguistic information, and from the final model, which incorporates syntactic and semantic information. []{data-label="final-output"}](graphics/final-output-full){width="4.7in"}
After experimenting with syntactically motivated grammar rules, we conducted experiments on the effects of incorporating semantic elements (e.g., named entities and modality/negation) into the translation grammars. In our devtest set our taggers tagged on average 3.5 named entities per sentence and 0.35 modality/negation markers per sentence. These were included by grafting modality, negation, and named-entity markers onto the parse trees. Individually, each of these made modest improvements over the syntactically-informed system alone. Grafting named entities onto the parse trees improved the Bleu score by 0.2 points. Modality/negation improved it by 0.3 points. Doing both simultaneously had an additive effect and resulted in a 0.5 Bleu score improvement over syntax alone. This improvement was the largest improvement that we got from anything other than the move from linguistically naïve models to syntactically informed models.
We used bootstrap resampling to test whether the differences in Bleu scores were statistically significant [@koehn:2004:EMNLP]. All of the results were a significant improvement over Hiero (at $p\leq0.01$). The difference between the syntactic system and the syntactic system with NEs as not significant ( $p=0.38$). The differences between the syntactic system and the syntactic system with MN, and between the syntactic system and the syntactic system with both MN and NEs were both significant at ($p\leq0.05$).
Figure \[final-output\] shows example output from the final SIMT system in comparison to the pre-SIMT results and the translation produced by a human (reference). An error analysis of this example output illustrates that SIMT enhancements have resulted in the elimination of misleading translation output in several cases:
1. [**pre-SIMT**]{}: China had the experience of Pakistan’s first nuclear bomb.\
[**SIMT**]{}: China has the first nuclear bomb test.\
[**reference**]{}: China has conducted the experiment of Pakistan’s first nuclear bomb.
2. [**pre-SIMT**]{}: the nuclear bomb in 1998 that Pakistan may experience\
[**SIMT**]{}: the experience of the atom bomb Pakistan in May 1998\
[**reference**]{}: the atom bomb, whose experiment was done in 1998 by Pakistan
3. [**pre-SIMT**]{}: He said that it is also present proof of that Dr. Abdul Qadeer Khan after the Chinese design\
[**SIMT**]{}: He said that there is evidence that Dr. Abdul Qadeer Khan has also used the Chinese design\
[**reference**]{}: He said that the proof to this also exists in that Dr. Abdul Qadeer Khan used the Chinese design
The article pertains to claims by Thomas Reid that China allowed Pakistan to detonate a nuclear weapon at its test site. However, in the first example above, the reader is potentially misled by the pre-SIMT output to believe that Pakistan launched a nuclear bomb on China. The SIMT output leaves out the mention of Pakistan, but correctly conveys the firm belief that the bomb event is a [ *test*]{} (closely resembling the term [*experiment*]{} in the human reference), not a true bombing event. This is clearly an improvement over the misleading pre-SIMT output.
In the second example, the pre-SIMT output misleads the reader to believe that Pakistan is (or will be) attacked, through the use of the phrase [*may experience*]{}, where [*may*]{} is poorly placed. (We note here that this is a date translation error, i.e., the month of [*May*]{} should be next to the year 1998, further adding to the potential for confusion.) Unfortunately, the SIMT output also uses the term [*experience*]{} (rather than [*experiment*]{}, which is in the human reference), but in this case the month is correctly positioned in the output, thus eliminating the potential for confusion with respect to the modality. The lack of a modal appropriately neutralizes the statement so that it refers to an abstract event associated with the atom bomb, rather than an attack on the country.
In the third example, where the Chinese design used by Dr. Abdul Qandeer Khan is argued to be proof of the nuclear testing relationship between Pakistan and China, the first pre-SIMT output potentially leads the reader to believe that Dr. Abdul Qadeer is after the Chinese design (not that he actually used it), whereas the SIMT output conveys the firm belief that the Chinese design has been used by Dr. Abdul Qadeer. This output very closely matches the human reference.
Note that even in the title of the article, the SIMT system produces much more coherent English output than that of the linguistically naïve system. The figure also shows improvements due to transliteration, which are described in [@Irvine-PBML]. The scores reported in Figure \[joshua-experiments\] do not include transliteration improvements.
Conclusions and Future Work {#conclusions}
===========================
We developed a modality/negation lexicon and a set of automatic MN taggers, one of which—the structure-based tagger—results in 86% precision for tagging of a standard LDC data set. The MN tagger has been used to improve machine translation output by imposing semantic constraints on possible translations in the face of sparse training data. The tagger is also an important component of a language-understanding module for a related project.
We have described a technique for translation that shows particular promise for low-resource languages. We have integrated linguistic knowledge into statistical machine translation in a unified and coherent framework. We demonstrated that augmenting hierarchical phrase-based translation rules with semantic labels (through “grafting”) resulted in a 0.5 Bleu score improvement over syntax alone.
Although our largest gains were from syntactic enrichments to the Hiero model, demonstrating success on the integration of semantic aspects of language bodes well for additional improvements based on the incorporation of other semantic aspects. For example, we hypothesize that incorporating relations and temporal knowledge into the translation rules would further improve translation quality. The syntactic grafting framework is well-suited to support the exploration of the impact of many different types of semantics on MT quality, though in this article we focused on exploring the impact of modality and negation.
An important future study is one that focuses on demonstrating whether further improvements in modality/negation identification are likely to lead to further gains in translation performance. Such a study would benefit from the inclusion of a more detailed manual evaluation to determine if modality and negation is adequately conveyed in the downstream translations. This work would be additionally enhanced through experimentation on other language pair(s) and larger corpora.
The work presented here represents the first small steps toward a full integration of MT and semantics. Efforts underway in DARPA’s GALE program have already demonstrated the potential for combining MT and semantics (termed [*distillation*]{}) to answer the information needs of monolingual speakers using multilingual sources. Proper recognition of modalities and negation is crucial for handling those information needs effectively. In previous work, however, semantic processing proceeded largely independently of the MT system, operating only on the translated output. Our approach is significantly different in that it combines syntax, semantics, and MT into a single model, offering the potential advantages of joint modeling and joint decision-making. It would be interesting to explore whether the integration of MT with syntax and semantics can be extended to provide a single-model solution for tasks such as cross-language information extraction and question answering, and to evaluate our integrated approach, e.g., using GALE distillation metrics.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank Aaron Phillips for help with conversion of the output of the entity tagger for ingest by the tree-grafting program. We thank Anni Irvine and David Zajic for their help with experiments on an alternative Urdu modality/negation tagger based on projection and training an HMM-based tagger derived from Identifinder [@Bikel:1999]. For their helpful ideas and suggestions during the development of the modality framework, we are indebted to Mona Diab, Eduard Hovy, Marge McShane, Teruko Mitamura, Sergei Nirenburg, Boyan Onyshkevych, and Owen Rambow. We also thank Basis Technology Corporation for their generous contribution of software components to this work. This work was supported, in part, by the Johns Hopkins Human Language Technology Center of Excellence (HLTCOE), by the National Science Foundation under grant IIS-0713448, and by BBN Technologies under GALE DARPA/IPTO Contract No. HR0011-06-C-0022. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of the sponsor.
[^1]: For non-constituent phrases, composite CCG-style categories are used [@Steedman:1999].
[^2]: In testing we found that grafting named entities first and modality/negation last yielded a slightly higher Bleu score than the reverse order.
[^3]: Future work could include exploring additional methods of resolving tag conflicts or combining tag types on single nodes, e.g. by inserting multiple intermediate nodes (effectively using unary rewrite rules) or by stringing tag names together.
[^4]: <http://www.itl.nist.gov/iad/mig/tests/mt/2008/doc/>
[^5]: These experiments were conducted on the devtest set, containing 883 Urdu sentences (21,623 Urdu words) and four reference translations per sentence. The Bleu score for these experiments is measured on uncased output.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Strontium-doped lanthanum nickelates La$_{2-x}$Sr$_{x}$NiO$_{4+\delta}$ (LSNO) and cuprates La$_{2-x}$Sr$_{x}$CuO$_{4}$ show remarkably different conductivity as insulating and superconductivity in the similar electronic structure and physical properties, and it is important to clarify the nature of hole doping. We performed resonant inelastic X-ray scattering (RIXS) study of LSNO at oxygen $K$-edge in order to investigate the charge excitations of doped holes. It newly found that the observed charge excitations are strongly localized from the little energy-dispersion in the O $K$-edge RIXS spectra. This localized character is associated with the absence of the superconductivity and could be an origin of wider hole density of charge-stripe ordered phase in nickelates than that in cuprates. Furthermore, for finite momentum, conspicuous energy position changes with temperature were observed at a charge-ordering transition temperature. In addition with the previous spectroscopic techniques, the possible origin of the localized charge excitations are also discussed.'
author:
- Kohei Yamagami
- Kenji Ishii
- Yasuyuki Hirata
- Keisuke Ikeda
- Jun Miyawaki
- Yoshihisa Harada
- Masanori Miyazaki
- Shun Asano
- Masaki Fujita
- Hiroki Wadati
title: |
Localized Charge Excitations of La$_{\bm{2-x}}$Sr$_{\bm x}$NiO$_{\bm{4+\delta}}$\
Revealed by Oxygen $\bm K$-edge Resonant Inelastic X-ray Scattering
---
Effects of carrier doping into insulating transition-metal oxides are a central issue in strongly correlated electron systems. Parent insulators are categorized into the following two Mott insulators or charge-transfer insulators [@MottCTGap; @TMOreview]. In the former, the on-site Coulomb repulsion ($U$) of the $d$ electrons is smaller than the energy of charge transfer ($\Delta$) from the oxygen 2$p$ to the transition-metal $d$ levels ($U<\Delta$), while, in the latter, $\Delta<U$ which is the case for the layered perovskite lanthanum cuprate La$_{2}$CuO$_{4}$ (LCO) as well as its isostructural nickelate La$_{2}$NiO$_{4+\delta}$ (LNO) [@OptLCO; @LSNOopticalI; @XPSLSNO]. These parent materials have weak inter-layer coupling of two-dimensional Heisenberg antiferromagnets with $S = 1$ (Ni$^{2+}$:$d^{8}$) for LNO or $S=1/2$ (Cu$^{2+}$:$d^{9}$) for LCO local spins, which shows three-dimensional antiferromagnetic behavior at 330 K and 270 K respectively [@LNONeel; @LCONeel]. When holes are doped into LNO and LCO by the Sr$^{2+}$ ion substitution for the La$^{3+}$ ions, La$_{2-x}$Sr$_{x}$NiO$_{4+\delta}$ (LSNO) and La$_{2-x}$Sr$_{x}$CuO$_{4}$ (LSCO), they predominantly occupy the oxygen 2$p$ orbitals, revealed by oxygen $K$-edge X-ray absorption spectroscopy (XAS) [@LSCOoxyXASI; @LSCOoxyXASII; @LSNOoxyXASI; @LSNOoxyXASII].
As an extension of XAS, resonant inelastic X-ray scattering (RIXS) allows to discuss the element- and bulk-sensitive charge and spin excitations for finite momentum [@RIXSreviewI; @RIXSreviewII]. In particular, O $K$-edge (1$s$-2$p$) RIXS is able to provide experimentally and theoretically information such as Zhang-Rice singlet, crystal field, and charge transfer excitations by electron correlation, complementary to $K$-edge RIXS of 3$d$ transition metal [@OKedgeRIXSThe; @OKedgeRIXSI; @NiORIXS; @OKedgeRIXSII; @OKedgeRIXSIII; @OKedgeRIXSIV; @OKedgeRIXSV; @LSCOoxyRIXS; @OKedgeRIXSVI; @OKedgeRIXSVII]. In the case of LSCO, O $K$-edge RIXS successfully probed charge dynamics originating from doped hole states which revealed the charge excitations with energy-momentum dispersions on the order of the transfer energy ($\sim$0.4 eV) [@LSCOoxyRIXS]. On the other hand, in the case of LSNO, Ni $K$-edge (1$s$-4$p$) RIXS studies have observed the energy-momentum dependent charge excitation at 1.5 eV due to the doped holes [@NiKedgeRIXS; @NiKedgeRIXStemp], whereas the low-energy charge excitation less than 1.0 eV observed by optical measurements [@LSNOopticalI; @LSNOopticalII; @LSNOopticalIII] was not observed due to the strong elastic peak intensity. O $K$-edge RIXS is expected to give the energy-momentum dispersion of the low-energy charge excitation and help us to understand the effects of hole doping in charge-transfer insulators.
With the discovery of superconductivity in an infinite-layer nickelate Nd$_{0.8}$Sr$_{0.2}$NiO$_{2}$ single-crystal thin film [@NSNOthinfilm], it is important to clarify the reason why LSNO does not exhibit superconductivity from the direct observation of hole doping. LSNO behaves as an insulator up to $x\sim0.8$ [@LSNOResistivity], while LSCO becomes superconducting at $x\sim0.05$ [@LSCOResistivity], suggesting the different properties of doped hole states between the two systems. The charge-spin stripe ordering derived from Ni$^{2+}$ and Ni$^{3+}$ sites in the NiO$_{2}$ plane observed by electron, X-ray and neutron diffraction techniques around $0.22\le n_{h}\le0.50$ ($n_{h}=x+2\delta$) [@NeuCOI; @NeuCOII; @EleCO; @XraCOI; @XraCOII]. Ordering transition temperatures, $T_{\rm CO}$ and $T_{\rm SO}$, show maximum ($T_{\rm CO}$ = 240 K, $T_{\rm SO}$ = 180 K) when $n_{h}$ = 0.33 [@TCOTSO]. Angle-resolved photoemission spectroscopy (ARPES) for metallic layered nickelate ($n_{h}>1.0$) indicates that the Ni $d_{x^{2}-y^{2}}$ and $d_{3z^{2}-r^{2}}$ multi-band feature with varying orbital character works against the emergence of the superconductivity [@ESNOARPESI; @ESNOARPESII]. However, since LSNO with charge-spin stripe ordering behaves as an insulator and is not suitable for ARPES, the energy-momentum dispersion of the charge dynamics as well as the spin dynamics is required to clarify the relationship among the charge ordering, and superconductivity.
In this paper, we performed O $K$-edge RIXS measurements of LSNO to investigate charge excitations of doped holes in momentum space. It revealed that the observed charge excitations are consistent with those in the optical conductivity and strongly localized from the energy-momentum dispersionless O $K$-edge RIXS spectra. Localized charge excitations are related to the absence of superconductivity and the wider hole density of charge-stripe ordered phase in the nickelates than that in the cuprates. Furthermore, for finite momentum, the energy position changes are observed at $T_{\rm CO}$. We discussed the possible origin of the localized charge excitations.
The O $K$-edge XAS and RIXS experiments were performed by using the HORNET spectrometer at BL07LSU of SPring-8 [@BL07LSU; @HORNET]. The X-ray spectra of single crystals of LSNO with ($x,\delta,n_{h}$) = ($0.00,0.05,0.10$), ($0.33,0.00,0.33$), prepared by the traveling solvent floating zone method, were measured at 30$-$350 K. The O $K$-edge XAS was measuremed in the partial fluorescence yield (PFY) mode by using a silicon drift detector. The RIXS spectra were taken with energy resolution of $\sim$140 meV, determined by the measurement of an elastic peak of a gold plate. The crystals were cleaved along the crystalline $ab$-plane (NiO$_{2}$ plane) in the air before the measurement. As shown in Fig. \[Fig.1\](a), the vertical-polarized X-rays ($\bm E$) with respected to the scattering plane, defined as incident and scattered X-ray vectors, were irradiated on the $ab$-plane (${\bm E}//ab$). The scattering angle (2$\theta$) was fixed to 135$^{\circ}$ and the crystalline $c$-axis was kept parallel to the scattering plane. The momentum transfer (${\bm q}={\bm k}_{\rm in}-{\bm k}_{\rm out}$) was tuned by changing the sample angles $\omega$ and $\phi$, where $\omega$ controls the magnitude of the projected momentum ($\bm q_{//}$) in $ab$-plane and the azimuthal angle $\phi$ determines the scattering plane. In our measurements, the \[$h, 0$\] and \[$h, h$\] directions were reached by fixing $\phi$ at 0$^{\circ}$ and 45$^{\circ}$ respectively and the sign of $\bm q_{//}$ was positive when $\omega$ was rotated clockwise from the specific angle corresponding to $\bm q_{//}$ = 0. The value of $\bm q_{//}$ were calculated from the degree of $\omega$, $|{\bm k}_{\rm in}|$, and the tetragonal lattice parameters of $a$ = $b$ = 3.82 Å [@LSNOlatticeI; @LSNOlatticeII], and expressed with the unit of $2\pi/a$. All the RIXS spectra were normalized by the total integrated RIXS intensity after subtracting the constant background.
![(Color online) (a) Experimental geometry of O $K$-edge RIXS measurements. ($h\nu_{\rm in}$, ${\bm k}_{\rm in}$) and ($h\nu_{\rm out}$, ${\bm k}_{\rm out}$) denote the photon energy and wave number vectors of incident and scattered X-rays respectively. (b) $n_{h}$ dependence of O $K$-edge XAS spectra for LSNO with ${\bm E}//ab$ in PFY mode. (c) $h\nu_{\rm in}$ dependence of O $K$-edge RIXS spectra at $\bm q_{//}$ = (0.13, 0). $h\nu_{\rm in}$ at 528.7 eV and 529.8 eV correspond \[A\] and \[B\] in O $K$-edge XAS spectra. The vertical solid lines at 0.6 eV and 1.2 eV denote the peak position of the Low- and High-E charge excitations, as discussed in the text.[]{data-label="Fig.1"}](Fig.1.eps){width="8.3cm"}
{width="18.1cm"}
Figure \[Fig.1\](b) shows the O $K$-edge XAS spectra with ${\bm E}//ab$ for LSNO with $n_{h}$ = 0.10 and 0.33. The spectra were normalized at 536 eV corresponding to O 1$s$ → 2$p$ main peak [@RSNOoxyXAS]. Since dipole selection rules indicate that only 1$s$ → 2$p_{x,y}$ transition are allowed for ${\bm E}//ab$, the Ni 3$d_{x^{2}-y^{2}}$ final state at 532.5 eV and the pre-edge peaks, labeled as A and B, at 528.7 eV and 529.8 eV were observed. From ligand field theory on a NiO$_{6}$ cluster, the peaks A and B were assigned as the high-spin ($S=1$) and the low-spin ($S=0$) states of Ni$^{2+}$ ions, respectively [@LSNOoxyXASII]. We confirmed the number of $n_{h}$ from the spectral intensity $\mu(n_{h})$ integrated from 525 to 530.9 eV, and obtained the value of $\mu(0.1)/\mu(0.33)=0.74$, which is consistent with the previous work [@LSNOoxyXASII].
In order to investigate the excitations of doped holes, we performed O $K$-edge RIXS at the incident photon energy ($h\nu_{\rm in}$) in pre-edge peaks from 528.4 eV to 529.8 eV. Figure \[Fig.1\](c) shows the $h\nu_{\rm in}$ dependence of O $K$-edge RIXS spectra of LSNO with $n_{h}$ = 0.10 and 0.33 at finite $\bm q_{//}$ as a function of the energy loss ($h\nu_{in}-h\nu_{out}$). The high-energy peak above 3 eV is assigned as a fluorescence due to the continuous shift to higher energy sides. On the other hand, excitations at 0.6 eV, 1.2 eV, and 2.0 eV corresponding to the energy scale below the charge transfer gap ($\sim$4.0 eV) for LNO [@XPSLSNO] are assigned as the Raman components due to the absence of the energy shift with $h\nu_{\rm in}$. These excitations are consistent with those of optical conductivity for LSNO [@LSNOopticalI; @LSNOopticalII; @LSNOopticalIII]. In addition, the observed charge excitations are in good agreement with the numerically exact-diagonalization based on the Hubbard model for two-dimensional doped nickelate [@Opticaltheory]. Their Hamiltonian take into account electron hopping ($t\sim0.3$ eV), on-site Coulomb interaction, and Hund’s-rule coupling ($J\sim1.4$ eV) between electrons in $e_{g}$ orbitals under octahedral symmetry, whereas crystal field splitting between $e_{g}$ orbitals, O 2$p$ orbitals, and electron-phonon interaction are neglected. It has been revealed that charge excitations at 0.6 eV, 1.2 eV, and 2.0 eV reflect the motion of carriers which keeps the $S=1$ configuration and changes doubly occupied sites from $S=1$ to $S=0$ configurations respectively. Therefore, we demonstrate that O $K$-edge RIXS detects the charge excitations originating from doped holes within LSNO.
From O $K$-edge RIXS spectrum at $h\nu_{\rm in}$ = 528.7 eV in Fig. \[Fig.1\](b), the intensity of the charge excitation at 0.6 eV clearly evolves with increasing $n_{h}$, whereas the intensity change of the charge excitation at 1.2 eV is small. Next, let us discuss the $\bm q_{//}$ and $n_{h}$ dependence of the charge excitations, especially at 0.6 eV and 1.2 eV (hereafter Low-E and High-E charge excitations), of LSNO at $h\nu_{\rm in}$ = 528.7 eV. In Figs. \[Fig.2\](a)-(c), at $T$ = 30 K, O $K$-edge RIXS spectra for LSNO with $n_{h}$ = 0.10, 0.33 along \[$h, 0$\] and \[$h, h$\] directions show little change with $\bm q_{//}$, suggesting that the observed charge excitations have the energy-momentum dispersionless. To investigate the behavior of these dispersions, we fitted the spectra by using multiple Gaussian. We considered not only the Low- and High-E charge excitations, but also the elastic peak, the low-energy tail in the elastic peak at 0.2 eV, the 2.0 eV charge excitation, and high-energy tail. The low-energy tail is due to phonon and magnetic components reported by the Raman scattering and Ni $L_{3}$-edge (2$p$-3$d$) RIXS [@LNORaman; @LSNORaman; @NiLedgeRIXS]. Figure \[Fig.2\](d) shows that multiple Gaussians can reproduce the experimental spectra. Note that the peak position of Low-E charge excitation for $n_{h}$ = 0.10 is estimated from the peak position of the spectral weight after subtraction of the others. We found that the Low-E charge excitation in $n_{h}$ = 0.33 is shifted to higher energy loss side by $\sim$0.1 eV than $n_{h}$ = 0.10, consistent with the hole doping dependence of the optical conductivity [@LSNOopticalI]. This energy shift was also observed at other $\bm q_{//}$, as shown in Figs. \[Fig.2\](e) and (f) (the fitting results for O $K$-edge RIXS spectra are summarized in supplemental materials [@Supplemental]). On the other hand, the peak position of Low- and High-E charge excitations for each $n_{h}$ are independent of $\bm q_{//}$. Therefore, we consider that the charge excitations of LSNO have localized characters.
The energy-momentum dispersionless charge excitations of doped holes for LSNO with $n_{h}=0.33$ are is quite different nature from those for LSCO [@LSCOoxyRIXS]. This localized charge excitations could lead to wider hole density of these charge-stripe ordered phases in nickelates than those in cuprates [@NSNOCO] and is related to the reason why LSNO does not show superconductivity. In the following, we argued possible explanations for the localized charge excitations of LSNO with $n_{h}=0.33$. One is that the doped holes form polarons originating from electron-phonon coupling. A previous optical studies have been suggested that the electron-phonon coupling is larger of LSNO than that of LSCO [@EPcouplingI]. Indeed, the four-band model taking Ni $3d_{x^{2}-y^{2}}, 3d_{3z^{2}-r^{2}}$ and O 2$p_{x,y}$ orbitals into account predicts that the polaron becomes stable if $\Delta\simeq U$ [@EPcouplingII], and X-ray photoemission (XPS) proposes that the polaron can exist stably in LSNO because of $\Delta\le U$ for LSNO [@XPSLSNO] and $\Delta\ll U$ for LSCO [@XPSLSCO]. This different coupling strength corresponds to the size of polarons, which is smaller for LSNO than that for LSCO. Our O $K$-edge RIXS results implies that the localized charge excitations reflects the presence of small polaron of LSNO. The other possible origin is the magnitude of the spin in parent materials. A lot of optical conductivity, XAS, and ARPES studies suggests that the doped hole enters into O 2$p_{\sigma}$ orbital hybridized with $d_{x^{2}-y^{2}}$ orbital [@LSNOoxyXASII; @ESNOARPESI; @ESNOARPESII; @RSNOoxyXAS; @LSNOopticalI; @LSNOopticalII; @LSNOopticalIII]. In addition, XPS shows that the spin of the doped hole nearest to the Fermi level is found to be antiparallel to the Ni$^{2+}$ spins [@XPSLSNO]. The hopping of the $d_{x^{2}-y^{2}}$-symmetry hole to the neighboring site is energetically unfavorable because of the antiferromagnetic coupling with the Ni spin. Furthermore, the quantum spin fluctuation of the antiferromagnetic states in LNO ($S=1$) is much weaker than in LCO ($S=1/2$), and it would lead to lower mobility to the doped holes within LSNO. The hole localization due to spin-polaron effects has also been observed by electron spin resonance, and suggested that this coupling of the hole to the spin waves could lead to a self-trapping of charges [@LNOESR].
![(Color online) $T$ dependence of O $K$-edge RIXS spectra for LSNO with $n_{h}$ = 0.33. (a) The spectral changing at $\bm q_{//}=(0.07, 0.07)$. The peak plots of charge excitations obtained by the multiple Gaussian and $2\Delta_{ \rm C}$ cited Ref. [@LSNOopticalII] are also shown in (b). (c) $\bm q_{//}=(h, h)$ dependence of O $K$-edge RIXS spectra at each temperatures, and the peak positions of charge excitations obtained by multiple Gaussian fittings are shown in (d). The horizontal dashed lines denote the energy positions of charge excitations. The error bar of the peak energy is within the size of the markers.[]{data-label="Fig.3"}](Fig.3.eps){width="8.7cm"}
As mentioned in the introduction, LSNO with $n_{h}$ = 0.33 has the charge-spin stripe ordering at $T_{\rm CO}$ = 240 K, $T_{\rm SO}$ = 180 K, while no-ordering exists in $n_{h}$ = 0.10 [@TCOTSO]. In order to investigate the behavior of the charge excitations of doped holes for LSNO with $n_{h}$ = 0.33 at $T_{\rm CO}$ and $T_{\rm SO}$, we performed temperature dependent O $K$-edge RIXS measurements at finite $\bm q_{//}$. In Fig. \[Fig.3\](a), O $K$-edge RIXS spectra of LSNO with $n_{h}$ = 0.33 at $\bm q_{//}$ = (0.07, 0.07) indicate that the Low-(High-)E charge excitation is dependent on (independent of) the temperature from 30 K to 350 K. For instance, compared between O $K$-edge RIXS spectra at 30 K and 350 K, the energy shift of $\sim$0.2 eV of the Low-E excitation was observed. The peak positions of charge excitations obtained by multiple Gaussian as a function of the temperature are summarized at Fig. \[Fig.3\](b). The details for multiple Gaussian fitting are summarized at supplemental materials [@Supplemental]. The peak position of the High-E excitation is mostly constant, while the Low-E charge excitation is clearly shifted to the lower energy loss side from 0.66 eV below $T_{\rm SO}$ to 0.45 eV above $T_{\rm CO}$. These temperature dependence was also observed at the others ${\bm q_{//}}=(h, h)$ in Fig. \[Fig.3\](c), and the charge excitations have localized character because the peak positions plotted in Fig. \[Fig.3\](d) are independent of $\bm q_{//}$. In all values of ${\bm q_{//}}$, it was revealed that the size of the energy shift 0.22 eV of Low-E charge excitation is consistent with the magnitude of the charge gap ($2\Delta_{\rm C}$) estimated from optical conductivity [@LSNOopticalII]. The ratio of the energy shift to the transition temperature ($2\Delta_{\rm C}/k_{B}T_{\rm CO}$) is $\sim$11, which is larger than the gap values for the charge- and spin-density-wave transitions $\sim$3.5 [@CDWSDW]. The temperature dependence of the Low-E charge excitation is not seen in O $K$-edge RIXS for LSCO [@LSCOoxyRIXS], and this difference is associate with the value of the optical gap reported by optical conductivity [@OptLCO; @LSNOopticalI; @LSNOopticalII].
In summary, we performed O $K$-edge RIXS measurements for LSNO with $n_{h}$ =0.10 and 0.33 and successfully observed the charge excitations of doped holes. The charge excitations are independent of the momentum transfer, reflecting the localized character. This nature is different from that for LSCO, showing that the localized charge excitations are associated with the absence of the superconductivity and the formation of the polaron and the spin fluctuation in LSNO. We demonstrate that O $K$-edge RIXS is an effective method to observe the charge excitations originating from doped holes within charge transfer insulators and expect the observation of delocalized charge excitations of an infinite-layer nickelate superconductor from momentum-resolved measurements.
We thank K. Yamazoe, and U. J. Ralph for supporting the experiments and K. Takubo for the flutiful discussions. The resonant inelastic X-ray scattering measurements at SPring-8 BL07LSU was carried out by the joint research in the Synchrotron Radiation Research Organization and the Institute for Solid State Physics, the University of Tokyo (Proposal No. 2018A7558). K. Yamagami would like to acknowledge the support from the Motizuki Fund of Yukawa Memorial Foundation.
[99]{} J. Zaanen, G. A. Sawatzky, and J. W. Allen, Phys. Rev. Lett. [**55**]{}, 418 (1985). M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys. [**70**]{}, 1039 (1998). S. Uchida, T. Ido, H. Takagi, T. Arima, Y. Tokura, and S. Tajima, Phys. Rev. B [**43**]{}, 7942 (1991). T. Ido, K. Magoshi, H. Eisaki, and S. Uchida, Phys. Rev. B [**44**]{}, 12094 (1991). H. Eisaki, S. Uchida, T. Mizokawa, H. Namatame, A. Fujimori, J. van Elp, P. Kuiper, G. A. Sawatzky, S. Hosoya, and H. K. Yoshida, Phys. Rev. B [**45**]{}, 12513 (1992). K. Nakajima, K. Yamada, S. Hosoya, T. Omata, and Y. Endoh, J. Phys. Soc. Jpn. [**62**]{}, 4438 (1993). D. Vaknin, S. K. Sinha, D. E. Moncton, D. S. Johnston, J. M. Newsan, C. R. Safinya, and H. E. King, Jr., Phys. Rev. Lett. [**58**]{}, 1196 (1987). C. T. Chen, F. Sette, Y. Ma, M. S. Hybertsen, E. B. Stechel, W. M. C. Foulkes, M. Schulter, S-W. Cheong, A. S. Cooper, L. W. Rupp, Jr., B. Batlogg, Y. L. Soo, Z. H. Ming, A. Krol, and Y. H. Kao, Phys. Rev. Lett. [**66**]{}, 104 (1991). C. T. Chen, L. H. Tjeng, J. Kwo, H. L. Kao, P. Rudolf, F. Sette, and R. M. Fleming Phys. Rev. Lett. [**68**]{}, 2543 (1992). P. Kuiper, J. van Elp, G. A. Sawatzky, A. Fujimori, S. Hosoya, and D. M. de Leeuw Phys. Rev. B [**44**]{}, 4570 (1991). E. Pellegrin, J. Zaanen, H.-J. Lin, G. Meigs, C. T. Chen, G. H. Ho, H. Eisaki, and S. Uchida, Phys. Rev. B [**53**]{}, 10667 (1996). L. J. P. Ament, M. van Veenendaal, M. van Veenendaal, J. P. Hill, and J. van den Brink, Rev. Mod. Phys. [**83**]{}, 705 (2011). K. Ishii, T. Toyama, and J. Mizuki, J. Phys. Soc. Jpn. [**82**]{}, 021015 (2013). S. Wakimoto, H. Kimura, K. Ishii, K. Ikeuchi, T. Adachi, M. Fujita, K. Kakurai, Y. Koike, J. Mizuki, Y. Noda, K. Yamada, A. H. Said, and Yu. Shvyd’ko, Phys. Rev. Lett. [**102**]{}, 157001 (2009). L. Simonelli, S. Huotari, M. Filippi, N. L. Saini, and G. Monaco, Phys. Rev. B [**81**]{}, 195124 (2010). T. Katsufuji, T. Tanabe, T. Ishikawa, Y. Fukuda, T. Arima, and Y. Tokura, Phys. Rev. B [**54**]{}, 14230 (1996). J. H. Jung, D.-W. Kim, T. W. Noh, H. C. Kim, H.-C. Ri, S. J. Levett, M. R. Lees, D. McK. Paul, and G. Balakrishnan, Phys. Rev. B [**64**]{}, 165106 (2001). K. Okada and A. Kotani, Phys. Rev. B [**65**]{}, 144530 (2002). Y. Harada, K. Okada, R. Eguchi, A. Kotani, H. Takagi, T. Takeuchi, and S. Shin, Phys. Rev. B [**66**]{}, 165104 (2002). L.-C. Duda, T. Schmitt, M. Magnuson, J. Forsberg, A. Olsson, and J. Nordgren, K. Okada, and A. Kotani, Phys. Rev. Lett. [**96**]{}, 067402 (2006). V. Bisogni, L. Simonelli, L. J. P. Ament, F. Forte, M. Moretti Sala, M. Minola, S. Huotari, J. van den Brink, G. Ghiringhelli, N. B. Brookes, and L. Braicovich, Phys. Rev. B [**85**]{}, 214527 (2012). V. Bisogni, M. Moretti Sala, A. Bendounan, N. B. Brookes, G. Ghiringhelli, and L. Braicovich, Phys. Rev. B [**85**]{}, 214528 (2013). C. Monney, V. Bisogni, K.-J. Zhou, R. Kraus, V. N. Strocov, G. Behr, J. Málek, R. Kuzian, S.-L. Drechsler, S. Johnston, A. Revcolevschi, B. Büchner, H. M. R[$\o$]{}nnow, J. van den Brink, J. Geck, and T. Schmitt, Phys. Rev. Lett. [**110**]{}, 087403 (2013). W. S. Lee, S. Johnston, B. Moritz, J. Lee, M. Yi, K. J. Zhou, T. Schmitt, L. Patthey, V. Strocov, K. Kudo, Y. Koike, J. van den Brink, T. P. Devereaux, and Z. X. Shen, Phys. Rev. Lett. [**110**]{}, 265502 (2013). S. Johnston, C. Monney, V. Bisogni, K.-J. Zhou, R. Kraus, G. Behr, V. N. Strocov, J. Málek, S.-L. Drechsler, J. Geck, T. Schmitt, and J. van den Brink, Nat. Commun. [**7**]{}, 10563 (2016). K. Ishii, T. Tohyama, S. Asano, K. Sato, M. Fujita, S. Wakimoto, K. Tustsui, S. Sota, J. Miyawaki, H. Niwa, Y. Harada, J. Pelliciari, Y. Huang, T. Schmitt, Y. Yamamoto, and J. Mizuki, Phys. Rev. B [**96**]{}, 115148 (2017). J. Schlappa, U. Kumar, K. J. Zhou, S. Singh, M. Mourigal, V. N. Strocov, A. Revcolevschi, L. Patthey, H. M. R[$\o$]{}nnow, S. Johnston, and T. Schmitt, Nat. Commun. [**9**]{}, 5394 (2018). D. Li, K. Lee, B. Y. Wang, M. Osada, S. Crossley, H. R. Lee, Y. Cui, Y. Hikita, and H. Y. Hwang Nature [**572**]{}, 624 (2019). S. Shinomori, Y. Okimoto, M. Kawasaki and Y. Tokura, J. Phys. Soc. Jpn., [**71**]{}, 705 (2002). H. Takagi, T. Ido, S. Ishibashi, M. Uota, S. Uchida, and Y. Tokura, Phys. Rev. B [**40**]{}, 2254 (1989). S.-H. Lee and S-W. Cheong, Phys. Rev. Lett. [**79**]{}, 2514 (1997). S. Anissimova, D. Parshall, G. D. Gu, K. Marty, M. D. Lumsden, S. Chi, J. A. Fernandez-Baca, D. L. Abernathy, D. Lamago, J. M. Tranquada, and D. Reznik, Nat. Commun. [**5**]{}, 3467 (2014). C. H. Chen, W. Cheong, and A. S. Cooper, Phys. Rev. Lett. [**71**]{}, 2461 (1993). A. Vigliante, M. von Zimmermann, J. R. Schneider, T. Frello, N. H. Andersen, J. Madsen, D. J. Buttrey, Doon Gibbs and J. M. Tranquada, Phys. Rev. B [**56**]{}, 8248 (1997). C-H. Du, M. E. Ghazi, Y. Su, I. Pape, P. D. Hatton, S. D. Brown, W. G. Stirling, M. J. Cooper, and S-W. Cheong, Rhys. Rev. Lett. [**84**]{}, 3911 (2000). H. Ulbrich and M. Braden, Physica C, [**481**]{}, 31 (2012). M. Uchida, K. Ishizaka, P. Hansmann, Y. Kaneko, Y. Ishida, X. Yang, R. Kumai, A. Toschi, Y. Onose, R. Arita, K. Held, O. K. Andersen, S. Shin, and Y. Tokura, Phys. Rev. Lett. [**106**]{}, 027001 (2011). M. Uchida, K. Ishizaka, P. Hansmann, X. Yang, M. Sakano, J. Miyawaki, R. Arita, Y. Kaneko, Y. Takata, M. Oura, A. Toschi, K. Held, A. Chainani, O. K. Andersen, S. Shin, and Y. Tokura, Phys. Rev. B [**84**]{}, 241109(R) (2011). S. Yamamoto, Y. Senba, T. Tanaka, H. Ohashi, T. Hirono, H. Kimura, M. Fujisawa, J. Miyawaki, A. Harasawa, T. Seike, S. Takahashi, N. Nariyama, T. Matsushita, M. Takeuchi, T. Ohata, Yukito Furukawa, K. Takeshita, S. Goto, Y. Harada, S. Shin, H. Kitamura, A. Kakizaki, M. Oshima, and I. Matsuda, J. Synchrotron Rad. [**21**]{}, 352 (2014). Y. Harada, M. Kobayashi, H. Niwa, Y. Senba, H. Ohashi, T. Tokushima, Y. Horikawa, S. Shin, and M. Oshima, Rev. Sci. Instrum. [**83**]{}, 013116 (2012). G. Wu, J. J. Neumeier, C. D. Ling, and Dimitri, and N. Argyriou Phys. Rev. B, [**65**]{}, 174113 (2002). C. H. Chen, S-W. Cheong, and A. S. Cooper, Phys. Rev. Lett. [**71**]{}, 2461 (1993). M. Uchida, Y. Yamasaki, Y. Kaneko, K. Ishizaka, J. Okamoto, H. Nakao, Y. Murakami, and Y. Tokura, Phys. Rev. B [**86**]{}, 165126 (2012). K. Tsutsui, W. Koshibae, and S. Maekawa, Phys. Rev. B [**59**]{}, 9729 (1999). G. Fabbris, D. Meyers, L. Xu, V. M. Katukuri, L. Hozoi, X. Liu, Z.-Y. Chen, J. Okamoto, T. Schmitt, A. Uldry, B. Delley, G. D. Gu, D. Prabhakaran, A. T. Boothroyd, J. van den Brink, D. J. Huang, and M. P. M. Dean, Phys. Rev. Lett. [**118**]{}, 156402 (2017). S. Sugai, M. Sato, T. Kobayashi, J. Akimitsu, T. Ito, H. Takagi, S. Uchida, S. Hosoya, T. Kajitani, and T. Fukuda, Phys. Rev. B [**42**]{}, 1045 (1990). K. Yamamoto, T. Katsufuji, T. Tanabe, and Y. Tokura, Phys. Rev. Lett. [**80**]{}, 1493 (1996). See Supplemental Material at http://link.aps.org/supplemental/xx.xxxx/xxxxxxxxxxx for the multiple Gaussian results for O $K$-edge RIXS spectra. A. W. Overhauser, Phys. Rev. [**128**]{}, 1437 (1962). K. Ishizaka, Y. Taguchi, R. Kajimoto, H. Yoshizawa, and Y. Tokura, Phys. Rev. B [**67**]{}, 184418 (2003). X-X. Bi and P. C. Eklund, Phys. Rev. Lett. [**70**]{}, 2625 (1993). J. Zaanen and P. B. Littlewood, Phys. Rev. B [**50**]{}, 7222 (1994). A. Fujimori, S. Takekawa, E. Takayama-Muromachi, Y. Uchida, A. Ono, T. Takahashi, Y. Okabe, and H. Katayama-Yoshida, Phys. Rev. B [**39**]{}, 2255 (1989). N. Poirot, R. A. Souza, and C. M. Smith, Solid State Sci. [**13**]{}, 1494 (2011).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Functions which are equivariant or invariant under the transformations of a compact linear group $G$ acting in a euclidean space $\real^n$, can profitably be studied as functions defined in the orbit space of the group. The orbit space is the union of a finite set of strata, which are semialgebraic manifolds formed by the $G$-orbits with the same orbit-type. In this paper, we provide a simple recipe to obtain rational parametrizations of the strata. Our results can be easily exploited, in many physical contexts where the study of equivariant or invariant functions is important, for instance in the determination of patterns of spontaneous symmetry breaking, in the analysis of phase spaces and structural phase transitions (Landau theory), in equivariant bifurcation theory, in crystal field theory and in most areas where use is made of symmetry adapted functions.
A physically significant example of utilization of the recipe is given, related to spontaneous polarization in chiral biaxial liquid crystals, where the advantages with respect to previous heuristic approaches are shown.
address: 'Dipartimento di Fisica,Università di Padova and INFN, Sezione di Padova I–35131 Padova, Italy '
author:
- G Sartori and G Valente
---
Introduction
============
The determination of properties of functions which are equivariant or invariant under the transformations of a compact linear group (hereafter abbreviated in CLG) $G$ is often a basic problem to solve in many physical contexts. $G$-invariant functions play an important role, for instance, in the determination of patterns of spontaneous symmetry breaking and structural phase transitions (Landau theory [@470; @620; @051; @557]), in equivariant bifurcation theory (see, for instance, [@Chossat] and references therein), in crystal field theory and in most areas of solid state theory.
It will not be essentially restrictive, in the following, to assume that $G$ is a matrix subgroup of the real group O(n). In fact, complex linear groups can always be transformed into real linear groups through a process of realification and compact real linear groups are equivalent to orthogonal linear groups.
An approach to the study of invariant functions, that fully exploits invariance and possible regularity properties, takes advantage of the fact that a $G$-invariant function $f$, defined on $\real^n$, is a constant along each orbit of $G$ and can, therefore, be considered as a function $\widehat f$ in the orbit space $\real^n/G$ of the action of $G$ in $\real^n$. Geometric invariant theory [@610; @611] suggests how to get, in principle, $\widehat f$ and $\real^n/G$.
The orbit space of a compact group can be realized as a connected semi-algebraic subset ([*i.e.*]{}, a subset defined by algebraic equalities and inequalities) of a Euclidean space $\real^q$. It turns out to be formed by the union of connected semialgebraic manifolds of different dimensions ([*primary strata*]{}). The images of $G$-orbits with the same orbit–type form [*isotropy–type strata*]{}, whose connected components are primary strata.
The properties of a function defined on $\real^n/G$, may critically depend on the geometry of this space, that has to be known explicitly. A simple way to obtain a substantial determination of the algebraic equations and inequalities defining the range of an orbit map and its strata has been suggested in [@020; @020a; @020b] (see also [@651; @650]). These relations can be expressed in the form of positivity and rank conditions of a matrix $\widehat P(p)$, whose elements are polynomial functions of $p\in\real^q$. In this way, the defining relations of the strata are obtained in the form of algebraic equations and inequalities.
In applications or, simply, to get a better understanding of the geometry of the stratum (its connectivity properties, its boundary, etc..), one often needs to solve explicitly these relations and, sometimes, this is very difficult to do using standard algorithms. In this paper we propose a general method to derive the equations defining the stratification of orbit spaces of CLG’s, in the form of explicit rational parametric relations.
3truemm This paper is organized as follows. In section 2 we summarize the results in the $P$-matrix approach to the characterization of the orbit spaces of CLG’s [@020; @020a]. In section 3 we present and prove our results, and in section 4, we illustrate them in a physically significant example, related to spontaneous polarization in chiral biaxial liquid crystals.
An overview of the geometry of linear group actions
===================================================
In this section, we shall recall, without proofs, some results concerning invariant theory and the geometry of orbit spaces of CLG’s (see, for instance, [@080; @710; @683] and references therein).
For our purposes, it will not be restrictive to assume that $G$ is a matrix subgroup of the group O(n), acting linearly in the Euclidean space $\real^{n}$ and thus defining the $G$-space $(G,\real^n)$.
We shall denote by $g\cdot x$ the action of $g\in G$ on $x\in\real^n$ and by $G_x$ the isotropy subgroup of $G$ at $x\in\real^n$. The isotropy subgroups of $G$ at points of the orbit $G\cdot x$ form the conjugacy class $(G_x)$ of $G_x$ in $G$, which identifies the [*orbit–type*]{} of (the points of) $G\cdot
x$. The set of points $x\in\real^n$ (of $G$-orbits) with the same orbit–type forms an [*isotropy–type stratum of*]{} $(G,\real^
n)$.
The [*orbit space*]{} of the action of $G$ in $\real^n$ is defined as the quotient space $\real^ n/G$, endowed with the quotient topology and differentiable structure. The image of a stratum of $\real^{n}$, through the canonical projection $\pi:\
\real^n\rightarrow\real^n/G $, defines an [*isotropy–type stratum*]{} of $\real^n/G$; the connected components of an [*isotropy–type stratum*]{} of $\real^n/G$ are iso–dimensional manifolds ([*primary strata*]{} of $\real^n/G$), but the orbit space is not a manifold.
Almost all the points of $\real^n/G$ belong to a unique stratum, the [*principal stratum*]{}, which is a connected open dense subset of $\real^ n/G$. The boundary of the principal stratum is the union of disjoint [*singular*]{} strata. Every stratum $\widehat S$ of $\real^n/G$ is open in its topological closure $\overline{\widehat S}$.
The following partial ordering can be introduced in the set of all the orbit–types: $(H)<(K)$, if $H$ is conjugate with a subgroup of $K$. If $(H_i)$ is the orbit–type of the stratum $\widehat
S_i$ of $\real^n/G$ , $i=1,2$, then $H_1<H_2$ iff $\widehat S_2$ is in the boundary of $\widehat S_1$; therefore, more peripheral strata of $\real^n/G$ have greater orbit–type. The number of distinct orbit–types of $G$ is finite and there is a unique minimum orbit–type, the [*principal orbit–type*]{}, corresponding to the principal stratum.
The ring $\real [x]^G$ of real $G$-invariant polynomial functions of $x$ is finitely generated [@286; @286a; @286b; @Nag]. There exists, therefore, a finite minimal collection of homogeneous $G$-invariant polynomials $p(x) = ( p_1(x), p_{2}(x), \ldots,
p_{q}(x) )$ (minimal integrity basis for the ring of $G$-invariant polynomials, henceforth abbreviated in MIB) such that any element $F\in\real[x]^G$ can be expressed as a polynomial function $\widehat F$ of $p(x)$:
[$${\widehat{F}(p(x)) \,=\, F(x), \, \forall x \in
\real^{n}\,.}\label{e2}$$]{} The number $q$ of elements of a minimal integrity basis and their homogeneity [*degrees*]{} $d_i$’s are only determined by the group $G$.
The group $G$ is said to be [*coregular*]{} if the elements of its MIB’s are algebraically independent. The elements of a MIB of a non coregular group satisfy a set of algebraic identities in $\real^n$: $\widehat F_A(p(x))\,=\,0$, $A=1,\dots$ and the associated set of equations
[$${ \widehat F_A(p)\,=\,0,\qquad A=1,\dots\,,}\label{e5}$$]{} defines an algebraic variety in $\real^q$, which is called the [*variety $\cal Z$ of the relations*]{} (among the elements of the MIB). If $G$ is coregular, ${\cal Z}=\real^q$.
Since $G$ is a compact group, the orbits of $G$ are separated by the elements of a MIB of $G$ (for an elementary proof see, for instance, [@020a]), [*i.e.*]{}, at least one element of a MIB of $G$ takes on different values at two distinct orbits. Thus, the elements of a MIB of $G$ yield a parametrization of the points of $\real^{n}/G$, that turns out to be also smooth, since the [*orbit map*]{} $p:\, \real^{n} \longrightarrow \real^{q}$, which maps all the points of $\real^{n}$ lying on an orbit of $G$ onto a single point of $\real^{q}$, induces a diffeomorphism of $\real^{n}/G$ onto a semialgebraic $q$–dimensional connected closed subset of $\real^{q}$. Like all semialgebraic varieties [@840], $p(\real^n)$ presents a natural stratification in connected semialgebraic sub-varieties, called [*primary strata*]{}, which turn out to be the connected components of the isotropy-type strata[^1].
A characterization of the image $p(\real^n)$ of the orbit space of $G$ as a semi-algebraic variety and of its primary strata can be easily obtained through a matrix $\widehat{P}(p)$, defined only in terms of the $G$-invariant Euclidean scalar products between the gradients of the elements of the MIB $\{p(x)\}$:
$$\label{matP}
P_{ab}(x) \,=\, \sum_{i=1}^n \,\partial_i p_a(x)\,
\partial_i p_b(x) \,=\, \widehat{P}_{ab}(p(x)), \hspace{3em}
a,b = 1, \ldots, q,$$
where in the last member, use has been made of Hilbert’s theorem, in order to express $P_{ab}(x)$ as a polynomial function of $p_1(x), \dots ,p_q(x)$.
The following theorem [@020; @020a; @020b] (see also [@651]) clarifies the meaning and points out the role of the matrix $\widehat{P}(p)$:
\[T1\] Let $G$ be a compact subgroup of the real group O(n), $p$ the orbit map $\real^n\rightarrow\real^q$ defined by the MIB $(p_1(x),p_2(x),\ldots ,p_q(x))$ and $\widehat{P}(p)$ the matrix defined in [(\[matP\])]{}. Then the range $p(\real^{n})$ of $p$ is the unique semialgebraic connected subset of the variety ${\cal
Z}\subseteq \real^q$ of the relations among the elements of the MIB where $\widehat{P}(p)$ is positive semi-definite. The $k$–dimensional primary strata of $p(\real^{n})$ are the connected components of the set $\{ p \in {\cal Z} \mid
\widehat{P}(p) \geq 0, \, {\rm rank} (\widehat{P}(p)) = k \}$; they are the images of the connected components of the k–dimensional isotropy–type strata of $\real^{n}/G$. In particular, the set of interior points of $p(\real^{n})$, where $\widehat{P}(p)$ has the maximum rank, is the image of the principal stratum.
In the following, we shall identify orbit spaces and their strata with their images through orbit maps.
If $G$ is coregular, the $p_i$’s, with range in the semialgebraic set $\Delta$ defined by the inequalities $\widehat P(p)>0$ (and no equation!) provide a one-to-one parametrization of the principal stratum of $\real^n/G$. If the stratum is singular, the $p_i$’s are not independent parameters, being bounded also by the equations defining the stratum. In applications or, simply, to get a better understanding of the geometry of the stratum (its connectivity properties, its boundary, etc. ...), one often needs to solve explicitly these equations, which are obtained in implicit form from the theorem recalled above. Often, this is very difficult to do using standard algorithms. In the following section, we propose a general method to derive the equations defining the singular strata of the orbit space of any CLG in the form of explicit rational parametric relations.
The mathematical apparatus just recalled offers effective tools, for instance, in the determination of possible patterns of spontaneous symmetry breaking, when the ground state of the system is determined by the absolute minimum of an invariant potential. Let us recall the basic ideas. In this context, the vector $x\in
\real^n$ is an order parameter and $G$ is the symmetry group of the potential $\Phi(\alpha; x)$ (free energy, or Higgs potential, for instance), expressed in terms, also, of parameters $\alpha$. The points $x_0(\alpha)$, where the function $\phi_\alpha(x) =
\Phi(\alpha; x)$ takes on its absolute minimum, determine the stable phase of the system, whose residual symmetry is defined by the isotropy subgroup of $G$ at $x_0$. Owing to the $G$-invariance of the potential, each of its stationary points is degenerate along the $G$-orbit through it. Since the isotropy subgroups of $G$ at points of the same orbit are conjugate in $G$, only the orbit–type of $x_0(\alpha)$ is physically relevant. [*Structural phase transitions*]{} take place when, by varying the values of the $\alpha$’s, the point $x_0(\alpha)$ is shifted to a different stratum.
If $\Phi(\alpha;x)$ is a sufficiently general function of the $\alpha$’s, by varying these parameters, the point $x_0(\alpha)$ can be shifted to any stratum of $\real^n/G$. So, [*the strata are in a one-to-one correspondence with the symmetry phases allowed by the $G$-invariance of the potential*]{}. On the contrary, extra restrictions on the form of the potential function, not coming from G-symmetry requirements (e.g., the assumption that the potential is a polynomial of low degree), can limit the number of allowed structural phases for the system in its ground state.
Parametrization of strata in orbit spaces of compact linear groups
==================================================================
Let $G$ be a compact linear group, acting orthogonally in $\real^n$ and $\{p\}$ be a related MIB. We shall prove that the possibility of parametrizing the principal stratum of the orbit space of a coregular group in terms of parameters related to the elements of a MIB can be extended to singular strata. The result stems from the proof of the following statement: [*the set of interior points of the topological closure of an isotropy–type stratum, with orbit–type $(H)$, is diffeomorphic to the principal stratum of the orbit space of a group space of the stabilizer of $H$ in $G$.*]{} If the stabilizer turns out to be coregular, the constructive proof of this proposition will provide a one-to-one rational parametrization of the stratum. Since, in view of possible applications, we are mainly interested in the possibility of getting a one-to-one parametrization of a stratum, in the presentation of our results, stress will be laid on this aspect of the matter.
Let $H$ be a proper isotropy subgroup of the $G$-space $(G,\real^n)$. We shall denote by $S$ the stratum of $(G,\real^n)$ with orbit–type $(H)$ and by $\widehat S$ its image $p(S)$ in the orbit space $\real^n/G$. Let us, also, define
$$V = \{x\in\real^n\mid G_x\supseteq H\},\qquad {\mathcal
V}=\{x\in\real^n \mid G_x = H\}\,. \label{1}$$
The non trivial sets $V$ and ${\mathcal V}$ have the following properties, which are more or less immediate consequences of their definitions:
1. $V =\{x\in\real^n \mid h\cdot x=x, h\in H\}$ is a linear subspace of $\real^n$, let us call $\nu$ its dimensions;
2. $V$ is the topological closure of $\mathcal V$: $V=\overline{\mathcal V}$;
3. ${\mathcal V}=S \cap V $ and every $G$-orbit lying in $S$ has at least a point in ${\mathcal V}$ so that $S = G\cdot {\mathcal V}$, $\overline{S}=G\cdot V$ (where the bar denotes topological closure) and, consequently,
[$${p({\mathcal V})={\widehat S},\qquad p(V)=\overline{\widehat
S}}\label{pV}$$]{} and
[$${{\rm rank}(P(x)\big|_{x\in{\mathcal V}})={\rm dim}(\widehat
S),\qquad {\rm rank}(P(x)\big|_{x\in V\backslash{\mathcal
V}})\,<\,{\rm dim}(\widehat S).}\label{rango}$$]{}
A $G$-orbit of $S$ may intersect $\mathcal V$ in one or more points. In the first case, every $G$-orbit of $S$ intersects $\mathcal V$ in a point and, owing to item [*(iii)*]{}, the coordinates of the points of ${\mathcal V}$ provide a one-to-one rational (in effect, linear) parametrizazion of the orbits of G lying in $S$, obtained by solving the system of linear equations $h\cdot x=x\ \forall h\in H$. The allowed range of $x$, $x\in{\mathcal V}$, is determined by the inequalities assuring that rank$(P(x))=$dim$(\widehat S)$. The parameters $x$, with range $\mathcal V$, would provide, in this case, a one-to-one parametrization of the stratum, thus solving our problem. In general, however, the intersection of a $G$-orbit, of orbit–type $(H)$, with the stratum $S$ does not reduce to a single point. So, a sounder analysis is required, that will be the object of the rest of this section[^2]. Two distinct points, $x$ and $g\cdot x$, $g\in G$, of the same $G$-orbit, lie in $\mathcal V$ iff $h\cdot g\cdot x = g\cdot x\
\forall h\in H$, that is, iff $g$ is in the stabilizer Stab$(H,
G)$ of $H$ in $G$. The intersection of a $G$-orbit of $S$ with ${\mathcal V}$ is, therefore, the Stab($H$, $G$)-orbit through $x$ and Stab($H$, $G$) is the largest subgroup of $G$ leaving $V$ invariant.
In the group space (Stab($H$, $G$),V), the isotropy subgroup at a point of general position is $H$. Therefore, the principal stratum $\Sigma$ satisfies the following relation:
[$${V\supset \Sigma\supseteq {\mathcal V}.}\label{ins}$$]{} It has to be noted that $\Sigma$ could contain $\mathcal V$ in a strict sense, since, at points of $V\backslash{\mathcal V}$, the conjugacy class in $G$ of the isotropy subgroup of Stab$(H,G)$ could be smaller than the conjugacy class in $G$ of the isotropy subgroup of $G$.
Now, $H$ is an invariant subgroup of Stab$(H,G)$ and a subgroup of all the isotropy subgroups of (Stab$(H,G),V)$. Thus, the action of Stab$(H,G)$ in $V$ defines a linear group $K$ (and the group space $(K,V)$), isomorphic to the quotient group Stab($H$, $G$)/$H$, through the relation
[$${(sH)\cdot v=s\cdot v,\qquad s\in {\rm Stab}(H,G),\ sH\in
{\rm Stab}(H,G)/H.}\label{K}$$]{} So, we can conclude that [*the intersection of a $G$-orbit of $S$ with ${\mathcal V}$ is an orbit of $(K,V)$*]{}. Obviously, the orbit spaces $V/K$ and $V/{\rm
Stab}(H,G)$ are isomorphic and can be identified.
We can rephrase the result just obtained, by claiming that the points of $\widehat S$ are in a one-to-one correspondence with the points of a, possibly proper, subset $\widehat{\mathcal V}$ of the principal stratum $\widehat\Sigma$ of the orbit space $V/K$ and the following relations hold true:
[$${\overline{\widehat S}\,=\,p(V)\,\supset\,
p(\Sigma)\,\supseteq\, p({\mathcal V})\, =\, \widehat S.}\label{pS}$$]{} Since, as stressed, we know how to parametrize a principal stratum, our problem is reduced to the determination of $\widehat{\mathcal V}$. We shall show that this can be easily done, making use of [(\[rango\])]{}.
To make simpler the solution of the problem, let us assume that an orthonormal basis has been introduced in $\real^n$ such that the first $\nu$ elements of the basis yield a basis for the vector space $V$. Then, if we denote by $V_\bot$ the orthogonal complement of $V$ in $\real^n$, the subspace $V_\bot$ is invariant under ($K$ and) $H$, owing to the orthogonality of the transformations of $G$. Since $V$ contains all the $H$-invariant vectors of $\real^n$, there is no non trivial $H$-invariant vector in $V_\bot$.
To attain our goal, we shall start from a convenient parametrization of the principal stratum $\widehat \Sigma$ of $V/K$ in terms of $l$ real parameters $\lambda$, related to a MIB $(\lambda_1(v),\dots ,\lambda_l(v))$ of the ring of polynomial $K$-invariant functions of $v \in V$. This parametrization will be global and one-to-one, if $K$ turns out to be coregular. In this case, the range of $\lambda$ has to be restricted to the positivity region of the $\widehat P$-matrix $\widehat\Lambda(\lambda)$, associated with the MIB $\{\lambda\}$:
[$${\widehat\Lambda_{\alpha\beta}(\lambda(v)) =
\Lambda_{\alpha\beta}(v) = \sum_{i=1}^\nu\,
\partial_i \lambda_\alpha(x_1,\dots ,x_\nu)\, \partial_i \lambda_\beta(x_1,\dots ,x_\nu).}\label{4}$$]{} As recalled in the previous section, in fact, the orbit space $V/K$ and its principal stratum $\widehat \Sigma$ can be identified, respectively, with the semialgebraic sets $\lambda(V)$ and $\lambda(\Sigma)$:
[$${\widehat\Sigma \,=\,
\lambda(\Sigma)\,=\,\{\lambda\in\real^l\mid \widehat
\Lambda(\lambda)>0\},\qquad V/K= \overline{\widehat
\Sigma}\,=\,\lambda(V)\,=\,\{\lambda\in\real^l\mid
\widehat\Lambda(\lambda)\ge 0\},}\label{20}$$]{} the second set being the closure of the first one. Moreover, the definition of $\widehat{\mathcal V}$ and equation [(\[ins\])]{} imply
[$${\widehat{\mathcal V}=\lambda({\mathcal
V})\subseteq\lambda(\Sigma)\,=\,\widehat\Sigma.}\label{inshat}$$]{}
If $K$ is not coregular, only a local one-to-one parametrization can be obtained for $\widehat \Sigma$, by eliminating redundant elements in the set of parameters $(\lambda_1,\dots ,\lambda_l)$, through the solution of the algebraic relation(s) among the elements of the MIB $\{\lambda\}$, and imposing convenient semi-positivity and rank conditions on the matrix $\widehat\Lambda(\lambda)$.
A one-to-one local or global parametrization of $\widehat \Sigma$ yields, obviously, also a local or global one-to-one parametrization of $\widehat {\mathcal V}$, provided that additional restrictions are imposed on the range of $\lambda$, whenever $\widehat {\mathcal V}$ is a proper subset of $\widehat\Sigma$. In this case, the correct bounds can be obtained in the following way (we shall only consider the case of a coregular $K$, the extension of the results to non coregular $K$’s is straightforward, but, as just stressed, may lead to a loss of globality).
When $x$ spans $V$, the elements of the MIB $(p_1(x),\dots,p_q(x))$ of $G$ define a set of $K$-invariant polynomial functions of $v=(x_1,\dots ,x_\nu)\in V$. Therefore, by the Hilbert theorem recalled in the previous section, they can be expressed as polynomial functions of $\lambda$, that is,
[$${p\big|_V=\phi\circ\lambda.}\label{phi}$$]{} Let us remark that, in our assumptions, there are no relations among the elements of the MIB $\{\lambda\}$ and, consequently, possible relations $F_\alpha(p)=0$ among the $p_i$’s are identically satisfied for $p=\phi(\lambda)$.
From [(\[phi\])]{} and [(\[ins\])]{}, one immediately obtains
[$${\overline{\widehat S}\,=\,p(V)\,=\,\phi(V/K)\,\supset\,
\phi(\widehat\Sigma)\,\supseteq\,\phi(\widehat{\mathcal
V})\,=\,p({\mathcal V})=\widehat S}\label{pins}$$]{} and, since $\widehat\Sigma$, being a principal stratum, is the set of interior points of $V/K$ and is connected, $\phi(\widehat\Sigma)$ will coincide with the set of interior points of the closure $\overline{\widehat S}$ of $\widehat S$ and will be connected. This set does not coincide with $\widehat S$ if $\overline{\widehat S}$ contains, in its interior, points representing bordering strata of $\widehat S$. This certainly happens if the set $\widehat S$ is not connected and, presumebly, also if it is not multiply connected[^3].
The identification of points $\lambda\in\widehat \Sigma$, if any, whose image $\phi(\lambda)\not\in\widehat S$, can be obtained in the following way.
Let $x=v\oplus v_\bot$ yield the decomposition of $x\in\real^n$ in its vector components $v\in V$ and $v_\bot\in V_\bot$: $v=(x_1,\dots ,x_\nu)$, $v_\bot=(x_{\nu+1},\dots ,x_n)$. Then, a $G$-invariant polynomial $f(x)$, can be thought of as a polynomial in $v$ and $v_\bot$ and it is easy to realize that it cannot contain linear terms in $v_\bot$, being $v$ invariant under $H$. As a consequence,
[$${\partial_i f(x)=0,\ {\rm for\ }x\in V\ {\rm and\
}i=\nu+1,\dots ,n.}\label{3}$$]{} Now, starting from the very definition of $P(x)$ (see [(\[matP\])]{}), for every $x=v\in V$ we obtain, using [(\[3\])]{} and the identity $p(v)=\phi(\lambda(v))$, [$${\begin{array}{rcl}
\widehat P_{ab}(p(v))&=&\sum_{i=1}^\nu\,\partial_i p_a(v)\,\partial_i p_b(v)\\
&&\\
&=& \left.
\sum_{\alpha,\beta=1}^l\,\partial_\alpha\phi_a(\lambda)\,
\partial_\beta\phi_b(\lambda)
\right|_{\lambda=\lambda(v)}\,\sum_{i=1}^\nu\,\partial_i
\lambda_\alpha(v)\,
\partial_i\lambda_\beta(v)\\
&&\\
&=& \left.
\left(J(\lambda)\,\widehat\Lambda(\lambda)\,J(\lambda)^{\rm
T}\right|_{\lambda=\lambda(v)}\right)_{ab} ,
\end{array}}\label{5}$$]{} where, the superscript $T$ denotes transposition and $J(\lambda)$ is the Jacobian matrix of the transformation $p=\phi(\lambda)$:
[$${J_{a\alpha}(\lambda)=\partial_a\phi_\alpha (\lambda),\qquad
a=1,\dots ,q,\ \ \alpha=1,\dots ,l.}\label{7}$$]{} So, we can conclude that, for all $\lambda\in V/K$,
[$${\widehat P(\phi(\lambda))=
J(\lambda)\,\widehat\Lambda(\lambda)\,J(\lambda)^{\rm T}.}\label{6}$$]{}
Equation [(\[6\])]{} leads to an easy calculation of the points $\lambda\in \widehat{\mathcal V}$, that is, of the points $\lambda\in \widehat\Sigma$ whose image $p=\phi(\lambda)$ is in $\widehat S$. In fact, from [(\[rango\])]{}, these points are characterized by the conditions $\lambda\in\widehat\Sigma$ and rank($\widehat P(\phi(\lambda)))=l$. Since, for $\lambda\in\widehat\Sigma$, the matrix $\widehat
\Lambda(\lambda)$ is positive definite with rank $l$, we can conclude that the range of $\lambda$ assuring the location of the point $p=\phi(\lambda)$ in $\widehat S$ is the semialgebraic subset $\Delta$ of $\real^l$, determined by the following inequalities:
[$${\Delta=\{\lambda\in\real^l\mid \widehat{\Lambda}(\lambda)>0
\ {\rm and}\ {\rm rank}(J(\lambda))=l\}.}\label{8}$$]{} For $\lambda\in\Delta$, the relation $p=\phi(\lambda)$ yields a global rational parametrization for $\widehat S$.
It will be worthwhile to stress that the parametrization we have suggested is, in some way, canonical: the unique arbitrariness in the choice of the parameters is related to the choice of the MIB’s.
An important byproduct of the result just proved is a simple test of the connection of $\widehat S$, that could be difficult or impossible to read directly from the equations of the stratum in implicit form. In fact, the condition that the boundary of $\overline{\widehat S}$ coincides with the boundary of $\widehat
S$ is equivalent to the condition rank($J(\lambda))=l$ for all $\lambda\in\widehat \Sigma$.
If $K$ is not coregular, there are polynomial relations $F_A(\lambda)=0$ among the elements of the MIB $(\lambda_1,\dots
,\lambda_l)$ and $l>$dim$(\widehat S)$. As already stressed, to obtain a one-to-one parametrization of the points of $\widehat S$ by means of the $\lambda_\alpha$’s, one has to eliminate the redundant parameters by solving the equations $F_A(\lambda)=0$. This may be feasible, but only locally and the resulting parametrization will not be global and possibly not rational.
An Example
==========
In this section we shall show how the parametrization technique works in a simple example. The notations will be the same defined in the previous sections.
We shall consider the orthogonal linear group $G$ defined by the action of the real group O(3) in the real eight–dimensional space spanned by the independent components $x_1,\dots ,x_5$ of a symmetric and traceless tensor $Q$ and the three components $x_6,x_7,x_8$ of a polar vector $\bi{P}$. To be specific, if $O$ is a generic $3\times 3$ real orthogonal matrix, the transformation rules of the $x_i$ are obtained from the following relations:
$$Q = \frac{1}{\sqrt{2}} \left(
\begin{array}{ccc}
-{\displaystyle\frac{2}{\sqrt{3}}} x_1& x_3 & x_4 \\
x_3 & {\displaystyle\frac{x_1}{\sqrt{3}}} - x_2 & x_5 \\
x_4 & x_5 & {\displaystyle\frac{x_1}{\sqrt{3}}} + x_2 \\
\end{array}
\right)$$
$$\begin{array}{rcll}
Q'_{\alpha\beta}&=& \sum_{\gamma\,\delta=1}^3 O_{\alpha\gamma}\,
Q_{\gamma\delta}\, O_{\beta\delta}& \\
&&&\alpha,\beta=1,2,3\,. \\
P'_{\alpha}&=& \sum_{\beta=1}^3 O_{\alpha\beta}\, P_{\beta}\;,
\end{array}\label{mu2}$$
It is worthwhile to remark that the reflection $I=\mbox{\rm diag}
(-1,-1,-1)\in$ O(3) reverses only the signs of the last three coordinates $(x_6,x_7,x_8)$ and, due to the symmetry of the tensor $Q$, each $G$-orbit contains points where $Q$ takes on a diagonal form ($x_3=x_4=x_5=0)$. This remark makes much easier the determination of “typical points" in strata, from which the orbit type of the stratum can be identified [@Li; @K].
The real orthogonal linear group $G$, just defined, is coregular (see [@711]) and its isotropy subgroup at a generic point of $\real^8$ is trivial. As a consequence, the principal orbits are three–dimensional manifolds, the principal stratum has dimensions five and there are five independent elements in a MIB (see, for instance [@683]), which can be chosen in the following way:
$$\begin{aligned}
p_1 &=&\Tr\, Q^{2} + \bi{P}\cdot
\bi{P} \nonumber\\
&=& \sum_{j=1}^{8} {x_j}^{2}\,,\nonumber\\
&&\nonumber\\
p_2 &=& \bi{P}\cdot \bi{P} \nonumber\\
&=& \sum_{j=6}^{8} {x_j}^{2}\,,\nonumber\\
&&\nonumber\\
p_3 &=& 6 \sqrt{2} \,\,\Tr\, Q^{3}\nonumber\\
&=& -2\, {\sqrt{3}}\, {{x_1}}^3 + 6\, {\sqrt{3}}\, {x_1}\, {{x_2}}^2
- 3\, {\sqrt{3}}\, {x_1}\, {{x_3}}^2 - 9\, {x_2}\, {{x_3}}^2
\label{BaseInt}\\
&& - 3\, {\sqrt{3}}\, {x_1}\, {{x_4}}^2+ 9\, {x_2}\, {{x_4}}^2+18\,
{x_3}\, {x_4}\, {x_5} + 6\, {\sqrt{3}}\, {x_1}\, {{x_5}}^2 \,,\nonumber\\
&& \nonumber\\
p_4 &=& 3 \sqrt{2} \,\sum_{\alpha\,\beta}P_\alpha Q_{\alpha\,\beta}
P_{\beta}\nonumber\\
&=& -2\,{\sqrt{3}}\, {x_1}\, {{x_6}}^2 +
6\, {x_3}\, {x_6}\, {x_7} + {\sqrt{3}}\, {x_1}\, {{x_7}}^2 -
3\, {x_2}\, {{x_7}}^2 \nonumber\\
&& + 6\, {x_4}\, {x_6}\, {x_8}+ 6\, {x_5}\, {x_7}\, {x_8}+ \sqrt{3}\,
{x_1}\, {{x_8}}^2+3\, {x_2}\, {{x_8}}^2\,,\nonumber\\
&&\nonumber\\
p_5 &=& 6 \,\sum_{\alpha\,\beta}P_\alpha {Q^{2}}_{\alpha\,\beta}
P_{\beta}\nonumber\\
&=& 4\,{{x_1}}^2\, {{x_6}}^2+ 3\, {{x_3}}^2\, {{x_6}}^2 +
3\, {{x_4}}^2\, {{x_6}}^2-2\, {\sqrt{3}}\, {x_1}\, {x_3}\,
{x_6}\,{x_7}
\nonumber\\
&&- 6 \,{x_2}\,{x_3}\,{x_6}\,{x_7}+ 6 \,{x_4}\,{x_5}\,{x_6}\,{x_7} +
{{x_1}}^2\, {{x_7}}^2 -
2\, {\sqrt{3}}\, {x_1}\, {x_2}\, {{x_7}}^2 \nonumber\\
&&+ 3\, {{x_2}}^2\, {{x_7}}^2 +
3\, {{x_3}}^2\, {{x_7}}^2 + 3\, {{x_5}}^2\, {{x_7}}^2 -
2\, {\sqrt{3}}\, {x_1}\, {x_4}\, {x_6}\, {x_8} \nonumber\\
&& + 6\, {x_2}\, {x_4}\, {x_6}\, {x_8}+6\, {x_3}\, {x_5}\, {x_6}\,
{x_8} + 6\, {x_3}\, {x_4}\, {x_7}\, {x_8} +
4\, {\sqrt{3}}\, {x_1}\, {x_5}\, {x_7}\, {x_8} \nonumber\\
&&+ {{x_1}}^2\, {{x_8}}^2
+2\,{\sqrt{3}}\, {x_1}\, {x_2}\, {{x_8}^2}+3\, {{x_2}}^2\, {{x_8}}^2 +
3\, {{x_4}}^2\, {{x_8}}^2 +3\, {{x_5}}^2\, {{x_8}}^2\;.\nonumber
\end{aligned}$$
The corresponding $\widehat P $-matrix elements can be easily calculated from their definition [(\[matP\])]{}: $$\begin{array}{rcl}
\widehat{P}_{1\,a} &=& 2 d_a p_a\;, \hspace{3em} 1 \leq a \leq 5 \\
\widehat{P}_{2\,2} &=& 4 p_2 \\
\widehat{P}_{2\,3} &=& 0 \\
\widehat{P}_{2\,4} &=& 4 p_4 \\
\widehat{P}_{2\,5} &=& 4 p_5 \\
\widehat{P}_{3\,3} &=& 108 \left(p_1 -p_2\right)^{2} \\
\widehat{P}_{3\,4} &=& 18\,\left( - 2\,p_1\,p_2+ 2 {p_2}^2+ p_5 \right) \\
\widehat{P}_{3\,5} &=& 12\,\left( \,p_2\,p_3+ p_1 \,p_4 -p_2 \, p_4 \right) \\
\widehat{P}_{4\,4} &=& 12 \,\left({p_2}^2 + p_5 \right)\\
\widehat{P}_{4\,5} &=& 4\,\left( \,p_2\,p_3+3\, p_1 \,p_4 -p_2 \, p_4 \right) \\
\widehat{P}_{5\,5} &=& \frac{4}{3} \left(p_3 \, p_4 + {p_4}^2 + 9\, p_1 p_5 \right)\;,
\end{array}$$ where the ordered set $(d_1,d_2,d_3,d_4,d_5)=(2,2,3,3,4)$ specifies the degrees of the polynomials of the MIB.
The determinant of the matrix $\widehat P(p)$ factorizes and only one of the two real irreducible factors, that we shall call $A(p)$, turns out to be [*active*]{} [@682], that is, to be relevant in the determination of the boundary of the orbit space $\real^8/G$:
$$\label{attivo}
\begin{array}{rcl}
A(p)&=& 3\,{{p_2}}^3\,{{p_3}}^2 + 18\,{p_1}\,{{p_2}}^2\,{p_3}\,{p_4} -
18\,{{p_2}}^3\,{p_3}\,{p_4} +
27\,{{p_1}}^2\,{p_2}\,{{p_4}}^2 \\
&& -
54\,{p_1}\,{{p_2}}^2\,{{p_4}}^2 + 27\,{{p_2}}^3\,{{p_4}}^2 +
{p_3}\,{{p_4}}^3 - 9\,{p_2}\,{p_3}\,{p_4}\,{p_5} \\
&&
-9\,{p_1}\,{{p_4}}^2\,{p_5} + 9\,{p_2}\,{{p_4}}^2\,{p_5} -
27\,{p_1}\,{p_2}\,{{p_5}}^2 + 27\,{{p_2}}^2\,{{p_5}}^2 +
9\,{{p_5}}^3\;.\\
\end{array}$$
The relations defining strata of dimension $<4$ in the orbit space of (O(3),$\real^8$) are summarized in . The isotropy subgroup lattice, with the possible phase transitions between bordering strata is shown in .
Parametrization of the strata
-----------------------------
The relations assuring that $\widehat P(p)\ge 0$ and has rank 4, define a unique four–dimensional stratum $\widehat S^{(4)}$ in the orbit space. Using well known matrix theory results, these conditions could be explicitly written, for instance, in the form $A(p)=0$ and $M_i(p)>0$, $i=1,\dots ,4$, where $M_i$ is the sum of the principal minors of order $i$ of the matrix $\widehat P(p)$: a cumbersome set of conditions that it is not worthwhile to write down explicitly.
Even in this simple example one immediately realizes that the choice of a minimal set of explicit algebraic relations providing a cylindrical decomposition [@Coste] for the semi-algebraic subset $\widehat S^{(4)}$ of $\real^{5}$ would be a really hard task (for the more peripheral strata, instead, the problem is much easier to solve). An immediate application of the results proved in the previous section, on the contrary, leads to a simple rational global parametrization of each stratum, as shown below.
### Stratum $\widehat S^{(4)}$.
A “typical point" of the stratum is $x_{\rm
t}=(1,1,0,0,0,0,1,1)$. The corresponding isotropy subgroup $H$ of $G$ is the $\integer_2$ group generated by the reflection representing the element diag$(-1,1,1)\in$ O(3) in $\real^8$: $$\mbox{\rm diag}(1,1,-1,-1,1,-1,1,1)\in G.$$
The vector space $V$ formed by the $H$-invariant vectors of $\real^8$ turns out to be five–dimensional: $$V = \left\{x\in \real^8 \;|\;x_3=x_4=x_6=0\right\}.$$
The elements of O(3), corresponding to elements of $\mbox{\rm
Stab }(H,G)$, are block-diagonal matrices of the form diag$(\pm
1,O)$, with $O\in $ O(2). Therefore, $\mbox{\rm Stab }(H,G)$ is isomorphic to a group $\integer_2 \times$O(2) and the quotient group $K=\mbox{\rm Stab }(H,G)/H$ is the representation in $V$ (induced by the representation $G$ of O(3)) of the O$_1(2)$ subgroup of O(3), formed by the rotations around the first axis.
Using coordinates $v=(x_1, x_2,x_5,x_7,x_8)$ for a vector $v\in
V$, the elements of $K$ turn out to be block diagonal matrices leaving invariant the following subspaces $V^{A}$, $1 \leq A \leq
3$: $$\begin{aligned}
V^{1}& = &\{(x_1,0,0,0,0) \in V | x_1 \in \real\}\,, \nonumber \\
V^{2}& = &\{(0,x_2,x_5,0,0) \in V | x_2\,,x_5 \in \real\} \,,\nonumber \\
V^{3}& = &\{(0,0,0,x_7,x_8) \in V | x_7\,,x_8 \in \real\}\,.\end{aligned}$$
A proper rotation $r(\phi)\in$ O$_1(2)$ of an angle $\phi$ and the reflection diag(-1,1)$\in {\rm O}_1(2)$ are represented in $V$ by the $5\times 5$ matrices $1\oplus r(-2\phi)\oplus r(\phi)$ and ${\rm diag}(1,-1,1,-1,1)$.
Noting that the complex variables $z_1=x_2+\rmi x_5$ and $z_2=x_7+\rmi x_8$ transform into $\exp(-2\rmi \phi)\,z_1$ and, respectively, $\exp(\rmi \phi)\,z_2$ under a rotation and into $z_1^*$ and, respectively, $-z_2^*$ under a reflection, it is easy to realize that a possible MIB for $(K,
V)$ is the following: $$\begin{aligned}
\lambda_1 &=& x_1 \,,\nonumber\\
\lambda_2 &=&|z_1|^2\,=\, x_2^2 + x_5^2\,,\nonumber\\
\lambda_3 &=& |z_2|^2\,=\,x_7^2 + x_8^2\,,\nonumber\\
\lambda_4 &=& 2 \, \mbox{\rm Re} (z_1 {z_2}^2)\,=\,2
\left( x_2 {x_7}^2 -2 x_5 x_7 x_8 - x_2 {x_8}^2 \right)\;.\end{aligned}$$
It is, now, easy to express the $p$’s in term of the $\lambda$’s, $p=\phi(\lambda)$, and to check that the following expressions, obtained in this way, identically satisfy the equation $A(p(\lambda))=0$ (see (\[attivo\])) entering the definition of the stratum $\widehat S^{(4)}$: $$\begin{aligned}
\label{parametrizzazione}
\phi_1(\lambda) &=& {\lambda_1}^2 + \lambda_2 + \lambda_3 \,,\nonumber\\
\phi_2(\lambda) &=& \lambda_3\,,\nonumber\\
\phi_3(\lambda) &=& -2 \, \sqrt{3}\,\lambda_1\,\left( {\lambda_1}^2 - 3 \lambda_2
\right)\,,\nonumber\\
\phi_4(\lambda) &=& \frac{\sqrt{3}}{2} \,\left( 2 \lambda_1 \lambda_3 - \sqrt{3}
\lambda_4 \right)\,, \nonumber\\
\phi_5(\lambda) &=& {\lambda_1}^{2} \lambda_3 + 3 \lambda_2
\lambda_3 - \sqrt{3} \lambda_1 \lambda_4.\end{aligned}$$
As explained in the previous section, since the group $K$ is coregular, the range $\Delta$ for $\lambda$ is the region where the $\widehat{\Lambda}(\lambda)$ associated to the MIB $\{\lambda\}$ is positive definite and the rank of the Jacobian matrix $J(\lambda)$ of the transformation $\phi(\lambda)$ is maximum (=4). The matrices $\widehat\Lambda(\lambda)$ and $J(\lambda)$ are easily calculated to be
$$\label{lamdilam}
\widehat{\Lambda}(\lambda) = \left(
\begin{array}{cccc}
1 & 0 & 0 & 0 \\
0 & 4 \lambda_2 & 0 & 2 \lambda_4 \\
0 & 0 & 4 \lambda_3 & 4 \lambda_4 \\
0 & 2 \lambda_4 & 4 \lambda_4 & 4\,\left( {\lambda_3}^2 + 4
\lambda_2 \lambda_3 \right)
\end{array}
\right)$$
and
$$J(\lambda) = \left(
\begin{array}{cccc}
2 \lambda_1 & 1 & 1 & 0 \\
0 & 0 & 1 & 0 \\
- 6\,\sqrt{3} \,\left({\lambda_1}^2 - \lambda_2 \right) &
6\,\sqrt{3}\, \lambda_1 & 0 & 0 \\
\sqrt{3} \lambda_3 & 0 & - \sqrt{3} \lambda_1 & -3/2 \\
2 \lambda_1 \lambda_3 - \sqrt{3} \lambda_4 & 3 \lambda_3 &
{\lambda_1}^2 + 3 \lambda_2 & - \sqrt{3} \lambda_1
\end{array}
\right)\;.$$
The conditions assuring the positivity of $\widehat{\Lambda}(\lambda)>0$ can be written in the form
$$\Delta = \left \{(\lambda_1,\lambda_2, \lambda_3, \lambda_4) \in
\real^4 \;|\; \lambda_2 >0 \; \mbox{and} \; \lambda_3 >0
\;\mbox{and}\; \lambda_4^2 < 4 \lambda_2 {\lambda_3}^2 \right \}$$
and the rank of $J(\lambda)$ turns out to equal 4 for all $\lambda \in \Delta$. So we can conclude that the stratum is connected, a piece of information that would be difficult to derive directly from the relations defining the stratum in implicit form ([*i.e.*]{} $A(p)=0$ and $M_i(p)>0$).
The parametrization obtained for the sub-peripheral stratum $\widehat S^{(4)}$ turns out to be useful, also, because bordering values of $\lambda$ immediately determine bordering values of $p$, corresponding to more peripheral strata of $\real^8/G$. The stratification of $ V/K$ is summarized in . We would like to remark, however, that there is not a one-to-one correspondence between singular strata of $(K,V)$ corresponding to bordering values of $\lambda$ and singular strata of $\real^n/G$, corresponding to the associated values of $p(\lambda)$. In fact, as was already noted, since the action of the stabilizer on the subspace $V^{1}$ is trivial, the invariant $\lambda_1$ has degree $1$, so it does not participate in the conditions defining the stratification of $ \left(K,V \right)$. That is why the parametrization procedure has to be applied [*stratum per stratum*]{}.
### Stratum $\widehat S^{(3)}$.
A typical point in this stratum is $x_{\rm t}=(1,1,0,0,0,1,0,0)$ and the isotropy subgroup at $x_{\rm t}$ is the $\integer_2 \times
\integer_2$ group represented by the O(3) elements $\mbox{\rm
diag}( 1, \pm1, \pm 1)$. The vector space left invariant by $H$ reduces to the three–dimensional space $V$:
$$V = \left\{x \in \real^8 \;|\;x_3=x_4=x_5=x_7=x_8=0\right\}\;.$$
The linear group $K$, acting in $V$, is a $\integer_2\times\integer_2$ group, generated by the matrices $\mbox{\rm diag}(1,-1,1)$ and $\mbox{\rm diag}(1,1,-1)$. A MIB for $\mbox{\rm Stab }(K,V)$ is $$(\lambda_1, \lambda_2, \lambda_3) =( x_1 ,
{x_2}^2,
{x_6}^2 )\;.$$ The associated matrix $\widehat\Lambda(\lambda)$ turns out to be [$${\widehat\Lambda(\lambda)=\mbox{\rm diag}(1, 4 \lambda_2, 4 \lambda_3)\,.}$$]{} Parametric equations, $p=\phi(\lambda)$, for the stratum can be written in the following form: $$\begin{aligned}
\label{parametrizzazione1}
\phi_1(\lambda) &=& {\lambda_1}^2+\lambda_2 + \lambda_3 \,,\nonumber\\
\phi_2(\lambda) &=& \lambda_3\,,\nonumber\\
\phi_3(\lambda) &=& -2\,\sqrt{3} \lambda_1 \left( {\lambda_1}^2 - 3 \lambda_2\right)\,,\nonumber\\
\phi_4(\lambda) &=& -2\,\sqrt{3} \lambda_1 \lambda_3 \,, \nonumber\\
\phi_5(\lambda) &=& 4 {\lambda_1}^2 \lambda_3\,,\end{aligned}$$ for $\lambda$ in the range $\Delta = \left \{(\lambda_1,\lambda_2,
\lambda_3) \in \real^3 \;|\; \lambda_3 >0 \;\mbox{\rm and} \;
\lambda_2 >0 \right \}$.
The rank of the jacobian matrix $J(\lambda)$ turns out to equal $3$ for all $\lambda \in
\Delta$, so that the stratum is connected.
### Stratum $S^{(2)}_A$.
A typical point is $x_{\rm t}=(1,0,0,0,0,1,0,0)$ and the isotropy subgroup $H$ of $G$ at $x_{\rm t}$ is the subgroup formed by the rotations (proper and improper) around the first axis, represented by block-diagonal matrices of the form diag$(1,O)$, with $O\in $ O(2). The vector space $V$ left invariant by $H$ is two–dimensional:
$$V = \left\{ x \in \real^8 \;
\mid\;x_2=x_3=x_4=x_5=x_7=x_8=0\right\}\,.$$
$K$ is the $\integer_2$ group generated by the matrix diag$(1,-1)$. A MIB for $(K,V)$ is the following: $$(\lambda_1, \lambda_2) =(x_1 , {x_6}^2) \,.$$ The associated matrix $\widehat\Lambda(\lambda)$ turns out to be [$${\widehat\Lambda(\lambda)=\mbox{\rm diag}
(1, 4 \lambda_2) \,.}$$]{} Parametric equations, $p=\phi(\lambda)$, for the stratum can be written in the following form: $$\begin{aligned}
\label{parametrizzazione2}
\phi_1(\lambda) &=& {\lambda_1}^2 + \lambda_2 \,,\nonumber\\
\phi_2(\lambda) &=& \lambda_2\,,\nonumber\\
\phi_3(\lambda) &=& -2 \, \sqrt{3}\,{\lambda_1}^{3}\,,\nonumber\\
\phi_4(\lambda) &=& - 2 \,\sqrt{3} \, \lambda_1 \lambda_2 \,, \nonumber\\
\phi_5(\lambda) &=& 4 {\lambda_1}^{2} \lambda_2 \,,\end{aligned}$$ for $\lambda$ in the range $\Delta = \left \{(\lambda_1,\lambda_2)
\in \real^2 \;|\; \lambda_2 >0 \right \}$. The rank of $J(\lambda)$ is equal to 2 on the whole $\Delta$.
### Stratum $S^{(2)}_B$.
A typical point is $x_{\rm t}=(1,1,0,0,0,0,0,0)$ and the isotropy subgroup $H$ of $G$ at $x_{\rm t}$ is the subgroup $\integer_2\times\integer_2\times\integer_2$ formed by the reflections of the axes in $\real^3$. The vector space left invariant by $H$ reduces to $$V = \left\{x\in \real^8 \mid
x_3=x_4=x_5=x_6=x_7=x_8=0\right\}\,.$$ The group $K$ is the finite group of order 6 generated by reflections with root system of type A$_2$ [@fini] with generators
$$\left( \begin{array}{cc}
-1/2 & \sqrt{3}/2 \\
\sqrt{3}/2 & 1/2
\end{array} \right) \,,\quad \quad
\left( \begin{array}{cc}
-1/2 & -\sqrt{3}/2 \\
\sqrt{3}/2 & -1/2
\end{array} \right)\;.$$
A MIB for $\mbox{\rm Stab }(H,G)$ is $$(\lambda_1,\lambda_2) = \left( {x_1}^2 + {x_2}^2\,,\,
x_1 ( {x_1}^2 - 3 \,{x_2}^2) \right)\,.$$ The associated matrix $\widehat\Lambda(\lambda)$ turns out to be [$${\widehat\Lambda(\lambda)=\left(\begin{array}{ll}
4 \lambda_1 & 6 \lambda_2 \\
6 \lambda_2 & 9 {\lambda_1}^2
\end{array}\right)\,.}$$]{} Parametric equations, $p=\phi(\lambda)$, for the stratum can be written in the following form: $$\label{parametrizzazione3}
\left(\phi_1(\lambda),\phi_2(\lambda),\phi_3(\lambda),
\phi_4(\lambda), \phi_5(\lambda) \right)= \left(\lambda_1 \,,
0\,, - 2\, \sqrt{3} \lambda_2\,, 0 \,, 0 \right)\,,$$ for $\lambda$ in the range $\Delta = \left \{(\lambda_1,\lambda_2)
\in \real^2 \;|\; \lambda_1 >0 \; \mbox{\rm and} \; {\lambda_1}^3-
{\lambda_2}^2>0 \right \}$. The rank of $J(\lambda)$ is equal to $2$ for all $\lambda\in\Delta$.
### Stratum $S^{(1)}$.
A typical point is $x_{\rm t}=(1,0,0,0,0,0,0,0)$ and the isotropy subgroup $H$ of $G$ at $x_{\rm t}$ consists of the O(3) block-diagonal matrices of the form diag$(\pm 1,O)$, with $O \in $ O(2). The vector space left invariant by $H$ reduces to $$V = \left\{x\in \real^8 \mid
x_2=x_3=x_4=x_5=x_6=x_7=x_8=0\right\}\,,$$ the group $K$ is trivial and a MIB for $\mbox{\rm Stab }(K,V)$ is simply $\lambda_1 = x_1$. The associated matrix $\widehat\Lambda(\lambda)$ reduces to $1$.
Parametric equations, $p=\phi(\lambda)$, for the stratum can be written in the following form: $$\label{parametrizzazione4}
\left(\phi_1(\lambda),\phi_2(\lambda),\phi_3(\lambda),\phi_4(\lambda),\phi_5(\lambda)\right) =
\left( {\lambda_1}^2 \,, 0\,, -2 \, \sqrt{3}\,{\lambda_1}^3\,,
0 \,, 0 \right)\,,$$ for $\lambda$ in the range $\Delta = \real$. One can verify that the rank of $J(\lambda)$ diminishes in the allowed range $\Delta$ for $\lambda_1=0$.
A physically interesting non coregular variant of the example
-------------------------------------------------------------
The example presented in the first part of this section has been suggested by a paper of Longa and Trebin [@1], devoted to the construction of a phenomenological theory of polar structures in chiral biaxial liquid crystals, through the exploitation of the properties, under transformations of a symmetry group SO(3), of a symmetric and traceless tensor order parameter field $Q$ and of a polar vector field $\bi{P}$.
In fact, the situation examined by Longa and Trebin can be recovered from the example worked about, considering the action in $\real^8$ of the subgroup SO(3) of O(3). In this case, we obtain a non-coregular linear group $G'$. To get a MIB for $G'$, a new $G'$-invariant polynomial $p_6(x)$, of degree 6, has to be added to the MIB (\[BaseInt\]): $$p_6 = 2\, \sqrt{2} \left(\bi{P} \wedge Q \,\bi{P}\right) \cdot
({Q}^2 \bi{P})\;.$$ The associated to the enlarged MIB can be constructed from the one shown in (\[BaseInt\]) by adding one more row and column: $$\begin{aligned}
\widehat P_{1\,6}&=& 12 p_6\,, \nonumber\\
\widehat P_{2\,6}&=& 6 p_6\,, \nonumber\\
\widehat P_{3\,6}&=& 0\,, \nonumber\\
\widehat P_{4\,6}&=& 0\,, \nonumber\\
\widehat P_{5\,6}&=& 12\, p_1 p_6\,, \nonumber\\
\widehat P_{6\,6}&=& -\frac{1}{9}\left[ -{{p_1}}^2\,{{p_4}}^2 +
{p_4}\,\left( {p_3} + {p_4} \right) \,{p_5} - 3\,{p_1}\,{{p_5}}^2
+
2\,{p_1}\,{p_2}\,\left( - {p_3}\,{p_4} - {{p_4}}^2 +
6\,{p_1}\,{p_5} \right) + \right. \nonumber\\
&& \left.\hspace{2em}-
{{p_2}}^2\,\left( {{p_3}}^2 - 2\,{p_3}\,{p_4} - 3\,{{p_4}}^2 +
12\,{p_1}\,{p_5} \right) \right]\,.\end{aligned}$$ The added element $p_6(x)$ is a “numerator invariant” for the Molien function [@Stan] and is, therefore, algebraically dependent on the other elements of the MIB. The relevant relation can be easily obtained from one of the irreducible polynomial factors of the enlarged $\widehat P$ matrix:
$$243 \,{p_6}^2 + A(p) =0\,,$$
where $A(p)$ is the same (see (\[attivo\])) as in the coregular case.
Being $G'$ a subgroup of $G$, the lattice of isotropy types of (SO(3),$\real^8$) is easily found from the following relation, holding true for all $x\in \real^8$:
[$${G'_x\, =\, G_x \bigcap G'\;.}\label{isotropy}$$]{}
An explicit calculation shows that, the number of strata of (SO(3),$\real^8$) is lower than the number of strata of (O(3),$\real^8$). In fact, the four–dimensional stratum $S^{(4)}$ of the second group space is part of the principal stratum of (SO(3),$\real^8$). The result is summarized in , where the notations are the same used for the generators in the O(3) case.
As explained at the end of section 2, if SO(3) is assumed to be the (largest) symmetry group of the free energy, the strata of $\real^8/G'$ are in a one-to-one correspondence with the structural phases of the system, [*i.e.*]{}, with the phases of the system that are identified only on the basis of their SO(3) symmetry. An easy comparison with the results summarized in table I (p. 3460) of [@1], shows that distinct polar states ($F_{\mbox{\rm ch}}$, $F_{Bg}$ and $F_{U2}$) defined by Longa and Trebin, lie on a same stratum (the principal stratum) of $(\mbox{\rm SO}(3),\real^8)$, meaning that their symmetry (orbit type) is the same: they form a unique structural phase. As stated above, the number of structural phases is increased if the symmetry group of the free energy is enlarged to O(3) (see our ), but also in this case the polar states $F_{Bg}$ and $F_{U2}$ lie on the same stratum $S^{(4)}$.
In their paper, Longa and Trebin make a big effort to determine the range of the orbit map and produce a classification of the polar states that, however, appears to be not based on their symmetry properties with respect to the symmetry group of the free energy. The effectiveness and mathematical rigor of the orbit space approach to the study of invariant functions clearly emerges from a comparison with the easy calculation, that led us, essentially, to more rigorous results, and with the additional important advantage of keeping, at every step, a clear correspondence between geometrical structures (strata) and symmetry of the corresponding physical configurations. In fact, the parametrization technique becomes very important for orbit spaces of group actions where more than three invariants are involved; here the geometrical intuition based on drawing the shape of the orbit space loses most of its efficiency.
References {#references .unnumbered}
==========
[pippo]{} Landau L D and Lifschitz E 1958 [*Statistical Physics*]{} (London: Pergamon) O’Raifeartaigh L 1979 [*Rep. Prog. Phys.*]{} [**49**]{} 159-223 Bernstein J 1974 [*Rev. Mod. Phys.*]{} [**46**]{} 7-48 Michel L 1980 [*Rev. Mod. Phys.*]{} 52 617-651 Chossat P and Lauterbach R 2000 [*Methods in equivariant Bifurcation Theory and Dynamical Systems*]{} (New Jersey, London, Singapore, Hong Kong: World Scientific) Mumford D 1965 [*Geometric invariant theory (Erg. Math., Bd. [**34**]{})*]{} (Berlin/Heidelberg/New York: Springer–Verlag) Munford D and Fogarty J 1982 [*Geometric invariant theory, 2nd ed. ( Ergeb. Math. Grenzgeb. (3) [**34**]{})*]{} (Berlin/Heidelberg/New York: Springer–Verlag) Abud M and Sartori G 1981 The geometry of orbit-space and natural minima of Higgs potentials 147-152 Abud M and Sartori G 1983 The geometry of spontaneous symmetry breaking 307-372 Sartori G 1983 A theorem on orbit structures (strata) of compact linear Lie groups 765-768 Procesi C and Schwarz G W 1985 Inequalities defining orbit spaces [*Invent. Math.*]{} [**81**]{} 539-554 Procesi C and Schwarz G W 1985 The geometry of orbit spaces and gauge symmetry breaking in supersymmetric gauge theories 117-121. Bredon G E 1972 [*Introduction to Compact Transformation Groups*]{} (New York: Academic Press) Schwarz G W 1980 Lifting smooth homotopies of orbit spaces [*Inst. Hautes Etudes Sci. Publ. Math.* ]{} [**51**]{}, 37-135 Sartori G 1991 Geometric invariant theory: a model-independent approach to spontaneous symmetry and/or supersymmetry breaking [*La Rivista del Nuovo Cimento*]{} [**14**]{}, 1-120. Hilbert D 1890 Ueber die Theorie der algebraischen Formen [*Math. Ann.*]{} [**36**]{} 473-534 Hilbert D 1893 Ueber die vollen Invariantensysteme [*Math. Ann.*]{} [**42**]{} 313-373 Noether E 1916 Der Endlichkeitssatz der Invarianten endlicher Gruppen [*Math. Ann.*]{} [**77**]{} 89-92 Nagata M 1964 Invariants of a group in an affine ring [*J. Math. Kyoto Univ.*]{} [**3**]{} 369-377 Whitney H 1957 Elementary structure of real algebraic varieties [*Ann. of Math.*]{} [**66**]{} 545-556. Jai Sam Kim 1984 Orbit spaces of low–dimensional representations of simple compact connected Lie groups and extrema of a group–invariant scalar potential [*J. Math. Phys.*]{} [**25**]{} 1694-1717. Sartori G and Talamini V 1994 Four dimensional orbit spaces of compact coregular linear groups [*J. of Group Theory in Physics*]{} [**2**]{} 13-39 \[http://xxx.lanl.gov/abs/hep-th/9512067\]. Sartori G and Valente G 1996 Orbit spaces of reflection groups with 2, 3 and 4 basic polynomial invariants 193-223 Sartori G and Talamini V 1991 Universality in orbit spaces of compact linear groups [*Commun. Math. Phys.*]{} [**139**]{} 559-588 Bochnak J, Coste M and Roy M–F 1998 [*Real Algebraic Geometry*]{} (Berlin Heidelberg New York: Springer–Verlag) Li L F 1974 Group theory of spontaneously broken gauge symmetries, [*Phys. Rev.*]{} [**D 9**]{} 1723-1739 Schwarz G W 1978 Representations of simple Lie groups with regular rings of invariants [*Invent. Math.*]{} [**49**]{} 167-191 Longa L and Trebin H 1990 Spontaneous polarization in chiral biaxial liquid crystals A [**42**]{} 3453-3469 Stanley R P 1979 Invariants of finite groups and their applications to combinatorics [*Bull. Amer. Math. Soc. New. Ser.*]{} [**1**]{} 475-511
[ccc]{} &&\
Stratum & Equalities & Inequalities\
&&\
&&\
$S^{(1)}$ & $p_2=p_4=p_5=12 \, {p_1}^3-{p_3}^2 = 0$ & $p_1>0$\
&&\
&&\
$S^{(2)}_{A}$ & $p_1 - p_2- {\displaystyle \frac{{p_4}^2}{12 \,
{p_2}^{2}}}= p_3 - {\displaystyle\frac{{p_4}^3}{12 \, {p_2}^{3}}}=
p_5- {\displaystyle\frac{{p_4}^2}{3 \, p_2}}=0$ & $p_2>0$\
&&\
&&\
$S^{(2)}_{B}$ & $p_2=p_4=p_5= 0$ & $12 \, {p_1}^3-{p_3}^2>0$\
&&\
&&\
$S^{(3)}$ & $p_3-p_4 \left( 3 - {\displaystyle \frac{3 p_1}{p_2}}
+ {\displaystyle
\frac{{p_4}^2}{3 {p_2}^3}}\right)=
p_5- {\displaystyle \frac{{p_4}^2}{3 \, p_2}}=0$ & $p_1>p_2>0\,,\;\;$\
& & $12\, p_1 {p_2}^2- 12\, {p_2}^3 - {p_4}^2 >0 $\
&&\
[ccc]{} &&\
Stratum & Equalities & Inequalities\
&&\
&&\
$\Sigma^{(2)}$ & $\lambda_3=\lambda_4= 0$ & $\lambda_2>0$\
&&\
&&\
$\Sigma^{(3)}$ & $ 4 \lambda_2 {\lambda_3}^{2}= {\lambda_4}^2$ & ${\lambda_3}>0$ and $\lambda_2 \geq 0$\
&&\
&&\
$\Sigma_p$ & & $\lambda_2>0$ and $\lambda_3>0$ and $ 4 \lambda_2 {\lambda_3}^{2}- {\lambda_4}^2 >0$\
&&\
[ccc]{} &&\
Stratum & typical point & isotropy subgroup or set of generators\
&&\
&&\
$S^{(5)}$ & (1,1,0,0,0,0,1,1,1) & $\{ \mbox{ e } \}$\
&&\
&&\
$S^{(3)}$ & (1,1,0,0,0,0,1,0,0) & $\{\mbox{\rm diag}(1,-1,-1)\}$\
&&\
$S^{(2)}_{A}$ & (1,0,0,0,0,0,1,0,0) & $\mbox{\rm diag}(1,O_{+}) $\
&&\
&&\
$S^{(2)}_{B}$ & (1,1,0,0,0,0,0,0,0) & $\{\mbox{\rm diag}(1,-1,-1),\mbox{\rm diag}(-1,1,-1)\}$\
&&\
&&\
$S^{(1)}$ & (1,0,0,0,0,0,0,0,0) & $\mbox{\rm diag}(1,O_{+})\cup\mbox{\rm diag}(-1,O_{-})$\
&&\
[cccc]{} Stratum & isotropy subgroup & G. R. of the [*typical point*]{} & Phase in [@1]\
$S^{(5)}$ & $\{ \mbox{\rm e} \}$ &
mm 0.3 (0 0) (9.375 9.375) (28.125 9.375) (9.375 9.375) (9.375 23.75) (9.375 9.375) (18.75 15.625) 0.01 (18.75 1.5625) (1.5625 1.5625) (18.75 1.5625) (26.5625 9.375) (18.75 15.625) (9.375 23.4375) (9.375 9.375) (18.75 1.5625)
& $F_{\rm ch}$\
&&&\
&&&\
$S^{(4)}$ & $\{$diag$(\pm 1,1,1)\}$ &
mm 0.3 (0 0) (9.375 9.375) (28.125 9.375) (9.375 9.375) (9.375 23.75) (9.375 9.375) (18.75 20.3125) 0.1 (18.75 20.3125) (18.75 9.375) (18.75 20.3125) (9.375 20.3125)
& $F_{Bg}$ ($F_{U2}$)\
&&&\
&&&\
$S^{(3)}$ & $\{$diag$(1,\pm 1,\pm 1)\}$ &
mm 0.3 (0 0) (9.375 9.375) (28.125 9.375) (9.375 9.375) (9.375 23.75) (18.75 4.6875) (18.75 20.3125)
& $F_{B||}$\
&&&\
&&&\
$S^{(2)}_{A}$ & $\{$diag$(1,O)$, $ O \in $ O$(2)\}$ &
mm 0.3 (9.375 9.375) (9.375 23.75) (18.75 4.6875) (18.75 20.3125)
& $F_{U1}$\
&&&\
&&&\
$S^{(2)}_{B}$ & $\{$diag$(\pm 1,\pm 1,\pm 1)\}$ &
mm 0.3 (0 0) (9.375 9.375) (28.125 9.375) (9.375 9.375) (9.375 23.75)
&\
&&&\
&&&\
$S^{(1)}$ & $\{$diag$(\pm 1,O)$, $ O \in $ O$(2)\}$ &
mm 0.3 (9.375 9.375) (9.375 23.75) 1 (10.0 23.75)
&\
mm
(80 160) h:C v:C
(85 150) h:C v:C
(85 144) (85 127)
h:C v:C (85 120)
(85 114) (47 97) (85 114) (111 97)
h:C v:C (47 90)
h:C v:C (111 90)
(47 83) (79 67) (111 83) (79 67)
h:C v:C (79 60)
(79 54) (79 37)
h:C v:C (79 30)
(79 24) (79 7)
h:C v:C (79 0)
[^1]: A simple example of a compact connected [*linear*]{} semialgebraic variety of $\real^3$ is yielded by a polyhedron: its interior points form a unique three–dimensional primary stratum (principal stratum), while two-, one- and zero–dimensional primary strata are formed, respectively, by the interior points of each face, by the interior points of each edge, by each vertex.
[^2]: Despite its being, generally, not one-to-one, the parametrization of the orbits of a stratum by means of the points of $V$ may be useful. It has been used, for instance, by Kim [@K] to parametrize the strata of a set of low dimensional orbit spaces of compact linear Lie groups.
[^3]: To our knowledge, no general proof exists even of simple connectivity of the principal stratum of a coregular compact group; however, principal strata of coregular low dimensional ($D\le 4$) orbit spaces can be checked to be multiply connected [@685; @fini].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This article presents the dynamic interpolation problem for locomotion systems evolving on a trivial principal bundle $Q$. Given an ordered set of points in $Q$, we wish to generate a trajectory which passes through these points by synthesizing suitable controls. The global product structure of the trivial bundle is used to obtain an induced Riemannian product metric on $Q$. The squared $L^2-$norm of the covariant acceleration is considered as the cost function, and its first order variations are taken for generating the trajectories. The nonholonomic constraint is enforced through the local form of the principal connection and the group symmetry is employed for reduction. The explicit form of the Riemannian connection for the trivial bundle is employed to arrive at the extremal of the cost function. The result is applied to generate a trajectory for the generalized Purcell’s swimmer - a low Reynolds number microswimming mechanism.'
address: 'Systems and Control Engineering Department, Indian Institute of Technology Bombay, Mumbai, India, 400072'
author:
- Sudin Kadam
- 'Ravi N. Banavar'
bibliography:
- 'bib\_IROS\_IFACJSC\_DMP\_formationControl.bib'
title: Variational dynamic interpolation for kinematic systems on trivial principal bundles
---
*Dynamic interpolation, principal bundle, Riemannian geometry, robotic locomotion*
Introduction
============
Locomotion relates to a variety of movements resulting in transportation from one place to another of biological and robotic systems. Our objective is to generate a trajectory of kinematic locomoting systems which passes through a given set of points. Many of the locomotion systems’ configuration space evolves on trivial principal fiber bundle. Some of these are also governed by nonholonomic constraints. Our work is inspired by the tracking problem for systems which are required to pass through certain ordered set of points on the configuration space. This is called as the dynamic interpolation problem and some of the early work on the topic is presented in [@jackson1991dynamic], [@lin1982formulation]. The idea of such interpolating curves on non-Euclidean spaces for applications to robotics first appears in [@noakes1989cubic]. A variational approach to such problems on Riemannian manifolds and compact connected Lie groups is presented in [@crouch1991geometry]. The further extensions to nonholonomic systems using sub-Riemannian geometry is explained in [@bloch1993nonholonomic], [@bloch2017variational], whereas reduction techniques using Lie group symmetries is presented in [@altafini2004reduction].
The present work extends the theory of dynamic interpolation to the principal kinematic form of systems which evolve on trivial principal bundle having a Riemannian structure and admit nonholonomic constraints. Minimization of the squared $L^2-$norm of covariant acceleration is considered as the optimality criterion to obtain the necessary conditions as a differential equation whose solution gives the interpolating curves. As part of the procedure, the expression for the covariant derivative on principal bundle is obtained and symmetry of the structure group is employed to reduce the configuration space which eliminates the group variable. Apart from the objective that the system should pass through the given set of points in the configuration space, the approach presented also gives the expression for the open loop control inputs through the minimizing curve in the base space of the fiber bundle.
Organization of the paper
-------------------------
In the next section we explain the topology of principal bundles that many locomotion systems exhibit and show that this can be treated as a Riemannian product manifold. We also explain the principal kinematic form of systems through the local form of principal connection in this section. Section $3$ explains the preliminaries of Riemannian geometry and also introduces the setting of the problem for the main result of the paper. We then present the dynamic interpolation problem for principal kinematic systems on trivial principal bundles and the ordinary differential equation for the resulting base and group curves is derived. We apply these result to the low Reynolds number generalized Purcell’s swimmer in section $4$.
Trivial principal bundles and the Riemannian structure
======================================================
The configuration space $Q$ of locomotion systems can usually be written as the product of two manifolds. One is the base manifold $M$ which describes the configuration of the internal shape variables of the mechanism, and the other part is a Lie group $G$ which represents the macro position of the body. Such systems’ configuration space admit the topology of a trivial principal fiber bundle[^1] [@ostrowski1996geometric].
Riemannian structure on trivial principal bundles
-------------------------------------------------
For a Lie group $G$ with $e$ as its identity, the set of left invariant vector fields on $G$ is isomorphic to its Lie algebra $\mathfrak{g} = T_eG$. By using the left invariance of the Riemannian metric[^2] and defining inner product $\langle \cdot, \cdot \rangle_{\mathfrak{g}}$ on the Lie algebra, we can give $G$ a Riemannian structure as $\langle u , v \rangle_{G} = \langle T_hL_{h^{-1}} u, T_hL_{h^{-1}} v \rangle_{\mathfrak{g}}, \: \forall h \in G \text{ and for } u,v \in T_hG$, [@do1992riemannian].
We consider a trivial principal bundle $(Q, \pi, M, G)$, $x \in M, \: g \in G$ with $\langle \cdot, \cdot \rangle_M$ as the Riemannian metric on $M$ and $\langle \cdot, \cdot \rangle_G$ as the left invariant Riemannian metric on $G$. Using the product structure $T(Q \times G) \cong TM \times TG \cong TM \times \mathfrak{g}$, $Q$ can be written as a Riemannian product manifold such that for $\dot{x}_1, \dot{x}_2 \in T_xM$ and $\dot{g}_1, \dot{g}_2 \in T_gG$, a metric $\langle \cdot, \cdot \rangle_Q$ on $Q$ is then defined as $$\label{gQ defintion}
\langle (\dot{x}_1, \dot{g}_1),\: (\dot{x}_2, \dot{g}_2) \rangle_{Q} = \langle \dot{x}_1, \dot{x}_2 \rangle_{M} + \langle \dot{g}_1, \dot{g}_2 \rangle_{G} = \langle \dot{x}_1, \dot{x}_2 \rangle_{M} + \langle g_1^{-1}\dot{g}_1, g_2^{-1}\dot{g}_2 \rangle_{\mathfrak{g}}.$$
Principal kinematic systems
---------------------------
In our work we consider the principal kinematic form of systems which evolve on trivial principal fiber bundle $Q \ni (x, g)$ and is defined by the kinematic equations given by $$\begin{aligned}
\dot{x} &=u, \label{kinematic_equations1} \\
\xi &= g^{-1}\dot{g} = - \mathbb{A}(x) u \label{kinematic_equations2}\end{aligned}$$ Here the local connection $\mathbb{A} : TM \mapsto \mathfrak{g}$ maps the points in the tangent bundle $TM$ of the shape space $M$ to the Lie algebra $\mathfrak{g}$ of $G$. We recall that $u \in T_xM$ is the control input which is the shape velocity $\dot{x}\in T_xM$. Also, equations , together define the nonholonomic constraint $\xi = - \mathbb{A}(x) \dot{x}$. Many robotic and biological locomotion systems are modelled using this paradigm in the kinematic domain [@ostrowski1996geometric], [@hatton2013geometric], [@bloch1996nonholonomic]. We also note that although the kinematic form of equations is quite common for many systems, even for dynamical systems, often the simplification in such kinematic form is useful for an analysis for applications such as path planning and trajectory design.
Dynamic interpolation
=====================
In this section we derive the equations of curves as the solution of the dynamic interpolation problem for the principal kinematic form of systems on trivial principle bundles.
Preliminaries on the variational principle
------------------------------------------
Let M be a smooth Riemannian manifold with the Riemannian metric $g^M$ denoted by $\langle \cdot,\cdot \rangle_{g^M} : T_pM \times T_pM \to \mathbb{R}$ at each point $p \in M$, where $T_pM$ is the tangent space of $M$ at $p$. A Riemannian connection $\overset{M}{\nabla}$ on $M$, is a map that assigns to any two smooth vector fields $X$ and $Y$ on $M$ a new vector field, $\overset{M}{\nabla}_XY$ on $M$. Given vector fields $X$, $Y$ and $Z$ on $M$, the vector field $R_M(X, Y )Z$ is given by $$\label{RicciTensor}
R_M(X, Y )Z = \overset{M}{\nabla}_X\overset{M}{\nabla}_Y Z - \overset{M}{\nabla}_Y \overset{M}{\nabla}_XZ - \overset{M}{\nabla}_{[X,Y ]}Z,$$ where $R_M$ is the curvature tensor on $M$ and $[X, Y ]$ denotes the Lie bracket of the vector fields $X$ and $Y$. For the properties of connection $\overset{M}{\nabla}$ and the curvature, we refer the reader to [@do1992riemannian].
Dynamic interpolation for principal kinematic systems
-----------------------------------------------------
Consider the set of all $C^1$ piecewise smooth time parameterized curves $q : [T_0, T_N] \to Q=M \times G$ on $(Q, \pi, M, G)$ such that $q(t)=\left(x(t),g(t)\right)$, where $x(t)$ is the curve on $M$ and $g(t)$ is the curve on $G$. Consider proper variations $\underline{q}$ of $q$ from a family of curves passing through fixed points $q_i, \: i = 0,1,...,N$ and fixed times $T_i$ where $T_0 < T_1 < \cdots < T_{N-1} < T_N$, defined by $\underline{q} : (-\epsilon, \epsilon) \times [T_0, T_N ] \to Q, \: (s,t) \mapsto \underline{q}(s,t) = q_s(t)=(x_s(t),g_s(t)), \: s \in (-\epsilon, \epsilon),\: \epsilon \in \mathbb{R}^+$ and $$\begin{aligned}
& \underline{q}(s,t)|_{s=0} = q(t) = (x(t), g(t) )\qquad \forall t \in [T_0, T_N], \label{condition1} \\
& \underline{q}(s,T_i) = q_i, \qquad \forall s \in (-\epsilon, \epsilon), \quad i = 0,1,...,N. \label{condition2}\end{aligned}$$ We call the curves with fixed $s,\: \underline{q}(s, t)|_{s=const}$ as main curves and those with fixed $t$, $\underline{q}(s, t)|_{t=const}$ as transverse curves. For a fixed $t \in [T_0, T_N]$, the variational vector field $\delta q$ on $Q$ is defined as $$\delta q(t) = \left( \delta x(t), \delta g(t) \right) = \evalat[\Big]{\frac{d}{ds}}{s=0}\:\underline{q}(s,t).$$ The left invariance of the vector fields $\dot{g}$ and $\delta g$ allows us to write them in terms of the pullback to the group identity. We call $\mathfrak{t}(t)$ and $\mathfrak{s}(t)$ as the $\mathfrak{g}$-valued infinitesimal variations corresponding to $\dot{g}(t)$ and $\delta g(t)$, respectively. The tangential and variational vector fields $\dot{q}(t)$ and $\delta q(t)$ can now be written in terms of these $\mathfrak{g}-$valued vector fields using the tangent lift of the group action $T_e\phi: G \times \mathfrak{g} \to \mathfrak{g}$ as $$\begin{aligned}
\dot{q}(t) &= (\dot{x}(t), \dot{g}(t)) = T_e\phi_{g(t)}(\dot{x}(t), \mathfrak{t}(t))= (\dot{x}(t), g(t)\mathfrak{t}(t) ) \text{\qquad \qquad \, (Tangential vector field)} \label{Tangential vector field}\\
\delta q(t) &= (\dot{x}(t), \delta g(t)) = T_e\phi_{g(t)}(\dot{x}(t), \mathfrak{s}(t)) = (\dot{x}(t),g(t)\mathfrak{s}(t)) \text{\qquad \:\:\:\:\:\: (Variational vector field)} \label{Variational vector field}\end{aligned}$$ We note that since the variation is proper, the variational vector field vanishes at all the interpolation points, i.e. $\delta q(T_i) = (\delta x(T_i),\: g^{-1}(T_i)\delta g(T_i)) = (\delta x(T_i),\: \mathfrak{s}(T_i)) = (0, 0)$ for $i = 0, ..., N$. For compactness of notation, in the subsequent parts of the paper we write $q_s(t),\:x_s(t),\:g_s(t)$ as $q_s,\:x_s,\:g_s,$ respectively. We now state the main result of the paper on the dynamic interpolation of principal kinematic systems.
A necessary condition for $q(t) = (x(t), g(t)), \: t \in [T_0, T_N]$ to be an extremal of the cost function $\mathcal{J}(q(t)) = \frac{1}{2} \int_{T_0}^{T_N} \langle \overset{Q}{\nabla}_{\dot{q}(t)} \dot{q}(t), \overset{Q}{\nabla}_{\dot{q}(t)} \dot{q}(t) \rangle_Q dt$ for the principal kinematic system defined by and over the class of $C^1$ paths $q_s(t)$ on $Q$ satisfying $$\label{BCs}
q_s(T_i) = q_{T_i} = (x_{T_i},\: g_{T_i})\:\:\: \forall i = 0, ... , N\: \text{ and }\: \dot{q}_s(T_0) = (\dot{x}_{T_0}, \dot{g}_{T_0}),\: \: \dot{q}_s(T_N) = (\dot{x}_{T_N}, \dot{g}_{T_N}) \quad \forall s \in (-\epsilon, \epsilon)$$ is that for every $t\in [T_{i-1},T_i], \: i = 1,...N$, the following equations hold - $$\begin{aligned}
& \overset{M}{\nabla^3}_{\dot{x}}\dot{x}+R_M \big( \overset{M}{\nabla}_{\dot{x}}\dot{x},\dot{x}\big) \dot{x} - \mathbb{A}^T(x) \Big( \dddot{\mathfrak{t}}+3\overset{G}{\nabla}_{\mathfrak{t}}\ddot{\mathfrak{t}} + 3 \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\dot{\mathfrak{t}} + \overset{G}{\nabla}_{\ddot{\mathfrak{t}}}\mathfrak{t} + 3\overset{G}{\nabla}\!\!\phantom{.}^{2}_{\mathfrak{t}}\dot{\mathfrak{t}} + 2\overset{G}{\nabla}_{\mathfrak{t}} \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\mathfrak{t} + \nonumber \\
& \qquad \qquad \qquad \qquad \qquad \qquad \quad \overset{G}{\nabla}_{\dot{\mathfrak{t}}} \overset{G}{\nabla}_{\mathfrak{t}} \mathfrak{t} + \overset{G}{\nabla}\!\!\phantom{.}^{3}_{\mathfrak{t}}\mathfrak{t} + R_G(\dot{\mathfrak{t}},\mathfrak{t})\mathfrak{t}+R_G(\overset{G}{\nabla}_\mathfrak{t} \mathfrak{t}, \mathfrak{t})\mathfrak{t} \Big) = 0, \label{FinalEquation} \\
&\mathfrak{t} = - \mathbb{A}(x) \dot{x}. \label{reconstructionEquation}\end{aligned}$$
We first state a few lemmas which will be used in the derivation of the result.
[(Variation on a Riemannian manifold [@crouch1991geometry])]{}\[Lemma1\] For a smooth curve $x(t) \in M$, $t \in [T_0, T_N ]$, such that $x(T_i) = x_i, \: \: i = 0, ... N,\: \dot{x}|_{t=T_0} = v_0$ and $\dot{x}|_{t=T_N} = v_N$, $\delta x(t)$ as the variational vector field corresponding to variations $x_s(t)$, the variation of $J(x(t)) = \frac{1}{2} \int_{T_0}^{T_N} \langle \overset{M}{\nabla}_{\dot{x}} \dot{x}, \overset{M}{\nabla}_{\dot{x}} \dot{x} \rangle dt$ is $$\begin{aligned}
\label{DynInterpRiemManifold}
\frac{d}{ds} \left( \frac{1}{2} \int_{T_0}^{T_N} \langle \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s, \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s \rangle dt \right) = &\int_{T_0}^{T_N} \Big \langle
\overset{G}{\nabla}\!\!\phantom{.}^{3}_{\dot{x}_s} \dot{x}_s + R(\delta x_s, \overset{M}{\nabla}_{\dot{x}_s}\dot{x}_s)\dot{x}_s, \: \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s \Big \rangle+ \nonumber \\
& \qquad \sum_{i=1}^{N} \left[ \left\langle \overset{M}{\nabla}_{\delta x_s}\dot{x}_s, \overset{M}{\nabla}_{\delta x_s}\dot{x}_s \right\rangle - \Big\langle \overset{M}{\nabla}_{\delta x_s}\dot{x}_s, \delta x_s \Big\rangle \right]_{T_{i-1}^+}^{T_{i}^-}.\end{aligned}$$
[(Variation on a Lie group [@altafini2004reduction])]{}\[DynInterpSDPLieGroup\] For a smooth curve $g(t) \in G$, a Lie group, with $\mathfrak{g}-$valued tangent vector field $g^{-1}(t) \dot{g}(t) = \mathfrak{t} \in \mathfrak{g}$ and $\mathfrak{s}$ as the $\mathfrak{g}-$valued variational vector field corresponding to proper variations of $g(t)$. The variation of $J(g_s(t)) = \frac{1}{2} \int_{T_0}^{T_N} \langle \overset{G}{\nabla}_{\dot{g}_s(t)} \dot{g}_s(t), \overset{G}{\nabla}_{\dot{g}_s(t)} \dot{g}_s(t) \rangle dt$ is given as $$\begin{aligned}
\frac{d}{ds} \left( \frac{1}{2} \int_{T_0}^{T_N} \langle \overset{G}{\nabla}_{\dot{g}_s(t)} \dot{g}_s(t), \overset{G}{\nabla}_{\dot{g}_s(t)} \dot{g}_s(t) \rangle dt \right) &= \int_{T_0}^{T_N} \Big \langle
\dddot{\mathfrak{t}}+3\overset{G}{\nabla}_{\mathfrak{t}}\ddot{\mathfrak{t}} + 3 \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\dot{\mathfrak{t}} + \overset{G}{\nabla}_{\ddot{\mathfrak{t}}}\mathfrak{t} + 3\overset{G}{\nabla}\!\!\phantom{.}^{2}_{\mathfrak{t}}\dot{\mathfrak{t}} + 2\overset{G}{\nabla}_{\mathfrak{t}} \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\mathfrak{t} + \overset{G}{\nabla}_{\dot{\mathfrak{t}}} \overset{G}{\nabla}_{\mathfrak{t}} \mathfrak{t} + \nonumber \\
& \qquad\qquad \overset{G}{\nabla}\!\!\phantom{.}^{3}_{\mathfrak{t}}\mathfrak{t} + R_G(\dot{\mathfrak{t}},\mathfrak{t})\mathfrak{t}+R_G(\overset{G}{\nabla}_\mathfrak{t} \mathfrak{t}, \mathfrak{t})\mathfrak{t},\: \mathfrak{s} \Big \rangle dt \label{DynInterpLieGroup}\end{aligned}$$
We now present the proof of theorem 1. The variation of the cost function $\mathcal{J}$ can be written as $$\begin{aligned}
\frac{d}{ds} \mathcal{J}(q_s(t)) &= \int_{T_0}^{T_N} \Big\langle \overset{Q}{\nabla}_{\delta q_s} \overset{Q}{\nabla}_{\dot{q}_s} \dot{q}_s, \overset{Q}{\nabla}_{\dot{q}_s} \dot{q}_s \Big\rangle_{Q} dt \\
&= \int_{T_0}^{T_N} \Big\langle \overset{Q}{\nabla}_{(\delta x_s, \delta g_s)} \overset{Q}{\nabla}_{(\dot{x}_s, \dot{g}_s)} (\dot{x}_s, \dot{g}_s), \overset{Q}{\nabla}_{(\dot{x}_s, \dot{g}_s)} (\dot{x}_s, \dot{g}_s) \Big\rangle_{Q} dt \\
&= \int_{T_0}^{T_N} \Big\langle \big( \overset{M}{\nabla}_{\delta x_s} \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s, \overset{G}{\nabla}_{\delta g_s} \overset{G}{\nabla}_{\dot{g}_s} \dot{g}_s \big), \big( \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s, \overset{G}{\nabla}_{\dot{g}_s} \dot{g}_s \big) \Big\rangle_{Q} dt \tag*{using the result in \ref{appendixB} }\\
&= \int_{T_0}^{T_N} \left[\Big \langle \overset{M}{\nabla}_{\delta x} \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s, \overset{M}{\nabla}_{\dot{x}_s} \dot{x}_s \Big\rangle_{M} + \Big\langle \overset{G}{\nabla}_{\delta g_s} \overset{G}{\nabla}_{\dot{g}_s} \dot{g}_s, \overset{G}{\nabla}_{\dot{g}_s} \dot{g}_s \Big \rangle_{G} \right] dt \tag*{from the metric defined in \eqref{gQ defintion} }\end{aligned}$$ Using lemma 1 and 2 in expanding the first and second terms respectively, and evaluating at $s=0$, we get $$\begin{aligned}
&\evalat[\Big]{\frac{d}{ds}}{s=0} \mathcal{J}(q_s(t)) = \int_{T_0}^{T_N} \Bigg [ \Big\langle \overset{M}{\nabla^3}_{\dot{x}}\dot{x}+R_M \Big( \overset{M}{\nabla}_{\dot{x}}\dot{x},\dot{x}\Big) \dot{x}, \delta x_s \Big\rangle_M + \Big\langle \dddot{\mathfrak{t}}+ 3\overset{G}{\nabla}_{\mathfrak{t}}\ddot{\mathfrak{t}} + 3 \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\dot{\mathfrak{t}} + \overset{G}{\nabla}_{\ddot{\mathfrak{t}}}\mathfrak{t} + 3\overset{G}{\nabla}\!\!\phantom{.}^{2}_{\mathfrak{t}}\dot{\mathfrak{t}} + 2\overset{G}{\nabla}_{\mathfrak{t}} \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\mathfrak{t} + \\
& \quad \overset{G}{\nabla}_{\dot{\mathfrak{t}}} \overset{G}{\nabla}_{\mathfrak{t}} \mathfrak{t} + \overset{G}{\nabla}\!\!\phantom{.}^{3}_{\mathfrak{t}}\mathfrak{t} + R_G(\dot{\mathfrak{t}},\mathfrak{t})\mathfrak{t}+R_G(\overset{G}{\nabla}_\mathfrak{t} \mathfrak{t}, \mathfrak{t})\mathfrak{t} , \mathfrak{s} \Big\rangle_{\mathfrak{g}} \Bigg ]_{s=0} dt + \sum_{i=1}^{N}\left[ \left\langle \overset{M}{\nabla}_{\delta x_s}\dot{x}_s, \overset{M}{\nabla}_{\delta x_s}\dot{x}_s \right\rangle_{s=0} - \Big\langle \overset{M}{\nabla}_{\delta x_s}\dot{x}_s, \delta x_s \Big\rangle_{s=0} \right]_{T_{i-1}^+}^{T_i^-}\end{aligned}$$ Since the variations are proper, $\delta x_s(t)|_{t=T_i} = 0$ and $\mathfrak{s}|_{t=T_i} = 0$. Hence, we get $$\begin{aligned}
\evalat[\Big]{\frac{d}{ds}}{s=0} \mathcal{J}(q_s(t)) &= \int_{T_0}^{T_N} \Bigg [ \left\langle \overset{M}{\nabla^3}_{\dot{x}}\dot{x}+R_M \left( \overset{M}{\nabla}_{\dot{x}}\dot{x},\dot{x}\right) \dot{x}, \delta x_s \right\rangle_M dt + \Big\langle \dddot{\mathfrak{t}}+3\overset{G}{\nabla}_{\mathfrak{t}}\ddot{\mathfrak{t}} + 3 \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\dot{\mathfrak{t}} + \overset{G}{\nabla}_{\ddot{\mathfrak{t}}}\mathfrak{t} + 3\overset{G}{\nabla}\!\!\phantom{.}^{2}_{\mathfrak{t}}\dot{\mathfrak{t}} + \\
& \qquad\qquad 2\overset{G}{\nabla}_{\mathfrak{t}} \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\mathfrak{t} + \overset{G}{\nabla}_{\dot{\mathfrak{t}}} \overset{G}{\nabla}_{\mathfrak{t}} \mathfrak{t} + \overset{G}{\nabla}\!\!\phantom{.}^{3}_{\mathfrak{t}}\mathfrak{t} + R_G(\dot{\mathfrak{t}},\mathfrak{t})\mathfrak{t}+R_G(\overset{G}{\nabla}_\mathfrak{t} \mathfrak{t}, \mathfrak{t})\mathfrak{t} , \mathfrak{s} \Big\rangle_{\mathfrak{g}} \Bigg ]_{s=0} dt\end{aligned}$$ For a principal kinematic system, the nonholonomic constraints are defined by $\xi = \mathfrak{t} = - \mathbb{A}(x) \dot{x}$. To obtain a curve which satisfies these nonholonomic constraints using the variational approach, the variations $\delta x$ and $\mathfrak{s}$ should also satisfy the constraints $\mathfrak{s} = - \mathbb{A}(x) \delta x$. The variational equations thus become $$\begin{aligned}
\evalat[\Big]{\frac{d}{ds}}{s=0} \mathcal{J}(q_s(t)) &= \int_{T_0}^{T_N}\Bigg [ \left\langle \overset{M}{\nabla^3}_{\dot{x}}\dot{x}+R_M \big( \overset{M}{\nabla}_{\dot{x}}\dot{x},\dot{x}\big) \dot{x}, \delta x_s \right\rangle_M dt + \Big\langle \dddot{\mathfrak{t}}+3\overset{G}{\nabla}_{\mathfrak{t}}\ddot{\mathfrak{t}} + 3 \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\dot{\mathfrak{t}} + \overset{G}{\nabla}_{\ddot{\mathfrak{t}}}\mathfrak{t} + 3\overset{G}{\nabla}\!\!\phantom{.}^{2}_{\mathfrak{t}}\dot{\mathfrak{t}} + \nonumber \\
& \qquad \qquad \:\: 2\overset{G}{\nabla}_{\mathfrak{t}} \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\mathfrak{t} + \overset{G}{\nabla}_{\dot{\mathfrak{t}}} \overset{G}{\nabla}_{\mathfrak{t}} \mathfrak{t} + \overset{G}{\nabla}\!\!\phantom{.}^{3}_{\mathfrak{t}}\mathfrak{t} + R_G(\dot{\mathfrak{t}},\mathfrak{t})\mathfrak{t}+R_G(\overset{G}{\nabla}_\mathfrak{t} \mathfrak{t}, \mathfrak{t})\mathfrak{t} , - \mathbb{A}(x) \delta x_s \Big\rangle_{\mathfrak{g}}\Bigg ]_{s=0} dt \\
&= \int_{T_0}^{T_N} \Big\langle \overset{M}{\nabla^3}_{\dot{x}}\dot{x}+R_M \big( \overset{M}{\nabla}_{\dot{x}}\dot{x},\dot{x} \big) \dot{x} - \mathbb{A}^T(x) \Big(\dddot{\mathfrak{t}}+3\overset{G}{\nabla}_{\mathfrak{t}}\ddot{\mathfrak{t}} + 3 \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\dot{\mathfrak{t}} + \overset{G}{\nabla}_{\ddot{\mathfrak{t}}} \mathfrak{t} + 3\overset{G}{\nabla}\!\!\phantom{.}^{2}_{\mathfrak{t}}\dot{\mathfrak{t}} + \nonumber \\
& \qquad \qquad 2\overset{G}{\nabla}_{\mathfrak{t}} \overset{G}{\nabla}_{\dot{\mathfrak{t}}}\mathfrak{t} + \overset{G}{\nabla}_{\dot{\mathfrak{t}}} \overset{G}{\nabla}_{\mathfrak{t}} \mathfrak{t} + \overset{G}{\nabla}\!\!\phantom{.}^{3}_{\mathfrak{t}}\mathfrak{t} + R_G(\dot{\mathfrak{t}},\mathfrak{t})\mathfrak{t}+R_G(\overset{G}{\nabla}_\mathfrak{t} \mathfrak{t}, \mathfrak{t})\mathfrak{t}\Big) , \delta x_s \Big\rangle_M dt\end{aligned}$$ Since $\evalat[\Big]{\frac{d}{ds}}{s=0} \mathcal{J}(q_s(t)) = 0$ for all the admissible variations $\delta q_s = (\delta x_s, \mathfrak{s})$ which satisfy the nonholonomic constraints , the result follows.
**Remark 3.1:** Consider $n_M$ and $n_G$ as the dimensions of $M$ and $G$, respectively. The equations and along with the group reconstruction equation $\mathfrak{t}(t) = g^{-1}(t)\dot{g}(t)$ give a set of $(n_M+n_G)-$dimensional fourth order equation in $x$ and $g$. These can be solved using the conditions to get the interpolating curve $q(t)$ for the $N$ intervals $[T_i, T_{i+1}]$ using the $4(n_M+n_G)N$ boundary conditions arising from
- The $(4n_M+4n_G)$ initial and terminal boundary conditions from and
- The interpolation conditions $q(T_i)^-=q(T_i)^+=q_{T_i}$ along with the $C^2-$smoothness conditions [@crouch1991geometry] $\dot{q}(T_i)^-=\dot{q}(T_i)^+, \: \ddot{q}(T_i)^-=\ddot{q}(T_i)^+, \: i = 1,...,N-1$ giving $(4n_M+4n_G)(N-1)$ conditions.
We also note that the procedure gives us the value of the control input $u = \dot{x}$ to trace the $C^2$ interpolating trajectory for the system from the given initial condition.
Example : The generalized Purcell’s swimmer
===========================================
[r]{}[6cm]{} {width="6cm"}
In this section we apply the dynamic interpolation results to a representative problem of great interest in the control community - the generalized Purcell’s swimmer which is a 3-link swimming mechanism moving in a fluid at low Reynolds number conditions [@kadam2018geometry], see Fig.\[Swimmer\]. The configuration space is a trivial principal bundle with the base space $SO(3) \times SO(3)$ representing the orientation of the coordinate frames $\{x_1,\: y_1,\: z_1 \},\: \{x_2,\: y_2,\: z_2 \} $ associated with the outer links $1$ and $2$ with respect to the base link frame $\{x_0,\: y_0,\: z_0 \}$ through shape variables $R_1, R_2 \in SO(3)$, respectively. The macro-position of the swimmer $g \in SE(3)$ is defined by location of the midpoint of the middle link and its orientation with respect to the inertial frame. We denote by $\xi \in \mathfrak{g}$ the body frame velocity of the base link such that $\xi_R$ and $\xi_T$ are the rotational and translational components of $\xi,$ respectively. $\hat{\omega}_1, \: \hat{\omega}_2 \in \mathfrak{so}(3)$ are the control inputs and represent the shape velocities of the outer links with respect to the base link. The swimmer motion can be written in a principal kinematic form using the Cox theory [@kadam2018geometry] with $\mathbb{A}:SO(3)\times SO(3) \times \mathfrak{so}(3) \times \mathfrak{so}(3) \to \mathfrak{se}(3)$.
Dynamic interpolation equations for the Purcell’s swimmer
---------------------------------------------------------
Using the expressions for the covariant derivative and curvature on connected and compact Lie groups[^3] and adaptation of theorem in \[appendixB\] to just the base space $M = SO(3) \times SO(3)$, which is a Riemannian product manifold, we get an expression for the first term in for the swimmer as $$\label{CovDer_Curv_on_base}
\overset{M}{\nabla^3}_{\dot{x}}\dot{x}+R \big( \overset{M}{\nabla}_{\dot{x}}\dot{x},\dot{x}\big) \dot{x} = \left[ \dddot{\omega}_1 - \ddot{\omega}_1 \times \omega_1, \: \dddot{\omega}_2 - \ddot{\omega}_2 \times \omega_2 \right]^T$$ Furthermore, for $\xi, \: \eta \in \mathfrak{se}(3)$ with metric tensor $\mathbb{I}$ on $\mathfrak{se}(3)$ as the identity matrix, the relation for covariant derivative of $\xi$ with respect to $•$ is $\eta$ is given as follows, [@altafini2004reduction] $$\label{AltafiniCovDerOnLieAlgebra}
\overset{G}\nabla_{\xi}\eta = \frac{1}{2} \left( [\xi,\eta] - \mathbb{I}^{-1} ( ad^*_{\xi} \mathbb{I} \eta + ad^*_{\eta} \mathbb{I} \xi \right) = \begin{bmatrix}
\frac{1}{2} \hat{\omega}_{\xi} \omega_{\eta} \\
\hat{\omega}_{\xi} v_{\eta}
\end{bmatrix}$$ For the generalized Purcell’s swimmer, using equations and , with $\mathbb{I}$ as the identity, we get the explicit expression for the terms in to get final set of equations as
$$\begin{aligned}
& \begin{bmatrix}
\dddot{\omega}_1 - \ddot{\omega}_1 \times \omega_1\\
\dddot{\omega}_2 - \ddot{\omega}_2 \times \omega_2
\end{bmatrix} - \mathbb{A}^T(R_1, R_2) \begin{bmatrix}
\dddot{\xi}_R + \frac{3}{2} \xi_R \times \ddot{\xi}_R + \frac{1}{2} \xi_R \times \xi_R \times \dot{\xi}_R + \frac{1}{2} \xi_R \times \dot{\xi}_R \times \xi_R + ( \dot{\xi}_R \times \xi_R ) \times \xi_R \\
\big( \dddot{\xi}_T + 3 \xi_R \times \ddot{\xi}_T + 3 \dot{\xi}_R \times \dot{\xi}_T + \ddot{\xi}_R \times \xi_T + \frac{5}{2} \xi_R \times \xi_R \times \dot{\xi}_T + ... \\
\qquad ...+\frac{7}{2} \xi_R \times \dot{\xi}_R \times \xi_T + 2 \dot{\xi}_R \times \xi_R \times \xi_T + \frac{1}{2} \xi_R \times \xi_R \times \xi_R \times \xi_T \big)
\end{bmatrix} = 0 \label{Purcell_final_eqn1} \\
& \begin{bmatrix}
\xi_R \\
\xi_T
\end{bmatrix} = -\mathbb{A}(R_1, R_2) \begin{bmatrix}
\omega_1 \\
\omega_2
\end{bmatrix} \label{Purcell_final_eqn2}\end{aligned}$$
Along with the smoothness $C^2$ smoothness conditions and the boundary conditions for $i \in \{0, 1, ... , N \}$ as $$\begin{aligned}
R_1(T_i)=\bar{R}_{1, T_i}, \:\:\:\: R_2(T_i)=\bar{R}_{2, T_i}, \:\: g(T_i)=\bar{g}_{T_i}, \tag*{(Conditions on the shape and group position)} \\
\omega_1(T_i)=\bar{\omega}_{1, T_i}, \:\:\:\: \omega_2(T_i)=\bar{\omega}_{2, T_i}, \:\:\:\: \xi(T_i)=\bar{\xi}_{T_i} \tag*{(Conditions on shape and group velocities)}\end{aligned}$$ Lastly, as part of future work, we hope to apply the results in this paper to other locomotion systems and formation control problem consisting of agents whose system model is of the principal kinematic form.
{#appendixA}
\[Levi-Civita connection on immersed submanifold\] Let $f : M^n \to \overline{M}^{n+k}$ be an immersion of differentiable manifold $M$ into a Riemmanian manifold $\overline{M}$. Assume that $M$ has an induced Riemannian metric. Let for $p \in M, \: U \subset M$ be an open neighbourhood of $p$ such that $f(U) \subset \overline{M}$ is an immersed submanifold of $\overline{M}$. Let $X, Y \in \mathfrak{X} (f(U))$ and $\overline{X}, \overline{Y}$ be their extensions[^4] on $\overline{M}$. We define $\nabla_XY(p) = (\overline{\nabla}_{\overline{X}}\overline{Y})^{T_M}$, where $\overline{\nabla}$ is a Riemannian connection on $\overline{M}$ and $(\overline{\nabla}_{\overline{X}}Y)^{T_M}$ is the tangential component of $\overline{\nabla}_{\overline{X}}Y$ to the manifold $M$. The claim is that $\nabla$ is the Riemannian connection on $M$.
We show that $\nabla$ satisfies the 2 properties in the Levi-Civita theorem as follows - **Symmetry**: For all $p \in M, \: f(p) \in f(M)$ with $x^i$ as the local coordinates of $M$ $$\begin{aligned}
(\nabla_XY - \nabla_YX)(p) &= \big( (\overline{\nabla}_{\overline{X}}\overline{Y})^{T_M} - (\overline{\nabla}_{\overline{Y}}\overline{X} )^{T_M} \big)(p) \nonumber \\
&= (\overline{\nabla}_{\overline{X}}\overline{Y} - \overline{\nabla}_{\overline{Y}}\overline{X})^{T_M}(p) \nonumber \\
&= [\overline{X}, \overline{Y}]^{T_M}(p) \nonumber
\\
&= \Bigg ( \sum_{i,j=1}^{n+k} \left( \overline{X}^i \frac{\partial \overline{Y}^j}{\partial x^i} - \overline{Y}^i \frac{\partial \overline{X}^j}{\partial x^i} \right)\frac{\partial}{\partial x_j} \Bigg )^{T_M}(p) \nonumber \\
&= \Bigg ( \sum_{i=1}^{n}\sum_{j=1}^{n+k} \left( X^i \frac{\partial \overline{Y}^j}{\partial x^i} - Y^i \frac{\partial \overline{X}^j}{\partial x^i} \right)\frac{\partial}{\partial x_j} \Bigg )^{T_M}(p) \tag*{as $X^i,\: Y^i = 0$ for $i = n+1,.., n+k$} \nonumber \\
&= \Bigg ( \sum_{i,j=1}^{n} \left( X^i \frac{\partial Y^j}{\partial x^i} - Y^i \frac{\partial X^j}{\partial x^i} \right)\frac{\partial}{\partial x_j} \Bigg )(p) \nonumber \\
&= [X, Y] \nonumber \end{aligned}$$ **Compatibility with the metric on $M$**: $$\begin{aligned}
X \langle Y, Z \rangle (p) &= \overline{X} \langle \overline{Y}, \overline{Z} \rangle (p) \nonumber \\
&= \langle \overline{\nabla}_{\overline{X}}\overline{Y}, \overline{Z} \rangle(p) + \langle \overline{Y}, \overline{\nabla}_{\overline{X}} \overline{Z} \rangle(p) \nonumber \\
&= \langle (\nabla_X Y)^{T_M}, Z \rangle (p) + \langle Y, (\nabla_X Z)^{T_M} \rangle(p) \nonumber \\
&= \langle \nabla_X Y, Z \rangle (p) + \langle Y, \nabla_X Z \rangle(p) \nonumber \end{aligned}$$ This shows that $\nabla$ is compatible with the Riemannian metric on $M$. Hence, using the Levi-Civita theorem [@do1992riemannian] $\nabla$ becomes the unique Riemannian connection on $M$ such that $\nabla_XY(p) = (\overline{\nabla}_{\overline{X}}\overline{Y})^{T_M}$.
{#appendixB}
\[Levi Civita connection on trivial principal bundle\] Let $M$ and Lie group $G$ be Riemannian manifolds and consider a trivial principal bundle $(Q, \pi, M ,G)$ as Riemannian product manifold $M \times G$ with the induced product metric $g^M \oplus g^{\mathfrak{g}}$. Let $\overset{M}{\nabla}$ be the Riemannian connection on $M$ and $\overset{G}{\nabla}$ be that on $G$. To prove that the Riemannian connection on $Q$ is given by $\overset{Q}{\nabla}_{(Y_1,Y_2)} (X_1, X_2) = (\overset{M}{\nabla}_{Y_1}X_1, \overset{G}{\nabla}_{Y_2}X_2)$ where $X_1, Y_1 \in \mathfrak{X}(M)$ and $X_2, Y_2 \in \mathfrak{X}(G)$.
Let $p \in M, \: q\in G$. Consider $f_M: M \to M \times \{q\}$ and $f_G : G \to \{p \} \times G$ as the canonical maps. We note that $f_1$ and $f_2$ are immersions of $M, \, G$ respectively into $M \times G$. Then using theorem \[Levi-Civita connection on immersed submanifold\] in \[appendixA\] and that $\overset{Q}{\nabla}$ is the Levi-Civita connection on $M \times G$, we write that $\overset{M}{\nabla}_{X_1}Y_1(p) = (\overset{Q}{\nabla}_{\overline{X}}\overline{Y})^{T_M}$ and $\overset{G}{\nabla}_{X_2}Y_2(p) = (\overline{\nabla}_{\overline{X}}\overline{Y})^{T_G}$, where $\overline{X}, \, \overline{Y}$ are the extensions of $X_1, X_2$ and $Y_1, Y_2$ defined as follows for all $(a,b) \in M \times G$ - $$\overline{X}(a,b) = (X_1(a,q), X_2(p,b)), \qquad \overline{Y}(a,b) = (Y_1(a,q), Y_2(p,b))$$ Thus, we get $$\begin{aligned}
(\overset{M}{\nabla}_{Y_1}X_1, \overset{G}{\nabla}_{Y_2}X_2) &= ((\overset{Q}{\nabla}_{\overline{Y}}{\overline{X}})^{T_{M}}, (\overset{G}{\nabla}_{\overline{Y}}{\overline{X}})^{T_G})) \\
&= \overset{Q}{\nabla}_{\overline{Y}}{\overline{X}} \\
&= \overset{Q}{\nabla}_{(Y_1, Y_2)}{(X_1,X_2})\end{aligned}$$
References {#references .unnumbered}
==========
[^1]: A *trivial principal fiber bundle* $(Q, \pi, M, G)$ with manifold $M$ as the base space and Lie group $G$ as the structure group is the manifold $Q = M \times G$ and $\pi: Q \to Q/G = M$ as the submersion together with a free left action of $G$ on $Q$ given by left translation in the group variable: $\Phi_h(x,g)=(x, hg)$ for $x \in M$ and $h,g \in G$.
[^2]: A Riemannian metric $\langle \cdot , \cdot \rangle$ on $G$ is said to be left invariant if every left translation is an isometry of $\langle \cdot , \cdot \rangle$.
[^3]: For a connected compact Lie group if $\nabla$ is the Riemannian connection associated with a left invariant metric and $X, Y$ and $Z$ are left invariant vector fields on the Lie group, then $\nabla_XY = \frac{1}{2}[X.Y]$ and the curvature $R(X,Y) Z = -\frac{1}{4}[[X,Y],Z]$, [@milnormorse].
[^4]: Let $\overline{M}$ be a Riemannian manifold and $M$ be its embedded submanifold. Let $X$ be a vector field on $M$, $X \in \mathfrak{X}(M)$. Then $\forall p \in \overline{M}$ and a vector field $\overline{X} \in \mathfrak{X}(\overline{U}$ such that $\overline{X}_p = X_p$/ $\overline{X}$ is called as the (local) extension of $X$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The estimation of viewpoints and keypoints effectively enhance object detection methods by extracting valuable traits of the object instances. While the output of both processes differ, i.e., angles vs. list of characteristic points, they indeed share the same focus on how the object is placed in the scene, inducing that there is a certain level of correlation between them. Therefore, we propose a convolutional neural network that jointly computes the viewpoint and keypoints for different object categories. By training both tasks together, each task improves the accuracy of the other. Since the labelling of object keypoints is very time consuming for human annotators, we also introduce a new synthetic dataset with automatically generated viewpoint and keypoints annotations. Our proposed network can also be trained on datasets that contain viewpoint and keypoints annotations or only one of them. The experiments show that the proposed approach successfully exploits this implicit correlation between the tasks and outperforms previous techniques that are trained independently.'
author:
- Pau Panareda Busto
- Juergen Gall
bibliography:
- '040.bib'
title: Joint Viewpoint and Keypoint Estimation with Real and Synthetic Data
---
Introduction {#sec:introduction}
============
Many camera-based applications need to identify and analyse certain object classes for a better understanding of their surroundings. While 2D object detection is often a starting point, it is usually required to extract more detailed information from the detected objects. For instance, 2D keypoints provide additional details regarding the shape of an object and the 3D viewpoint provides the information about the orientation of an object. Both tasks, however, are correlated since the locations of the 2D keypoints depend on the orientation of the object and the 2D keypoints are a cue for the 3D orientation. In this work, we exploit this implicit correlation and introduce a joint model for 3D viewpoint and 2D keypoint estimation. The proposed network generalises the human pose estimator by Wei et al. [@POSE_Wei16] to multiple objects and it is trained jointly for the two tasks. For the 3D viewpoint estimation, we propose a simple yet effective multi-granular viewpoint classification approach.
The labelling process for training our network requires nonetheless large amounts of accurate labelled data. While human annotations excel in annotating object instances by bounding boxes, they fail to accurately estimate fine 3D viewpoints [@DA_Busto15]. The same applies for annotating keypoints, which require pixel precision and a correct handling of occlusions. In order to alleviate the collection of training data, we propose two solutions. Firstly, we design our network such that it can be trained with images from different datasets. The datasets can provide annotations for only viewpoints, keypoints or both. Secondly, we make use of synthetic data to increase the amount of training samples since computer generated images are a quick way to collect many training samples, as well as precise ground truth. Specifically, we introduce a novel synthetic dataset that includes not only viewpoints, but also accurate keypoints.
We evaluate our method on 12 popular classes of the *ObjectNet3D* [@DATA_Yu16] dataset, which contains both viewpoint and keypoint annotations. We demonstrate that our method outperforms current well established methods for multi-class viewpoint and keypoint estimation.
Related Work
============
Viewpoint Estimation
--------------------
We divide viewpoint estimation techniques in two categories: regression methods that compute the pose by optimising in a continuous space [@POSE_Fenzi13; @POSE_He14; @OBJ_Pepik15] and classification-based methods that simplify the span of viewpoints into a limited set of discrete bins. From the latter, focus of our work, Liebelt and Schmid [@OBJ_Liebelt10] optimise a multi-view linear SVM and classify local features to select the winning viewpoint based on a voting approach. Busto et al. [@PANAREDABUSTO201875; @DA_Busto15] also use linear SVMs in a one-vs-all multi-class approach to refine coarse annotated viewpoints. Su et al. [@SYN_Su15] propose a classification-based CNN model with one bin per degree, i.e. 360 bins for the azimuth angle, and a Gaussian function that spreads the optimisation to neighbouring bins. The training phase uses millions of synthetic samples to compensate the fine viewpoint representation. A coarser discretisations was proposed by Tulsiani and Malik [@POSE_Tulsiani15], which showed better accuracies when trained on real data. Massa et al. [@POSE_Massa16] concluded that classification-based approaches obtain better viewpoint accuracies than regression techniques when jointly trained with an object detector in different popular CNN architectures. It has also been shown by Ghodrati et al. [@POSE_Ghodrati14] that these methods using global features extracted from the 2D bounding boxes outperform more complex methods trained on 3D data. More recently, Divon and Tal [@POSE_Divon18] introduced a triplet loss to increase the dissimilarity of viewpoints that are far apart. Viewpoint estimation can also benefit from 3D object detections, as shown by Kehl et al. [@POSE_Kehl17], who extended a popular real-time object detector with 3D viewpoint predictions.
Close in spirit to our work, other approaches already used the spatial information of keypoints to estimate accurate viewpoints. Torki and Elgammal [@POSE_Torki11] learn a regression function to compute the azimuth angle of vehicles based on pre-computed local features and their spatial arrangements. Pepik et al. [@OBJ_Pepik12] extend the deformable part model [@OBJ_Pedro10] to 3D objects, optimising at the same time the location and the viewpoint of the object for a fixed number of bins. Concretely for hand pose estimation, Zimmermann and Brox [@POSE_Zimmermann17] compute the camera parameters by using keypoint confidence maps as input of the network. A deep regression technique is presented by Wu et al. [@POSE_Wu16], where 2D keypoints are used to estimate the camera parameters after concatenating several fully connected layers. Lately, Grabner et al. [@POSE_Grabner18] use the Perspective-n-Point algorithm to extract the viewpoint from a detected 3D bounding box.
Keypoint Estimation
-------------------
Research in keypoint estimation has mostly been centred on human articulated poses [@POSE_Belagiannis17; @POSE_Chu17; @POSE_Tompson15; @POSE_Toshev14]. In this paper, we expand the CNN model for human pose estimation proposed by Wei et al. [@POSE_Wei16], who optimised confidence maps for each keypoint. This model appends the later portion of the network several times, i.e., the input of the new stage comes from the output of the previous one, creating larger receptive fields. The deeper the stacked network the more it suppresses ambiguities and better captures the spatial layout of the keypoints. Newell et al. [@POSE_Newell16] refined this architecture by adding transposed convolutions at the end of each stage for finer confidence maps.
Previous to our work, keypoint estimation in rigid objects has already been in focus. Long et al. [@POSE_Long14] initially addressed the capabilities of CNNs for keypoint estimation by dividing the last convolutional layer in smaller cells and training each keypoint as an independent class in a multi-class SVM. Moving towards a purely neural network approach, Tulsiani and Malik [@POSE_Tulsiani15] concatenate the spatial information in a fully connected layer and only activate through the network those receptive fields that include the corresponding keypoints. The prediction is further refined with independently computed viewpoints. The human pose estimation by [@POSE_Newell16] has been modified by [@POSE_Pavlakos17; @POSE_Zhou18] to detected 3D keypoints of multiple rigid classes to consequently estimate the translation and rotation of the object by fitting the keypoints into a shape model.
Synthetic Data
--------------
Synthetic dataset has been used for many years to easily increase the amount of training samples in object detection tasks [@SYN_Peng15; @SYN_Pishchulin11]. In recent years, new datasets based on computer generated models with accurate 3D pose information have been proposed. For instance, ShapeNet [@DATA_Chang15] provides a large dataset of 3D graphics models for hundreds of object classes. Its drawback comes from the low quality of most of their 3D models. From another perspetive, other approaches [@POSE_Wang18; @DATA_Xiang14] compensate the lack of photo-realism in synthetic images by aligning 3D models with real samples for accurate 3D pose annotations.
Joint Viewpoint and Keypoint Estimation {#technical}
=======================================
In this work, we propose a multi-task network that leverages 3D viewpoint and 2D keypoint estimation. We assume that an object has been already detected and our goal is to estimate the keypoints as well as the viewpoint. Our network is trained for all object classes $C = \{c_1, \ldots, c_{|C|}\}$ where the number of keypoints per object class $K_c$ varies. A second important aspect of the network is that it can be trained on various types of data including real and synthetic data at the same time. Since the data might be annotated for only one of the two tasks, $\mathcal{M}$ denotes the set of training samples with viewpoint and 2D keypoint annotations, $\mathcal{N}$ denotes the set with only viewpoint annotations and $\mathcal{O}$ the set with only keypoint annotations. An overview of the proposed CNN architecture is presented in Figure \[fig:cnn\]. We first discuss the parts that are relevant for keypoint estimation.
![ Overview of the proposed multi-class CNN for joint viewpoint and keypoint estimation. The network uses a multi-stage architecture. The first row shows the first stage, which predicts for each keypoint per class a heatmap. For the later stages (second row), the features of the first and the previous stage after the last ReLU are used as input. At each stage, an L2-loss is used, which compares the predicted heatmaps for the class of the training sample to the ground truth heatmaps. After the last stage, additional layers for viewpoint estimation are added (third row). We use a multi-resolution loss where fully connected layers map the 128x28x28 features to nine vectors corresponding to three different discretisations (15$^\circ$, 30$^\circ$, 60$^\circ$) of azimuth (az), elevation (el) and tilt (ti). []{data-label="fig:cnn"}](cnn.pdf){width="\textwidth"}
Keypoint Estimation
-------------------
The proposed network is a multi-stage architecture with intermediate loss functions after each stage and the first part is similar to the convolutional pose machines [@POSE_Wei16], which is a multi-stage network for 2D human pose estimation. The cropped image of an detected object is fed to a VGG-16 model [@CNN_Simonyan14] and additional convolutional layers are used to generate heatmaps for each keypoint and each object class. In total, we have $\sum_{c\in C} K_c$ heatmaps, where $K_c$ denotes the number of keypoints of the $c$-th class. Since the object class $c$ is known for an image during training, the $L_2$-loss is computed only for the heatmaps of the corresponding class. At the first stage $s=1$, the loss is therefore given by $$\label{eq:keyloss}
\mathcal{L}_{kp_s} = \sum_{x_i\in\{\mathcal{M},\mathcal{O}\}} \frac{1}{K_{c_i}}\sum_{k=1}^{K_{c_i}}\norm{y_{i,k} - f_s(x_i)_{c,k}}^2_2 \text{,}$$ where $x_i$ denotes a training sample from the set $\mathcal{M}$ or $\mathcal{O}$ and $f_s(x_i)$ denotes all heatmaps that are predicted for the stage $s$. The estimated heatmap for the $k$-th keypoint of class $c$ is then denoted by $f_s(x_i)_{c,k}$ and $y_{i,k}$ is the corresponding ground-truth heatmap for the training sample $x_i$. The L2-loss is computed over all pixels in the heatmap, but we write $\norm{a-b}^2_2$ instead of $\sum_{\omega\in\Omega}\norm{a(\omega)-b(\omega)}^2_2$.
As in [@POSE_Wei16], we do not use one stage but 6 stages. For each stage except of the first one, we use the heatmaps of the previous stage and the feature maps of the first stage after the last ReLU layer as input. Since heatmaps are computed at each stage $s$, we sum the loss functions over all stages, i.e., $\sum_s \mathcal{L}_{kp_s}$.
Viewpoint Estimation
--------------------
As shown in Figure \[fig:cnn\], the proposed network not only predicts the 2D keypoints but also the 3D viewpoint encoded by the three angles $\{\phi, \psi, \theta \}$, which denote azimuth ($\phi \in [0^{\circ},360^{\circ}]$), elevation ($\psi \in [-90^{\circ},90^{\circ}]$) and in-plane rotation ($\theta \in [-180^{\circ},180^{\circ}]$), respectively. We opt for a classification-based approach to estimate the viewpoints and discretise each angle using a bin size of $15^{\circ}$. We obtain the probabilities for each bin by a fully connected layer and a softmax layer for each angle. The cross-entropy loss for bin size $b=15^{\circ}$ is then given by $$\label{eq:viewloss}
\mathcal{L}_{vp_{b}} = \sum_{x_i\in\{\mathcal{M},\mathcal{N}\}} \sum_{v \in \{\phi, \psi, \theta\}} -\log \left ( f_b(x_i)_{c,v,v_i} \right ) \text{,}
$$ where $x_i$ denotes a training sample from the set $\mathcal{M}$ or $\mathcal{N}$, $v_i$ denotes the ground-truth bin for angle $v$ and $f_b(x_i)$ denotes the vector with the bin probabilities for all classes and angles. The estimated probability for the $v_i$-th bin of class $c$ and angle $v$ is then denoted by $f_b(x_i)_{c,v,v_i}$.
In addition, the network predicts during training the viewpoint for each class for two coarser discretisations of the angles, namely for $60^{\circ}$ and $30^{\circ}$. In this way, the coarse discretisations guide the network to the correct bin of the finer discretisation and improve the accuracy as we will show as part of the experimental evaluation. The multi-task loss for the network is then expressed as $$\mathcal{L} = \sum_s\mathcal{L}_{kp_s} + \sum_b\mathcal{L}_{vp_b}\text{.}$$
Since we aim at a finer viewpoint prediction than $15^{\circ}$, we upsample the estimated viewpoint probabilities to an angular resolution of $1^{\circ}$ during inference. To this end, we interpolate the probabilities by applying a cubic filter [@MATH_Keys81] as illustrated in Figure \[fig:vp\]. For the azimuth and the in-plane rotation, we convolve the discrete bins as a circular array.
![ Using a cubic filter, the probabilities of the viewpoint quantised at $15^{\circ}$ are upsampled to an angle resolution of $1^{\circ}$. Note that we have 24 bins for azimuth and $\theta$ since they are circular, but only 13 bins for elevation where the 7th bin is centred at zero elevation and the outer bins have only $7.5^{\circ}$. []{data-label="fig:vp"}](viewpoint.pdf){width="10cm"}
Experiments
===========
In this section we evaluate the performance of our method, denoted as JVK (*Joint Viewpoint and Keypoints*), and compare its results with several popular viewpoint and keypoint estimation algorithms. We train our network for 12 popular object categories, i.e., $|C|=12$, namely: *airplane*, *bicycle*, *boat*, *bottle*, *bus*, *car*, *chair*, *diningtable*, *motorbike*, *sofa*, *train* and *tvmonitor*. We then evaluate our method on the test images of the *ObjectNet3D* [@DATA_Yu16] dataset. The source code is available at <https://github.com/Heliot7/viewpoint-cnn-syn>.
Datasets
--------
### ObjectNet3D [@DATA_Yu16]
is a large dataset that contains real images of 100 object categories. From all of them, the 12 classes that we selected include not just viewpoints from aligned 3D shapes, but also manually annotated keypoints. The selected subset is evenly separated between training and test data with 11421 and 11327 images, respectively. Most of the classes contain between 500 and 1000 samples in every set. The classes *bottle* and *diningtable* are above 1000 samples and *car* above 2000 samples.
### ShapeNet [@DATA_Chang15]
is a large-scale dataset of 3D shapes whose most relevant subset contains the 12 object categories, providing a considerable amount of models for each class. Although this setting allows for an extensive image dataset with a great variety of object orientations, the low quality of the renderings produce training samples that greatly differ from real images. This dataset only provides 3D viewpoints, automatically generated from the camera parameters in the image rendering. For our experiments, we make use of all models and generate 100000 images per class with random camera viewpoints, i.e., 1200000 images in total.
### New Synthetic Data:
In this work, we introduce a new synthetic dataset from 3D graphics models for the 12 object categories. For each class, we collect 10 graphics models with higher levels of realism and more detailed meshes compared to ShapeNet. In addition to the 3D viewpoint annotations that are directly extracted from the camera rotation, we go one step further and introduce automatically generated 2D keypoints. In order to easily obtain keypoints from synthetic data, we firstly set deformable spheres in the 3D rendered model locations that we consider to be valid using the keypoints from ObjectNet3D as reference. Figure \[fig:syn\_a\] shows some 3D graphics models with spheres placed as keypoints. Then, we project the centre of each sphere to pixel coordinates for a given camera orientation to create the 2D keypoints. For the projection, we take occlusions into account. We generate synthetic data with 10000 samples per class with random orientations. Examples of rendered images are illustrated in Figure \[fig:syn\_b\] with the 2D bounding boxes and the visible 2D keypoints. The resulting images also include a background image from the KITTI dataset [@DATA_Geiger12].
Network configuration
---------------------
We train the proposed CNN model for a total of 150000 iterations when using only real images for training, 250000 iterations when including one of the two synthetic datasets and 350000 iteration for all 3 datasets. The weight decay is set to 0.0005 and the learning rate to 0.00005, which is multiplied by $0.1$ every 100000 iterations. The input image will be cropped in all experiments to 224x224 pixels while preserving the aspect ratio. The batch contains 20 samples per iteration where we sample uniformly across the datasets if we use more than one for training. In addition, standard data augmentation techniques are employed during the training of the network: flipping, in-plane rotation $[-45^{\circ},45^{\circ}]$, image scaling (0.4,1.0) and translation. However, we only add the transformed image if the intersection over union of the transformed bounding box compared to the original one is above 0.8.
For the test phase, we will extract the samples of each object class using their annotated 2D bounding boxes, i.e., without any prior object detector. We run 5 passes with different scaling factors and average all of them to obtain the final confidence map of keypoints and 3D viewpoints.
From our model, we analyse two modifications. In JVK-KP, we only train the keypoint estimation, ignoring the viewpoint extension. Then, JVK denotes the standard network for both keypoint and viewpoint sections. We also modify the training datasets that we utilise, combining the real samples from ObjectNet3D [@DATA_Yu16] with manually labelled viewpoints and keypoints (Re), ShapeNet [@DATA_Chang15] images with only viewpoints (Sh) and our novel synthetic dataset with generated viewpoints and keypoints (Sy).
Keypoint estimation
-------------------
To measure the quality of our keypoint localisation, we use the PCK\[$\alpha = 0.1$\] evaluation introduced by Yang and Ramanan [@POSE_Yang11]. An estimated keypoint is valid if the Euclidean distance with respect to the corresponding ground truth is below $\alpha \times max(h,w)$, where $h$ and $w$ are the height and width of the object’s bounding box, respectively.
As a baseline, we compare our method with the popular keypoint estimation for rigid objects [@POSE_Tulsiani15] (VpKp). We report the results of VpKp with 192x192 input resolution (192), 384x384 input resolution (384), both resolutions trained one after the other (192-384) and in a setting where the viewpoint is first estimated for the low resolution and used as input to refine the keypoints for the higher resolution (pLike).
We report the results in Table \[table:o3d\_kps\]. Firstly, we observe that JVK-KP (Re), which uses the same real data as in VpKp, already outperforms all variations of VpKp. For instance, our method has $+2.2\%$ accuracy compared to VpKp (pLike). In contrast to VpKp that requires several sequential steps and higher resolutions, we only require a small amount of forward passes of our network with rescaled images. If we compare our modifications, we see a comparable improvement when including synthetic images with only keypoints, JVK-KP (Re-Sy), or only viewpoints, JVK (Re-Sh). This shows the benefits of estimating 3D viewpoint and 2D keypoints jointly. The network trained with all three training datasets (Re-Sy-Sh) obtains the best overall PCK accuracy, which is $+0.7\%$ higher compared to the result without Shapenet (Re-Sy).
[max width=]{}
aero bike boat bottle bus car chair dtable mbike sofa train tv
-- ----------------------------------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ----------
VpKp [@POSE_Tulsiani15] (192) 74.4 80.6 60.7 81.9 80.7 89.6 71.1 52.4 78.0 76.2 57.4 47.1 70.8
VpKp [@POSE_Tulsiani15] (384) 80.1 88.6 70.7 90.0 93.7 96.5 76.7 65.4 85.2 89.1 68.7 78.7 82.0
VpKp [@POSE_Tulsiani15] (192-384) 84.1 90.0 74.4 91.3 94.4 97.5 84.9 73.3 87.4 91.0 71.3 80.1 85.0
VpKp [@POSE_Tulsiani15] (pLike) 82.7 90.7 69.2 92.6 95.8 95.6 89.5 76.3 85.9 92.5 72.0 80.3 85.3
JVK-KP (Re) 85.7 92.7 74.8 **94.5** 98.1 98.4 89.4 83.9 89.7 93.8 73.4 75.7 87.5
JVK (Re-Sh) 87.9 94.7 75.3 94.3 **98.6** **98.5** 89.6 **84.5** 90.6 **94.0** 75.0 77.0 88.3
JVK-KP (Re-Sy) 87.7 95.2 73.6 93.9 97.8 **98.5** 90.1 81.5 91.3 93.5 **75.2** 83.4 88.5
JVK (Re-Sy) 88.8 95.2 75.1 93.6 98.0 **98.5** 90.9 83.6 91.2 93.8 73.3 82.3 88.7
JVK (Re-Sy-Sh) **89.5** **95.9** **77.1** 93.9 98.2 **98.5** **91.5** 83.3 **93.0** 93.9 74.2 **84.0** **89.4**
Viewpoint estimation
--------------------
[max width=]{}
aero bike boat bottle bus car chair dtable mbike sofa train tv
-- --------------------------------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ----------
Regression (Re) [@POSE_Massa16] 79.9 81.0 66.7 93.3 92.8 96.7 90.8 79.3 83.0 96.1 94.9 89.7 87.0
VpKp (Re) [@POSE_Tulsiani15] 88.7 79.4 74.3 91.7 96.7 96.3 92.2 82.3 80.8 95.4 95.7 83.1 88.0
Render4CNN (Sh) [@SYN_Su15] 71.0 76.1 45.1 83.7 86.3 89.9 88.5 63.0 68.4 90.4 82.3 92.3 78.1
Class-15 (Re) 83.6 77.0 71.9 89.6 95.4 95.0 90.4 84.8 76.6 95.4 93.5 79.1 86.0
Class-15-30-60 (Re) 85.8 81.5 71.9 92.4 96.1 95.9 92.7 85.5 81.1 95.1 94.6 83.7 87.9
Class-15-30-60 upsamp. (Re) 86.7 82.5 73.5 92.8 95.9 96.6 93.1 85.7 81.6 96.0 94.5 85.2 88.7
Class (Re-Sy) 88.4 85.8 76.5 94.5 96.9 96.8 95.6 86.5 88.5 96.5 94.3 87.5 90.7
Class (Re-Sh) **91.5** 85.4 80.3 94.5 97.6 97.3 97.5 86.8 86.6 97.8 95.5 89.9 91.7
Class (Re-Sy-Sh) 90.7 85.7 **81.0** 93.8 98.0 97.1 **97.9** 88.3 88.5 97.9 94.6 90.3 92.0
JVK (Re) 86.3 85.1 79.0 94.5 98.5 97.8 92.2 87.7 87.5 97.1 95.1 87.5 90.7
JVK (Re-Sy) 89.8 **88.9** 78.6 **95.5** 98.3 97.4 93.5 87.3 90.5 97.2 94.0 88.9 91.7
JVK (Re-Sh) 87.7 86.8 80.6 95.1 97.8 **98.3** 96.2 **89.2** **91.3** 98.1 94.5 92.0 **92.3**
JVK (Re-Sy-Sh) 87.8 87.0 79.8 95.0 **98.7** 97.5 96.0 86.6 90.7 **98.3** **95.8** **92.7** 92.2
Regression (Re) 13.4 16.7 18.6 8.2 4.3 4.8 9.9 11.5 16.4 9.1 6.4 13.0 11.0
VpKp (Re) [@POSE_Tulsiani15] 12.2 16.0 15.4 12.7 6.8 8.9 11.6 11.1 16.8 12.3 8.0 14.0 12.2
Render4CNN (Sh) [@SYN_Su15] 14.9 18.6 35.5 11.4 8.2 7.5 9.5 17.4 20.1 12.9 13.0 14.6 15.3
Class-15 (Re) 13.0 17.0 15.8 10.0 5.9 8.1 10.3 9.3 18.1 11.7 8.1 15.0 11.9
Class-15-30-60 (Re) 11.7 15.2 15.2 9.3 5.8 8.0 9.7 9.5 17.3 11.3 8.0 14.1 11.3
Class-15-30-60 upsamp. (Re) 9.8 13.8 13.6 8.6 4.5 5.5 7.6 7.3 15.6 9.4 6.9 13.2 9.7
Class (Re-Sy) 9.0 12.5 12.5 8.0 4.2 5.1 7.2 6.8 13.0 8.6 6.1 11.4 8.7
Class (Re-Sh) **8.0** 11.5 11.2 8.4 4.2 4.9 6.9 6.7 13.0 8.3 6.0 10.5 8.3
Class (Re-Sy-Sh) 8.3 10.9 **10.8** **7.4** 4.2 4.4 6.9 6.5 12.3 7.9 6.0 10.2 8.0
JVK (Re) 8.5 11.2 12.3 7.5 4.1 **3.7** 7.3 6.1 12.4 8.1 **5.5** 9.7 8.0
JVK (Re-Sy) 8.3 **10.0** 12.0 **7.4** **3.6** **3.7** **6.5** 6.0 **11.5** 7.7 5.6 **8.9** **7.6**
JVK (Re-Sh) 8.4 10.4 11.2 **7.4** 4.0 3.9 **6.5** **5.6** 12.1 **7.5** 5.7 9.6 7.7
JVK (Re-Sy-Sh) 8.1 10.7 11.4 7.6 4.0 3.8 7.2 6.0 11.7 7.7 5.9 9.5 7.8
We evaluate our viewpoint estimation using two widely used metrics. The first metric [@POSE_Tulsiani15] is the geodesic distance between the ground truth and predicted rotation matrices from $\phi$, $\psi$ and $\theta$, which is given by $$\Delta(R_{gt},R_{pred}) = \frac{||log(R_{gt}^TR_{pred})||_F}{\sqrt{2}}.$$ The viewpoint is considered to be correct if the distance is below $\frac{\pi}{6}$ rad ($Acc{\frac{\pi}{6}}$). The second measure is the median error (MedError).
For this evaluation against other CNN-based approaches, we take as baseline a standard regression approach by [@POSE_Massa16], where continous angles are seen as a circular array and represented in $\mathbb{R}^2$. VpKp [@POSE_Tulsiani15] proposes a classification-based viewpoint with also several discretisation levels. Then, Render4CNN [@SYN_Su15] presents a very fine discretisation with Gaussian filters to leverage the neighbouring bins by using millions of synthetic images. Finally, we re-train a VGG-16 [@CNN_Simonyan14] model for testing different classification-based configurations (Class): with only one level of discretisation (15$^{\circ}$), our proposed approach with 3 quantisations with $15^{\circ}$, $30^{\circ}$ and $60^{\circ}$, and including the upsampling with cubic filtering (upsamp.).
The evaluation results for all the presented baselines and our configurations are shown in Table \[table:o3d\_vp\]. Generally, we observe that the regression technique obtains similar results compared to other classification-based techniques. However, the cubic interpolation provides a significant reduction in median error and accuracy that favors classification approaches. Compared to the same configuration without upsampling, the error is reduced by $-1.6^{\circ}$ and the accuracy increases by $+0.8\%$. The fine discretisaton of Render4CNN fails to compute robust viewpoints and ends up being the worst performing method by a large margin. Using real images from ObjectNet3D would not solve the problem, since the amount of training samples is too scarce for the large number of bins per angle. Class-15-30-60 outperforms Class-15, showing that learning several angle quantisations at the same time provides better results. When we compare JVK with Class, we observe that including a specific network for keypoint estimation allows for better viewpoint accuracies and reduced angle errors. JVK (Re) demonstrates to be superior compared Class upsampling (Re) by $+2\%$ in accuracy and $-1.7^{\circ}$ in the median error. Although the gap is significantly smaller when training the networks with synthetic data, JVK trained with additional synthetic data achieves the best overall results. Specifically, the results of JVK trained on our new synthetic data are comparable to the ones using ShapeNet, but employing 10 times less samples. The better quality and additional labelled data of our dataset play an important role in improving the overall results.
Qualitative results
-------------------
For completeness, we also show some qualitative results in Figure \[fig:quali\]. For each class, we show the results for the first three test images of ObjectNet3D [@DATA_Yu16]. We observe that the predicted 2D keypoints and 3D viewpoints are in alignment. The majority of the few wrongly estimated keypoints and viewpoints are due to lateral symmetries of objects.
![ Qualitative results for the proposed approach JVK (Re-Sy-Sh). The directional arrow represents the projected 3D viewpoint. Blue (dots) and red (crosses) denote correct and wrong estimations based on the PCK\[$\alpha=0.1$\] or Acc($\pi/6$) measure, respectively. []{data-label="fig:quali"}](quali.png){width="\textwidth"}
Conclusion
==========
In this paper we have presented an approach for joint viewpoint and keypoint estimation for multiple rigid object classes. The approach includes a simple yet effective branch for viewpoint estimation with different discretisation levels and cubic upsampling that produce more accurate results. In contrast to previous methods that train a separate approach for each task, we have shown that viewpoint and keypoint estimation benefit from each other. Our approach also handles different kinds of training datasets containing real or synthesized images, as well as datasets where only one of the tasks is annotated. We evaluated our approach on ObjectNet3D where it outperforms previous approaches.
Acknowledgement {#acknowledgement .unnumbered}
===============
The work has been supported by the ERC Starting Grant ARCA (677650).
| {
"pile_set_name": "ArXiv"
} |
****
Sensitivity of the EDELWEISS WIMP search to spin-dependent interactions
[The EDELWEISS Collaboration:]{}\
A. Benoit$^{1}$, L. Bergé$^{2}$, J. Blümer$^{3,4}$, A. Broniatowski$^{2}$, B. Censier$^{2}$, L. Chabert$^{5}$, M. Chapellier$^{6}$, G. Chardin$^{7}$, S. Collin$^{2}$, M. De Jésus$^{5}$, H. Deschamps$^{7}$, P. Di Stefano$^{5}$, Y. Dolgorouky$^{2}$, D. Drain$^{5}$, L. Dumoulin$^{2}$, K. Eitel$^{4}$, M. Fesquet$^{7}$, S. Fiorucci$^{7}$, J. Gascon$^{5}$, G. Gerbier$^{7}$, C. Goldbach$^{8}$, M. Gros$^{7}$, R. Gumbsheimer$^{4}$, M. Horn$^{4}$, A. Juillard$^{2}$, A. de Lesquen$^{7}$, M. Luca$^{5}$, J. Mallet$^{7}$, S. Marnieros$^{2}$, L. Mosca$^{7}$, X.-F. Navick$^{7}$, G. Nollez$^{8}$, P. Pari$^{6}$, M. Razeti$^{5}$, V. Sanglard$^{5}$, L. Schoeffel$^{7}$, M. Stern$^{5}$, V. Villar$^{7}$
[$^{1}$Centre de Recherche sur les Très Basses Températures, SPM-CNRS, BP 166, 38042 Grenoble, France\
$^{2}$Centre de Spectroscopie Nucléaire et de Spectroscopie de Masse, IN2P3-CNRS, Université Paris XI, bât 108, 91405 Orsay, France\
$^{3}$Institut für Experimentelle Kernphysik, Universität Karlsruhe (TH), Gaedestr. 1,\
76128 Karlsruhe, Germany\
$^{4}$Forschungszentrum Karlsruhe, Institut für Kernphysik, Postfach 3640, 76021 Karlsruhe, Germany\
$^{5}$Institut de Physique Nucléaire de Lyon-UCBL, IN2P3-CNRS, 4 rue Enrico Fermi,\
69622 Villeurbanne Cedex, France\
$^{6}$CEA, Centre d’Études Nucléaires de Saclay, DSM/DRECAM, 91191 Gif-sur-Yvette Cedex, France\
$^{7}$CEA, Centre d’Études Nucléaires de Saclay, DSM/DAPNIA, 91191 Gif-sur-Yvette Cedex, France\
$^{8}$Institut d’Astrophysique de Paris, INSU-CNRS, 98 bis Bd Arago, 75014 Paris, France ]{}
[**Abstract**]{}
The EDELWEISS collaboration is searching for WIMP dark matter using natural Ge cryogenic detectors. The whole data set of the first phase of the experiment contains a fiducial exposure of 4.8 kg.day on $^{73}$Ge, the naturally present (7.8%), high-spin Ge isotope. The sensitivity of the experiment to the spin-dependent WIMP-nucleon interactions is evaluated using the model-independent framework proposed by Tovey [*et al.*]{} [@tovey]. It is shown that the EDELWEISS sensitivity for the WIMP-neutron coupling is competitive when compared with results of other spin-sensitive WIMP dark matter experiments. The current experimental limits lie however two orders of magnitude higher than the most optimistic SUSY models.
[**PACS classification codes:**]{} 95.35.+d, 14.80.Ly, 98.80.Es, 29.40.Wk.
[**Introduction**]{}\
In direct searches for the Galactic cold dark matter in the form of Weakly Interacting Massive Particles (WIMPs), the experimental signature is the observation of nuclear recoils induced by WIMP scattering off nuclei in a terrestrial detector. If the WIMP is assumed to be the neutralino, the lightest supersymmetric particle, two types of couplings with matter have to be considered: scalar (spin-independent) and axial-vector (spin-dependent) [@jung].
For sufficiently heavy nuclei, the spin-independent interaction is expected to give the dominant contribution [@jung], due to the coherent enhancement approximately proportional to $A^2$, the square of the mass number. Experimental results are thus most often given as exclusion curves for the spin-independent WIMP-nucleon diffusion cross-section (see for instance and references therein).
For spin-dependent couplings, this enhancement is not present due to cancellation effects between paired nucleons within the nucleus. For example, in the single-particle shell model [@good] the nuclear spin comes from the unpaired proton or neutron. Even with high-spin nuclei the sensitivity of direct detection experiments on the WIMP-nucleon cross-section is orders of magnitude lower than for the spin-independent case (see Refs. [@giuliani] and [@savage] for recent reviews). For axial-vector coupling, indirect detection experiments searching for high-energy neutrinos produced by WIMP annihilations in the Sun provide much higher sensitivity than direct detection ones [@savage]: the capture rate of WIMPs in the Sun core is governed by the scattering rate on protons (low mass, high specific spin).
For spin-dependent interactions, comparisons between WIMP-nucleon cross-section limits given by various experiments is problematic: the comparison of results obtained using targets with either an unpaired neutron or an unpaired proton relies on the assumptions concerning the ratio of the WIMP-neutron to the WIMP-proton cross-sections. This is particularly obvious in the framework of the simple single-particle model. Although it is expected that this ratio should be ${\cal O}(1)$ in minimal supersymmetric models [@bednyakov2], Tovey [*et al.*]{} [@tovey] have proposed a method to extract limits on the spin-dependent interaction in the more general case where there is no constraint on this ratio. For a given WIMP mass, exclusion curves are obtained in the plane of the effective WIMP-neutron and WIMP-proton couplings. The same type of exclusion curves has been computed very recently [@savage] using a more rigorous formalism. Evaluations of current results of spin-dependent WIMP searches have been given using both techniques .
The EDELWEISS collaboration searches for WIMP dark matter using natural Ge cryogenic detectors . The whole data set of the first phase of the experiment contains a fiducial exposure of 4.8 kg.day on $^{73}$Ge, the naturally present (7.8%), high-spin Ge isotope. The aim of this paper is to compare, for the spin-dependent interaction and using the model-independent framework of Ref. [@tovey], EDELWEISS limits with those of the most representative high spin target experiments. It will be shown that, despite the very low $^{73}$Ge content of natural Ge, the EDELWEISS sensitivity for the WIMP-neutron coupling is competitive.
\
[**Experimental data**]{}\
The EDELWEISS experiment is set in the Laboratoire Souterrain de Modane (LSM) in the Fréjus tunnel connecting France and Italy. Detectors used by the experiment are 320 g Ge phonon and ionization cryogenic detectors placed in a dilution cryostat at a regulated temperature of 17 mK. The experimental setup is described elsewhere .
The main characteristic of the EDELWEISS detectors is the simultaneous measurement of the phonon and the ionization signals. The ionization signal is measured by Al electrodes sputtered on each side of the crystal and the phonon signal by a NTD heat sensor glued onto one electrode. The measurement of both signals provides a very good event-by-event discrimination between nuclear and electronic recoils, typically more than 99.9$\%$ above 15 keV. More details on the detectors and their performances can be found in Ref. [@edw2]. Following the first published data , a new set of three detectors was operated in the cryostat, all of them having either a Ge or Si amorphous layer for better charge collection.
Between 2000 and 2003, four physics runs have been performed with different detectors, trigger conditions and efficiencies. All these runs have been re-analysed with an uniform analysis threshold of 15 keV and using the efficiency versus recoil energy function of each run. A detailed presentation of the experimental data and a thorough discussion of their analysis is given in Ref. [@edw4]. Only the most relevant features are summarized hereafter.
In 2000 and 2002, 13.6 kg.day (after the fiducial volume cut) were accumulated with two detectors . Two events compatible with nuclear recoils have been recorded in 2000 above the new analysis threshold of 15 keV, and two in 2002. In 2003 two other runs have been performed with new detectors but with two different triggers. In the first run, the trigger was the fast ionization signal (run 2003i, fiducial exposure 25.7 kg.day). In the second run, the trigger was the phonon signal (run 2003p, fiducial exposure 22.7 kg.day). This latter trigger condition improves the efficiency at low energy: the baseline resolution is better on the phonon channel and the full recoil energy deposition $E_R$ is recorded, instead of the quenched fraction $E_I = Q.E_R$ in the ionization channel. In the runs 2003i and 2003p, the numbers of observed events compatible with nuclear recoils above the analysis threshold are 17 and 19, respectively. This apparent increase of the raw nuclear recoil rates in the new 2003 runs is explained by the significant increase in efficiency at low energy in the new data set (see Ref. [@edw4] for details). Only three events lie in the energy interval from 30 to 100 keV, a result consistent with the previous data sets . The entire set of data for the EDELWEISS-I experiment consists therefore of a fiducial exposure of 62.0 kg.day with a total of 40 events compatible with nuclear recoils above 15 keV. Conservatively considering all these events as WIMP interactions, a 90 $\%$ C.L. upper limit on the WIMP-nucleon cross-section as a function of the WIMP mass is derived with the Optimum Interval Method [@yellin].
For the spin-dependent coupling, one must consider the exposure on $^{73}$Ge (4.8 kg.day) and a specific form factor. Here, the form factor of Dimitrov [*et al.*]{} [@dimitrov] is used, with the usual approximation that the isoscalar, isovector and interference form factors are identical in order to make the form factor independent of the WIMP-nucleon couplings [@tovey]. The resulting uncertainty on the cross-section is within $\pm$15% up to a WIMP mass of 1 TeV/c$^2$. Other calculations of spin-dependent scattering on $^{73}$Ge are briefly presented hereafter. The resulting 90 $\%$ C.L. limits on the WIMP-nucleus cross-section $\sigma_A$ for WIMP masses between 20 GeV/c$^2$ and 1 TeV/c$^2$ are listed in table \[resultat\]. These limits are conservative as they neglect any contribution from the spin-independent coupling.
\
[**Model-independent Exclusion Limits**]{}\
For spin-dependent interactions the WIMP-nucleus cross-section $\sigma_A$ at zero momentum transfer can be approximated by the expression (see for instance [@jung]): $$\sigma_A={32\over \pi}G_F^2\mu_A^2(a_p\left<S_p\right>+a_n\left<S_n\right>)^2{J+1\over J},
\eqno (1)$$ where $G_F$ is the Fermi coupling constant, $\mu_A$ the WIMP-nucleus reduced mass, $a_{p,n}$ the effective WIMP-proton(neutron) couplings, $\left<S_{p,n}\right>$ the expectation values of the proton(neutron) spins within the nucleus and $J$ the total nuclear spin. WIMP-nucleon cross-sections $\sigma^{lim(A)}_{p,n}$ in the limit $a_{n,p}=0$ respectively, are defined as [@tovey]: $$\sigma^{lim(A)}_{p,n}={3\over 4}{J\over J+1}{\mu_{p,n}^2\over \mu_A^2}
{\sigma_A\over \left<S_{p,n}\right>^2},\eqno (2)$$ where $\mu_{p,n}$ is the WIMP-proton(neutron) reduced mass and $\sigma_A$ the WIMP-nucleus cross-section limit (at 90% CL) deduced from the experiment.
It is shown in [@tovey] that the [*allowed*]{} values of $a_p$ and $a_n$ for a particular WIMP mass obey the inequality: $$\left({a_p\over \sqrt{\sigma^{lim(A)}_p}}\pm{a_n\over \sqrt{\sigma^{lim(A)}_n}}\right)^2
\leq {\pi\over 24G_F^2\mu_p^2}.\eqno (3)$$
The sign between parentheses is that of ${\left<S_p\right>\over \left<S_n\right>}$. Equation (3) defines two parallel straight lines in the $(a_n,a_p)$ plane, the slope of which is ${-\left<S_n\right>\over \left<S_p\right>}$. The allowed values of $a_p$ and $a_n$ are within the band defined by these two lines (while in [@savage] an extremely elongated ellipse is found). For experiments with two active nuclei with different ${\left<S_p\right>\over \left<S_n\right>}$ ratios the combination of the two bands gives rise to a closed elliptical contour.
Nuclear spin structure calculations have been recently reviewed by Bednyakov and Šimkovic [@bednyakov]: for $^{73}$Ge the two most comprehensive spin structure analyses are from Ressel et al. [@ressel] and from Dimitrov et al. [@dimitrov] (see table 2). The corresponding form factors are very similar. In these calculations the neutron spin ${\left<S_n\right>}$ varies by $\sim$20% depending on the nuclear stucture model. The value of the ratio ${\left<S_n\right>\over \left<S_p\right>}$ is model dependent, but is always much greater than unity. Thus the odd-N, even-Z nucleus $^{73}$Ge nucleus is mainly sensitive to $a_n$ only, and in that sense is complementary to other detectors made out of odd-Z material such as $^{23}$Na, $^{127}$I, $^7$Li, $^{19}$F or $^{27}$Al.
Table 1 gives the values of the spin-dependent WIMP-nucleon cross-sections deduced from our experimental $\sigma_A$ values using eq. (2) and the $\left<S_{p,n}\right>$ taken from [@dimitrov].
Plots of the WIMP-nucleon cross-sections $\sigma^{lim(A)}_{p,n}$ versus the WIMP mass are shown in Fig. 1 and 2. Figures 3 and 4 show the regions in the $(a_n,a_p)$ plane allowed by EDELWEISS, using eq. (3) and the experimental values of table 1, for two illustrative WIMP masses $M_{\chi}=$ 50 and 500 GeV/c$^2$. A comparison is made with other experiments representative of various techniques: NaI scintillators (NAIAD experiment [@ahmed]), fluored bolometers (Kamioka LiF [@miuchi]), sapphire bolometers (CRESST-I [@anglo]), freon droplets (SIMPLE [@girard], PICASSO [@picasso]) and the other natural Ge cryogenic experiment (CDMS-II [@CDMS]). A more exhaustive comparison can be found in the review of Giuliani [@giuliani]. All experiments use the same dark matter halo model to extract $\sigma_A$. Form factors, values of $\left<S_{p,n}\right>$ specific to each nucleus and WIMP-nucleon cross-sections $\sigma^{lim(A)}_{p,n}$ are given in [@ahmed], [@miuchi], [@girard] and [@picasso] .
For the CRESST-I (sapphire) experiment, relevant cross-sections can be deduced from the published data using $\left<S_{p,n}\right>$ values from [@engel] for $^{27}$Al (neglecting the $^{16}$O spin sensitivity). For the CDMS experiment limits on spin-dependent nucleon cross-sections are derived in Ref. [@savage], including the coupling dependency in the form factor rather than taking the isoscalar, isovector and interference terms to be identical. For consistency, $\sigma^{lim(A)}_{p,n}$ are recomputed from the CDMS data (no event in 52.6 kg.day raw exposure in Ge detectors over 10 to 100 keV recoil energy) using the experimental efficiency quoted in Ref. [@savage], form factor and $\left<S_{p,n}\right>$ values from Ref. [@dimitrov]. The differences in the $a_n$ allowed regions defined by the two calculations are insignificant in the limit $a_p=0$.
The DAMA/Xe experiment [@bernabei] makes use of a $^{129}$Xe target, an other example of odd-N, even-Z nucleus, as $^{73}$Ge. The $\sigma^{lim(A)}_p$ can be calculated from the reported $\sigma^{lim(A)}_n$ cross-section. However this result deserves a special comment. The value of the quenching factor for liquid Xenon is taken in [@bernabei] as $Q=$ 0.44, i.e. more than a factor of 2 greater than other measured values . Adopting these more recent and precise values shifts the nuclear recoil energy threshold of the DAMA/Xe experiment from 30 keV to at least 60 keV. For $M_{\chi}=$ 50 GeV/c$^2$ the WIMP event rate at the threshold is then divided by a factor of 15 and even more if the fast decrease of the form factor with recoil energy is taken into account. The cross-section limit is underestimated by the same factor. This considerable uncertainty on the actual recoil energy scale does not allow any reliable comparison and the published results of DAMA/Xe are not shown on the figure.
Fig. 3 and 4 show that cryogenic natural Ge experiments such as EDELWEISS and CDMS give the most stringent limits on $a_n$, at roughly the same level as the $a_p$ limits given by odd-Z detectors. The low $^{73}$Ge content of natural Ge is balanced by the very low level of nuclear recoil backgrounds achieved in the cryogenic detectors and by the high neutron nuclear spin value of $^{73}$Ge. The allowed region of minimal extension in the $(a_n,a_p)$ plane is given by the combination of experiments with respectively high $\left<S_n\right>$ and high $\left<S_p\right>$ nuclei. The best limit on $a_n$ is presently set by the CDMS experiment [@CDMS]. As shown in Ref. [@savage] the limit on $a_p$ set by the neutrino observatories Super-K [@superK] and Baksan [@baksan] is more than a factor of 10 better than current direct detection results.
\
[**Conclusions**]{}\
For a WIMP mass between 50 and 500 Gev/c$^2$, current limits given by direct detection experiments are $|a_{n,p}|<$ few units, or equivalently $\sigma_{p,n}^{SD}<$ few pb. The maximal values given by SUSY models lie two orders of magnitude lower (see for instance ). Only a new generation of experiments designed to gain two orders of magnitude in sensitivity will be able to reach the SUSY predictions for $\sigma_{p,n}^{SD}$. EDELWEISS-II is one of these forthcoming experiments [@edw5].
In the experimentally constrained versions of Supersymmetry, the neutralino is not a pure gaugino and one expects $|a_p|/|a_n|\sim{\cal O}(1)$ (for example see Ref. [@bednyakov2]). To go beyond these constraints and fix limits on $a_p$ by the direct detection technique, high sensitivity experiments, using efficient background rejection techniques and high $\left<S_p\right>$ nuclear targets are needed. A multi nuclear target experiment, with powerful background discrimination capability and complementary proton vs neutron spin values would constitute an interesting new approach. Setting decisive constraints on the SUSY parameters for spin-dependent WIMP-nucleon interactions remains a considerable experimental challenge for direct detection experiments.
\
[**Acknowledgments** ]{}\
It is a pleasure to thank Franco Giuliani and Chris Savage for stimulating discussions. The help of the technical staff of the Laboratoire Souterrain de Modane and of the participating laboratories is gratefully acknowledged. This work has been partially funded by the EEC Network program under contract HPRN-CT-2002-00322.
[99]{} D.R. Tovey [*et al.*]{}, Phys. Lett. B [**488**]{} (2000) 17. G. Jungman, M. Kamionkowski and K. Griest, Physics Reports [**267**]{} (1996) 195. A. Morales [*et al.*]{}, Phys. Lett. B [**532**]{} (2002) 8. A. Benoit [*et al.*]{}, Phys. Lett. B [**545**]{} (2002) 43. D.S. Akerib [*et al.*]{}, Phys. Rev. Lett. [**93**]{} (2004) 211301 G. Angloher [*et al.*]{}, arXiv:astro-ph/0408006, accepted for publication in Astropart. Phys. M.W. Goodman and E. Witten, Phys. Rev. D [**31**]{} (1986) 3059. F. Giuliani, Phys. Rev. Lett. [**93**]{} (2004) 161301 C. Savage, P. Gondolo and K. Freese, Phys. Rev. D [**70**]{} (2004) 123513. V.A. Bednyakov and H.V. Klapdor-Kleingrothaus, Phys. Rev. D [**70**]{} (2004) 096006. A. Benoit [*et al.*]{}, Phys. Lett. B [**513**]{} (2001) 15. O. Martineau [*et al.*]{}, Nucl. Instr. Meth. A [**530**]{} (2004) 426 V. Sanglard [*et al.*]{}, arXiv:astro-ph/0503265, submitted to Phys. Rev. D S. Yellin, Phys. Rev. D [**66**]{} (2002) 032005 V.I. Dimitrov, J. Engel and S. Pittel, Phys. Rev. D [**51**]{} (1995) R291. V.A. Bednyakov and F. Šimkovic, arXiv:hep-ph/0406218. T. Ressel [*et al.*]{}, Phys. Rev. D [**48**]{} (1993) 5519. B. Ahmed [*et al.*]{}, Astropart. Phys. [**19**]{} (2003) 691. K. Miuchi [*et al.*]{}, Astropart. Phys. [**19**]{} (2003) 135. G. Angloher [*et al.*]{}, Astropart. Phys. [**18**]{} (2002) 43. F. Giuliani and T.A. Girard, Phys. Lett. B [**588**]{} (2004) 151. M. Barnabé-Heider [*et al.*]{}, arXiv:hep-ex/0502028, submitted to Phys. Lett. B J. Engel [*et al.*]{}, Phys. Rev. C [**52**]{} (1995) 2216. R. Bernabei [*et al.*]{}, Phys. Lett. B [**436**]{} (1998) 379. F. Arneodo [*et al.*]{}, Nucl. Instr. Meth. A [**449**]{} (2000) 147. D. Akimov [*et al.*]{}, Phys. Lett. B [**524**]{} (2002) 245. S. Desai [*et al.*]{}, Phys. Rev. D [**70**]{} (2004) 083523. O.V. Suvorova, arXiv:hep-ph/9911415. J. Ellis [*et al.*]{}, Eur. Phys. J. C [**24**]{} (2002) 311. S. Profumo and C.E. Yaguna, Phys. Rev. D [**70**]{} (2004) 095004. P.C.F. Di Stefano [*et al.*]{}, in: Proc. 6th UCLA Symposium on Sources and Detection of Dark Matter and Dark Energy in the Universe, February 18-20, 2004, Marina del Rey, California, to be published in New Astronomy Reviews
$M_{\chi}$(GeV/c$^2$) $\sigma_A$(pb) $\sigma^{lim(A)}_p$(pb) $\sigma^{lim(A)}_n$(pb)
----------------------- ---------------- ------------------------- -------------------------
20 769. 1759. 11.1
30 304. 395. 2.50
40 201. 181. 1.14
50 192. 133. 0.845
60 199. 114. 0.721
80 238. 103. 0.651
100 291. 104. 0.661
200 648. 149. 0.945
400 1482. 262. 1.66
500 1917. 319. 2.02
600 2358. 377. 2.39
800 3247. 494. 3.13
1000 4140. 611. 3.87
: \[resultat\] *Values of the WIMP-nucleus cross-section limit $\sigma_A$ (at 90% CL) deduced from the experiment. The corresponding $\sigma^{lim(A)}_{p,n}$ cross-sections (eq. (2)) are calculated using $\left<S_{p,n}\right>$ values from [@dimitrov].*
$\left<S_p\right>$ $\left<S_n\right>$
----------------------------------- -------------------- --------------------
Ressel [*et al.*]{} [@ressel] 0.011 0.468
unquenched values
Ressel [*et al.*]{} [@ressel] 0.009 0.372
quenched values
Dimitrov [*et al.*]{} [@dimitrov] 0.030 0.378
: \[vfidt\] *Spin values for $^{73}$Ge ($J=9/2$). See [@bednyakov] for a critical evaluation of the various calculations.*
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The effect of retardation in the atom-field interaction on the dynamics of entanglement in a double Jaynes-Cummings system is investigated. We consider large cavities in which a finite time necessary for light to travel between the atoms and the cavity mirrors may result in retardation effects. Our results demonstrate the qualitatively new behaviour observable in the time evolution of entanglement when the retardation effects are included. Solutions for single and double excitation in the system are presented. We follow the temporal evolution of an initial entanglement and find that the evolution is affected drastically by the retardation effects. In particular, the harmonic oscillations of the atomic populations and the concurrence, characteristic of single-mode Jaynes-Cummings systems, are suppressed when the retardation effects are included. The process of revival of the entanglement degrades with an increasing number of the cavity modes to which the atoms are coupled. It is also found that the effect of the retardation on the doubly excited states is more drastic than on the single excitation states that at relatively short times, the retardation leads to a complete distortion of entanglement carried by a doubly excited state.'
address: |
$^{1}$Department of Physics, Faculty of Science, Taibah University, P.O. Box 30002, Madinah, Saudi Arabia\
$^{2}$The National Center for Mathematics and Physics, KACST, P.O. Box 6086, Riyadh 11442, Saudi Arabia
author:
- 'Smail Bougouffa$^{1}$ and Zbigniew Ficek$^{2}$'
date:
-
-
title: 'Effect of retardation in the atom-field interaction on entanglement in a double Jaynes-Cummings system'
---
Introduction
============
Partly because of the successful experimental realisation of quantum gates, and the opportunity it provides for applications in quantum information and quantum computation, the problem of the entanglement evolution and transfer in a network of quantum systems has recently received a great deal of attention [@10; @11]. The theoretical description of such systems is generally complicated, in particular when one includes decoherence processes resulting from the coupling of a system to external environment. The complete treatments of these situations are available. However, the physics of the coherent processes involved in the transfer of entanglement are often lost in details or masked by the incoherent processes [@12; @14; @15; @16; @18; @19; @20].
Although less important from the practical situation, studying the evolution and transfer of entanglement in simpler systems such as Jaynes-Cummings models [@JC63; @pa63] involving single atoms interacting with a single-mode cavity field is significant in terms of giving greater insight into the fundamentals of the transfer process. In particular, the relation between the form of entangled states and the efficiency of the transfer can be studied. A large number of studies of such simple systems have been carried out for independent [@Eberly1; @Eberly2; @Eberly3; @yelattice; @sainz; @Ficek1; @S10; @TMSZ10; @SA11] and also coupled Jaynes-Cummings models [@nr07; @oi08; @ew10; @dz12; @gl12; @xf12; @pe97; @sm06; @pl07; @zh09; @zy10; @sl12]. In these models simple analytical formulas can be derived for the state vector or density operator of the system and the relations between the coherence properties and the efficiency of transferring a given state can be studied.
It has been demonstrated that the evolution of singly and doubly excited entangled states in independent or coupled Jaynes-Cummings systems can be significantly different that the singly excited states can be continuously and completely transferred between the atoms and the cavity modes whereas doubly excited states may suffer an abrupt loss during the evolution, the phenomenon known as sudden death of entanglement [@13; @17; @col]. Although the single-mode approximation of the interaction of an atom with a cavity field is often entirely adequate to describe a practical situation, there are circumstances when the interaction of the atom with a large number of modes cannot be neglected, for example when sizes of cavities are much larger than the wavelength of the atomic transition. In this case, a finite number of the cavity modes could fit into the atomic linewidth leading to an effective multi-mode interaction of the atom with the cavity field. The atom can then be re-excited at finite time and at multiple finite times required for a photon emitted by the atom to cycle through the cavity and be reabsorbed. The finite interaction time between the atom and the cavity field, arising from a finite time required for the photon to travel between the atom and the cavity mirrors, could lead to retardation in the excitation of the atom [@42; @43; @44; @Meyster1; @gf12].
In this paper, we show how various results relating to the evolution of entanglement in a system composed of two isolated Jaynes-Cummings models can be affected by the retardation. We calculate concurrence, the measure of entanglement between two qubits, and investigate the effect of retardation on the evolution of single and double excitation entangled states. We find that the evolution differs qualitatively from that predicted before in the absence of the retardation. In particular, the evolution is essentially nonoscillatory. A simple physical interpretation is given based on the energy-time uncertainty relation.
The paper is organised as follows. We begin in section \[sec1\] by introducing the model and explaining in details the meaning of retardation in the atom-field interaction. In section \[sec2\] we derive equations of motion for the probability amplitudes of single and double excitation states. Following the description of single and double excitation systems, in section \[sec3\] we derive expressions for the concurrence between the atoms. We pay particular attention to the role of retardation in modifying the evolution of an initial entanglement. We show that the evolution of entangled states is affected drastically by the retardation effects that the harmonic oscillations of the concurrences characteristic of the single-mode interaction are suppressed in the presence of the retardation effects. At the same time we show that the evolution of the concurrence from an initial state in which atoms are entangled follows the evolution of the atomic population. The numerical results for various special cases of the time evolution of the concurrence are illustrated. Finally, in section \[Conclusion\] we summarize our results.
The system and meaning of retardation {#sec1}
=====================================
We consider a system composed of two independent Jaynes-Cummings (JC) cavities [@JC63; @pa63]. Each JC system is composed of a one-dimensional cavity containing an atom located at a position $\vec{x}$ inside the cavity modes. The atoms are modelled as two-level systems with the excited ${| e_{i} \rangle}$ and ground ${| g_{i} \rangle}$ states separated by the transition frequency $\omega_a$, as illustrated in Fig. \[fig1\]. The cavities are considered as composed of a multi-mode field with frequency difference between adjacent modes (free spectral range) such that multiple modes are supported within the atomic resonance line width. The need to consider a multi-mode rather than single-mode cavity results from the time-energy uncertainty relation that a finite interaction time between the atom and the cavity field leads to unavoidable spread of the achievable energy distribution.
![(Color online) Schematic diagram of the system considered. The total system is composed of two independent Jaynes-Cummings systems, each containing a cavity and an atom located inside a one-dimensional, in general, multi-mode cavity field.[]{data-label="fig1"}](jcfig1){width="0.5\linewidth"}
The interaction Hamiltonian between the atom and the multi-mode field, in the rotating-wave approximation, is of the form $$H_{af} = i\hbar\sum_{n}\sum_{j=1}^{2}\left(g_{jn}(\vec{x}_j)a_{n}S^{+}_{j} - \rm{H.c.}\right) ,\label{eq8}$$ where $g_{jn}(\vec{x}_j)$ is the position-dependent Rabi frequency which determines the strength of the coupling of the $j$th atom to the mode $n$ of the cavity field. The Rabi frequency is proportional to the density of the cavity modes to with the atom is coupled and can be written in a form $$g_{jn}(\vec{x}_j) = g(\omega_{n})\left(\vec{d}_{j}\cdot
\hat{e}_{n}\right) {\rm e}^{i \vec{k}_{n}\cdot\vec{x}_j} ,\label{eq9}$$ where $g(\omega_{n})$ characterise the density of the cavity modes of frequencies $\omega_{n}$ and $\hat{e}_{n}$ is the unit polarization vector of the mode $n$.
The function can be identified as the Airy function of the cavity so that the form of $g(\omega_{n})$ depends on the type of the cavity. For a high-finesse optical cavity the function is of appreciable magnitude only at a single frequency $\omega_{c}$ called the cavity resonance. The function can be approximated by a delta function $\delta(\omega_{c})$ if the cavity is composed of perfectly reflecting mirrors, or by a Lorentzian of bandwidth $\kappa$, if the cavity losses through the mirrors with a rate $\kappa$ are included. For larger cavities (microwave cavities) with plane perfectly reflecting mirrors, the appropriate form of the function $g(\omega_{n})$ is $$g(\omega_{n}) = \sqrt{\frac{\omega_{n}}{2\hbar\epsilon_0 L}} ,\label{eq9a}$$ where $\omega_{n}=2 \pi n c/L$ is in the frequency of the modes set by the periodic boundary conditions of the cavity of length $L$. Similar results can be obtained for a confocal cavity where resonance frequencies of the transverse modes either overlap or fall exactly halfway between the longitudinal mode resonances [@yariv]. If the linewidth $\Gamma$ of an atom located inside the cavity is much larger than the frequency difference between the modes, $\Gamma\gg c/L$, and the cavity bandwidth, $c/L\gg \kappa$, the atom may couple to $n$ discrete modes fitting inside the atomic linewidth and differing in frequency by $c/L$. In this case, the atom can be re-excited at finite time $L/c$ and also at multiples of $L/c$. In other words, multiple finite times are required for a photon emitted by the atom to cycle through the cavity and be reabsorbed by the atom. Thus, the retardation may result in the excitation arising from a finite time required for the photon to travel between the atom and the cavity mirrors.
Evolution of entangled states {#sec2}
=============================
The main motivation of considering the multimode interaction between the atom and the cavity field is to study the drastic departure of the dynamics of entanglement from the usual single-mode behaviour [@Eberly1; @Eberly2; @Eberly3; @yelattice; @sainz; @Ficek1]. We consider situations when only single and double excitations are present in the system. In the case of single excitation, we assume that either the atoms or the cavity modes are initially prepared in a single excitation entangled state and discuss in details the dynamics of entanglement for two initial states $${| \psi(0) \rangle} = \left(\cos\theta\,{| e_{1}g_{2} \rangle} +\sin\theta\,{| g_{1}e_{2} \rangle}\right)\otimes{| \{0\}_\mu\{0\}_\nu \rangle} ,\label{eq24}$$ which is a tensor product of a single excitation entangled state of the atoms and the vacuum states of both cavities, and $${| \psi(0) \rangle} = \left(\cos\theta\,{| \{1\}_{\mu}\{0\}_{\nu} \rangle} + \sin\theta\,{| \{0\}_{\mu}\{1\}_{\nu} \rangle}\right)\otimes{| g_{1}g_{2} \rangle} ,\label{eq25}$$ which is a tensor product of a single excitation entangled state of the cavities and the atomic ground states. Here, $\theta$ determines the degree of entanglement in the initial state, and we use the index $\mu$ to label modes of the cavity $a$, and $\nu$ those of the cavity $b$.
In the case of double excitation, we choose an initial state of the form $$\begin{aligned}
{| \phi(0) \rangle} = \left(\cos\theta\,{| g_1g_2 \rangle}+ \sin\theta\,{| e_{1}e_{2} \rangle}\right)\otimes{| \{0\}_\mu\{0\}_\nu \rangle} ,\label{eq26}\end{aligned}$$ which is a tensor product of a double excitation entangled state of the atoms and the vacuum states of both cavities.
Evolution of single-excitation states {#case-single}
-------------------------------------
First, suppose that only a single excitation is present in the system. In this case, the Schrödinger equation $$i\hbar\frac{\partial {| \psi(t) \rangle}}{\partial t} = H_{af}{| \psi(t) \rangle} ,\label{eq13}$$ for the evolution of the state vector $$\begin{aligned}
{| \psi(t) \rangle} &=& C_{1}(t)\,{| e_{1} g_{2}\{0\}_{\mu}\{0\}_{\nu} \rangle} + C_{2}(t)\,{| g_{1} e_{2}\{0\}_{\mu}\{0\}_{\nu} \rangle}\nonumber\\
&+& \left(\sum_{\mu}C_{a\mu}(t)\,{| \{1\}_{\mu}\{0\}_{\nu} \rangle}
+\sum_{\nu}C_{b\nu}(t)\,{| \{0\}_{\mu}\{1\}_{\nu} \rangle}\right)\otimes{| g_{1}g_{2} \rangle} ,\label{eq27}\end{aligned}$$ leads to a simple set of coupled differential equations for the probability amplitudes $$\begin{aligned}
\dot{C}_{1}(t) &=& \sum_{\mu}g_{1\mu}C_{a\mu}(t) , \label{eq28} \\
\dot{C}_{a\mu}(t) &=& -i\Delta_{\mu} C_{a\mu}(t) - g^{\ast}_{1\mu}C_{1}(t) ,\label{eq29}\end{aligned}$$ and $$\begin{aligned}
\dot{C}_{2}(t) &=& \sum_{\nu}g_{2\nu} C_{b\nu}(t) , \label{eq30} \\
\dot{C}_{b\nu}(t) &=& -i\Delta_{\nu} C_{b\nu}(t) - g^{\ast}_{2\nu}C_{2}(t) ,\label{eq31}\end{aligned}$$ where sums are taken over the discrete modes of the cavities, $\Delta_{\mu}=\omega_{\mu}-\omega_a$ and $\Delta_{\nu}=\omega_{\nu}-\omega_a$ are detunings of the mode frequencies $\omega_{\mu}$ and $\omega_{\nu}$ of the cavity $a$ and $b$ from the atomic transition frequency $\omega_a$, respectively. For simplicity, we assume that $\omega_{a}$ coincides with the central mode frequency of each cavity. Also we note that the state (\[eq27\]) involves superpositions of the cavity modes oscillating with amplitudes $C_{a\mu}(t)$ and $C_{b\nu}(t)$.
The formal integration of (\[eq29\]) and (\[eq31\]) yields $$\begin{aligned}
C_{a\mu}(t) = C_{a\mu}(0){\rm e}^{-i\Delta_{\mu}t} -
\int_{0}^{t}dt^{\prime}g^{\ast}_{1\mu}C_{1}(t^{\prime}){\rm
e}^{-i\Delta_{\mu}(t-t^{\prime})} ,\label{eq32}\\
C_{b\nu}(t) = C_{b\nu}(0){\rm e}^{-i\Delta_{\nu}t} -
\int_{0}^{t}dt^{\prime}g^{\ast}_{2\nu}C_{2}(t^{\prime}){\rm
e}^{-i\Delta_{\nu}(t-t^{\prime})} .\label{eq33}\end{aligned}$$
On substituting for $C_{a\mu}(t)$ from (\[eq32\]) in (\[eq28\]), and for $C_{b\nu}(t)$ from (\[eq33\]) in (\[eq30\]), we arrive at the following two integro-differential equations $$\begin{aligned}
\dot{C}_{1}(t) &=& \sum_{\mu}g_{1\mu} C_{a\mu}(0){\rm e}^{-i\Delta_{\mu}t} - \int_{0}^{t}dt^{\prime}
\sum_{\mu}|g_{1\mu}|^{2}C_{1}(t-t^{\prime})\,{\rm e}^{-i\Delta_{\mu}t^{\prime}} , \label{eq34} \\
\dot{C}_{2}(t) &=& \sum_{\nu}g_{2\nu} C_{b\nu}(0){\rm e}^{-i\Delta_{\nu}t} - \int_{0}^{t}dt^{\prime}
\sum_{\nu}|g_{2\nu}|^{2}C_{2}(t-t^{\prime})\,{\rm e}^{-i\Delta_{\nu}t^{\prime}} .\label{eq35}\end{aligned}$$ The effect of retardation in the atom-field interaction is is clearly visible in terms of the detunings $\Delta_\mu$ and $\Delta_{\nu}$, which in general are all different, because the frequencies $\omega_{\mu}$ and $\omega_{\nu}$ are different. Since a large cavity of the length $L$ contains a finite number of discrete modes whose frequencies differ by $2 \pi c/L$, the detunings also differ by integer multiples of $2\pi c/L$. Thus, the dynamics of the system response will exhibit sharp peaks at discrete times due to the constructive interference of all modes.
Evolution of double-excitation states
-------------------------------------
We now turn to the problem of determining of how the retardation in the atom-field interaction can affect the evolution of double-excitation states. For this purpose we consider the situation when the excitations are evenly redistributed through the system, i.e., each of the Jaynes-Cummings cavities contains a single excitation. In this case, the state vector of the system can be written as $$\begin{aligned}
{| \phi(t) \rangle} &=& D_{11}(t)|e_{1}e_{2}\{0\}_{\mu}\{0\}_{\nu}\rangle\!+\!\sum_{\nu}D_{2\nu}(t)|e_{1}g_{2}\{0\}_{\mu}\{1\}_{\nu}\rangle\!+\!\sum_{\mu}D_{3\mu}(t){| g_{1}e_{2}\{1\}_{\mu}\{0\}_{\nu} \rangle} \nonumber\\
&+& \!\left(D_{00}(t){| \{0\}_{\mu}\{0\}_{\nu} \rangle}
+\sum_{\mu}\sum_{\nu}D_{4\mu\nu}(t){| \{1\}_\mu\{1\}_{\nu} \rangle}\right)\otimes{| g_{1}g_{2} \rangle} ,\label{eq37}\end{aligned}$$ where $\{m\}_{\mu}\{n\}_{\nu}$ denotes the state of the cavity modes with $m$ photons in the mode $\mu$ of the cavity $a$, and $n$ photons in the mode $\nu$ of the cavity $b$. We have also included the auxiliary state ${| g_{1}g_{2}\{0\}_{\mu}\{0\}_{\nu} \rangle}$ with zero excitation that as one can see from (\[eq37\]) creates superpositions of the cavity modes crucial for the presence of an entanglement in the system.
On making use the Schrödinger equation, we arrive at the following set of coupled equations of motion for the probability amplitudes $$\begin{aligned}
\dot{D}_{00}(t) &=& 0 ,\nonumber\\
\dot{D}_{11}(t) &=& -\sum_{\nu} g^{*}_{2\nu} D_{2\nu}(t)
-\sum_{\mu} g^{*}_{1\mu} D_{3\mu}(t) ,\nonumber\\
\dot{D}_{2\nu}(t) &=& -i\Delta_{\nu}D_{2\nu}(t) + g_{2\nu}\tilde{D}_{11}(t)
-\sum_{\mu} g^{*}_{1\mu} D_{4\mu\nu}(t) ,\nonumber\\
\dot{D}_{3\mu}(t) &=& -i\Delta_{\mu}D_{3\mu}(t) + g_{1\nu}D_{11}(t)
-\sum_{\nu} g^{*}_{2\nu}D_{4\mu\nu}(t) ,\nonumber\\
\dot{D}_{4\mu\nu}(t &=& -i(\Delta_{\mu} + \Delta_{\nu})D_{4\mu\nu}(t)
+ g_{1\mu}D_{2\nu}(t)+g_{2\nu}D_{3\mu}(t) .\label{eq38}\end{aligned}$$ The equations (\[eq38\]) are cumbersome because the involvement of many modes of the cavities. Therefore, we will use numerical methods to solve them for a given initial condition.
Evolution of entanglement {#sec3}
=========================
Since we are interested in the dynamics of entanglement between the atoms and also between the cavity modes that behaviour like two-qubit systems, we shall consider concurrence as a measure of entanglement. The concurrence is defined as [@W98] $$\mathcal{C} = \rm{max}\{0,\sqrt{\lambda_1}-\sqrt{\lambda_2}-\sqrt{\lambda_3}-\sqrt{\lambda_4}\} ,\label{eq10}$$ where $\lambda_i$ are the eigenvalues (in descending order) of the Hermitian matrix $R=\rho\tilde{\rho}$, in which $\rho$ is the density matrix of the system, $\tilde{\rho} = \sigma_y\otimes\sigma_{y}\,\rho^{\ast}\sigma_y\otimes\sigma_y$ and $\sigma_y$ is a Pauli matrix. The concurrence varies between $0$ when qubits are separable and $1$ when they are maximally entangled.
In the case of a single excitation present in the system, the reduced density operator of the atoms $\rho_{AB}={\rm Tr}_{ab}(\rho)$ can be written in the basis of the product states, $|1\rangle=|e_{1}e_{2}\rangle,|2\rangle=|e_{1}g_{2}\rangle,|3\rangle=|g_{1}e_{2}\rangle,|4\rangle=|g_{1}g_{2}\rangle$, in which it has a form $$\begin{aligned}
\rho_{AB} &=& \left(
\begin{array}{cccc}
0 & 0 & 0 & 0\\
0 &|C_{1}(t)|^{2} & C^{\ast}_{1}(t)C_{2}(t) & 0\\
0 & C_{1}(t)C^{\ast}_{2}(t) & |C_{2}(t)|^{2} & 0\\
0 & 0 & 0 & \sum_{\mu}|C_{a\mu}(t)|^{2}+\sum_{\nu}|C_{b\nu}(t)|^{2}
\end{array}\right) ,\label{eq11}\end{aligned}$$ from which we can easy find the concurrence $$\mathcal{C}_{AB}(t) = 2\left|C_{1}(t)C_{2}^{\ast}(t)\right| = 2|\rho_{23}(t)| .\label{eq12}$$ Thus, the necessary and sufficient condition for entanglement between the atoms is the present of the one-photon coherence $|\rho_{23}(t)|$.
In the case of a double excitation, again by tracing over the cavity modes, the reduced density operator of the atoms written in the basis of the product states states takes a matrix form $$\begin{aligned}
\rho_{AB} &=& \left(\!
\begin{array}{cccc}
|D_{11}(t)|^{2} & 0 & 0 & D_{11}(t)D_{00}^{*}(t)\\
0 &\sum_{\nu}\!|D_{2\nu}(t)|^{2} & 0 & 0\\
0 & 0 & \sum_{\mu}\!|D_{3\mu}(t)|^{2} & 0\\
D_{11}^{*}(t)D_{00}(t) & 0 & 0 &
|D_{00}(t)|^{2}\!+\!\sum_{\nu\mu}\!|D_{4\mu\nu}(t)|^{2}
\end{array}\!\right) ,\label{eq42}\end{aligned}$$ from which it is straightforward to evaluate the concurrence $$\begin{aligned}
\mathcal{C}_{AB}(t) = 2\, {\rm max}\left\{0,\tilde{\mathcal{C}}_{AB}(t)\right\} ,\label{eq43}\end{aligned}$$ where $$\begin{aligned}
\tilde{\mathcal{C}}_{AB}(t) = |D_{11}(t)||D^*_{00}(t)| -\sqrt{\left(\sum_{\nu}|D_{2\nu}|^2\right)\left(\sum_{\mu}|D_{3\mu}|^2\right)} .\label{eq44}\end{aligned}$$ The entangled properties of the system are clearly exhibited by the off-diagonal elements of the density matrix (\[eq42\]). It is apparent from (\[eq44\]) that the presence of the auxiliary state with zero excitation, $D_{00}(t)\neq 0$, is necessary to produce entanglement between the atoms.
The case of single-mode cavities
--------------------------------
Let us first present the results for the evolution of entanglement in the simplified double Jaynes-Cummings system composed of two single-mode $(\mu =\nu =1)$ cavities [@Eberly1; @Eberly2; @Eberly3; @yelattice; @sainz; @Ficek1]. Figure \[fig2\] shows the time evolution of the concurrences for several different values of the parameter $\theta$. It is seen that an initial entanglement periodically oscillates in time and vanishes at certain discrete times $t_{n} = n\pi/2,\, (n=1,3,5,\ldots)$ when it is completely transferred to the cavity modes. This behaviour is seen for both maximally $(\theta =\pi/4)$ and non-maximally entangled states $(\theta\neq \pi/4)$, but the effect of going off maximally entangled initial states is clearly to decrease the amount of entanglement [@Meyster1; @Eberly1; @Eberly2; @Eberly3; @Ficek1].
Similar behaviour of the periodic exchange of entanglement is observed for double excitation states with a difference that in this case the phenomenon of the sudden death of entanglement is observed. Therefore, we may conclude that in the case of single-mode cavities the periodic oscillatory behaviour of the concurrence can be observed as resulting from an instantaneous exchange of an excitation between the atoms and the cavity modes. This is true, however, only for sufficiently small cavities in which a light signal can pass between the atom and the cavity boundaries in a time short compared with the time $\Delta t$ required for appreciate (secular) changes in the atomic states. In other words, the length $L$ of the cavity must be much smaller than $c\Delta t$.
The case of multi-mode cavities
-------------------------------
We now examine the effect of retardation on the dynamics of entanglement for different numbers of modes and for two different initial states. Figure \[fig4\] shows the time evolution of the concurrence for the initial state (\[eq24\]) in which the atoms are initially entangled, whereas in Fig. \[fig5\] the initial state is (\[eq25\]) in which the cavity modes are initially entangled. The concurrence is determined from (\[eq12\]), where the time evolution of the probability amplitudes is found by solving numerically the sets of coupled differential equations (\[eq34\]) and (\[eq35\]).
The evolution of the concurrence in the presence of the retardation is seen to be qualitatively different from the previous case of the instantaneous exchange of an excitation between the atoms and the cavity modes, shown in Fig. \[fig2\]. Firstly, the evolution is not oscillatory with time, and secondly the initial entanglement is not preserved in time, it decreases as time progresses. At the initial time, the concurrence decays almost exponentially to zero and remains zero until a finite time at which a nonzero entanglement suddenly revivals. The process repeats with increasing distortions of the concurrence. The effect of increasing number of modes to which the atom couples is to decrease the periodicity of the entanglement revival. In other words, the process of revival of the entanglement degrades with an increasing number of modes to which the atoms are coupled. Equivalently, we may say that the memory effects of the atom-field interaction degrade with an increasing number of the field modes.
Suppose now that the system starts at $t=0$ from the state (\[eq25\]) in which the cavity modes are entangled and the atoms are in their ground states. In this case, the transfer of the initial entanglement may depend strongly on the number of the cavity modes to which the atoms are coupled. Figure \[fig5\] shows the time evolution of the concurrence between the atoms for two different numbers of cavity modes. The effect of increasing the number of modes is seen to decrease the efficiency of transferring entanglement from the field to the atoms, namely, the entanglement between the atoms decreases in magnitude and occurs in more restricted intervals of time. This is readily understood if it is recalled that increasing the number of modes leads to a smaller efficiency of transferring an initial entanglement existing between two modes of the cavities to the atoms. One may notice that the concurrence exhibits collapses and revivals on time scales much larger than the Rabi period, the phenomenon characteristics of the Jaynes-Cummings system [@SM83].
It is interesting to note that the evolution of the concurrence from an initial state in which atoms are entangled follows the evolution of the atomic population. To see it, we plot in Fig. \[fig6\] the time evolution the probabilities $|C_{1}(t)|^2$ and $|C_{2}(t)|^2$ for the same parameters as in Fig. \[fig4\]. Clearly, the initially excited atoms decay almost exponentially in time [@Meyster1]. However, at the particular times that correspond to $nL/c$, where $n$ is an integer, a sudden change (jump) in the probabilities occurs. These are just the times when the radiation field emitted into the cavity modes returns to the atoms. Correspondingly, the entanglement revivals and remains nonzero for a finite time the excitation resides in the atoms.
The sudden jumps continue in time. However, the periodic maxima of the populations are reduced in magnitude and become more broadened as $t$ increases. This effect can be explained using the energy-time uncertainty arguments and is readily understood if it is recalled that the excitation wave packet spreads during the evolution, that the excitation becomes less localized as time progresses. This delocalisation of the excitation, it turns out, is sufficient to wipe out the entanglement.
Figure \[fig7\] shows the time evolution of the concurrence for two initial conditions of maximally and non-maximally entangled doubly excited state (\[eq26\]). At very early times, the concurrence decays almost exponentially to zero, remains zero for a finite period of time and then revivals. The concurrence remains nonzero for a finite time, and the process repeats with increasing distortion. The distortion is significantly much stronger than that observed previously for the evolution of the single excitation states. The reason is that the evolution of the excitations occurs in two JC cavities with no fixed relation between them. At short times the excitations in the two JC cavities are unresolved resulting in the revival of the entanglement, but after a long time, the excitations can be resolved which gives no inkling of correlations between the atoms.
Summary {#Conclusion}
=======
We have studied the effect of retardation on the evolution of entanglement in a system composed of two independent Jaynes-Cummings cavities. The retardation effects could be important in large cavities where the mode spacing is small enough such that an atom located inside the cavity can concurrently couple to many modes of the cavity field. We have used the concurrence to quantify entanglement between the atoms and the cavity modes and have shown that the multi-mode structure of the field inside the cavity affects considerably the harmonic evolution of the atomic population and entanglement characteristic of single-mode Jaynes-Cummings systems. We have considered both single and double excitation cases and have shown that the retardation leads to sudden death of the initial entanglement and abrupt revival at times corresponding to intervals required for a round trip of the excitation in the cavity. This demonstrates that in the presence of the retardation, the complete transfer of entanglement during the evolution of the independent JC systems never occurs, which is the essential feature of single mode Jaynes-Cummings systems. Finally, we have shown that single-excitation entangled states are more robust against the retardation than doubly-excited states.
Acknowledgments {#acknowledgments .unnumbered}
===============
This work was supported by a research grant from the King Abdulaziz City for Science and Technology (KACST).
References {#references .unnumbered}
==========
[50]{}
Olmschenk S, Matsukevich D N, Maunz P, Hayes D, Duan L M and Monroe C 2009 Science **323** 486
Feng M and Wang X 2002 J. Opt. B: Quantum and Semiclassical Opt. [**4**]{} 283
Ficek Z 2010 Front. Phys. China [**5**]{} 26
Ann K and Jaeger G 2009 Foundations of Physics [**39**]{} 790
Zyczkowski K, Horodecki P, Horodecki M and Horodecki R 2001 Phys. Rev. A [**65**]{} 012101
Dodd P J and Halliwell J J 2004 Phys. Rev. A [**69**]{} 052105
Almeida M P, de Melo F, Hor-Meyll M, Salles A, Walborn S P, Ribeiro P H S and Davidovich L 2007 Science [**316**]{} 579
Salles A, de Melo F, Almeida M P, Hor-Meyll M, Walborn S P, Souto Ribeiro P H and Davidovich L 2008 Phys. Rev. A [**78**]{} 022322
Laurat J, Choi K S, Deng H, Chou C W and Kimble H J 2007 Phys. Rev. Lett. [**99**]{} 180504
Jaynes E T and Cummings F W 1963 Proc. IEEE **51** 89
Paul H 1963 Ann. der Phys. **466** 411
Yonac M and Eberly J H 2006 J. Phys. B: At. Mol. Opt. Phys. **39** S621
Yonac M and Eberly J H 2007 J. Phys. B: At. Mol. Opt. Phys. **40** S45
Yonac M and Eberly J H 2010 Phys. Rev. A **82** 022321
Yonac M and Eberly J H 2008 Opt. Lett. **33** 270
Sainz I and Bjork G 2007 Phys. Rev. A **76** 042313
Chan S, Reid M D and Ficek Z 2009 J. Phys. B: At. Mol. Opt. Phys. **42** 065507
Bougouffa S 2010 Optics Comm. **283** 2989
Tahira R, Ikram M, Bougouffa S and Zubairy M S 2010 J. Phys. B: At. Mol. Opt. Phys. **43** 035502
Bougouffa S and Hindi A 2011 Phys. Scr. **T143** 014006
Nohama F K and Roversi J A 2007 J. Mod. Opt. [**54**]{} 1139
Ogden C D, Irish E K and Kim M S 2008 Phys. Rev. A [**78**]{} 063805
El-Orany F A A and Wahiddin M R B 2010 J. Phys. B: At. Mol. Opt. Phys. **43** 085502
Dong Y L, Zhu S Q and You W L 2012 Phys. Rev. A [**85**]{} 023833
Guo Y, Li J, Zhang T and Paternostro M 2012 Phys. Rev. A [**86**]{} 052315
Xue P, Ficek Z and Sanders B C 2012 Phys. Rev. A [**86**]{} 043826
Pellizzari T 1997 Phys. Rev. Lett. [**79**]{} 5242
Serafini A, Mancini S and Bose S 2006 Phys. Rev. Lett. [**96**]{} 010503
Peng P and Li F L 2007 Phys. Rev. A [**75**]{} 062320
Zheng S B 2009 Appl. Phys. Lett. [**94**]{} 154101
Zheng S B, Yang Z B and Xia Y 2010 Phys. Rev. A [**81**]{} 015804
Sun L H and Li G X 2012 Phys. Rev. A [**85**]{} 065801
Yu T and Eberly J H 2004 Phys. Rev. Lett. [**93**]{} 140404
Yu T and Eberly J H 2009 Science [**323**]{} 598
Cole J H 2010 J. Phys. A: Math. Theor. [**43**]{} 135301
Milonni P W and Knight P L 1974 Phys. Rev. A [**10**]{} 1096
Arecchi F T and Courtens E 1970 Phys. Rev. A [**2**]{} 1730
Gie$\beta$en H, Berger J D, Mohs G, Meystre P and Yelin S F 1996 Phys. Rev. A [**53**]{} 2816
Goldstein E V and Meystre P 1997 Phys. Rev. A **56** 5135
Gulfam Q, Ficek Z and Evers J 2012 Phys. Rev. A [**86**]{} 022325
Yariv A 1991 Optical Electronics 4th Edition, Holt, Rinehart and Winston, Philadelphia, p. 736
Wootters K 1998 Phys. Rev. Lett. **80** 2248
Sanchez-Mondragon J J, Narozhny N B and Eberly J H 1983 Phys. Rev. Lett. **51** 550
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Dong Huang, Jian-Huang Lai, and Chang-Dong Wang, [^1]'
bibliography:
- 'tkde\_2014\_short.bib'
title: |
Robust Ensemble Clustering Using\
Probability Trajectories
---
ensemble clustering technique has recently been drawing increasing attention due to its ability to combine multiple clusterings to achieve a probably better and more robust clustering [@Fred05_EAC; @LiT08; @yang11_tkde; @iam_on11_linkbased; @LiT11; @iamon12_tkde; @yi_icdm12; @ren13_icdm]. Despite the fact that many ensemble clustering approaches have been developed in recent years, there are still two limitations to most of the existing approaches.
First, the existing approaches mostly overlook the problem of uncertain links (or unreliable links) in the ensemble, which may mislead the overall consensus process. In the general formulation of the ensemble clustering problem [@Fred05_EAC; @iam_on11_linkbased; @iamon12_tkde; @yi_icdm12], we have no access to the original data features, as only the different types of relational information are available. The most basic relational information are the links between objects which reflect how objects are grouped in the same cluster or different clusters in the ensemble [@Fred05_EAC]. Based on the object-object links, coarser grains of links can be defined, e.g., the links between objects and clusters, the links between clusters, the links between base clusterings, etc. The links of one type or different types can be further used to form the similarity matrix [@Fred05_EAC; @yang11_tkde; @iam_on11_linkbased], construct the graph model [@strehl02; @fern04_bipartite; @Mimaroglu11_pr], or define the optimization problem [@LiT08; @cristofor02; @topchy05].
A link between two data objects denotes that they appear in the same cluster in one or more base clusterings. The links between objects are typically represented by the co-association (CA) matrix [@Fred05_EAC; @yang11_tkde; @wang09_pr]. In the similarity graph induced by the CA matrix, each node corresponds to a data object and the weight of each link corresponds to an entry in the CA matrix. However, previous approaches generally overlook the different reliability of the links (or the entries in the CA matrix) and may suffer from the collective influence of the unreliable links. As an example, we construct an ensemble of $10$ base clusterings for the $\emph{MNIST}$ dataset (see Section \[sec:construct\_base\] for more details). Figure \[fig:link\_weight\_distribution\] illustrates the distribution of the link weights in the similarity graph induced by the CA matrix, and Table \[table:link\_weight\_distribution\] shows the percentages of the links with different weights that make *correct* decisions. Note that a “link” with zero-weight does not count as a link here. If there is a link between objects $x_i$ and $x_j$ in the graph AND $x_i$ and $x_j$ are in the same class in the ground-truth, then we say the link between $x_i$ and $x_j$ makes a *correct* decision. As shown in Table \[table:link\_weight\_distribution\], the links with greater weights are much more likely to make correct decisions and generally more reliable than the small-weight links. But unfortunately, the small-weight links, which are probably unreliable (or uncertain), make up the majority of the graph links (see Fig. \[fig:link\_weight\_distribution\]). When the number of the uncertain links is large, the collective influence of them may mislead the ensemble clustering process or even lead to deteriorative clustering results [@yi_icdm12]. It remains an open problem how to effectively and efficiently deal with the uncertain links (or probably unreliable links) and thereby enhance the robustness and accuracy of the consensus results.
[ [![The distribution of the link weights for the similarity graph induced by the CA matrix for the *MNIST* dataset[]{data-label="fig:link_weight_distribution"}](Figures/pie2 "fig:"){width="0.7\linewidth"}]{}]{}
[m[0.7cm]{}<|m[2.4cm]{}<|m[4.6cm]{}<]{} Weight &$\frac{\#\textrm{Links with the weight}}{\#\textrm{All (non-zero) Links}}$ &$\frac{\#\textrm{Links with the weight that make correct decisions}}{\#\textrm{Links with the weight}}$\
$1$ &$2\%$ &$83\%$\
$0.9$ &$2\%$ &$73\%$\
$0.8$ &$2\%$ &$65\%$\
$0.7$ &$3\%$ &$58\%$\
$0.6$ &$3\%$ &$50\%$\
$0.5$ &$4\%$ &$43\%$\
$0.4$ &$6\%$ &$35\%$\
$0.3$ &$9\%$ &$27\%$\
$0.2$ &$18\%$ &$18\%$\
$0.1$ &$51\%$ &$9\%$\
Second, the existing ensemble clustering approaches mostly lack the ability to utilize global structure information to refine the local links. In the classical evidence accumulation clustering [@Fred05_EAC] and some of its extensions [@yang11_tkde; @wang09_pr], the CA matrix reflects the local (or *direct*) relationship, i.e., the co-occurrence relationship, between objects, yet generally neglects the *indirect* relationship inherent in the ensemble. Let $x_i$ and $x_j$ be two nodes in the graph. If there is a (non-zero) link between $x_i$ and $x_j$, then we say that there is a *direct* relationship between them. If there exist $\xi$ nodes (with $\xi\geq 1$), say, ${x'}_{1},\cdots,{x'}_{\xi}$, such that there are links between $x_i$ and ${x'}_{1}$, between ${x'}_{\xi}$ and $x_j$, and between ${x'}_{k}$ and ${x'}_{k+1}$ for any $1\leq k <\xi$, then $x_i$ and $x_j$ are indirectly connected by the $\xi$ nodes. Here we say there is a $\xi$-step *indirect* relationship between $x_i$ and $x_j$. We refer to the entirety of the direct relationships and the different steps of indirect relationships as the *global structure information*. Three sample graphs are illustrated in Fig. \[fig:global\_local1\], Fig. \[fig:global\_local2\], and Fig. \[fig:global\_local3\], respectively. Although the direct link weights between $x_1$ and $x_2$ and between $x_2$ and $x_3$ remain unchanged across the three sample graphs in Fig. \[fig:global\_locals\], the global structures of the three graphs are very different. The local links may be affected by the noise and the inherent complexity of the real-world datasets, while the global structure information is more robust to the potential local errors and can provide an alternative way to explore the relationship between objects. The relationship between objects lies not only in the direct connections, but also in the indirect connections [@iam_on11_linkbased]. The key problem here is how to exploit the global structure information in the ensemble effectively and efficiently and thereby improve the final clustering results.
Aiming to tackle the aforementioned two limitations, in this paper, we propose an ensemble clustering approach based on sparse graph representation and probability trajectory analysis.
We introduce the concept of microclusters as a compact representation for the ensemble data, which is able to greatly reduce the problem size and facilitate the computation. To deal with the uncertain links, we propose a $k$-nearest-neighbor-like strategy termed elite neighbor selection (ENS) to identify the uncertain links and build a sparse graph termed $K$-elite neighbor graph ($K$-ENG) that preserves only a small number of probably reliable links. Two microclusters are elite neighbors if the link between them has a large weight (see Definition \[def:K\_EN\]). We argue that the use of a small number of probably reliable links is capable of reflecting the overall structure of the graph and may lead to much better and more robust clustering results than using all graph links without considering their reliability. According to our experimental analysis, we find that only preserving several percent or even less than one percent of the links via the ENS strategy can result in significant improvements to the final consensus results compared to using all links in the original graph.
Having constructed the $K$-ENG graph, we proceed to exploit the global structure information to refine local links by means of random walks. The random walk technique has proved to be a powerful tool for finding community structures (or cluster structures) in the field of community discovery [@newman04; @pons_rw_05; @lai_PRE10; @neiwalk14_tkde]. The random walks are performed on the sparse graph $K$-ENG and each node in $K$-ENG is treated as a start node for a random walker. We propose a new transition probability matrix that simultaneously considers the link weights and the node sizes. By analyzing the probability trajectories of the random walkers starting from different initial nodes, we derive a novel and dense similarity measure termed probability trajectory based similarity (PTS) from the sparse graph. The PTS explores the pair-wise relationships by capturing the global structure information in the sparse graph via random walk trajectories. Based on PTS, two consensus functions are further proposed, namely, probability trajectory accumulation (PTA) and probability trajectory based graph partitioning (PTGP).
The overall process of our approach is illustrated in Fig. \[fig:flowchart\]. Given the ensemble of base clusterings, we first map the data objects to a set of microclusters and compute the microcluster based co-association (MCA) matrix. With each microcluster treated as a node, we construct the microcluster similarity graph (MSG) according to the MCA matrix. Then, the ENS strategy is performed on the MSG and the sparse graph $K$-ENG is constructed by preserving a small number of probably reliable links. The random walks are conducted on the $K$-ENG graph and the PTS similarity is obtained by comparing random walk trajectories. Having computed the new similarity matrix, any pair-wise similarity based clustering methods can be used to achieve the consensus clustering. Typically, we propose two novel consensus functions, termed PTA and PTGP, respectively. Note that PTA is based on agglomerative clustering, while PTGP is based on graph partitioning. Figure \[fig:ensize9in1\] summarizes the average performances (over nine real-world datasets) of the proposed PTA and PTGP methods and four existing methods, i.e., MCLA [@strehl02], GP-MGLA [@huang14_weac], EAC [@Fred05_EAC], and WEAC [@huang14_weac], with varying ensemble sizes. Each method is run 20 times on each dataset and their average NMI scores are shown in Fig. \[fig:ensize9in1\]. By dealing with the uncertain links and the global structure information, the proposed PTA and PTGP methods exhibit a significant advantage in clustering robustness (to various datasets and ensemble sizes) over the baseline methods. Please see Section \[sec:experiment\] for more extensive details of our experimental evaluation.
[ [![Flow diagram of the proposed approach.[]{data-label="fig:flowchart"}](Figures/flowchart3 "fig:"){width="0.67\linewidth"}]{}]{}
The main contributions of our approach are summarized as follows:
1. Our approach addresses the issue of uncertain links in an effective and efficient manner. We propose to identify the uncertain links by the ENS strategy and build a sparse graph with a small number of probably reliable links. Our empirical study shows the advantage of using only a small number of probably reliable links rather than all graph links regardless of their reliability.
2. Our approach is able to incorporate global information to construct more accurate local links by exploiting the random walk trajectories. The random walkers driven by a new probability transition matrix are utilized to explore the graph structure. A dense similarity measure is further derived from the sparse graph $K$-ENG using probability trajectories of the random walkers.
3. Extensive experiments are conducted on a variety of real-world datasets. The experimental results show that our approach yields significantly better performance than the state-of-the-art approaches w.r.t. both clustering accuracy and efficiency.
[ [![Average performances (in terms of NMI) of different approaches over nine different real-world datasets with varying ensemble size $M$.[]{data-label="fig:ensize9in1"}](Figures/EnSizesIII/Ensize9in1 "fig:"){width="0.65\linewidth"}]{}]{}
The rest of this paper is organized as follows. The related work is introduced in Section \[sec:related\_work\]. The formulation of the ensemble clustering problem is provided in Section \[sec:formulation\]. The proposed ensemble clustering approach is described in Section \[sec:our\_framework\]. The experimental results are reported in Section \[sec:experiment\]. Section \[sec:conclusion\] concludes the paper.
related work {#sec:related_work}
============
There is a large amount of literature on ensemble clustering in the past few decades [@vega_pons11_survey]. The existing ensemble clustering approaches can be categorized into three main classes, that is, (i) the pair-wise similarity based approaches [@Fred05_EAC; @yang11_tkde; @wang09_pr; @li07], (ii) the median partition based approaches [@cristofor02; @topchy05], and (iii) the graph partitioning based approaches [@strehl02; @fern04_bipartite; @abdala10_icpr; @ren13_icdm].
The pair-wise similarity based approaches represent the ensemble information by some pair-wise similarity measure [@Fred05_EAC; @yang11_tkde; @wang09_pr; @li07]. The evidence accumulation clustering (EAC) proposed by Fred and Jain [@Fred05_EAC] is probably the best-known pair-wise similarity based approach. In EAC, a co-association (CA) matrix is constructed by counting how many times two objects occur in the same cluster in the ensemble of multiple base clusterings. By treating the CA matrix as a new similarity matrix, clustering algorithms, such as the agglomerative clustering methods, can be further utilized to obtain the consensus clustering. Li et al. [@li07] proposed a hierarchical clustering algorithm to construct the consensus clustering using the CA matrix. The concept of normalized edges is introduced in [@li07] to measure the similarity between clusters. Wang et al. [@wang09_pr] extended the EAC method by taking the cluster sizes into consideration and proposed the probability accumulation method.
The median partition based approaches aim to find a clustering (or partition) that maximizes the similarity between this clustering and all of the base clusterings, which can be viewed as finding the median point of the base clusterings [@cristofor02; @topchy05]. Due to the huge space of all possible clusterings, it is generally infeasible to find the optimal solution for the median partition problem. In fact, the median partition problem is NP-complete [@topchy05]. Cristofor and Simovici [@cristofor02] proposed to find an approximative solution for the ensemble clustering problem by exploiting the genetic algorithm, in which clusterings are represented as chromosomes. Topchy et al. [@topchy05] proposed to formulate the median partition problem into a maximum likelihood problem and solved it by the EM algorithm.
The graph partitioning based approaches are another main category of ensemble clustering [@ren13_icdm; @strehl02; @fern04_bipartite]. Strehl and Ghosh [@strehl02] formulated the ensemble clustering problem into a graph partitioning problem and proposed three ensemble clustering approaches: cluster-based similarity partitioning algorithm (CSPA), hypergraph partitioning algorithm (HGPA), and meta-clustering algorithm (MCLA). Fern and Brodley [@fern04_bipartite] formulated the clustering ensemble into a bipartite graph by treating both clusters and objects as graph nodes and obtained the consensus clustering by partitioning the bipartite graph. Ren et al. [@ren13_icdm] proposed to assign weights to data objects with regard to how difficult it is to cluster them and presented three graph partitioning algorithms based on the weighted-object scheme, that is, weighted-object meta clustering (WOMC), weighted-object similarity partition (WOSP) clustering, and weighted-object hybrid bipartite (WOHB) graph partition clustering.
Despite the significant success, there are still two limitations to most of the existing ensemble clustering approaches. First, the existing approaches mostly overlook the problem of uncertain links which may mislead the consensus process. Second, most of them lack the ability to incorporate global structure information to refine local links accurately and efficiently. Recently, some efforts have been made to address these two limitations. To deal with the uncertain links, Yi et al. [@yi_icdm12] proposed an ensemble clustering approach based on global thresholding and matrix completion. However, using global thresholds may lead to *isolated* nodes, i.e., all of the links connected to a node may be cut out, due to the lack of local adaptivity. Moreover, in the approach of [@yi_icdm12], a parameter $C$ is used to scale the noise term in the objective function (see [@yi_icdm12] for more details) and plays a sensitive and crucial role for yielding a good consensus clustering. Without knowing the ground-truth in advance, tuning the parameter $C$ for [@yi_icdm12] is very difficult and computationally expensive. To exploit the indirect relationships in the ensemble, Iam-On et al. [@iam_on11_linkbased; @iamon12_tkde] proposed to refine the CA matrix by considering the shared neighbors between clusters. The approaches in [@iam_on11_linkbased] and [@iamon12_tkde] utilize the common neighborhood information, i.e., the $1$-step indirect relationships, and have not gone beyond the $1$-step indirect relationships to explore the more comprehensive structure information in the ensemble. To utilize multi-step indirect structure information, in the work of [@iamon08_icds], Iam-on et al. proposed to refine pair-wise links by the SimRank similarity (SRS), which, however, suffers from its high computational complexity and is not feasible for large datasets (see Fig. \[fig:time\_complexity\]). Different from [@yi_icdm12] and [@iamon08_icds], in this paper, we propose to tackle these two limitations in a unified and efficient manner. We present the elite neighbor selection strategy to identify the uncertain links by locally adaptive thresholds and build a sparse graph with a small number of probably reliable links, which has shown its advantage compared to using the whole original graph without considering the reliability of the links (see Section \[sec:experiment\]). To explore the global structure information of the graph, we exploit the random walk process with a new transition probability matrix. By analyzing the probability trajectories of random walkers, a novel and dense similarity measure termed PTS is derived from the sparse graph. Specifically, based on the PTS, two ensemble clustering algorithms are further proposed, termed PTA and PTGP, respectively.
Problem Formulation {#sec:formulation}
===================
-0.2 in \[table:notations\]
--------------------- ---------------------------------------------------------------------------------------
$N$ Number of data objects in a dataset
$x_i$ Data object
$\mathcal{X}$ Dataset of $N$ objects, $\mathcal{X}=\{x_1,\cdots,x_N\}$
$n^k$ Number of clusters in the $k$-th base clustering
$C^k_j$ The $j$-th cluster in the $k$-th base clustering
$\pi^k$ The $k$-th base clustering, $\pi^k=\{C^k_1,\cdots,C^k_{n^k}\}$
$M$ Number of base clusterings
$\Pi$ Ensemble of $M$ base clusterings, $\Pi=\{\pi^1,\cdots,\pi^M\}$
$\pi^*$ Consensus clustering
$N_c$ Total number of clusters in $\Pi$
$C_i$ The $i$-th cluster in $\Pi$
$\mathcal{C}$ Set of clusters in $\Pi$, $\mathcal{C}=\{C_1,\cdots,C_{N_c}\}$
$\tilde{N}$ Number of microclusters
$y_i$ The $i$-th microcluster
$\mathcal{Y}$ Set of microclusters, $\mathcal{Y}=\{y_1,\cdots,y_{\tilde{N}}\}$
$\tilde{n}_i$ Number of data objects in $y_i$
$b_{ij}$ Number of times $x_i$ and $x_j$ occur in the same cluster in $\Pi$
$a_{ij}$ Entry of co-association (CA) matrix
$A$ CA matrix, $A=\{a_{ij}\}_{N\times N}$
$\tilde{b}_{ij}$ Number of times $y_i$ and $y_j$ occur in the same cluster in $\Pi$
$\tilde{a}_{ij}$ Entry of microcluster based co-association (MCA) matrix
$\tilde{A}$ MCA matrix, $\tilde{A}=\{\tilde{a}_{ij}\}_{\tilde{N}\times \tilde{N}}$
$\tilde{G}$ Microcluster similarity graph (MSG)
$\tilde{V}$ Node set of $\tilde{G}$. Note that $\tilde{V}=\mathcal{Y}$
$\tilde{L}$ Link set of $\tilde{G}$
$\tilde{w}_{ij}$ Weight between two nodes in $\tilde{G}$
$\bar{G}$ $K$-elite neighbor graph ($K$-ENG)
$\bar{V}$ Node set of $\bar{G}$. Note that $\tilde{V}=\mathcal{Y}$
$\bar{L}$ Link set of $\bar{G}$
$\bar{w}_{ij}$ Weight between two nodes in $\bar{G}$
$p_{ij}$ (1-step) transition probability from $y_i$ to $y_j$
$P$ (1-step) transition probability matrix, $P=\{p_{ij}\}_{\tilde{N}\times\tilde{N}}$
$p^T_{ij}$ $T$-step transition probability from $y_i$ to $y_j$
$P^T$ $T$-step transition probability matrix, $P^T=\{p^T_{ij}\}_{\tilde{N}\times\tilde{N}}$
$p^T_{i:}$ The $i$-th row of $P^T$, $p^T_{i:}=\{p^T_{i1},\cdots,p^T_{i\tilde{N}}\}$
$PT^T_i$ Probability trajectory of a random walker starting from
node $y_i$ with length $T$
$PTS_{ij}$ Probability trajectory based similarity between $y_i$ and $y_j$
$\mathcal{R}^{(0)}$ Set of the initial regions for PTA,
$\mathcal{R}^{(0)}=\{R^{(0)}_1,\cdots,R^{(0)}_{|\mathcal{R}^{(0)}|}\}$
$S^{(0)}$ Initial similarity matrix for PTA,
$S^{(0)}=\{s^{(0)}_{ij}\}_{|\mathcal{R}^{(0)}|\times |\mathcal{R}^{(0)}|}$
$\mathcal{R}^{(t)}$ Set of the $t$-step regions for PTA,
$\mathcal{R}^{(t)}=\{R^{(t)}_1,\cdots,R^{(t)}_{|\mathcal{R}^{(t)}|}\}$
$S^{(t)}$ The $t$-step similarity matrix for PTA,
$S^{(t)}=\{s^{(t)}_{ij}\}_{|\mathcal{R}^{(t)}|\times |\mathcal{R}^{(t)}|}$
$\ddot{G}$ Microcluster-cluster bipartite graph (MCBG)
$\ddot{N}$ Number of nodes in $\ddot{G}$
$\ddot{V}$ Node set of $\ddot{G}$
$\ddot{L}$ Link set of $\ddot{G}$
$\ddot{w}_{ij}$ Weight between two nodes in $\ddot{G}$
--------------------- ---------------------------------------------------------------------------------------
: Summary of notations
In Table \[table:notations\], we summarize the notations that are used throughout the paper. Let $\mathcal{X}=\{x_1,\cdots,x_N\}$ be a dataset of $N$ objects. Given $M$ partitions of $\mathcal{X}$, each treated as a base clustering, the goal is to find a consensus clustering $\pi^*$ that summarizes the information of the ensemble. The ensemble is denoted as $\Pi=\{\pi^1,\cdots,\pi^M\}$, where $\pi^k=\{C^k_1,\cdots,C^k_{n^k}\}$ is the $k$-th base clustering. Let $Cls^k(x_i)$ be the cluster in $\pi^k$ that contains object $x_i$. If $x_i\in C^k_j$, then $Cls^k(x_i)=C^k_j$. Let $\mathcal{C}=\{C_1,\cdots,C_{N_c}\}$ be the set of clusters in all of the $M$ base clusterings. Obviously, $N_c=\sum_{k=1}^M n^k$.
With regard to the difference in the input information, there are two formulations of the ensemble clustering problem. In the first formulation, the ensemble clustering system takes both the clustering ensemble $\Pi$ and the data features of $\mathcal{X}$ as inputs [@vega_pons10; @vega_pons_PRL11]. In the other formulation, the ensemble clustering system takes only the clustering ensemble $\Pi$ as input and has no access to the original feature vectors of the dataset [@Fred05_EAC; @iam_on11_linkbased; @iamon12_tkde; @yi_icdm12]. In this paper, we comply with the latter formulation of the ensemble clustering problem, which is also the common practice for most of the existing ensemble clustering algorithms [@vega_pons11_survey]. That is, in our formulation, the *input* of the ensemble clustering system is the ensemble $\Pi$, and the *output* is the consensus clustering $\pi^*$.
Ensemble Clustering Using Probability Trajectories {#sec:our_framework}
==================================================
In this section, we describe the proposed ensemble clustering approach based on sparse graph representation and probability trajectory analysis.
The overall process of the proposed approach is illustrated in Fig. \[fig:flowchart\]. The microclusters are used as a compact representation for the clustering ensemble. The microcluster based co-association (MCA) matrix for the ensemble is computed and a microcluster similarity graph (MSG) is constructed from the MCA matrix with the microclusters treated as graph nodes. In order to deal with the uncertain links, we propose a $k$-nearest-neighbor-like pruning strategy termed elite neighbor selection (ENS), which is able to identify the uncertain links by locally adaptive thresholds. A sparse graph termed $K$-elite neighbor graph ($K$-ENG) is then constructed with only a small number of probably reliable links. The ENS strategy is a crucial step in our approach. We argue that using a small number of probably reliable links may lead to significantly better consensus results than using all graph links regardless of their reliability. The random walk process driven by a new transition probability matrix is performed on the $K$-ENG to explore the global structure information. From the sparse graph $K$-ENG, a dense similarity measure termed PTS is derived by exploiting the probability trajectories of the random walkers. Two novel consensus functions are further proposed, termed PTA and PTGP, respectively. In the following, we describe each step of our framework in detail.
-0.1in \[table:mc\_example\]
------- --------- --------- -------
$\pi^1$ $\pi^2$
$x_1$ $C_1^1$ $C_1^2$ $y_1$
$x_2$ $C_1^1$ $C_1^2$ $y_1$
$x_3$ $C_1^1$ $C_1^2$ $y_1$
$x_4$ $C_1^1$ $C_2^2$ $y_2$
$x_5$ $C_2^1$ $C_2^2$ $y_3$
$x_6$ $C_2^1$ $C_2^2$ $y_3$
$x_7$ $C_2^1$ $C_3^2$ $y_4$
$x_8$ $C_2^1$ $C_3^2$ $y_4$
------- --------- --------- -------
: An example of the microcluster representation
Microcluster Based Representation {#sec:MCA}
---------------------------------
In this paper, we propose to discover the object relationship by analyzing the probability trajectories of the random walkers. One hurdle in conducting random walks is the computational complexity. In a graph of $N$ nodes, it takes $O(TN^2)$ operations to calculate $T$ steps of random walks, which limits its application in large datasets. A practical strategy is to use a larger granularity than the original objects to reduce the number of graph nodes. In the field of image segmentation, superpixels are often adopted as the primitive objects [@CVPR12_Li]. However, with neither the spatial constraints of the image data nor access to the original data features, the conventional superpixel segmentation methods [@CVPR12_Li] are not applicable for generating the primitive segments for the ensemble clustering problem. In this paper, we introduce the concept of microclusters as a compact representation for the ensemble. The objects $x_i$ and $x_j$ are defined to be in the same microcluster if and only if they occur in the same cluster for *all* of the $M$ base clusterings, i.e., for $k=1,\cdots,M$, $Cls^k(x_i)=Cls^k(x_j)$.
\[def:MC\] Let $y$ be a set of objects. The set $y$ is a microcluster if and only if (i) $\forall x_i, x_j\in y$ and $k\in\mathbb{Z}$ s.t. $1\leq k \leq M$, $Cls^k(x_i)=Cls^k(x_j)$, and (ii) $\forall x_i\in y$ and $x_j \not\in y$, $\exists k\in\mathbb{Z}$ s.t. $1\leq k \leq M$, $Cls^k(x_i)\neq Cls^k(x_j)$.
Given the clustering ensemble, we can produce a set of $\tilde{N}$ non-overlapping microclusters, denoted as $$\mathcal{Y}=\{y_1,\cdots,y_{\tilde{N}}\}.$$ Intuitively, the set of microclusters is produced by intersecting the $M$ base clusterings. As it is defined, there are no clues to distinguish the objects in the same microcluster given the information of the ensemble $\Pi$. In our work, the microclusters are utilized as the primitive objects. For any microcluster $y_i\in\mathcal{Y}$ and cluster $C_j^k\in \pi^k$, it holds that *either* every object in $y_i$ is in $C_j^k$ *or* no object in $y_i$ is in $C_j^k$. Because the microclusters are treated as primitive objects, in the following, if all objects in $y_i$ are in $C_j^k$, we write it as $y_i\in C_j^k$ rather than $y_i\subseteq C_j^k$; otherwise, we write it as $y_i\not\in C_j^k$ rather than $y_i\not\subseteq C_j^k$. Let $Cls^k(y_i)$ denote the cluster in $\pi^k$ that contains the microcluster $y_i$. If $y_i\in C^k_j$, then $Cls^k(y_i) = C^k_j$.
In Table \[table:mc\_example\] and Fig. \[fig:mc\_examples\], we show an example of a dataset with eight objects to describe the construction of the microclusters. There are two base clusterings in the example, namely, $\pi^1$ and $\pi^2$, which consist of two and three clusters, respectively. With regard to Definition \[def:MC\], four microclusters can be produced using $\pi^1$ and $\pi^2$, which correspond to the intersection of $\pi^1$ and $\pi^2$ (as illustrated in Fig. \[fig:mc\_example3\] and \[fig:mc\_example4\]).
\[def:CA\] The co-association (CA) matrix of the clustering ensemble $\Pi$ is defined as $$A=\{a_{ij}\}_{N\times N},$$ where $$a_{ij}=\frac{b_{ij}}{M}$$ and $b_{ij}$ denotes how many times the objects $x_i$ and $x_j$ occur in the same cluster among the $M$ base clusterings.
\[def:MCA\] The microcluster based co-association (MCA) matrix of the clustering ensemble $\Pi$ is defined as $$\label{eq:MCA}
\tilde{A}=\{\tilde{a}_{ij}\}_{\tilde{N}\times \tilde{N}},$$ where $$\tilde{a}_{ij}=\frac{\tilde{b}_{ij}}{M}$$ and $\tilde{b}_{ij}$ denotes how many times the microclusters $y_i$ and $y_j$ occur in the same cluster among the $M$ base clusterings.
\[thm:MCA\_CA\] For all $x_i,x_j\in\mathcal{X},y_k,y_l\in\mathcal{Y}$ such that $x_i\in y_k$ and $x_j\in y_l$, it holds that $a_{ij}=\tilde{a}_{kl}$.
Given $x_i,x_j\in\mathcal{X},y_k,y_l\in\mathcal{Y}$, we have $a_{ij}=b_{ij}/M$ and $\tilde{a}_{kl}=\tilde{b}_{kl}/M$ according to Definitions \[def:CA\] and \[def:MCA\]. To prove $a_{ij}=\tilde{a}_{kl}$, we need to prove $b_{ij}=\tilde{b}_{kl}$. Given $x_i\in y_k$ and $x_j\in y_l$, for any base clustering $\pi^m\in \Pi$, if $y_k$ and $y_l$ are in the same cluster in $\pi^m$, then $x_i$ and $x_j$ are in the same cluster in $\pi^m$. Thus we have $b_{ij}\geq \tilde{b}_{kl}$. If $y_k$ and $y_l$ are in different clusters in $\pi^m$, then $x_i$ and $x_j$ are in different clusters. Thus we have $b_{ij}\leq\tilde{b}_{kl}$. Because $b_{ij}\geq \tilde{b}_{kl}$ and $b_{ij}\leq\tilde{b}_{kl}$, we have $b_{ij}=\tilde{b}_{kl}$, which leads to $a_{ij}=\tilde{a}_{kl}$.
Similar to the conventional CA matrix [@Fred05_EAC] (see Definition \[def:CA\]), the microcluster based co-association (MCA) matrix is computed by considering how many times two microclusters occur in the same cluster in $\Pi$. Then the microcluster similarity graph (MSG) is constructed based on the MCA matrix (see Definition \[def:MSG\]). By using the microclusters, the size of the similarity graph is reduced from $N$ to $\tilde{N}$.
\[def:MSG\] The microcluster similarity graph (MSG) is defined as $$\tilde{G}=(\tilde{V},\tilde{L}),$$ where $\tilde{V}=\mathcal{Y}$ is the node set and $\tilde{L}$ is the link set. The weight of the link between the nodes $y_i$ and $y_j$ is defined as $$\tilde{w}_{ij}=\tilde{a}_{ij}.$$
Elite Neighbor Selection {#sec:ENS}
------------------------
In this section, we introduce the elite neighbor selection (ENS) strategy to deal with the problem of uncertain links. The uncertain links are the connections in the similarity graph that are of low confidence (typically with small weights). One key issue here is how to decide a *proper* threshold to classify low-confidence and high-confidence and thereby identify the uncertain links in the graph. Yi et al. [@yi_icdm12] proposed to use global thresholds to identify the uncertain entries in the similarity matrix, which, however, has several drawbacks in practical applications. First, it neglects the local structure of the similarity graph and may lead to isolated nodes. Second, it is also a difficult task to find a proper global threshold for different clustering ensembles due to the inherent complexity of real-world datasets. Instead of using global thresholds, in this paper, we propose a $k$-nearest-neighbor-like strategy, termed ENS, to identify the uncertain links by locally adaptive thresholds.
\[def:K\_ENT\] The $K$-elite neighbor threshold for a node $y_i$ in the MSG is defined as the value of the $K$-th largest link weight connected to $y_i$, which is denoted as $Thres_{K}(y_i)$.
Having constructed the MSG graph (see Definition \[def:MSG\]), we define the $K$-elite neighbor threshold, denoted as $Thres_{K}(y_i)$, for each node $y_i$ in the graph (see Definition \[def:K\_ENT\]). Given two node $y_i$ and $y_j$ in the MSG graph, if $y_i$ is one of the top-$K$ neighbors of $y_j$ OR $y_j$ is one of the top-$K$ neighbors of $y_i$, then $y_i$ and $y_j$ are referred to as the $K$-elite neighbors for each other. Formally, the definition of the $K$-elite neighbors is given in Definitions \[def:K\_EN\]. Note that the $K$-elite neighbor relationship is symmetric, i.e., $y_i\in K$-EN($y_j$) is equivalent to $y_j\in K$-EN($y_i$) (see Theorem \[thm:K-EN\_sym\]).
\[def:K\_EN\] Given two nodes $y_i$ and $y_j$ in the MSG graph, $y_i$ is a $K$-elite neighbor ($K$-EN) of $y_j$ if and only if $\tilde{w}_{ij}\geq$ $Thres_{K}(y_i)$ or $\tilde{w}_{ij}\geq$ $Thres_{K}(y_j)$. The set of the $K$-elite neighbors ($K$-ENs) of $y_i$ is denoted as $K$-EN($y_i$).
\[thm:K-EN\_sym\] The $K$-elite neighbor relationship is symmetric, i.e., for all $y_i,y_j\in\mathcal{Y}$, $y_i$ is a $K$-elite neighbor of $y_j$ if and only if $y_j$ is a $K$-elite neighbor of $y_i$ .
Given $y_i\in K$-EN($y_j$), it holds that $\tilde{w}_{ij}\geq$ $Thres_{K}(y_i)$ or $\tilde{w}_{ij}\geq$ $Thres_{K}(y_j)$. Thus we have $y_j\in K$-EN($y_i$) according to Definition \[def:K\_EN\]. Given $y_i\not\in K$-EN($y_j$), it holds that $\tilde{w}_{ij}<$ $Thres_{K}(y_i)$ and $\tilde{w}_{ij}<$ $Thres_{K}(y_j)$. Thus we have $y_j\not\in K$-EN($y_i$).
\[def:K\_ENG\] The $K$-elite neighbor graph ($K$-ENG) is defined as $$\bar{G}=(\bar{V},\bar{L}),$$ where $\bar{V}=\mathcal{Y}$ is the node set and $\bar{L}$ is the link set. The weight of the link between the nodes $y_i$ and $y_j$ is defined as $$\bar{w}_{ij}=\begin{cases}
\tilde{w}_{ij}, &\text{if $y_i\in K$-EN($y_j$),}\\
0,&\text{otherwise}.
\end{cases}$$
The $K$-elite neighbor graph ($K$-ENG) is constructed by preserving a certain number of probably reliable links in the MSG. The link between two nodes, say, $y_i$ and $y_j$, is preserved if and only if $y_i$ is a $K$-elite neighbor of $y_j$, i.e., $y_i\in K$-EN($y_j$). In our experimental study, we have shown that setting $K$ to a small value, e.g., in the interval of $[5,20]$, which preserves only several percent or even less than one percent of the links in the MSG, can lead to much better and more robust clustering results than preserving a great portion or even all of the links. In the following steps, the small number of probably reliable links are exploited by the random walk process and a dense pair-wise measure is derived from the sparse graph $K$-ENG.
From Sparse Graph to Dense Similarity {#sec:link_propagation}
-------------------------------------
To discover the cluster structure from the sparse graph $K$-ENG, we use the random walks to explore the graph and propose to derive a dense pair-wise similarity measure based on the probability trajectories of random walkers. A random walker is a dynamic process that randomly transits from one node to one of its neighbors with a certain transition probability. The random walk technique has been widely used in the field of community discovery [@newman04; @pons_rw_05; @lai_PRE10; @neiwalk14_tkde] due to its ability of finding community structures, or cluster structures, in a graph. As pointed out in [@lai_PRE10], the random walkers that start from the same cluster are more likely to have similar patterns when visiting the graph. In other words, the random walkers that start from the same cluster are more likely to have similar trajectories when they randomly walk on the graph than the random walkers that start from different clusters. Based on the random walk technique, in this paper, we propose to discover the latent relationships between graph nodes by analyzing the probability trajectories of the random walkers that start from different nodes.
The nodes in the $K$-ENG are microclusters, each of which consists of a certain number of original objects. Here, we refer to the number of objects in a microcluster as the size of the microcluster. In a link-weighted graph, the transition probability from a node to one of its neighbors is generally proportional to the weight of the link between them [@newman04; @pons_rw_05; @lai_PRE10]. In the graph model of [@newman04], [@pons_rw_05], and [@lai_PRE10], all of the nodes are identical, which is different from the $K$-ENG in that the nodes in $K$-ENG may be of different sizes or even significantly different sizes. There is a need to distinguish the nodes of different sizes when deciding the transition probabilities.
In Fig. \[fig:node\_weight1\], we illustrate a graph with three nodes, each being a microcluster. The nodes $y_1$, $y_2$, and $y_3$ consist of four, one, and two objects, respectively. There are two (microcluster-microcluster) links in this graph which are equally weighted at $0.7$. When treating the three nodes equally, the transition probabilities from $y_3$ to $y_1$ and from $y_3$ to $y_2$ are equally $0.5$, i.e., $p_{31}=p_{32}=0.7/(0.7+0.7)=0.5$. However, the microcluster-microcluster links are representative of a certain number of hidden object-object links. With respect to the object-object relationships, there are $4\times 2=8$ and $1\times 2 = 2$ hidden object-object links between $y_1$ and $y_3$ and between $y_2$ and $y_3$, respectively (see Fig. \[fig:node\_weight2\]). The weights of the object-object links correspond to the entries of the CA matrix (see Definition \[def:CA\]), whereas the weights of the microcluster-microcluster links correspond to the entries of the MCA matrix (see Definition \[def:MCA\]). According to Theorem \[thm:MCA\_CA\], the weight of every object-object link between $y_1$ and $y_3$ (or between $y_2$ and $y_3$) is equal to that of the microcluster-microcluster link between $y_1$ and $y_3$ (or between $y_2$ and $y_3$), that is to say, the weight of every object-object link in the graph illustrated in Fig. \[fig:node\_weight2\] is equal to $0.7$. If we perform random walks on the object granularity rather than the microcluster granularity, the probability of walking from one of the objects in $y_3$ to one of the objects in $y_1$ would be four times as great as the probability of walking from one of the objects in $y_3$ to one of the objects in $y_2$.
To reflect the hidden object-object connections in a microcluster based graph, the construction of the transition probability matrix needs to take into consideration the sizes of the microclusters. Specifically, the transition probability from a microcluster node, say, $y_i$, to one of its neighbors should be proportional to the size of this neighbor if the weights of the links between $y_i$ and all of its neighbors are equal. In our scenario, both the weights between $y_i$ and its neighbors and the sizes of its neighbors may be different. Let $\tilde{n}_i$ be the size of a microcluster $y_i$. The transition probability from $y_i$ to one of its neighbors, say, $y_j$, is defined to be proportional to the sum of the weights of all hidden object-object links between $y_i$ and $y_j$, i.e., $\tilde{n}_i\cdot \tilde{n}_j\cdot \bar{w}_{ij}$. Because $\tilde{n}_i$ is a constant given the node $y_i$, the probability of walking from $y_i$ to one of its neighbors $y_j$ therefore should be proportional to $\tilde{n}_j\cdot \bar{w}_{ij}$. Formally, the definition of the transition probability matrix on the $K$-ENG is given in Definition \[def:transit\_matrix\].
\[def:transit\_matrix\] Let $P=\{p_{ij}\}_{\tilde{N}\times\tilde{N}}$ be the transition probability matrix of the random walk on the $K$-ENG. The transition probability from node $y_i$ to node $y_j$ is defined as $$\label{eq:transit_matrix}
p_{ij}=\frac{\tilde{n}_j\cdot\bar{w}_{ij}}{\sum_{k\neq i}\tilde{n}_k\cdot\bar{w}_{ik}},$$ where $\tilde{n}_j$ is the number of objects in $y_j$.
The random walk process is driven by the transition probability matrix P. Let $P^T=\{p^T_{ij}\}_{\tilde{N}\times\tilde{N}}$ be the $T$-step transition probability matrix, i.e., probability distribution at step $T$, where $p^T_{ij}$ is the probability that a random walker starting from node $y_i$ arrives at node $y_j$ at step $T$. The probability distribution at step $1$ is obviously the transition probability matrix, i.e., $P^1=P$. The probability distribution at step $T$ is computed as $P^T=P\cdot P^{T-1}$, for $T\geq 2$.
Let $p^T_{i:}=\{p^T_{i1},\cdots,p^T_{i\tilde{N}}\}$ denote the probability distribution of node $y_i$ at step $T$, which is the $i$-th row of $P^T$ and represents the probability of going from node $y_i$ to each node in the graph by the random walk process at step $T$. The relationship between node $y_i$ and node $y_j$ can be studied by comparing their probability distributions at a certain step [@pons_rw_05]. However, the probability distributions at different steps reflect different scales of information for the graph structure. Using a single step of probability distribution as the feature of a node overlooks the properties of this node at different scales. In order to take advantage of multi-scale information in the graph, we propose to exploit the probability trajectory for describing the random walk process starting from each node, which considers the probability distributions from step $1$ to step $T$ rather than a single step. The formal definition of the probability trajectory is given in Definition \[def:PT\].
\[def:PT\] The probability trajectory of a random walker starting from node $y_i$ with length $T$ is defined as a $T\tilde{N}$-tuple: $$PT^T_i=\{p^1_{i:},p^2_{i:},\cdots,p^T_{i:}\},$$ where $p^T_{i:}$ is the probability distribution of node $y_i$ at step $T$.
The probability trajectory of a random walker starting from a given node is a $T\tilde{N}$-tuple and can be viewed as a feature vector for the node. We further define a pair-wise similarity measure based on the probability trajectory representation.
\[def:PTS\] The probability trajectory based similarity (PTS) between node $y_i$ and node $y_j$ is defined as $$\label{eq:PTS}
PTS_{ij}=Sim(PT^T_i,PT^T_j),$$ where $Sim(u,v)$ is a similarity measure between two vectors $u$ and $v$.
In fact, any similarity measure can be used in Eq. (\[eq:PTS\]). In our work, we use the cosine similarity as the similarity measure, which effectively captures the relationship between the random walk trajectories. The cosine similarity between two vectors $u$ and $v$ is computed as follows: $$\label{eq:cosine}
Sim_{cos}(u,v)=\frac{<u,v>}{\sqrt{<u,u>\cdot <v,v>}},$$ where $<u,v>$ is the inner product of $u$ and $v$.
Therefore we obtain the new similarity measure PTS using the probability trajectory of the random walker starting from each node. Specifically, the random walks are performed on the sparse graph $K$-ENG which preserves only a small number of probably reliable links by the ENS strategy. The probability trajectories on the $K$-ENG are used as the feature vectors for the graph nodes, which incorporates multi-scale graph information into a $T\tilde{N}$-tuple by the different steps of the random walks. Theoretically, it is possible that the $K$-ENG may consist of more than one connected component, in which case we can perform the random walk on each connected component of the graph separately and then map the random walk trajectories at each component back to the whole graph to facilitate the computation.
For clarity, the algorithm of computing PTS is given in Algorithm 1.
**Algorithm 1 (Computation of Probability Trajectory Based Similarity)**\
Consensus Functions
-------------------
Having generated the new similarity measure PTS, the next step is to obtain the consensus clustering. Here, any clustering algorithm based on pair-wise similarity can be applied to the PTS measure to obtain the final clustering. Typically, we propose two different types of consensus functions based on PTS, termed probability trajectory accumulation (PTA) and probability trajectory based graph partitioning (PTGP), respectively.
### Probability Trajectory Accumulation (PTA) {#sec:PTA}
In this section, we introduce the consensus function termed PTA, which is based on hierarchical agglomerative clustering.
To perform hierarchical clustering, the set of microclusters are treated as the initial regions and the PTS is used as the similarity measure to guide the region merging process. Let $\mathcal{R}^{(0)}=\{R^{(0)}_1,\cdots,R^{(0)}_{|\mathcal{R}^{(0)}|}\}$ denote the set of initial regions, where $R^{(0)}_j = y_j$ and $|\mathcal{R}^{(0)}|=\tilde{N}$. Let $S^{(0)}=\{s^{(0)}_{ij}\}_{|\mathcal{R}^{(0)}|\times |\mathcal{R}^{(0)}|}$ be the initial similarity matrix, where $s^{(0)}_{ij}=PTS_{ij}$.
In each step, the two regions with the highest similarity are merged into a new and bigger region and thus the number of regions decrements by one. Then the similarity matrix for the new set of regions will be computed w.r.t. average-link (AL), complete-link (CL), or single-link (SL). Let $\mathcal{R}^{(t)}=\{R^{(t)}_1,\cdots,R^{(t)}_{|\mathcal{R}^{(t)}|}\}$ be the set of generated regions in the $t$-step, for $t=1,2,\cdots,\tilde{N}-1$, where $|\mathcal{R}^{(t)}|$ is the number of regions in $\mathcal{R}^{(t)}$. Note that each region contains one or more microclusters. We write it as $y_i\in R^{(t)}_j$ if microcluster $y_i$ is in region $R^{(t)}_j$. Let $S^{(t)}=\{s^{(t)}_{ij}\}_{|\mathcal{R}^{(t)}|\times |\mathcal{R}^{(t)}|}$ be the similarity matrix for $\mathcal{R}^{(t)}$, which can be computed w.r.t. AL, CL, or SL. That is
$$s^{(t)}_{ij}=\begin{cases}
\frac{1}{|R^{(t)}_i|\cdot |R^{(t)}_j|}\sum_{y_k\in R^{(t)}_i,y_l\in R^{(t)}_j}PTS_{kl}, &\text{If Method=AL,}\\
\sum_{y_k\in R^{(t)}_i,y_l\in R^{(t)}_j}PTS_{kl}, &\text{If Method=CL,}\\
\max_{y_k\in R^{(t)}_i,y_l\in R^{(t)}_j}PTS_{kl}, &\text{If Method=SL,}\\
\end{cases}$$
where $|R^{(t)}_i|$ is the number of microclusters in $R^{(t)}_i$.
The region merging process is performed iteratively and the number of regions decrements by one in each step. Obviously, after the $(\tilde{N}-1)$-step, there will be one region left, which contains the entire set of the microclusters. Then we have a dendrogram, i.e., a hierarchical representation of clusterings. Each level in the dendrogram represents a clustering with a certain number of clusters (or regions). The final clustering is obtained by specifying a level for the dendrogram.
An advantage of agglomerative clustering is that it can efficiently generate a hierarchy of clusterings where each level represents a clustering with a certain number of clusters. However, the region merging process is inherently local and greedy. A mistaken merging may lead to increasing errors in the following merging steps. The similarity measure determines the region merging order and plays a crucial role in agglomerative clustering. In our work, the PTS measure is able to deal with the uncertain links and incorporate the global structure information in the ensemble, which is beneficial for improving the accuracy and robustness of the agglomerative clustering. The experimental results also show the advantage of the PTA method (base on PTS) compared to other pair-wise similarity based methods [@Fred05_EAC; @iam_on11_linkbased; @yi_icdm12; @iamon08_icds; @huang14_weac] (see Section \[sec:comp\_ensemb\] and Table \[table:compare\_ce\]).
For clarity, the PTA algorithm is summarized in Algorithm 2.
**Algorithm 2 (Probability Trajectory Accumulation)**\
### Probability Trajectory Based Graph Partitioning (PTGP) {#sec:PTGP}
In this section, we introduce the consensus function termed PTGP, which is based on bipartite graph formulation.
A bipartite graph is constructed by treating both clusters and microclusters as nodes. As illustrated in Fig. \[fig:MCBG\], there are no links between two clusters or between two microclusters. A link between two nodes exists if and only if one of the nodes is a cluster and the other is a microcluster. The weight of the link between a microcluster and a cluster is decided by the similarity between them. Here, we define the similarity between a microcluster $y_i$ and a cluster $C_j$ as the average PTS measure between $y_i$ and the microclusters in $C_j$. The formal definition is given as follows.
[ [![The microcluster-cluster bipartite graph (MCBG)[]{data-label="fig:MCBG"}](Figures/MCBGII "fig:"){width="0.8\linewidth"}]{}]{}
\[def:mc\_similarity\] The similarity between a microcluster $y_i$ and a cluster $C_j$ is defined as $$\begin{aligned}
\label{eq:mc_similarity}
Sim_{mc}(y_i, C_j) &= Sim_{mc}(C_j, y_i)\nonumber\\
&=\frac{1}{|C_j|}\sum_{y_k\in C_j}PTS_{ik},\end{aligned}$$ where $|C_j|$ is the number of microclusters in $C_j$.
\[def:MCBG\] The microcluster-cluster bipartite graph (MCBG) is defined as $$\ddot{G}=(\ddot{V},\ddot{L}),$$ where $\ddot{V}=\mathcal{Y}\bigcup \mathcal{C}=\{\ddot{v}_1,\dots,\ddot{v}_{\ddot{N}}\}$ is the node set, $\ddot{N}=\tilde{N}+N_c$ is the number of nodes and $\ddot{L}$ is the link set. The weight of the link between nodes $\ddot{v}_i$ and $\ddot{v}_j$ is defined as $$\label{eq:mcbg_weight}
\ddot{w}_{ij}=\begin{cases}
Sim_{mc}(\ddot{v}_i,\ddot{v}_j), &\text{if } \ddot{v}_i\in \mathcal{Y}, \ddot{v}_j\in\mathcal{C}\\
&\text{or }\ddot{v}_i\in\mathcal{C}, \ddot{v}_j\in \mathcal{Y}, \\
0,&\text{otherwise}.
\end{cases}$$
By treating both clusters and microclusters as nodes, the microcluster-cluster bipartite graph (MCBG) is constructed according to Definition \[def:MCBG\]. With regard to the bipartite structure of the MCBG, the efficient graph partitioning method Tcut [@CVPR12_Li] can be used to partition the graph into a certain number of disjoint sets of nodes. The microcluster nodes in the same segment are treated as a cluster and thus the final consensus clustering can be obtained by mapping the microclusters back to the data objects.
For clarity, we summarize the PTGP algorithm in Algorithm 3.
**Algorithm 3 (Probability Trajectory Based Graph Partitioning)**\
Experiments {#sec:experiment}
===========
In this section, we conduct experiments using ten real-world datasets. All of the experiments are conducted in Matlab R2014a 64-bit on a workstation (Windows Server 2008 R2 64-bit, 8 Intel 2.40 GHz processors, 96 GB of RAM).
Datasets and Evaluation Method {#sec:dataset_and_eval}
------------------------------
In our experiments, we use ten real-world datasets, namely, *Multiple Features* (*MF*), *Image Segmentation* (*IS*), *MNIST*, *Optical Digit Recognition* (*ODR*), *Landsat Satellite* (*LS*), *Pen Digits* (*PD*), *USPS*, *Forest Covertype* (*FC*), *KDD99-10P* and *KDD99*. The *MNIST* dataset and the *USPS* dataset are from [@lecun98] and [@Dueck_AP_PHDThesis:09], respectively. The *KDD99* dataset is from the UCI KDD Archive [@uci_kdd_archive99], whereas *KDD99-10P* is a $10\%$ subset of *KDD99*. The other six datasets are from the UCI machine learning repository [@Bache+Lichman:2013]. The details of the benchmark datasets are given in Table \[table:datasets\].
We use the normalized mutual information (NMI) [@strehl02] to evaluate the quality of the consensus clusterings, which provides a sound indication of the shared information between two clusterings. Note that a higher NMI indicates a better test clustering.
-0.1 in \[table:datasets\]
[p[1.6cm]{}<|p[1.5cm]{}<p[1.3cm]{}<p[1.3cm]{}<]{} Dataset &\#Object &\#Attribute &\#Class\
*MF* &2,000 &649 &10\
*IS* &2,310 &19 &7\
*MNIST* &5,000 &784 &10\
*ODR* &5,620 &64 &10\
*LS* &6,435 &36 &6\
*PD* &10,992 &16 &10\
*USPS* &11,000 &256 &10\
*FC* &11,340 &54 &7\
*KDD99-10P* &49,402 &41 &23\
*KDD99* &494,020 &41 &23\
Construction of Ensembles {#sec:construct_base}
-------------------------
To evaluate the effectiveness of the proposed approach over various combinations of base clusterings, we construct a pool of a large number of base clusterings. In our experiments, we use the $k$-means algorithm and the rival penalized competitive learning (RPCL) [@xu93_rpcl] algorithm to construct the base clustering pool. The $k$-means and RPCL algorithms are performed repeatedly with random initializations and parameters. The numbers of initial clusters for $k$-means and RPCL are randomly chosen in the interval of $[2,ub]$, where $ub=\min\{\sqrt{N}/2,50\}$ is the upper bound of the number of clusters and $N$ is the number of objects in the dataset. By running $k$-means and RPCL $100$ times respectively, a pool of $200$ base clusterings is obtained for each benchmark dataset.
For each run of the proposed methods and the baseline ensemble clustering methods, we generate the ensemble by randomly drawing $M$ base clusterings from the base clustering pool. Unless specially mentioned, the ensemble size $M=10$ is used in this paper. To rule out the factor of *getting lucky occasionally*, the average performances of the proposed methods and the baseline methods are evaluated and compared over a large number of runs.
Parameter Analysis {#sec:sens_paras}
------------------
-0.05 in \[table:avg\_MC\]
[m[2.3cm]{}<|m[1cm]{}<m[1cm]{}<m[1.1cm]{}<m[1.1cm]{}<m[1.1cm]{}<m[1.1cm]{}<m[1.2cm]{}<m[1.1cm]{}<m[1.5cm]{}<m[0.9cm]{}<]{} $Dataset$ &*MF* &*IS* &*MNIST* &*ODR* &*LS* &*PD* &*USPS* &*FC* &*KDD99-10P* &*KDD99*\
\#Node (i.e.,$\tilde{N}$) &$242$ &$297$ &$1,438$ &$899$ &$1,064$ &$1,095$ &$2,975$ &$1,837$ &$230$ &$301$\
\#Link &$12,978$ &$19,778$ &$468,453$ &$153,949$ &$219,753$ &$208,146$ &$1,825,938$ &$583,369$ &$19,260$ &$34,509$\
-0.05 in \[table:RatioPL\]
[m[1.2cm]{}<|m[1.5cm]{}<|m[0.9cm]{}<m[0.9cm]{}<m[0.9cm]{}<m[0.9cm]{}<m[0.9cm]{}<m[0.9cm]{}<m[0.9cm]{}<m[0.9cm]{}<]{} &1 &2 &4 &8 &16 &32 &64 &ALL\
&*MF* &$2.4\%$ &$3.4\%$ &$6.6\%$ &$11.4\%$ &$21.1\%$ &$43.5\%$ &$82.1\%$ &$100.0\%$\
&*IS* &$2.0\%$ &$3.2\%$ &$5.9\%$ &$10.5\%$ &$19.4\%$ &$37.9\%$ &$70.1\%$ &$100.0\%$\
&*MNIST* &$0.6\%$ &$0.9\%$ &$1.7\%$ &$2.9\%$ &$5.0\%$ &$8.7\%$ &$16.2\%$ &$100.0\%$\
&*ODR* &$1.0\%$ &$1.6\%$ &$3.0\%$ &$4.9\%$ &$8.5\%$ &$15.0\%$ &$31.0\%$ &$100.0\%$\
&*LS* &$0.8\%$ &$1.1\%$ &$2.3\%$ &$3.8\%$ &$6.9\%$ &$12.4\%$ &$23.9\%$ &$100.0\%$\
&*PD* &$0.9\%$ &$1.2\%$ &$2.4\%$ &$4.0\%$ &$7.1\%$ &$13.5\%$ &$28.1\%$ &$100.0\%$\
&*USPS* &$0.3\%$ &$0.5\%$ &$1.0\%$ &$1.7\%$ &$2.9\%$ &$5.0\%$ &$8.7\%$ &$100.0\%$\
&*FC* &$0.5\%$ &$0.7\%$ &$1.5\%$ &$2.4\%$ &$4.4\%$ &$8.0\%$ &$15.2\%$ &$100.0\%$\
&*KDD99-10P* &$1.7\%$ &$2.1\%$ &$4.2\%$ &$9.2\%$ &$23.9\%$ &$53.4\%$ &$90.0\%$ &$100.0\%$\
&*KDD99* &$1.6\%$ &$1.9\%$ &$3.6\%$ &$7.2\%$ &$14.3\%$ &$34.0\%$ &$68.9\%$ &$100.0\%$\
-0.05 in \[table:comp\_para\_PTA\_KT\]
[m[1.5cm]{}<|m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<|m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<]{} $K$ &1 &2 &4 &8 &16 &32 &64 &ALL &\
$T$ & &1 &2 &4 &8 &16 &32 &64 &128\
*MF* &0.585 &0.597 &0.630 &0.627 &0.605 &0.560 &0.538 &0.535 &0.613 &0.617 &0.619 &0.620 &0.620 &0.620 &0.619 &0.615\
*IS* &0.574 &0.612 &0.632 &0.615 &0.595 &0.592 &0.610 &0.609 &0.613 &0.616 &0.611 &0.612 &0.611 &0.612 &0.611 &0.612\
*MNIST* &0.556 &0.575 &0.585 &0.584 &0.589 &0.574 &0.538 &0.486 &0.567 &0.585 &0.582 &0.591 &0.593 &0.591 &0.592 &0.592\
*ODR* &0.770 &0.792 &0.819 &0.820 &0.813 &0.798 &0.757 &0.713 &0.790 &0.816 &0.813 &0.817 &0.816 &0.817 &0.812 &0.810\
*LS* &0.599 &0.595 &0.615 &0.618 &0.621 &0.612 &0.586 &0.539 &0.600 &0.613 &0.618 &0.620 &0.620 &0.627 &0.632 &0.637\
*PD* &0.714 &0.715 &0.757 &0.765 &0.761 &0.735 &0.698 &0.678 &0.722 &0.759 &0.768 &0.765 &0.762 &0.762 &0.761 &0.760\
*USPS* &0.560 &0.565 &0.585 &0.590 &0.591 &0.579 &0.570 &0.442 &0.574 &0.587 &0.588 &0.592 &0.595 &0.594 &0.594 &0.587\
*FC* &0.235 &0.237 &0.247 &0.254 &0.250 &0.237 &0.221 &0.199 &0.246 &0.245 &0.257 &0.255 &0.254 &0.250 &0.250 &0.250\
*KDD99-10P* &0.565 &0.599 &0.624 &0.659 &0.664 &0.592 &0.562 &0.561 &0.677 &0.681 &0.686 &0.677 &0.671 &0.668 &0.676 &0.670\
*KDD99* &0.538 &0.560 &0.613 &0.682 &0.700 &0.644 &0.570 &0.562 &0.687 &0.693 &0.698 &0.700 &0.691 &0.689 &0.694 & 0.693\
-0.05 in \[table:comp\_para\_PTGP\_KT\]
[m[1.5cm]{}<|m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<|m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<m[0.50cm]{}<]{} $K$ &1 &2 &4 &8 &16 &32 &64 &ALL &\
$T$ & &1 &2 &4 &8 &16 &32 &64 &128\
*MF* &0.609 &0.610 &0.627 &0.621 &0.596 &0.536 &0.499 &0.495 &0.619 &0.617 &0.619 &0.616 &0.610 &0.607 &0.599 &0.597\
*IS* &0.610 &0.615 &0.623 &0.616 &0.589 &0.582 &0.603 &0.594 &0.615 &0.618 &0.614 &0.611 &0.613 &0.611 &0.608 &0.616\
*MNIST* &0.575 &0.577 &0.585 &0.587 &0.587 &0.576 &0.533 &0.471 &0.581 &0.583 &0.585 &0.586 &0.588 &0.589 &0.589 &0.587\
*ODR* &0.809 &0.815 &0.822 &0.830 &0.821 &0.799 &0.745 &0.679 &0.812 &0.817 &0.824 &0.824 &0.824 &0.821 &0.818 &0.814\
*LS* &0.604 &0.607 &0.615 &0.623 &0.623 &0.621 &0.586 &0.505 &0.615 &0.616 &0.616 &0.615 &0.626 &0.629 &0.631 &0.631\
*PD* &0.746 &0.747 &0.759 &0.762 &0.756 &0.727 &0.696 &0.649 &0.749 &0.751 &0.759 &0.765 &0.766 &0.761 &0.760 &0.758\
*USPS* &0.575 &0.570 &0.578 &0.586 &0.582 &0.572 &0.556 &0.404 &0.566 &0.570 &0.576 &0.580 &0.580 &0.588 &0.584 &0.574\
*FC* &0.224 &0.227 &0.235 &0.238 &0.237 &0.234 &0.218 &0.195 &0.229 &0.231 &0.232 &0.236 &0.237 &0.237 &0.236 &0.234\
*KDD99-10P* &0.618 &0.620 &0.627 &0.643 &0.646 &0.591 &0.524 &0.521 &0.673 &0.675 &0.665 &0.659 & 0.655 & 0.656 &0.664 &0.662\
*KDD99* &0.620 &0.622 &0.624 &0.686 &0.701 &0.653 &0.571 &0.567 &0.709 &0.713 &0.720 &0.722 &0.721 & 0.698 & 0.696 & 0.704\
In this paper, we propose two ensemble clustering methods, termed PTA and PTGP, respectively. There are two parameters, namely, $K$ and $T$, in the proposed methods. The parameter $K$ specifies how many neighbors of a node will be treated as elite neighbors and preserved. The smaller the parameter $K$ is, the sparser the $K$-ENG will be. The parameter $T$ is the length of the probability trajectories.
We first test the influence of parameter $K$ on the proportion of preserved links in the $K$-ENG. The average information of the MSG over 20 runs on each dataset is shown in Table \[table:avg\_MC\]. Because the microclusters are generated by intersecting multiple base clusterings, the number of microclusters is affected by the total number of objects as well as the shapes of cluster boundaries. For the *KDD99-10P* and *KDD99* datasets, the number of microclusters is less than $1\%$ of the number of the original objects. For the other datasets, the number of microclusters is averagely about $10\%$ to $20\%$ of the number of the original objects, i.e., using microclusters as nodes reduces the graph size by about $80\%$ to $90\%$. The number of links in the MSG is also given in Table \[table:avg\_MC\]. Note that a link between two nodes exists if and only if the weight between them is non-zero, i.e., a “link” with zero-weight does not count as a link here. By cutting out the probably unreliable links in the MSG via the ENS strategy, the sparse graph $K$-ENG is constructed. The ratio of preserved links (RatioPL) is defined as $$\label{eq:ratioPL}
\mathrm{RatioPL} = \frac{\#\text{Links in $K$-ENG}}{\#\text{Links in MSG}}.$$
For each parameter setting of $K$ and $T$, we run the proposed methods $20$ times with the ensemble of base clusterings randomly drawn from the base clustering pool (see Section \[sec:construct\_base\]) at each time. The RatioPL with respect to different values of $K$ is shown in Table \[table:RatioPL\]. When $K=\text{ALL}$, all links in MSG are preserved. The average NMI scores of PTA and PTGP with varying $K$ and $T$ are reported in Tables \[table:comp\_para\_PTA\_KT\] and \[table:comp\_para\_PTGP\_KT\], respectively. The performances of PTA and PTGP are consistently good when $K$ is set in the interval of $[5,20]$ on the benchmark datasets and significantly better than setting $K=\text{ALL}$. As can be seen in Tables \[table:RatioPL\], \[table:comp\_para\_PTA\_KT\] and \[table:comp\_para\_PTGP\_KT\], preserving a small proportion of the links via the ENS strategy can lead to significantly better performance than using all graph links by setting $K=ALL$.
As shown in Tables \[table:comp\_para\_PTA\_KT\] and \[table:comp\_para\_PTGP\_KT\], the performances of the proposed PTA and PTGP methods are robust over various parameter settings. Setting $K$ and $T$ to moderate values, e.g., both in the interval of $[5,20]$, leads to consistently good performances on the benchmark datasets. Empirically, it is suggested that the parameters $K$ and $T$ be set in the interval of $[\sqrt{\tilde{N}}/5,\sqrt{\tilde{N}}]$, where $\tilde{N}$ is the number of the graph nodes in the $K$-ENG. In the following of this paper, we set both $K$ and $T$ to the floor of $\sqrt{\tilde{N}}/2$ in all experiments on the benchmark datasets.
Comparison against Base Clusterings {#sec:comp_base}
-----------------------------------
The purpose of ensemble clustering is to combine multiple base clusterings into a probably better and more robust clustering. In this section, we compare the proposed PTA (associated with average-link) and PTGP methods against the base clusterings. Figure \[fig:base\_comp\] illustrates the average NMI scores and the variances of the proposed methods and the base clusterings over 100 runs. As shown in Fig. \[fig:base\_comp\], for the benchmark datasets, the proposed PTA and PTGP algorithms produce overall more accurate clusterings than the base clusterings. Especially, for the *ODR*, *LS*, *PD*, *KDD99-10P*, and *KDD99* datasets, the proposed methods achieve significant improvements in terms of NMI compared to the base clusterings.
[ [![Average performances in terms of NMI of our methods and the base clusterings over $100$ runs.[]{data-label="fig:base_comp"}](Figures/basecomp10 "fig:"){width="0.95\linewidth"}]{}]{}
Comparison against Other Ensemble Clustering Approaches {#sec:comp_ensemb}
-------------------------------------------------------
In this section, we evaluate the effectiveness and robustness of the proposed PTA and PTGP methods by comparing them against ten ensemble clustering methods, namely, cluster-based similarity partitioning algorithm (CSPA) [@strehl02], hypergraph partitioning algorithm (HGPA) [@strehl02], meta- clustering algorithm (MCLA) [@strehl02], seeded random walk (SRW) [@abdala10_icpr], graph partitioning with multi-granularity link analysis (GP-MGLA) [@huang14_weac], evidence accumulation clustering (EAC) [@Fred05_EAC], SimRank similarity based method (SRS) [@iamon08_icds], weighted connected-triple method (WCT) [@iam_on11_linkbased], ensemble clustering by matrix completion (ECMC) [@yi_icdm12], and weighted evidence accumulation clustering (WEAC) [@huang14_weac].
Of the ten baseline ensemble clustering methods, CSPA, HGPA, MCLA, SRW, and GP-MGLA are graph partitioning based methods, while EAC, SRS, WCT, ECMC, and WEAC are pair-wise similarity based methods. The pair-wise similarity based methods construct a similarity matrix based on the ensemble information. For each of the pair-wise similarity based methods and the proposed PTA method, we use three agglomerative clustering methods to obtain the final clusterings, namely, average-link (AL), complete-link (CL), and single-link (SL). Thus, each pair-wise similarity based method is associated with three sub-methods. For the other baseline methods, we use the parameter settings as suggested by the authors in the their papers [@Fred05_EAC; @iam_on11_linkbased; @strehl02; @iamon08_icds; @abdala10_icpr; @huang14_weac].
-0.1 in \[table:compare\_ce\]
[|m[1.8cm]{}<|m[0.96cm]{}<m[0.96cm]{}<|m[0.96cm]{}<m[0.96cm]{}<|m[0.96cm]{}<m[0.96cm]{}<|m[0.96cm]{}<m[0.96cm]{}<|m[0.96cm]{}<m[0.96cm]{}<|]{} & & & & &\
&Best-$k$&True-$k$ &Best-$k$&True-$k$ &Best-$k$&True-$k$ &Best-$k$&True-$k$ &Best-$k$&True-$k$\
PTA-AL &**0.631** &**0.614** &0.623 &0.607 &**0.585** &**0.578** &**0.813** &**0.804** &**0.626** &**0.622**\
PTA-CL &**0.629** &**0.611** &0.620 &**0.609** &**0.582** &**0.577** &0.805 &0.793 &0.595 &0.584\
PTA-SL &0.612 &0.522 &0.616 &0.521 &0.523 &0.103 &0.760 &0.533 &0.552 &0.114\
PTGP &**0.626** &**0.613** &**0.625** &**0.611** &0.581 &**0.576** &**0.819** &**0.813** &**0.627** &**0.625**\
CSPA &0.597 &0.591 &0.605 &0.605 &0.493 &0.486 &0.726 &0.723 &0.511 &0.475\
HGPA &0.485 &0.231 &0.500 &0.457 &0.423 &0.120 &0.643 &0.353 &0.406 &0.324\
MCLA &0.617 &0.596 &0.623 &**0.609** &0.536 &0.518 &0.785 &0.770 &0.550 &0.518\
SRW &0.468 &0.197 &0.509 &0.175 &0.393 &0.126 &0.514 &0.135 &0.434 &0.124\
GP-MGLA &0.618 &0.604 &0.613 &0.608 &0.569 &0.557 &**0.807** &**0.798** &0.615 &**0.607**\
EAC-AL &0.603 &0.578 &0.612 &0.605 &0.570 &0.555 &0.792 &0.772 &0.596 &0.569\
EAC-CL &0.572 &0.508 &0.622 &0.442 &0.460 &0.203 &0.651 &0.389 &0.459 &0.282\
EAC-SL &0.531 &0.173 &0.542 &0.413 &0.021 &0.002 &0.257 &0.099 &0.079 &0.002\
SRS-AL &0.613 &0.584 &0.614 &0.603 &0.575 &0.557 &0.794 &0.772 &0.603 &0.583\
SRS-CL &0.587 &0.547 &**0.625** &0.585 &0.554 &0.534 &0.771 &0.744 &0.536 &0.453\
SRS-SL &0.484 &0.189 &0.579 &0.358 &0.017 &0.002 &0.130 &0.003 &0.065 &0.001\
WCT-AL &0.605 &0.579 &0.617 &0.606 &**0.585** &0.562 &0.800 &0.774 &**0.617** &0.603\
WCT-CL &0.579 &0.546 &**0.634** &**0.612** &0.561 &0.529 &0.774 &0.741 &0.540 &0.459\
WCT-SL &0.596 &0.247 &0.599 &0.415 &0.028 &0.002 &0.318 &0.132 &0.213 &0.002\
ECMC-AL &0.588 &0.333 &0.593 &0.277 &0.554 &0.132 &0.780 &0.328 &0.584 &0.018\
ECMC-CL &0.591 &0.520 &0.560 &0.411 &0.446 &0.276 &0.617 &0.454 &0.402 &0.204\
ECMC-SL &0.416 &0.166 &0.560 &0.200 &0.041 &0.015 &0.350 &0.177 &0.071 &0.001\
WEAC-AL &0.606 &0.583 &0.610 &0.605 &0.577 &0.569 &0.799 &0.785 &0.608 &0.596\
WEAC-CL &0.581 &0.520 &0.618 &0.431 &0.463 &0.194 &0.643 &0.371 &0.456 &0.235\
WEAC-SL &0.602 &0.268 &0.605 &0.417 &0.039 &0.002 &0.348 &0.126 &0.228 &0.002\
& & & & &\
&Best-$k$&True-$k$ &Best-$k$&True-$k$ &Best-$k$&True-$k$ &Best-$k$&True-$k$ &Best-$k$&True-$k$\
PTA-AL &**0.757** &**0.732** &**0.583** &**0.565** &**0.243** &**0.232** &**0.644** &**0.525** &**0.654** &**0.510**\
PTA-CL &0.749 &**0.733** &0.552 &0.530 &0.230 &**0.214** &**0.659** &0.509 &**0.683** &0.489\
PTA-SL &0.700 &0.445 &0.499 &0.051 &**0.247** &0.011 &0.636 &**0.545** &0.635 &**0.545**\
PTGP &**0.755** &**0.738** &**0.568** &**0.551** &**0.239** &**0.220** &**0.647** &**0.527** &**0.664** &**0.535**\
CSPA &0.669 &0.661 &0.481 &0.469 &0.213 &0.199 &N/A &N/A &N/A &N/A\
HGPA &0.584 &0.193 &0.407 &0.017 &0.167 &0.103 &0.311 &0.155 &N/A &N/A\
MCLA &0.699 &0.676 &0.519 &0.488 &0.229 &0.204 &0.622 &0.305 &0.621 &0.044\
SRW &0.469 &0.109 &0.467 &0.112 &0.198 &0.047 &N/A &N/A &N/A &N/A\
GP-MGLA &**0.754** &0.731 &0.560 &0.547 &0.227 &0.189 &0.631 &0.503 &0.623 &0.462\
EAC-AL &0.740 &0.699 &0.550 &0.526 &0.221 &0.194 &0.629 &0.510 &N/A &N/A\
EAC-CL &0.615 &0.382 &0.424 &0.169 &0.207 &0.073 &0.601 &0.504 &N/A &N/A\
EAC-SL &0.367 &0.013 &0.011 &0.001 &0.032 &0.002 &0.525 &0.218 &N/A &N/A\
SRS-AL &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A\
SRS-CL &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A\
SRS-SL &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A\
WCT-AL &0.752 &0.695 &**0.563** &0.533 &0.229 &0.199 &N/A &N/A &N/A &N/A\
WCT-CL &0.694 &0.621 &0.516 &0.476 &0.217 &0.189 &N/A &N/A &N/A &N/A\
WCT-SL &0.575 &0.015 &0.011 &0.001 &0.033 &0.001 &N/A &N/A &N/A &N/A\
ECMC-AL &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A\
ECMC-CL &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A\
ECMC-SL &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A &N/A\
WEAC-AL &0.751 &0.716 &0.561 &**0.548** &0.222 &0.194 &0.629 &0.499 &N/A &N/A\
WEAC-CL &0.606 &0.349 &0.429 &0.146 &0.208 &0.065 &0.610 &0.502 &N/A &N/A\
WEAC-SL &0.609 &0.029 &0.011 &0.001 &0.067 &0.002 &0.562 &0.275 &N/A &N/A\
[ [![The number of times of each method to be ranked in the top 3 (in terms of NMI) across the 20 columns in Table \[table:compare\_ce\].[]{data-label="fig:comp_top1top3"}](Figures/ranktopIII3 "fig:"){width="0.98\columnwidth"}]{}]{}
We run the proposed PTA and PTGP methods and the baseline methods 100 times on each dataset. If a method is computationally infeasible to be performed on a dataset, the corresponding NMI score will be labeled as “N/A”. For each run, the ensemble of base clusterings is randomly drawn from the base clustering pool (see Section \[sec:construct\_base\]). As the number of clusters of the consensus clustering needs to be pre-specified for the baseline methods and the proposed methods, to compare their clustering results in a fair way, we use two criteria to choose cluster numbers in the experiments, namely, best-$k$ and true-$k$. In the best-$k$ criterion, the cluster number that leads to the best performance is specified for each method. In the true-$k$ criterion, the true number of classes of the dataset is specified for each method.
The average performances of the proposed PTA and PTGP methods and the baseline methods over 100 runs are reported in Table \[table:compare\_ce\]. Each pair-wise similarity based method is associated with one of the three agglomerative clustering methods, namely, AL, CL, and SL. As shown in Table \[table:compare\_ce\], PTA-AL achieves the highest NMI scores for the *MF*, *MNIST*, and *USPS* w.r.t. both best-$k$ and true-$k$ and almost the highest NMI scores for the *ODR*, *LS*, *PD*, *FC*, *KDD99-10P*, and *KDD99* datasets. The proposed PTGP method achieves the highest scores for the *ODR* and *LS* datasets w.r.t. both best-$k$ and true-$k$ and almost the highest scores for the *MF*, *IS*, *PD*, *USPS*, *FC*, *KDD99-10P*, and *KDD99* datasets. To compare the performance of the test methods in a clearer way, Fig. \[fig:comp\_top1top3\] shows the number of times of each method being ranked in the top 3 (in terms of NMI) in Table \[table:compare\_ce\]. Out of the 20 columns in Table \[table:compare\_ce\], PTGP and PTA-AL are ranked in the top 3 (among the 24 test methods) 19 times and 18 times, respectively, while the best baseline method is ranked in the top 3 only 4 times (see Fig. \[fig:comp\_top1top3\]). As can be seen in Table \[table:compare\_ce\] and Fig. \[fig:comp\_top1top3\], the proposed PTA and PTGP methods achieve the overall best performance in clustering accuracy and robustness compared to the baseline methods across a variety of datasets.
It is worth mentioning that the PTGP method significantly outperforms the other five graph partitioning based methods, namely, CSPA, HGPA, MCLA, SRW, and GP-MGLA. Specifically, the PTGP method yields higher, or even significantly higher, NMI scores than the other graph partitioning based methods on all of the benchmark datasets (see Table \[table:compare\_ce\]). Also, we compare the PTA method to the other five pair-wise similarity based methods, namely, EAC, SRS, WCT, ECMC, and WEAC, w.r.t. the same agglomerative clustering method. As shown in Table \[table:compare\_ce\], the PTA-AL method achieves the best performance among the pair-wise similarity based methods (all associated with AL) on all of the benchmark datasets. When considering CL or SL, the advantages of PTA become even greater. The PTA-CL method significantly outperforms the other pair-wise similarity based methods associated with CL on all benchmark datasets except *IS*. The PTA-SL method achieves *far* better consensus results than the other pair-wise similarity based methods associated with SL on all of the benchmark datasets. With the ability of handling uncertain links and incorporating global information to construct more accurate local links, the proposed PTA and PTGP methods perform significantly better than the baseline ensemble clustering methods on the benchmark datasets.
Robustness to Ensemble Size $M$ {#sec:ensemble_size}
-------------------------------
In this section, we further evaluate the robustness of the proposed methods with varying ensemble sizes $M$. For each ensemble size $M$, we run the PTA and PTGP methods and the baseline methods 20 times and report their average performances in Fig. \[fig:comp\_Msize\]. Here, all pair-wise similarity based methods are associated with AL. The PTA and PTGP methods produce consistently good results with different ensemble sizes. As shown in Fig. \[fig:comp\_Msize\], the PTA and PTGP methods yield the best or nearly the best performance with varying $M$ for the benchmark datasets. Especially, for the *MF*, *IS*, *PD*, *FC*, *KDD99-10P*, and *KDD99* datasets, the PTA and PTGP methods exhibit significant advantages in the robustness to varying ensemble sizes over the baseline methods. Further, we illustrate the average performances of different approaches over nine datasets, KDD99 not included, in Fig. \[fig:ensize9in1\], which is in fact the average of the first nine sub-figures in Fig. \[fig:comp\_Msize\], i.e., the sub-figures from Fig. \[fig:comp\_Msize1\] to Fig. \[fig:comp\_Msize9\]. Figure \[fig:ensize9in1\] provides an average view to compare our methods and the baseline methods across datasets, which demonstrates the advantage of our methods in the robustness to various datasets and ensemble sizes. In particular, the advantage of the proposed methods becomes even greater when the ensemble size gets larger, e.g., when the ensemble size goes beyond 20 (see Fig. \[fig:ensize9in1\]).
Execution Time {#sec:comp_time}
--------------
In this section, we evaluate the time performances of the proposed PTA and PTGP methods and the baseline methods with varying data sizes. The experiments are conducted on varying subsets of the *KDD99* dataset. The sizes of the subsets range from $0$ to the full size $494,020$. The execution times of the proposed methods and the baseline methods w.r.t. varying data sizes are illustrated in Fig. \[fig:time\_complexity\]. Note that the time lines of PTA and PTGP almost overlap with each other due to their very similar time performances. Besides that, the time lines of EAC and WEAC also nearly overlap with each other. As shown in Fig. \[fig:time\_complexity\], the proposed PTA and PTGP methods exhibit a significant advantage in efficiency over the baseline methods. Especially, the proposed PTGP and PTA methods consume 3.22 seconds and 3.70 seconds, respectively, to process the entire dataset of KDD99 which consists of nearly half a million objects, whereas eight out of the ten baseline methods are not even computationally feasible to process such large-scale datasets.
Conclusion {#sec:conclusion}
==========
In this paper, we propose a novel ensemble clustering approach based on sparse graph representation and probability trajectory analysis. The microclusters are exploited as primitive objects to speedup the computation. We present the ENS strategy to identify uncertain links in a locally adaptive manner and construct a sparse graph with a small number of probably reliable links. It has been shown that the use of a small number of probably reliable links can lead to significantly better clusterings than using all graph links regardless of their reliability. To explore the global structure information in the ensemble, we utilize the random walks driven by a new transition probability matrix that considers the link weights and the node sizes simultaneously. A novel and dense similarity measure termed PTS is derived from the sparse graph $K$-ENG by analyzing the probability trajectories of the random walkers. Based on PTS, we further propose two consensus functions, termed PTA and PTGP, respectively. Extensive experiments have been conducted on ten real-world datasets. The experimental results show that our approach significantly outperforms the state-of-the-art approaches in both clustering accuracy and efficiency.
\[fig:time\_complexity\] 0.2in
Acknowledgments {#acknowledgments .unnumbered}
===============
Acknowledgment {#acknowledgment .unnumbered}
==============
The authors would like to thank the anonymous reviewers for their insightful comments and suggestions which helped enhance this paper significantly. This project was supported by NSFC (61173084 & 61502543), National Science & Technology Pillar Program (No. 2012BAK16B06), Guangdong Natural Science Funds for Distinguished Young Scholar, the GuangZhou Program (No. 201508010032), and the PhD Start-up Fund of Natural Science Foundation of Guangdong Province, China (No. 2014A030310180).
[Dong Huang]{} received his B.S. degree in computer science in 2009 from South China University of Technology, China. He received his M.Sc. degree in computer science in 2011 and his Ph.D. degree in computer science in 2015, both from Sun Yat-sen University, China. He joined South China Agricultural University in 2015 as an Assistant Professor with College of Mathematics and Informatics. His research interests include data mining and pattern recognition. He is a member of the IEEE.
[Jian-Huang Lai]{} received the M.Sc. degree in applied mathematics in 1989 and the Ph.D. degree in mathematics in 1999 from Sun Yat-sen University, China. He joined Sun Yat-sen University in 1989 as an Assistant Professor, where he is currently a Professor with the Department of Automation of School of Information Science and Technology, and Dean of School of Information Science and Technology. His current research interests include the areas of digital image processing, pattern recognition, multimedia communication, wavelet and its applications. He has published more than 100 scientific papers in the international journals and conferences on image processing and pattern recognition, such as IEEE TPAMI, IEEE TKDE, IEEE TNN, IEEE TIP, IEEE TSMC-B, Pattern Recognition, ICCV, CVPR, IJCAI, ICDM and SDM. Prof. Lai serves as a Standing Member of the Image and Graphics Association of China, and also serves as a Standing Director of the Image and Graphics Association of Guangdong. He is a senior member of the IEEE.
[Chang-Dong Wang]{} received his Ph.D. degree in computer science in 2013 from Sun Yat-sen University, China. He is currently an assistant professor at School of Mobile Information Engineering, Sun Yat-sen University. His current research interests include machine learning and pattern recognition, especially focusing on data clustering and its applications. He has published over 30 scientific papers in international journals and conferences such as IEEE TPAMI, IEEE TKDE, IEEE TSMC-C, Pattern Recognition, Knowledge and Information System, Neurocomputing, ICDM and SDM. His ICDM 2010 paper won the Honorable Mention for Best Research Paper Awards. He won 2012 Microsoft Research Fellowship Nomination Award. He was awarded 2015 Chinese Association for Artificial Intelligence (CAAI) Outstanding Dissertation. He is a member of the IEEE.
[^1]: The MATLAB code and experimental data of this work are available at: [https://www.researchgate.net/publication/284259332]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Using the Davies–Helffer–Sjöstrand functional calculus based on almost analytic extensions, we address the following problem: Given self-adjoint operators $S_j$, $j=1,2$, in ${{\mathcal H}}$, and functions $f$ in an appropriate class, for instance, $f \in C_0^{\infty}({{\mathbb{R}}})$, how to control the norm $\|f(S_2) - f(S_1)\|_{{{\mathcal B}}({{\mathcal H}})}$ in terms of the norm of the difference of resolvents, $\big\|(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}$, for some $z_0 \in {{\mathbb{C}}}\backslash{{\mathbb{R}}}$. We are particularly interested in the case where ${{\mathcal B}}({{\mathcal H}})$ is replaced by a trace ideal, ${{\mathcal B}}_p({{\mathcal H}})$, $p \in [1,\infty)$.'
address:
- 'Department of Mathematics, University of Missouri, Columbia, MO 65211, USA'
- 'Mathematics Department, The University of Tennessee at Chattanooga, 415 EMCS Building, Dept. 6956, 615 McCallie Ave, Chattanooga, TN 37403, USA'
author:
- Fritz Gesztesy
- Roger Nichols
title: Some Applications Of Almost Analytic Extensions To Operator Bounds in Trace Ideals
---
[^1] [^2]
Introduction
============
[s1]{} Yurij M. Berezansky’s contributions to analysis in general, and areas such as functional analysis, operator theory, spectral and inverse spectral theory, harmonic analysis, analysis in spaces of functions of an infinite number of variables, stochastic calculus, mathematical physics, quantum field theory, integration of nonlinear evolution equations, in particular, are legendary and of a lasting nature. The list of fields his ground breaking work changed in dramatic fashion can easily be continued in many directions as is demonstrated by the extraordinary breadth revealed in his highly influential monographs [@Be68]–[@BSU96a]. Since operator theoretic methods frequently play a role in his research interests, we hope our modest contribution to operator bounds in trace ideals will create some joy for him.
This paper has its origins in the following question: Given self-adjoint operators $S_j$, $j=1,2$, in ${{\mathcal H}}$, and functions $f$ in an appropriate class, for instance, $f \in C_0^{\infty}({{\mathbb{R}}})$, how to control the norm $\|f(S_2) - f(S_1)\|_{{{\mathcal B}}({{\mathcal H}})}$ in terms of the norm of the difference of resolvents, $\big\|(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}$, for some $z_0 \in {{\mathbb{C}}}\backslash{{\mathbb{R}}}$? In particular, the question is just as natural with ${{\mathcal B}}({{\mathcal H}})$ replaced by a trace ideal, ${{\mathcal B}}_p({{\mathcal H}})$, $p \in [1,\infty)$.
In fact, our interest in these questions stems from computations of the Witten index (a suitable extension of the Fredholm index) for a class of non-Fredholm model operators $${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{} = {\frac}{d}{dt} + {{\boldsymbol{A}}}, \quad
\operatorname{dom}({{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{})= W^{1,2}({{\mathbb{R}}}; {{\mathcal H}}) \cap \operatorname{dom}({{\boldsymbol{A}}}), {\label}{1.1}$$ in the Hilbert space $L^2({{\mathbb{R}}}; {{\mathcal H}})$, where $$\begin{aligned}
&({{\boldsymbol{A}}}f)(t) = A(t) f(t) \, \text{ for a.e.\ $t\in{{\mathbb{R}}}$,} {\notag}\\
& f \in \operatorname{dom}({{\boldsymbol{A}}}) = \bigg\{g \in L^2({{\mathbb{R}}};{{\mathcal H}}) \,\bigg|\,
g(t)\in \operatorname{dom}(A(t)) \text{ for a.e.\ } t\in{{\mathbb{R}}}; {\label}{1.2} \\
& \quad t \mapsto A(t)g(t) \text{ is (weakly) measurable;} \,
\int_{{{\mathbb{R}}}} dt \, \|A(t) g(t)\|_{{{\mathcal H}}}^2 < \infty\bigg\}, {\notag}\end{aligned}$$ with $A(t)$, $t\in {{\mathbb{R}}}$, a family of self-adjoint operators in ${{\mathcal H}}$ with asymptotes $A_{\pm}$ (in norm resolvent sense). Interesting concrete examles for $A_{\pm}$ are given by massless Dirac-type operators in ${{\mathcal H}}= L^2({{\mathbb{R}}}^d)$, $d \in {{\mathbb{N}}}$ (the latter are known to be non-Fredholm), see, [@CGLPSZ14]–[@CGPST14]. More precisely, given the sequence of self-adjoint operators ${{\boldsymbol{H}}}_{j,n}$, ${{\boldsymbol{H}}}_j$ in $L^2({{\mathbb{R}}}; dt; {{\mathcal H}})$, $j=1,2$, $n \in {{\mathbb{N}}}$, and self-adjoint operators $A_{+,n}$, $A_+$, $A_-$ in ${{\mathcal H}}$, $n\in {{\mathbb{N}}}$, and a Pushnitski-type relation between the spectral shift functions $\xi(\, \cdot \,; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})$ and $\xi(\, \cdot \,; A_{+,n}, A_-)$ for the pairs, $({{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})$ and $(A_{+,n}, A_-)$ of the form $$\xi(\lambda; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n}) = \begin{cases}
{\frac}{1}{\pi} \int_{- \lambda^{1/2}}^{\lambda^{1/2}}
{\frac}{\xi(\nu; A_{+,n}, A_-) \, d \nu}{(\lambda - \nu^2)^{1/2}} \,
\text{ for a.e.~$\lambda > 0$,} \\
0, \; \lambda < 0,
\end{cases} \; n\in{{\mathbb{N}}}, {\label}{1.3}$$ we were interested in performing the limit $n \to \infty$ in to obtain the analogous relation for the limiting spectral shift functions $\xi(\, \cdot \,; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1)$ and $\xi(\, \cdot \, ; A_+ , A_-)$ corresponding to the limiting pairs $({{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1)$ and $(A_+, A_-)$, respectively. The latter is instrumental in computing the Witten index for ${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}$. The task of performing the limit $n \to \infty$ in is considerably complicated since due to the nature of the approximations involved, no suitable bounds on $\xi(\, \cdot \,; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})$ and $\xi(\, \cdot \,; A_{+,n}, A_-)$ (independent of $n \in {{\mathbb{N}}}$) are readily available. To circumvent this difficulty one can resort to a distributional approach considering $$\int_{{{\mathbb{R}}}} d \lambda \, \xi(\lambda; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n}) f'(\lambda)
= {\frac}{1}{\pi} \int_{{{\mathbb{R}}}} d\nu \, \xi(\nu; A_{+,n},A_-) F'(\nu), \quad n \in {{\mathbb{N}}}, {\label}{1.4}$$ where $f \in C_0^{\infty}({{\mathbb{R}}})$ is arbitrary, and $F' \in C_0^{\infty}({{\mathbb{R}}})$ is given by $$F'(\nu) = \int_{\nu^2}^{\infty} d\lambda \, f'(\lambda) (\lambda - \nu^2)^{-1/2}, \quad \nu \in {{\mathbb{R}}}.
{\label}{1.5}$$ Focusing now on the left-hand side of , one recalls Krein’s trace formula, $${\operatorname{tr}}_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}; {{\mathcal H}}))} (f({{\boldsymbol{H}}}_{2,n}) - f({{\boldsymbol{H}}}_{1,n})) =
\int_{[0,\infty)} d \lambda \, \xi(\lambda; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n}) f'(\lambda), \quad
f \in C_0^{\infty}({{\mathbb{R}}}). {\label}{1.6}$$ Thus, given control of resolvents in the form $$\begin{aligned}
\begin{split}
& \lim_{n\to\infty} \big\|\big[({{\boldsymbol{H}}}_{2,n} - z \, {{\boldsymbol{I}}})^{-m_2} - ({{\boldsymbol{H}}}_{1,n} - z \, {{\boldsymbol{I}}})^{-m_2}\big] \\
& \hspace*{1cm} - [({{\boldsymbol{H}}}_2 - z \, {{\boldsymbol{I}}})^{-m_2} - ({{\boldsymbol{H}}}_1 - z \, {{\boldsymbol{I}}})^{-m_2}\big]
\big\|_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}; {{\mathcal H}}))} = 0, \quad z \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}. {\label}{1.7}
\end{split} \end{aligned}$$ one can hope to control $$\begin{aligned}
\begin{split}
& \lim_{n\to\infty} \|[f({{\boldsymbol{H}}}_{2,n}) - f({{\boldsymbol{H}}}_{1,n})]
- [f({{\boldsymbol{H}}}_2) - f({{\boldsymbol{H}}}_1)]\|_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}; {{\mathcal H}}))} =0, {\label}{1.8}
\end{split} \end{aligned}$$ and hence obtain $$\begin{aligned}
& \lim_{n \to \infty} \int_{[0,\infty)} d \lambda \, \xi(\lambda; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n}) f'(\lambda)
= \lim_{n \to \infty} {\operatorname{tr}}_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}; {{\mathcal H}}))} (f({{\boldsymbol{H}}}_{2,n}) - f({{\boldsymbol{H}}}_{1,n})) {\notag}\\
& \quad = {\operatorname{tr}}_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}; {{\mathcal H}}))} (f({{\boldsymbol{H}}}_2) - f({{\boldsymbol{H}}}_1))
= \int_{[0,\infty)} d \lambda \, \xi(\lambda; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1) f'(\lambda), \quad
f \in C_0^{\infty}({{\mathbb{R}}}). {\label}{1.9}\end{aligned}$$ Together with controlling the limit $n\to\infty$ on the right-hand side of , this leads to $$\int_{{{\mathbb{R}}}} d \lambda \, \xi(\lambda; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1) f'(\lambda)
= {\frac}{1}{\pi} \int_{{{\mathbb{R}}}} d\nu \, \xi(\nu; A_+,A_-) F'(\nu). {\label}{1.10}$$ Without going into further details we note that in turn can be used to prove the limiting relation in and the latter leads to a compution of the semigroup regularized Witten index, $W_s({{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{})$, of ${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}$: Assuming that $0$ is a left and a right Lebesgue point of $\xi(\, \cdot \,; A_+,A_-)$, denoting the corresponding values by $\xi(0_{\pm}; A_+,A_-)$, the semigroup regularized Witten index is found to be $$\begin{aligned}
W_s({{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}) &:= \lim_{t \uparrow \infty} \operatorname{tr}_{L^2({{\mathbb{R}}};{{\mathcal H}})}
\big(e^{- t {{\boldsymbol{H}}}_1} - e^{-t {{\boldsymbol{H}}}_2}\big) {\notag}\\
&\,= [\xi(0_-; A_+,A_-) + \xi(0_+; A_+,A_-)]/2, \end{aligned}$$ see, for instance, [@CGLPSZ14]–[@CGPST14]. (We here use the semigroup regularized Witten index rather than the resolvent regularised one as the former is applicable in the case of $d$-dimensional Dirac-type operators $A_{\pm}$, $d\in{{\mathbb{N}}}$.) We trust this sufficiently illustrates our interest in using control of resolvents of self-adjoint operators to gain control over their $C_0^{\infty}$-functions.
We also note a further complication lies in the fact that when studying multi-dimensional Dirac-type operators $A_{\pm}$, resolvents alone are not sufficient in the trace class context and hence sufficiently high powers (depending on the space dimension involved) of resolvents have to be employed.
Our principal tool to gain control over $C_0^{\infty}$-functions of $S$ in terms of (powers of) resolvents of $S$ is furnished by a suitable application of almost analytic extensions $\widetilde{f}_{\ell,\sigma}$ of $f$ in the form of a Davies–Helffer–Sjöstrand functional calculus [@Da95a], [@Da95 Ch. 2], [@HS89 Proposition 7.2], of the form $$f(S) = \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
{\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(z) (S - z I_{{{\mathcal H}}})^{-1}, {\label}{1.11}$$ and a refinement due to Khochman [@Kh07] of the type $$f(S) = \frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z)
(z-z_0)^m (S - z_0I_{{{\mathcal H}}})^{-m} (S - zI_{{{\mathcal H}}})^{-1}, \quad m \in {{\mathbb{N}}}\cup \{0\}. {\label}{1.12}$$ to be discussed in some detail in Section \[s2\]. Section \[s3\] contains our principal results and some applications. Finally, Appendix \[sA\] recalls various useful facts concerning (powers of) resolvents.
We conclude with some comments on the notation employed in this paper: Let ${{\mathcal H}}$ be a separable complex Hilbert space, $(\cdot,\cdot)_{{{\mathcal H}}}$ the scalar product in ${{\mathcal H}}$ (linear in the second argument), and $I_{{{\mathcal H}}}$ the identity operator in ${{\mathcal H}}$.
Next, if $T$ is a linear operator mapping (a subspace of) a Hilbert space into another, then $\operatorname{dom}(T)$ and $\ker(T)$ denote the domain and kernel (i.e., null space) of $T$. The spectrum and resolvent set of a closed linear operator in a Hilbert space will be denoted by $\sigma(\cdot)$ and $\rho(\cdot)$, respectively. t
The convergence of bounded operators in the strong operator topology (i.e., pointwise limits) will be denoted by $\operatorname*{s-lim}$.
The Banach spaces of bounded and compact linear operators on a separable complex Hilbert space ${{\mathcal H}}$ are denoted by ${{\mathcal B}}({{\mathcal H}})$ and ${{\mathcal B}}_\infty({{\mathcal H}})$, respectively; the corresponding $\ell^p$-based trace ideals will be denoted by ${{\mathcal B}}_p ({{\mathcal H}})$, their norms are abbreviated by $\|\cdot\|_{{{\mathcal B}}_p({{\mathcal H}})}$, $p \geq 1$. Moreover, $\operatorname{tr}_{{{\mathcal H}}}(A)$ denotes the corresponding trace of a trace class operator $A\in{{\mathcal B}}_1({{\mathcal H}})$.
The symbol $C^{\infty}_0({{\mathbb{R}}})$ represents $C^{\infty}$-functions of compact support on ${{\mathbb{R}}}$; continuous functions on ${{\mathbb{R}}}$ vanishing at infinity are denoted by $C_{\infty}({{\mathbb{R}}})$.
Basic Facts on Almost Analytic Extensions And The Functional Calculus For Self-Adjoint Operators
================================================================================================
[s2]{} In this preparatory section we briefly recall the basics of almost analytic extensions and the ensuing functional calculus for self-adjoint operators, following Davies’ detailed treatment in [@Da95a], [@Da95 Ch. 2].
One introduces the class $S^{\beta}({{\mathbb{R}}})$, $\beta \in {{\mathbb{R}}}$, consisting of all functions $f \in C^{\infty}({{\mathbb{R}}})$ such that $$f^{(m)}(x) \underset{|x|\to\infty}{=} {O}(\big\langle x\rangle^{\beta - m}\big), \quad
m \in {{\mathbb{N}}}_0, {\label}{2.1}$$ where $\langle z \rangle = \big(|z|^2 + 1\big)^{1/2}$, $z\in {{\mathbb{C}}}$. Then in obvious notation, with “$\cdot$” denoting pointwise multiplication, $S^{\beta}({{\mathbb{R}}}) \cdot S^{\gamma}({{\mathbb{R}}}) \subseteq S^{\beta + \gamma}({{\mathbb{R}}})$, $\beta, \gamma \in {{\mathbb{R}}}$, and the space $${{\mathcal A}}({{\mathbb{R}}}) = \bigcup_{\beta < 0} S^{\beta}({{\mathbb{R}}}) {\label}{2.2}$$ is an algebra under pointwise multiplication with $$C_0^{\infty}({{\mathbb{R}}}) \subset {{\mathcal A}}({{\mathbb{R}}}). {\label}{2.3}$$ In particular, $f \in {{\mathcal A}}({{\mathbb{R}}})$ implies $f \in C_{\infty}({{\mathbb{R}}})$ (the continuous functions vanishing at $\pm \infty$) and $f^{(m)} \in L^1({{\mathbb{R}}})$, $m \in {{\mathbb{N}}}$.
Given $f \in {{\mathcal A}}({{\mathbb{R}}})$, one defines an [*almost analytic extension*]{} ${\widetilde}f_{\ell, \sigma}$, of $f$ to ${{\mathbb{C}}}$ by $${\widetilde}f_{\ell,\sigma} (z) = \sigma(x,y) \sum_{k=0}^{\ell} {\frac}{f^{(k)}(x) (iy)^k}{k!}, \quad
z = x + iy \in {{\mathbb{C}}}, \; \ell \in {{\mathbb{N}}}, {\label}{2.4}$$ where $$\sigma (x,y) = \tau(y/\langle x \rangle), \; x, y \in {{\mathbb{R}}}, \quad \tau \in C_0^{\infty}({{\mathbb{R}}}), \;
\tau (s) = \begin{cases} 1, & |s| \leq 1, \\ 0, & |s| \geq 2. \end{cases} {\label}{2.5}$$ The precise structure of ${\widetilde}f_{\ell, \sigma}$ will not be important and other expresssions for it are possible (cf., [@Da95a]).
We note the formula $$\begin{aligned}
{\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(z)
&= {\frac}{1}{2} \bigg({\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial x}(z)
+ i {\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial y}(z)\bigg) {\notag}\\
&= {\frac}{1}{2} [\sigma_x(x,y) + \sigma_y(x,y)] \sum_{k=0}^{\ell} {\frac}{f^{(k)}(x) (iy)^k}{k!}
+ {\frac}{1}{2} \sigma(x,y) {\frac}{f^{\ell + 1}(x) (iy)^{\ell}}{\ell !}, {\label}{2.6} \\
& \hspace*{9.1cm} z \in {{\mathbb{C}}}, {\notag}\end{aligned}$$ implying the crucial fact, $$\bigg|{\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(x + iy) \bigg| \underset{y \downarrow 0}{=}
{O}\big(|y|^{\ell}\big), {\label}{2.7}$$ in particular, $${\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(x) = 0, \quad x \in {{\mathbb{R}}}. {\label}{2.8}$$
Following Helffer and Sjöstrand [@HS89 Proposition 7.2], particularly, in the form presented by Davies [@Da95a], [@Da95 Ch. 2], one then establishes a functional calculus for self-adjoint operators $S$ in a complex, separable Hilbert space ${{\mathcal H}}$ via the formula $$f(S) = \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
{\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(z) (S - z I_{{{\mathcal H}}})^{-1}. {\label}{2.9}$$ Since the integrand is norm continuous, the integral in is norm convergent, in particular, one notes that and , together with the standard estimate $\big\|(S - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} \leq |{\text{\rm Im}}(z)|^{-1}$, overcome the apparent singularity of the integrand in for $z \in \sigma(S) \subseteq {{\mathbb{R}}}$ (cf. also ).
The justification for calling this a functional calculus follows upon proving the following facts:\
$\bullet$ The left-hand side of is independent of the choice of $\ell \in {{\mathbb{N}}}$ and the precise form of $\sigma$ in .\
$\bullet$ If $f \in C_0^{\infty}({{\mathbb{R}}})$ with $\operatorname{supp}(f) \cap \sigma(S) = \emptyset$, then $f(S)=0$.\
$\bullet$ If $f, g \in {{\mathcal A}}({{\mathbb{R}}})$, then $(fg)(S) = f(S) g(S)$, $f(S)^* = {\overline}f(S)$, $\|f(S)\|_{{{\mathcal B}}({{\mathcal H}})} = \|f\|_{L^{\infty}({{\mathbb{R}}})}$.\
$\bullet$ Let $z \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$ and $f_z(x) = (x - z)^{-1}$, then $f_z \in {{\mathcal A}}({{\mathbb{R}}})$ and $f_z(S) = (S - z I_{{{\mathcal H}}})^{-1}$.
In addition, we note that Khochman [@Kh07] proved the following extension of :
\[[@Kh07]\] [l2.1]{} Let $m \in {{\mathbb{N}}}$, $f\in C_0^{\infty}({{\mathbb{R}}})$, and suppose that $S$ is self-adjoint in ${{\mathcal H}}$. Then, $$f(S) = \frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z)
(z-z_0)^m (S - z_0I_{{{\mathcal H}}})^{-m} (S - zI_{{{\mathcal H}}})^{-1}. {\label}{2.10}$$
We will employ (in fact, rederive) in the proof of Theorem \[t3.7\]. Next, we discuss another extension focusing on semigroups rather than powers of resolvents.
[l2.2]{} Let $t > 0$, $f\in C_0^{\infty}({{\mathbb{R}}})$, and suppose that $S$ is self-adjoint and bounded from below in ${{\mathcal H}}$. Then, $$f(S) = \frac{1}{\pi} \int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z) e^{tz} e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}. {\label}{2.11}$$
We start by noting that if $f,g\in C_0^{\infty}({{\mathbb{R}}})$, it is proved in [@Da95 p. 28] that $${\label}{D.17}
\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{(fg)}_{\ell'',\sigma''}}{\partial \overline{z}}(z)
(S - z I_{{{\mathcal H}}})^{-1} = \int_{{{\mathbb{C}}}}dxdy\,
\frac{\partial( \tilde{f}_{\ell',\sigma'}\tilde{g}_{\ell,\sigma})}{\partial \overline{z}}(z)
(S - z I_{{{\mathcal H}}})^{-1}.$$
Next, suppose that $f\in C_0^{\infty}({{\mathbb{R}}})$ and let $E_t \in C_0^{\infty}({{\mathbb{R}}})$ denote a function which coincides with $e^{tx}$ on an open interval $I$ with $\operatorname{supp}(f)\subset I$. Then $$\widetilde{E}_{t, \ell,\sigma}(z) = \sigma(x,y)e^{tx} \sum_{k=0}^{\ell}\frac{(i t y)^k}{k!},
\quad z=x+iy,\; x\in I.$$ Let $f_{\ell',\sigma'}$ denote an almost analytic extension of $f$. Setting $g= f E_t$, with $\widetilde{g}_{\ell'',\sigma''}$ an almost analytic extension of $g$, in light of the identity $f(S) = g(S) e^{- t S}$, one infers from the Davies–Helffer–Sjöstrand functional calculus applied to $g$, $$\begin{aligned}
f(S)&= \frac{1}{\pi}\int_{{{\mathbb{C}}}} dxdy\, \frac{\partial \widetilde{g}_{\ell'',\sigma''}}{\partial \overline{z}}(z) e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}{\notag}\\
&=\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial (\widetilde{f}_{\ell',\sigma'}\widetilde{E}_{t,\ell,\sigma})}{\partial \overline{z}}(z) e^{- t S}(S - z I_{{{\mathcal H}}})^{-1} {\notag}\\
&=\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{f}_{\ell',\sigma'}}{\partial \overline{z}}(z)\widetilde{E}_{t, \ell,\sigma}(z)e^{- t S}(S - z I_{{{\mathcal H}}})^{-1} {\notag}\\
&\quad +\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \widetilde{f}_{\ell',\sigma'}(z)
\frac{\partial \widetilde{E}_{t,\ell,\sigma}}{\partial \overline{z}}(z) e^{- t S}(S - z I_{{{\mathcal H}}})^{-1} {\notag}\\
&=\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy \, \frac{\partial \widetilde{f}_{\ell'',\sigma''}}{\partial \overline{z}}(z)\sigma(x,y)e^{t x} \bigg(\sum_{k=0}^{\ell}\frac{(i t y)^k}{k!}\bigg)e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}{\notag}\\
&\quad +\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \widetilde{f}_{\ell',\sigma'}(z)\Bigg\{\frac{1}{2}\bigg[ \sigma_x(x,y)e^{t x} \bigg(\sum_{k=0}^{\ell}\frac{(i t y)^k}{k!}\bigg) {\notag}\\
& \hspace*{4.05cm} +\sigma(x,y) e^{t x} \bigg(\sum_{k=0}^{\ell}\frac{(i t y)^k}{k!}\bigg)\bigg]{\notag}\\
&\quad +\frac{i}{2}\bigg[ \sigma(x,y)e^{t x} it\bigg(\sum_{k=0}^{\ell-1}\frac{(i t y)^k}{k!}\bigg)+\sigma_y(x,y)e^{t x} \bigg(\sum_{k=0}^{\ell}\frac{(i t y)^k}{k!} \bigg)
\bigg]\Bigg\} {\notag}\\
& \hspace*{6.6cm} \times e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}. {\label}{2.14} \end{aligned}$$ Exploiting the fact that $f(S)$ is independent of $\ell$, we now take the limit $\ell\to \infty$ in . Since $\widetilde{f}_{\ell',\sigma'}$ has compact support and takes care of the singularity of the resolvent, $e^{- t S}$ is bounded and $z$-independent, and the exponential series converges uniformly on compact sets, one may pass the limit under the integral to obtain $$\begin{aligned}
f(S)&=\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{f}_{\ell',\sigma'}}{\partial \overline{z}}\sigma(x,y)e^{tz} e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}{\notag}\\
&\quad +\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \widetilde{f}_{\ell',\sigma'}(z)\frac{\partial(\sigma(x,y)e^{t z})}{\partial \overline{z}}e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}{\notag}\\
&=\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial}{\partial\overline{z}}\big(\widetilde{f}_{\ell',\sigma'}(z)\sigma(x,y)e^{t z} \big)e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}{\notag}\\
&=\frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \frac{\partial}{\partial\overline{z}}\big(\widetilde{f}_{\ell',\widehat{\sigma}}(z)e^{t z} \big)e^{- t S}(S - z I_{{{\mathcal H}}})^{-1},\end{aligned}$$ where $\widehat{\sigma}=\sigma'\sigma$ (which corresponds to choosing $\widehat{\tau}=\tau' \tau$). It is a simple matter to verify that $$\frac{\partial}{\partial\overline{z}}\big(\widetilde{f}_{\ell',\widehat{\sigma}}(z)e^{t z} \big) = \frac{\partial \widetilde{f}_{\ell,\widehat{\sigma}}}{\partial \overline{z}}e^{t z},$$ which then shows $$f(S) = \frac{1}{\pi} \int_{{{\mathbb{C}}}}dxdy\, \frac{\partial \widetilde{f}_{\ell',\widehat{\sigma}}}{\partial \overline{z}}e^{t z} e^{- t S}(S - z I_{{{\mathcal H}}})^{-1}$$ and hence (renaming $\ell'$ and $\widehat \sigma$).
Historically, the idea of almost analytic (resp., pseudo-analytic) extensions appeared in Hörmander [@Ho69] and Dynkin [@Dy75], [@Dy93], Melin and Sjöstrand [@MS75] (see also [@DS99 Ch. 8], [@MP89 Sect. III.6] for expositions and [@JN94] for an alternative approach). The functional calculus was used by Helffer and Sjöstrand in their seminal 1989 paper on Schrödinger operators with magnetic fields [@HS89], which in turn was the basis for the systematic treatment by Davies [@Da95a], [@Da95 Ch. 2]. Since these early developments, there has been a large body of literature in connection with spectral theory for Schrödinger and Dirac-type operators applying this functional calculus. While a complete list of references in this context is clearly beyond the scope of this paper, we want to illustrate the great variety of applications that rely on this functional calculus a bit and hence refer to [@CCO02], [@Di06], [@DD14], [@DP03], [@DP10], [@DS96], [@DZ12], [@FHS10], [@FGM08], [@Ge92], [@Ge08], [@Gr04], [@HS90], [@Kh07], [@ORS13], [@PRS13], [@SZ91], [@Sk92], and the references cited therein.
While we here exclusively focus on linear operators in a Hilbert space, this functional calculus applies to operators in Banach spaces with real spectrum, see, for instance, [@BF03], [@Cl12], [@Da95a], [@Da95b], [@GMP02], [@GP97]. Extensions to the case where the spectrum is contained in the unit circle or contained in finitely-many smooth arcs were also treated in [@Dy75].
Some Applications
=================
[s3]{} In this section we apply the almost analytic extension method and its ensuing functional calculus for self-adjoint operators to derive various norm bounds and convergence properties of operators in trace ideals.
We start with the following estimates established in the proof of [@Da95 Theorem 2.6.2] (more precisely, is proved in [@Da95], but then the rest of Lemma \[l3.1\] is obvious):
[l3.1]{} Let $z_0 \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$, $f \in {{\mathcal A}}({{\mathbb{R}}})$ and suppose that $S_j$, $j=1,2$, are self-adjoint in ${{\mathcal H}}$. Then $$\begin{aligned}
\begin{split}
\| f(S_2) - f(S_1)\|_{{{\mathcal B}}({{\mathcal H}})} & \leq {\frac}{8}{\pi} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg| {\frac}{\big[|z_0|^2 + |z|^2\big]}{|{\text{\rm Im}}(z)|^2} \\
& \quad \times \big\|(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}. {\label}{3.3}
\end{split} \end{aligned}$$ In addition, if for some $p \in [1,\infty)$, $\big[(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big] \in {{\mathcal B}}_p({{\mathcal H}})$ for some $($and hence for all$)$ $z_0 \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$, then $$[f(S_2) - f(S_1)] \in {{\mathcal B}}_p({{\mathcal H}}) {\label}{3.4}$$ and $$\begin{aligned}
\begin{split}
\| f(S_2) - f(S_1)\|_{{{\mathcal B}}_p({{\mathcal H}})} & \leq {\frac}{8}{\pi} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg| {\frac}{\big[|z_0|^2 + |z|^2\big]}{|{\text{\rm Im}}(z)|^2} \\
& \quad \times \big\|(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}_p({{\mathcal H}})}. {\label}{3.5}
\end{split} \end{aligned}$$ If $\big[(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big] \in {{\mathcal B}}_{\infty}({{\mathcal H}})$, the inclusion extends to $p=\infty$.
Combining , , and one obtains, $$f(S_2) - f(S_1) = \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z) \big[(S_2 - z I_{{{\mathcal H}}})^{-1}
- (S_1 - z I_{{{\mathcal H}}})^{-1}\big] {\label}{3.6}$$ and hence $$\begin{aligned}
& \|f(S_2) - f(S_1)\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg| \big\|(S_2 - z I_{{{\mathcal H}}})^{-1}
- (S_1 - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg|
\big\|(S_2 - z_0 I_{{{\mathcal H}}}) (S_2 - z I_{{{\mathcal H}}})^{-1} {\notag}\\
& \qquad \times \big[(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big]
(S_1 - z_0 I_{{{\mathcal H}}}) (S_1 - z I_{{{\mathcal H}}})^{-1} \big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg|
\big\|(S_2 - z_0 I_{{{\mathcal H}}}) (S_2 - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \qquad \times \big\|(S_1 - z_0 I_{{{\mathcal H}}}) (S_1 - z I_{{{\mathcal H}}})^{-1} \big\|_{{{\mathcal B}}({{\mathcal H}})}
\big\|\big[(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big]\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \bigg({\frac}{8}{\pi} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg| {\frac}{\big[|z_0|^2 + |z|^2\big]}{|{\text{\rm Im}}(z)|^2}\bigg)
{\notag}\\
& \qquad \times \big\|\big[(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big]\big\|_{{{\mathcal B}}({{\mathcal H}})}.
{\label}{3.7} \end{aligned}$$ Precisely the same chain of estimates applies to ${{\mathcal B}}({{\mathcal H}})$ replaced by ${{\mathcal B}}_p({{\mathcal H}})$, relying on the ideal properties of ${{\mathcal B}}_p({{\mathcal H}})$, $p \in [1,\infty)$. The case $p=\infty$ in is a consequence of the norm convergent integral on the right-hand side of .
Combined with a Stone–Weierstrass approximation argument and the fact that $\|f(S)\|_{{{\mathcal B}}({{\mathcal H}})} = \|f\|_{L^\infty({{\mathbb{R}}})}$, $f\in {{\mathcal A}}({{\mathbb{R}}})$, Lemma \[l3.1\] yields the following well-known fact, recorded, for instance, in [@Da95 Theorem 2.62], [@RS80 Theorem V.III.20(a)]:
[l3.2]{} Let $S_n$, $n \in {{\mathbb{N}}}$, and $S$ be self-adjoint in ${{\mathcal H}}$, and suppose that $S_n$ converges to $S$ in norm resolvent sense as $n \to \infty$. Then, $$\lim_{n \to \infty} \|f(S_n) - f(S)\|_{{{\mathcal B}}({{\mathcal H}})} = 0 {\label}{3.8}$$ for all $f \in C_{\infty}({{\mathbb{R}}})$.
[r3.3]{} We note that the functional calculus based on almost analytic extensions is not the only possible approach to address estimates such as and . As a powerful alternative we mention the theory of double operator integrals (DOI), which can prove stronger inequalities of the following type (cf. [@BS03], [@CGLPSZ14b], [@Ya05]): Given $m \in {{\mathbb{N}}}$ odd and $p\in [1,\infty)$, there exist constants $a_1,a_2 \in \mathbb{R} \backslash \{0\}$ and $C=C(f,m, a_1, a_2) \in (0,\infty)$ such that $$\begin{aligned}
\begin{split}
& \big\|f(A)-f(B)\|_{{{\mathcal B}}_p({{\mathcal H}})}\leq C\big(\big\|(A-a_1iI_{{{\mathcal H}}})^{-m} - (B-a_1iI_{{{\mathcal H}}})^{-m}\big\|_{{{\mathcal B}}_p({{\mathcal H}})} {\label}{3.9} \\
&\quad +\big\|(A-a_2iI_{{{\mathcal H}}})^{-m} - (B-a_2iI_{{{\mathcal H}}})^{-m}\big\|_{{{\mathcal B}}_p({{\mathcal H}})}\big),
\quad f \in C^{\infty}_0({{\mathbb{R}}}),
\end{split} \end{aligned}$$ which permits the use of differences of higher powers $m \in \mathbb{N}$ of resolvents to control the $\| \cdot \|_{\mathcal{B}_p(\mathcal{H})}$-norm of the left-hand side $[f(A)-f(B)]$ for $f \in C^{\infty}_0({{\mathbb{R}}})$. In fact, this extends to a much larger class of functions $f$, see [@CGLNPS16] for details.
Moreover, repeatedly differentiating $$f(\lambda) = \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
{\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(z) (\lambda - z)^{-1}, \quad
\lambda \in {{\mathbb{R}}}, {\label}{3.10}$$ with respect to $\lambda$ yields $$f^{(m-1)}(S) = \pi^{-1} (-1)^{m-1} (m-1)! \int_{{{\mathbb{C}}}} dx dy \,
{\frac}{\partial {\widetilde}f_{\ell,\sigma}}{\partial {\overline}z}(z) (S - z)^{-m}, \quad
\lambda \in {{\mathbb{R}}}. {\label}{3.11}$$ This leads to estimates of the type but with $f$ replaced by $f^{(m-1)}$. $\diamond$
We recall a useful result:
[l3.4]{} Let $p\in[1,\infty)$ and assume that $R,R_n,T,T_n\in{{\mathcal B}}({{\mathcal H}})$, $n\in{{\mathbb{N}}}$, satisfy $\operatorname*{s-lim}_{n\to\infty}R_n = R$ and $\operatorname*{s-lim}_{n\to \infty}T_n = T$ and that $S,S_n\in{{\mathcal B}}_p({{\mathcal H}})$, $n\in{{\mathbb{N}}}$, satisfy $\lim_{n\to\infty}\|S_n-S\|_{{{\mathcal B}}_p({{\mathcal H}})}=0$. Then $\lim_{n\to\infty}\|R_n S_n T_n^\ast - R S T^\ast\|_{{{\mathcal B}}_p({{\mathcal H}})}=0$.
This follows, for instance, from [@Gr73 Theorem 1], [@Si05 p. 28–29], or [@Ya92 Lemma 6.1.3] with a minor additional effort (taking adjoints, etc.).
Next, we describe a typical convergence result:
[t3.5]{} Let $S_{j,n}$, $n \in {{\mathbb{N}}}$, and $S_j$, $j=1,2$, be self-adjoint in ${{\mathcal H}}$, and assume that $S_{j,n}$ converges in strong resolvent sense as $n \to \infty$ to $S_j$, $j=1,2$, respectively.\
$(i)$ Suppose that for some $($and hence for all$)$ $z_0 \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$, $$\begin{aligned}
\begin{split}
& \lim_{n \to \infty} \big\|\big[(S_{2,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_2 - z_0 I_{{{\mathcal H}}})^{-1}\big] \\
& \hspace*{1cm}
- \big[(S_{1,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big] \big\|_{{{\mathcal B}}({{\mathcal H}})}
=0. {\label}{3.12}
\end{split} \end{aligned}$$ Then, $$\lim_{n \to \infty} \|[f(S_{2,n}) - f(S_{1,n})] - [f(S_2) - f(S_1)]\|_{{{\mathcal B}}({{\mathcal H}})} = 0, \quad
f \in C_0^{\infty}({{\mathbb{R}}}). {\label}{3.13}$$ $(ii)$ Let $p \in [1,\infty)$ and suppose that for some $($and hence for all$)$ $z_0 \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$, $$\begin{aligned}
& \big[(S_{2,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0 I_{{{\mathcal H}}})^{-1}\big],
\big[(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big] \in {{\mathcal B}}_p({{\mathcal H}}), {\notag}\\
& \hspace*{10cm} n \in {{\mathbb{N}}}, {\label}{3.14} \end{aligned}$$ and $$\begin{aligned}
\begin{split}
& \lim_{n \to \infty} \big\|\big[(S_{2,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_2 - z_0 I_{{{\mathcal H}}})^{-1}\big] \\
& \hspace*{1cm}
- \big[(S_{1,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big] \big\|_{{{\mathcal B}}_p({{\mathcal H}})}
=0. {\label}{3.15}
\end{split} \end{aligned}$$ Then, $$\lim_{n \to \infty} \|[f(S_{2,n}) - f(S_{1,n})] - [f(S_2) - f(S_1)]\|_{{{\mathcal B}}_p({{\mathcal H}})} = 0, \quad
f \in C_0^{\infty}({{\mathbb{R}}}). {\label}{3.16}$$
As usual, a combination of identity , Lemma \[l3.4\], and the assumed strong resolvent convergence of $S_{j,n}$ to $S_j$ as $n \to \infty$, $j=1,2$, proves sufficiency of the conditions and for just one $z_0 \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$. Thus, assumption actually implies $$\begin{aligned}
\begin{split}
& \lim_{n \to \infty} \big\|\big[(S_{2,n} - z I_{{{\mathcal H}}})^{-1} - (S_2 - z I_{{{\mathcal H}}})^{-1}\big] \\
& \hspace*{1cm}
- \big[(S_{1,n} - z I_{{{\mathcal H}}})^{-1} - (S_1 - z I_{{{\mathcal H}}})^{-1}\big] \big\|_{{{\mathcal B}}({{\mathcal H}})}
=0, \quad z \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}. {\label}{3.17}
\end{split} \end{aligned}$$
Next, mimicking , one obtains, $$\begin{aligned}
& [f(S_{2,n}) - f(S_{1,n})] - [f(S_2) - f(S_1)] {\notag}\\
& \quad = \pi^{-1} \int_{{{\mathbb{C}}}} dx dy \,
{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z) \Big[ \big[(S_{2,n} - z I_{{{\mathcal H}}})^{-1}
- (S_{1,n} - z I_{{{\mathcal H}}})^{-1}\big] {\label}{3.18} \\
& \hspace*{4.2cm} - \big[(S_2 - z I_{{{\mathcal H}}})^{-1} - (S_1 - z I_{{{\mathcal H}}})^{-1}\big]\Big],
\quad f \in C_0^{\infty}({{\mathbb{R}}}), {\notag}\end{aligned}$$ and hence, $$\begin{aligned}
& \lim_{n \to \infty} \|[f(S_{2,n}) - f(S_{1,n})] - [f(S_2) - f(S_1)] \|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \pi^{-1} \lim_{n \to \infty} \int_{{{\mathbb{C}}}} dx dy \,
\bigg|{\frac}{\partial {\widetilde}f_{2,\sigma}}{\partial {\overline}z}(z)\bigg| \Big\|\big[(S_{2,n} - z I_{{{\mathcal H}}})^{-1}
- (S_{1,n} - z I_{{{\mathcal H}}})^{-1}\big] {\label}{3.19} \\
& \hspace*{3cm} - \big[(S_2 - z I_{{{\mathcal H}}})^{-1} - (S_1 - z I_{{{\mathcal H}}})^{-1}\big]\Big\|_{{{\mathcal B}}({{\mathcal H}})},
\quad f \in C_0^{\infty}({{\mathbb{R}}}). {\notag}\end{aligned}$$ In this context one observes that $f \in C_0^{\infty}({{\mathbb{R}}})$ implies ${\widetilde}f_{2,\sigma} \in C_0^{\infty}({{\mathbb{C}}})$. Since the exceptional set ${{\mathbb{R}}}$ in has $dxdy$-measure zero (in addition to the fact that by , $\partial {\widetilde}f_{2,\sigma}/ \partial {\overline}z$ vanishes on ${{\mathbb{R}}}$), an application of the Lebesgue dominated convergence theorem to interchange the limit $n \to \infty$ with the integral on the right-hand side of requires establishing an $n$-independent integrable majorant of the integrand in . Employing identity and estimate , this majorant can be obtained as follows: $$\begin{aligned}
& \big\|\big[(S_{2,n} - z I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z I_{{{\mathcal H}}})^{-1}\big]
- \big[(S_2 - z I_{{{\mathcal H}}})^{-1} - (S_1 - z I_{{{\mathcal H}}})^{-1}\big]\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \big\|(S_{2,n} - z I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \qquad + \big\|(S_2 - z I_{{{\mathcal H}}})^{-1} - (S_1 - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq \big\|(S_{2,n} - z_0 I_{{{\mathcal H}}})(S_{2,n} - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \qquad \quad \times \big\|(S_{2,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}
{\notag}\\
& \qquad \quad \times \big\|(S_{1,n} - z_0 I_{{{\mathcal H}}})(S_{1,n} - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \qquad + \big\|(S_2 - z_0 I_{{{\mathcal H}}})(S_2 - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \qquad \quad \times \big\|(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}
{\notag}\\
& \qquad \quad \times \big\|(S_1 - z_0 I_{{{\mathcal H}}})(S_1 - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \quad \leq 8 \big[|z_0|^2 + |z|^2\big] |{\text{\rm Im}}(z)|^{-2}
\Big[\big\|(S_{2,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\notag}\\
& \hspace*{4.5cm} + \big\|(S_2 - z_0 I_{{{\mathcal H}}})^{-1} - (S_1 - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}\Big]
{\notag}\\
& \quad \leq 8 \big[|z_0|^2 + |z|^2\big] |{\text{\rm Im}}(z)|^{-2} C(z_0), {\label}{3.20} \end{aligned}$$ for some $0 < C(z_0) < \infty$, independent of $n \in {{\mathbb{N}}}$, since by assumption , $$\big\|(S_{2,n} - z_0 I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0 I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} \leq {\widetilde}C(z_0)
{\label}{3.21}$$ for some $0 < {\widetilde}C(z_0) < \infty$, independent of $n \in {{\mathbb{N}}}$. Together with the properties , of $\partial {\widetilde}f_{2,\sigma}/ \partial {\overline}z$, this establishes the sought integrable majorant, independent of $n \in {{\mathbb{N}}}$, and thus permits the interchange of the limit $n \to \infty$ with the integral on the right-hand side of . This completes the proof of .
The proof of proceeds exactly along the same lines employing once more the ideal properties of ${{\mathcal B}}_p({{\mathcal H}})$, $p \in [1,\infty)$.
We remark in passing that double operator integral techniques permit one to enlarge the class of functions $f$ to which Theorem \[t3.5\] applies (cf. also Remark \[r3.8\]).
[r3.5a]{} The proof to Theorem \[t3.5\] uses dominated convergence and given , the crucial observation is that $${\label}{D.1}
\int_{{{\mathbb{C}}}}dxdy\, \bigg|\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z)
\bigg|\frac{|z_0|^2+|z|^2}{|{\text{\rm Im}}(z)|^2}<\infty,$$ which is obvious if $f\in C_0^{\infty}({{\mathbb{R}}})$, since then $\widetilde{f}_{\ell,\sigma}$ is compactly supported. However, it is possible to once again prove if $f\in S^{\beta}({{\mathbb{R}}})$ for some $\beta<-1$. Indeed, following [@Da95 p. 25], and setting $$U = \{(x,y)\, |\, \langle x \rangle < |y| < 2\langle x \rangle\},\quad V = \{(x,y)\, |\, 0 < |y| < 2\langle x \rangle\},$$ one infers $$|\sigma_x + i \sigma_y|\leq c \langle x \rangle^{-1}\chi_U(x,y),\quad z=x+iy\in {{\mathbb{C}}},$$ for some constant $c>0$. Then for $f\in S^{\beta}({{\mathbb{R}}})$, $$\begin{aligned}
\begin{split}
\bigg|\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z)\bigg|
&\leq C\Bigg\{\sum_{k=0}^{\ell}\langle x \rangle ^{\beta-k-1}|y|^k \chi_U(x,y)\Bigg\} + C\langle x\rangle^{\beta-\ell-1}|y|^{\ell}\chi_V(x,y), \\
&\hspace*{6.4cm}\quad z=x+iy\in {{\mathbb{C}}},
\end{split} \end{aligned}$$ where $C>0$ is an appropriate constant. Therefore, $$\begin{aligned}
\bigg|\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z) \bigg|
\frac{|z_0|^2+|z|^2}{|{\text{\rm Im}}(z)|^2}&\leq C\Bigg\{\sum_{k=0}^{\ell}\langle x \rangle ^{\beta-k-1}|y|^k \chi_U(x,y) \frac{|z_0|^2+|x|^2+|y|^2}{|y|^2}\Bigg\} {\notag}\\
&\quad + C\langle x\rangle^{\beta-\ell-1}|y|^{\ell}\chi_V(x,y) \frac{|z_0|^2+|x|^2+|y|^2}{|y|^2}{\notag}\\
&\leq C2^{\ell}\Bigg\{\sum_{k=0}^{\ell}\langle x \rangle ^{\beta-1} \chi_U(x,y) \frac{|z_0|^2+5\langle x\rangle^2}{\langle x\rangle^2}\Bigg\} {\notag}\\
&\quad + C\langle x\rangle^{\beta-\ell-1}|y|^{\ell-2}\chi_V(x,y) \big\{|z_0|^2+5\langle x\rangle^2\big\}{\notag}\\
&\leq C2^{\ell}\Bigg\{\sum_{k=0}^{\ell}\langle x \rangle ^{\beta-3} \chi_U(x,y) \big\{|z_0|^2+5\langle x\rangle^2\big\}\Bigg\} {\notag}\\
&\quad + C2^{\ell-2}\langle x\rangle^{\beta-3}\chi_V(x,y) \big\{|z_0|^2+5\langle x\rangle^2\big\}{\notag}\\
&\leq \widehat{C}\langle x\rangle^{\beta-1}\big\{\chi_U(x,y) + \chi_V(x,y) \big\},\quad z=x+iy\in {{\mathbb{C}}}\backslash {{\mathbb{R}}},{\label}{D.5}\end{aligned}$$ where $\widehat{C} =\widehat{C}(z_0) >0$ is a constant. Here, we used $|z_0|^2+5\langle x\rangle^2< \widetilde{C}\langle x\rangle^2$, for an appropriate constant $\widetilde{C}=\widetilde{C}(z_0)>0$. Given , holds if $\beta<-1$ since $$\begin{aligned}
&\int_{{{\mathbb{C}}}}dxdy\, \langle x\rangle^{\beta-1}\big\{\chi_U(x,y) + \chi_V(x,y) \big\} {\notag}\\
&\quad = \int_{-\infty}^{\infty}dx\, \langle x\rangle^{\beta-1}\, \int_{-\infty}^{\infty}dy\, \big\{\chi_U(x,y) + \chi_V(x,y) \big\}{\notag}\\
&\quad = 2 \int_{-\infty}^{\infty}dx\, \langle x\rangle^{\beta-1}\, \bigg\{ \int_0^{2\langle x\rangle}dy + \int_{\langle x\rangle}^{2\langle x\rangle}dy\bigg\}{\notag}\\
&\quad = 6\int_{-\infty}^{\infty}dx\, \langle x\rangle^{\beta}<\infty.\end{aligned}$$ Thus, the majorant is integrable as long as $f\in S^{\beta}({{\mathbb{R}}})$ for some $\beta<-1$ and hence Theorem \[t3.5\] extends from $C_0^{\infty}({{\mathbb{R}}})$ to $S^{\beta}({{\mathbb{R}}})$, $\beta<-1$.
Next, we briefly apply Theorem \[t3.5\] to a concrete $(1+1)$-dimensional example treated in great detail in [@CGLPSZ14] and [@CGLPSZ14a] by alternative methods.
[e3.7]{} Assuming the real-valued functions $\phi, \theta$ satisfy $$\begin{aligned}
& \phi \in AC_{{\operatorname{loc}}}({{\mathbb{R}}}) \cap L^{\infty}({{\mathbb{R}}}) \cap L^1({{\mathbb{R}}}),
\; \phi' \in L^{\infty}({{\mathbb{R}}}), \\
\begin{split}
& \theta \in AC_{{\operatorname{loc}}}({{\mathbb{R}}}) \cap L^{\infty}({{\mathbb{R}}}), \;
\theta' \in L^{\infty}({{\mathbb{R}}}) \cap L^1({{\mathbb{R}}}), \\
& \lim_{t \to \infty} \theta (t) = 1, \; \lim_{t \to - \infty} \theta (t) = 0,
\end{split} \end{aligned}$$ we introduce the family of self-adjoint operators $A(t)$, $t \in {{\mathbb{R}}}$, in $L^2({{\mathbb{R}}})$, $$A(t) = - i {\frac}{d}{dx} + \theta(t) \phi, \quad
\operatorname{dom}(A(t)) = W^{1,2}({{\mathbb{R}}}), \; t \in {{\mathbb{R}}}, {\label}{9.3}$$ and its self-adjoint asymptotes as $t \to \pm \infty$, $$A_+ = - i {\frac}{d}{dx} + \phi, \quad A_- = - i {\frac}{d}{dx}, \quad
\operatorname{dom}(A_{\pm}) = W^{1,2}({{\mathbb{R}}}).$$ In addition, we introducing the operator $d/dt$ in $L^2\big({{\mathbb{R}}}; dt; L^2({{\mathbb{R}}};dx)\big)$ by $$\begin{aligned}
& \bigg({\frac}{d}{dt}f\bigg)(t) = f'(t) \, \text{ for a.e.\ $t\in{{\mathbb{R}}}$,} {\notag}\\
& \, f \in \operatorname{dom}(d/dt) = \big\{g \in L^2\big({{\mathbb{R}}};dt;L^2({{\mathbb{R}}})\big) \, \big|\,
g \in AC_{{\operatorname{loc}}}\big({{\mathbb{R}}}; L^2({{\mathbb{R}}})\big), \\
& \hspace*{6.3cm} g' \in L^2\big({{\mathbb{R}}};dt;L^2({{\mathbb{R}}})\big)\big\} {\notag}\\
& \hspace*{2.3cm} = W^{1,2} \big({{\mathbb{R}}}; dt; L^2({{\mathbb{R}}}; dx)\big). \label{2.ddtR} \end{aligned}$$ Next, we agree to identify $L^2\big({{\mathbb{R}}}; dt; L^2({{\mathbb{R}}}; dx)\big)$ with $L^2({{\mathbb{R}}}^2; dt dx)$ $($denoting the latter by $L^2({{\mathbb{R}}}^2)$ for brevity$)$ and introduce ${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}$ in $L^2({{\mathbb{R}}}^2)$ by $${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{} = {\frac}{d}{dt} + {{\boldsymbol{A}}},
\quad \operatorname{dom}({{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{})= W^{1,2}({{\mathbb{R}}}^2),$$ with ${{\boldsymbol{A}}}$ defined as in and $A(t)$, $t \in {{\mathbb{R}}}$, given by . Moreover, we introduce the nonnegative, self-adjoint operators ${{\boldsymbol{H}}}_j$, $j=1,2$, in $L^2({{\mathbb{R}}}^2)$ by $${{\boldsymbol{H}}}_1 = {{\boldsymbol{D}}}_{{{\boldsymbol{A}}}}^{*} {{\boldsymbol{D}}}_{{{\boldsymbol{A}}}}^{}, \quad
{{\boldsymbol{H}}}_2 = {{\boldsymbol{D}}}_{{{\boldsymbol{A}}}}^{} {{\boldsymbol{D}}}_{{{\boldsymbol{A}}}}^{*}.$$
As shown in [@CGLPSZ14], the assumptions on $\phi$ and $\theta$ guarantee that $$\big[(A_+ - z I)^{-1} - (A_- - z I)^{-1}\big] \in {{\mathcal B}}_1\big(L^2({{\mathbb{R}}})\big),
\quad z \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}{\label}{9.13}$$ $($for simplicity, we adopt the abbreviation $I = I_{L^2({{\mathbb{R}}})}$ throughout this example$)$, and thus, the spectral shift function $\xi(\, \cdot \, ; A_+, A_-)$ for the pair $(A_+, A_-)$ exists and is well-defined up to an arbitrary additive real constant, satisfying $$\xi(\, \cdot \, ; A_+, A_-) \in L^1\big({{\mathbb{R}}}; (\nu^2 + 1)^{-1} d\nu\big).$$ Introducing $\chi_n(A_-) = n (A_-^2 + n^2 I)^{-1/2}$ and $A_{+,n} = A_- + \chi_n(A_-) \phi \chi_n(A_-)$, $n \in {{\mathbb{N}}}$, the fact $$A_{+,n} - A_- = \chi_n(A_-) \phi \chi_n(A_-) \in {{\mathcal B}}_1\big(L^2({{\mathbb{R}}})\big), \quad n \in {{\mathbb{N}}},$$ implies that also the spectral shift functions $\xi(\, \cdot \, ; A_{+,n}, A_-)$, $n \in {{\mathbb{N}}}$, exist and are uniquely determined by $$\xi(\, \cdot \, ; A_{+,n}, A_-) \in L^1({{\mathbb{R}}}; d\nu), \quad n \in {{\mathbb{N}}}.$$ In fact, as has been shown in [@CGLPSZ14], the open constant in $\xi(\, \cdot \, ; A_+, A_-)$ can naturally be determined via the limiting procedure $$\lim_{n \to \infty} \xi(\nu; A_{+,n}, A_-) = {\frac}{1}{2 \pi} \int_{{{\mathbb{R}}}} dx \, \phi(x)
= \xi(\nu; A_+, A_-), \quad \nu \in {{\mathbb{R}}}. {\label}{9.14}$$ In particular, $\xi(\, \cdot \, ; A_+, A_-)$ turns out to be constant in this example.
Replacing $A(t)$ by $A_n(t) = A_- + \chi_n(A_-) \theta(t) \phi \chi_n(A_-)$, $n \in {{\mathbb{N}}}$, $t \in {{\mathbb{R}}}$, and hence, ${{\boldsymbol{A}}}$ by ${{\boldsymbol{A}}}_n$, ${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}$ by ${{\boldsymbol{D}}}_{{{\boldsymbol{A}}}^{}_n}$, ${{\boldsymbol{H}}}_j$ by ${{\boldsymbol{H}}}_{j,n}$, $j=1,2$, $n \in{{\mathbb{N}}}$, one verifies the facts, $$\begin{aligned}
& \big[({{\boldsymbol{H}}}_2 - z \, {{\boldsymbol{I}}})^{-1} - ({{\boldsymbol{H}}}_1 - z \, {{\boldsymbol{I}}})^{-1}\big] \in
{{\mathcal B}}_1\big(L^2({{\mathbb{R}}}^2)\big), \quad z \in {{\mathbb{C}}}\backslash [0,\infty), \\
& \big[({{\boldsymbol{H}}}_{2,n}-z \, {{\boldsymbol{I}}})^{-1} - ({{\boldsymbol{H}}}_{1,n}-z \, {{\boldsymbol{I}}})^{-1}\big] \in {{\mathcal B}}_1\big(L^2({{\mathbb{R}}}^2)\big),
\quad n \in {{\mathbb{N}}}, \; z \in {{\mathbb{C}}}\backslash [0,\infty), \end{aligned}$$ showing that the spectral shift functions $\xi(\, \cdot \, ; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1)$ and $\xi(\, \cdot \, ; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})$ for the pairs $({{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1)$ and $({{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1)$, $n \in {{\mathbb{N}}}$, respectively, are well-defined and satisfy $$\xi(\, \cdot \, ; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1), \, \xi(\, \cdot \, ; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})
\in L^1\big({{\mathbb{R}}}; (\lambda^2 + 1)^{-1} d\lambda\big), \quad n \in {{\mathbb{N}}}.$$ Since ${{\boldsymbol{H}}}_j\geq 0$, ${{\boldsymbol{H}}}_{j,n} \geq 0$, $n \in {{\mathbb{N}}}$, $j=1,2$, one uniquely introduces $\xi(\,\cdot\,; {{\boldsymbol{H}}}_2,{{\boldsymbol{H}}}_1)$ and $\xi(\, \cdot \, ; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})$, $n\in{{\mathbb{N}}}$, by requiring that $$\xi(\lambda; {{\boldsymbol{H}}}_2,{{\boldsymbol{H}}}_1) = 0, \quad
\xi(\, \cdot \, ; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n}) = 0, \quad \lambda < 0, \; n \in {{\mathbb{N}}}.$$ As shown in [@CGLPSZ14], one can now prove the following intimate connection between $\xi(\, \cdot \,; A_{+,n}, A_-)$ and $\xi(\, \cdot \, ; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n})$, $n \in {{\mathbb{N}}}$, the Pushnitski-type formula [@GLMST11], [@Pu08], $$\xi(\lambda; {{\boldsymbol{H}}}_{2,n}, {{\boldsymbol{H}}}_{1,n}) = {\frac}{1}{\pi} \int_{- \lambda^{1/2}}^{\lambda^{1/2}}
{\frac}{\xi(\nu; A_{+,n}, A_-) d \nu}{(\lambda - \nu^2)^{1/2}} \, \text{ for a.e.~$\lambda > 0$,
$n\in{{\mathbb{N}}}$.} {\label}{9.22}$$ Moreover, as shown in [@CGLPSZ14] and [@CGLPSZ14a], one indeed has the convergence property $$\begin{aligned}
\begin{split}
& \lim_{n\to\infty} \big\|\big[({{\boldsymbol{H}}}_{2,n} - z \, {{\boldsymbol{I}}})^{-1} - ({{\boldsymbol{H}}}_{1,n} - z \, {{\boldsymbol{I}}})^{-1}\big] \\
& \hspace*{1cm} - [({{\boldsymbol{H}}}_2 - z \, {{\boldsymbol{I}}})^{-1} - ({{\boldsymbol{H}}}_1 - z \, {{\boldsymbol{I}}})^{-1}\big]
\big\|_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}^2))} = 0, \quad z \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}. {\label}{9.22a}
\end{split} \end{aligned}$$ Thus, Theorem \[t3.5\] applies and hence yields, $$\lim_{n \to \infty} \|[f({{\boldsymbol{H}}}_{2,n}) - f({{\boldsymbol{H}}}_{1,n})] - [f({{\boldsymbol{H}}}_2) - f({{\boldsymbol{H}}}_1)]\|_{{{\mathcal B}}_1(L^2({{\mathbb{R}}}^2))}
= 0, \quad f \in C_0^{\infty}({{\mathbb{R}}}). {\label}{9.22b}$$ This, in turn permits one to take the limit $n \to \infty$ in , implying $$\xi(\lambda; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1) = {\frac}{1}{\pi} \int_{- \lambda^{1/2}}^{\lambda^{1/2}}
{\frac}{\xi(\nu; A_+, A_-) d \nu}{(\lambda - \nu^2)^{1/2}} \, \text{ for a.e.~$\lambda > 0$,
$n\in{{\mathbb{N}}}$.} {\label}{9.22c}$$ Equation combined with yields $$\xi(\lambda; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1) = \xi (\nu; A_+, A_-) = {\frac}{1}{2 \pi} \int_{{{\mathbb{R}}}} dx \, \phi(x) {\label}{9.23}$$ for a.e. $\lambda > 0$ and a.e. $\nu \in {{\mathbb{R}}}$. As a consequence of , the semigroup regularized Witten index $W_s({{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{})$ of the non-Fredholm operator ${{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}$ exists and equals $$W_s({{\boldsymbol{D}}}_{{\boldsymbol{A}}}^{}) = \xi(0_+; {{\boldsymbol{H}}}_2, {{\boldsymbol{H}}}_1) =
\xi(0; A_+, A_-) = {\frac}{1}{2 \pi} \int_{{{\mathbb{R}}}} dx \, \phi(x). {\label}{9.24}$$ This yields an alternative proof of the principal Witten index results in [@CGLPSZ14] and [@CGPST14].
The following result provides an extension of Theorem \[t3.5\] to higher powers of resolvents (necessitated by applications to $d$-dimensional Dirac-type operators as hinted at in the introduction).
[t3.7]{} Let $S_{j,n}$, $n \in {{\mathbb{N}}}$, and $S_j$, $j=1,2$, be self-adjoint in ${{\mathcal H}}$, and assume that $S_{j,n}$ converges in strong resolvent sense as $n \to \infty$ to $S_j$, $j=1,2$, respectively. Suppose that for some $m\in {{\mathbb{N}}}$ and some $p\in [1,\infty)$, $$\begin{aligned}
\big[(S_{2,n}-zI_{{{\mathcal H}}})^{-m} - (S_{1,n}-zI_{{{\mathcal H}}})^{-m}\big], \big[(S_2-zI_{{{\mathcal H}}})^{-m} - (S_1-zI_{{{\mathcal H}}})^{-m}\big]\in&{{\mathcal B}}_p({{\mathcal H}}),{\notag}\\
z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\, n\in {{\mathbb{N}}}.& {\label}{3.22}\end{aligned}$$ If $$\begin{aligned}
&\lim_{n\rightarrow \infty}\big\|\big[(S_{2,n}-zI_{{{\mathcal H}}})^{-m} - (S_{1,n}-zI_{{{\mathcal H}}})^{-m}\big] {\notag}\\
&\qquad - \big[(S_2-zI_{{{\mathcal H}}})^{-m} - (S_1-zI_{{{\mathcal H}}})^{-m}\big]\big\|_{{{\mathcal B}}_p({{\mathcal H}})}=0,\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}}, {\label}{3.23}\end{aligned}$$ and for some $z_0\in {{\mathbb{C}}}\backslash {{\mathbb{R}}}$, $$\begin{aligned}
&(S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-1} \big],{\notag}\\
&\quad (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[(S_2 - z_0I_{{{\mathcal H}}})^{-1} - (S_1 - z_0)^{-1} \big]\in {{\mathcal B}}_p({{\mathcal H}}),\quad n\in {{\mathbb{N}}}, {\label}{3.23a} \end{aligned}$$ with $$\begin{aligned}
\lim_{n\rightarrow \infty}&\big\| (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-1}
- (S_{1,n} - z_0I_{{{\mathcal H}}})^{-1} \big] {\notag}\\
&\quad - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[(S_2 - z_0I_{{{\mathcal H}}})^{-1} - (S_1 - z_0)^{-1} \big]\big\|_{{{\mathcal B}}_p({{\mathcal H}})} = 0,{\label}{3.24}\end{aligned}$$ then $${\label}{3.25}
\lim_{n \to \infty} \|[f(S_{2,n}) - f(S_{1,n})] - [f(S_2) - f(S_1)]\|_{{{\mathcal B}}_p({{\mathcal H}})} = 0, \quad
f \in C_0^{\infty}({{\mathbb{R}}}).$$ In addition, these results hold upon systematically replacing ${{\mathcal B}}_p({{\mathcal H}})$ by ${{\mathcal B}}({{\mathcal H}})$ in –.
Let $f\in C_0^{\infty}({{\mathbb{R}}})$ be fixed and $\widetilde{f}_{\ell,\sigma}\in C_0^{\infty}({{\mathbb{R}}}^2)$ a compactly supported almost analytic extension. Following a device due to Khochman [@Kh07], one introduces $${\label}{3.26}
g(x) = f(x)(x-z_0)^m,$$ concluding $g\in C_0^{\infty}({{\mathbb{R}}})$. By the Davies–Helffer–Sjöstrand functional calculus applied to $g$, one obtains for any self-adjoint operator $S$ in ${{\mathcal H}}$, $${\label}{3.27}
g(S) = \frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\,
\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z)(z-z_0)^m (S-zI_{{{\mathcal H}}})^{-1},$$ and hence, $$\begin{aligned}
f(S) &= (S - z_0I_{{{\mathcal H}}})^{-m}g(S) {\notag}\\
&= \frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\,
\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z) (z-z_0)^m
(S - z_0I_{{{\mathcal H}}})^{-m} (S - zI_{{{\mathcal H}}})^{-1}.{\label}{3.28}\end{aligned}$$ Applying multiple times choosing $H\in \{S_2,S_1,S_{2,n},S_{1,n}\}$, one infers $$\begin{aligned}
&\big\| \big[f(S_{2,n}) - f(S_{1,n}) \big] - \big[f(S_2) - f(S_1) \big]\big\|_{{{\mathcal B}}_p({{\mathcal H}})}{\notag}\\
&\quad \leq \frac{1}{\pi}\int_{{{\mathbb{C}}}}dxdy\, \bigg|\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z) (z-z_0)^m\bigg|{\notag}\\
&\qquad \times \big\|\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{1,n} - zI_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\qquad \quad - \big[(S_2 - z_0I_{{{\mathcal H}}})^{-m}(S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}(S_1 - zI_{{{\mathcal H}}})^{-1} \big] \big\|_{{{\mathcal B}}_p({{\mathcal H}})},{\notag}\\
&\hspace*{10cm} n\in {{\mathbb{N}}}.{\label}{3.29}\end{aligned}$$ In order to prove the convergence claim in , the idea is to take the limit $n\to \infty$ and apply dominated convergence in . However, doing so requires one to obtain an $n$-independent integrable majorant for the expression under the integral in and then to show that the integrand converges to zero pointwise with respect to $z$ as $n\to \infty$. In order to carry this out, one expresses the difference in the $\|\,\cdot\,\|_{{{\mathcal B}}_p({{\mathcal H}})}$-norm in as follows: $$\begin{aligned}
&\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{1,n} - zI_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\qquad - \big[(S_2 - z_0I_{{{\mathcal H}}})^{-m}(S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}(S_1 - zI_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\quad = (S_{2,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{2,n} - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad + (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}(S_{1,n} - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad - \big[ (S_2 - z_0I_{{{\mathcal H}}})^{-m}(S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}(S_2 - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad \quad + (S_1 - z_0I_{{{\mathcal H}}})^{-m}(S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}(S_1 - zI_{{{\mathcal H}}})^{-1}\big]{\notag}\\
&\quad = \big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-m} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m} \big](S_{2,n} - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad + (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - zI_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\qquad -\big\{ \big[ (S_2 - z_0I_{{{\mathcal H}}})^{-m} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big](S_2 - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad \quad + (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[ (S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - zI_{{{\mathcal H}}})^{-1}\big] \big\}{\notag}\\
&\quad = \big\{\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-m} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m} \big](S_{2,n} - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad \quad - \big[ (S_2 - z_0I_{{{\mathcal H}}})^{-m} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big](S_2 - zI_{{{\mathcal H}}})^{-1}\big\}{\notag}\\
&\qquad + \big\{(S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - zI_{{{\mathcal H}}})^{-1}
{\label}{3.30} \\
&\qquad \quad - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[ (S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - zI_{{{\mathcal H}}})^{-1}\big]\big\},\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\; n\in {{\mathbb{N}}}. {\notag}\end{aligned}$$ For the first term in braces after the final equality in , one has a bound of the type $$\begin{aligned}
&\big\|\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-m} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m} \big](S_{2,n} - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad - \big[ (S_2 - z_0I_{{{\mathcal H}}})^{-m} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big](S_2 - zI_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}_p({{\mathcal H}})}{\notag}\\
&\quad \leq \widetilde{C}(z_0)|{\text{\rm Im}}(z)|^{-1},\quad z\in {{\mathbb{C}}}\backslash {{\mathbb{R}}},\, n\in {{\mathbb{N}}},{\label}{3.31}\end{aligned}$$ for a constant $\widetilde{C}(z_0)>0$ which does not depend on $n\in {{\mathbb{N}}}$ or $z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}}$. The estimate in follows at once from the triangle inequality, basic properties of the Schatten–von Neumann trace ideals, the standard resolvent estimate $${\label}{3.32}
\|(S - zI_{{{\mathcal H}}})^{-1}\|_{{{\mathcal B}}({{\mathcal H}})}\leq |{\text{\rm Im}}(z)|^{-1},\quad z\in {{\mathbb{C}}}\backslash {{\mathbb{R}}},$$ for an arbitrary self-adjoint operator $S$ in ${{\mathcal H}}$, and assumption . Moreover, by Lemma \[3.4\], one also infers $$\begin{aligned}
&\lim_{n\to \infty}\big\|\big[(S_{2,n} - z_0I_{{{\mathcal H}}})^{-m} - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m} \big](S_{2,n} - zI_{{{\mathcal H}}})^{-1}{\notag}\\
&\qquad \quad - \big[ (S_2 - z_0I_{{{\mathcal H}}})^{-m} - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big](S_2 - zI_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}_p({{\mathcal H}})} =0,\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}}.{\label}{3.33}
\end{aligned}$$ For the second term in braces after the final equality in , $$\begin{aligned}
&(S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - zI_{{{\mathcal H}}})^{-1} {\notag}\\
&\qquad \quad - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[ (S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - zI_{{{\mathcal H}}})^{-1}\big]{\notag}\\
&\quad = (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[(S_1 - zI_{{{\mathcal H}}})^{-1} - (S_2 - zI_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\qquad - (S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{1,n} - zI_{{{\mathcal H}}})^{-1} - (S_{2,n} - zI_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\quad = \{I_{{{\mathcal H}}} + (z-z_0)(S_1 - zI_{{{\mathcal H}}})^{-1} \}(S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[(S_1 - z_0I_{{{\mathcal H}}})^{-1} - (S_2 - z_0I_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\qquad \quad \times \{I_{{{\mathcal H}}} + (z-z_0)(S_2 - zI_{{{\mathcal H}}})^{-1} \} - \{I_{{{\mathcal H}}} + (z-z_0)(S_{1,n} - zI_{{{\mathcal H}}})^{-1} \}{\notag}\\
&\qquad \qquad \times(S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{1,n} - z_0I_{{{\mathcal H}}})^{-1} - (S_{2,n} - z_0I_{{{\mathcal H}}})^{-1} \big]{\notag}\\
&\qquad \qquad \times \{I_{{{\mathcal H}}} + (z-z_0)(S_{2,n} - zI_{{{\mathcal H}}})^{-1} \},\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\, n\in {{\mathbb{N}}}.{\label}{3.34}\end{aligned}$$ Using , one finds for any self-adjoint operator $H$, $$\begin{aligned}
\|I_{{{\mathcal H}}} + (z_0 - z)(H - zI_{{{\mathcal H}}})^{-1}\|_{{{\mathcal B}}({{\mathcal H}})} &\leq 1 + (|z_0| + |z|)|{\text{\rm Im}}(z)|^{-1}{\notag}\\
&\leq 2 (|z_0| + |z|)|{\text{\rm Im}}(z)|^{-1},\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}}.{\label}{3.35}\end{aligned}$$ As a result of , and , $$\begin{aligned}
&\big\|(S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - zI_{{{\mathcal H}}})^{-1} {\notag}\\
&\qquad - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[ (S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - zI_{{{\mathcal H}}})^{-1}\big]\big\|_{{{\mathcal B}}_p({{\mathcal H}})}{\notag}\\
&\leq \widehat{C}(z_0)\frac{|z_0|^2 + |z|^2}{|{\text{\rm Im}}(z)|^2},\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\, n\in {{\mathbb{N}}},{\label}{3.36}\end{aligned}$$ for a constant $\widehat{C}(z_0)>0$ which does not depend on $n\in {{\mathbb{N}}}$ or $z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}}$. Moreover, and another application of Lemma \[l3.4\] immediately imply $$\begin{aligned}
&\lim_{n\to \infty}\big\|(S_{1,n} - z_0I_{{{\mathcal H}}})^{-m}\big[(S_{2,n} - zI_{{{\mathcal H}}})^{-1} - (S_{1,n} - zI_{{{\mathcal H}}})^{-1} {\notag}\\
&\qquad \quad - (S_1 - z_0I_{{{\mathcal H}}})^{-m}\big[ (S_2 - zI_{{{\mathcal H}}})^{-1} - (S_1 - zI_{{{\mathcal H}}})^{-1}\big]\big\|_{{{\mathcal B}}_p({{\mathcal H}})}=0, \quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}}.{\label}{3.38}\end{aligned}$$ The estimates in and show that away from ${{\mathbb{R}}}$, which has $dxdy$-measure equal to zero, the integrand in is bounded above by $$\begin{aligned}
\big|z-z_0\big|^m\bigg|\frac{\partial \widetilde{f}_{\ell,\sigma}}{\partial \overline{z}}(z)\bigg| \bigg(\frac{\widetilde{C}(z_0)}{|{\text{\rm Im}}(z)|} + \widehat{C}(z_0)\frac{[|z_0|^2 + |z|^2]}{|{\text{\rm Im}}(z)|^2} \bigg),\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\end{aligned}$$ which is integrable with $\ell=2$ in light of and the fact that $\widetilde{f}_{\ell,\sigma}$ is compactly supported. Therefore, taking the limit $n\to \infty$ on both sides of and then applying dominated convergence in combination with , , and yields .
Clearly, the proof remains valid with ${{\mathcal B}}_p({{\mathcal H}})$ replaced by ${{\mathcal B}}({{\mathcal H}})$.
[r3.8]{} Although applicable to the Witten index computation described in the introduction, Theorem \[t3.7\] is far from optimal. Indeed, upon communicating Theorem \[t3.7\] to G. Levitina, D. Potapov, and F. Sukochev, they subsequently pointed out to us [@LPS15] that an application of the double operator integral method permits one to extend the classes of functions $f$ to the one employed in [@Ya05], and more importantly, the DOI approach permits one to dispense with the conditions and altogether. (Conditions and are clearly an artifact of the resolvent term $(S - zI_{{{\mathcal H}}})^{-1}$ in formula ). This will be further pursued elsewhere [@CGLNPS16].
[r3.6]{} While we exclusively focused on applications to self-adjoint operators $S$, as long as $\sigma(T)\subset {{\mathbb{R}}}$ and the singularity of the resolvent $(T - z I_{{{\mathcal H}}})^{-1}$ of $T$ as $z$ approaches the spectrum is uniformly bounded by $|{\text{\rm Im}}(z)|^{-N}$ for some fixed $N \in {{\mathbb{N}}}$, choosing $\ell \in {{\mathbb{N}}}$ sufficiently large in ${\widetilde}f_{\ell,\sigma}$, one can handle such classes of non-self-adjoint operators $T$, particularly, operators in Banach spaces with real spectrum. In fact, a functional calculus for the case of a non-self-adjoint operator $T$ with $\sigma(T) \subset {{\mathbb{R}}}$ and a resolvent that satisfies an estimate of the type $$\|(T - z I_{{{\mathcal H}}})^{-1}\|_{{{\mathcal B}}({{\mathcal H}})} \leq c|{\text{\rm Im}}(z)|^{-1}\bigg(\frac{\langle z\rangle}{|{\text{\rm Im}}(z)|} \bigg)^{\alpha},\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},$$ for some $c>0$ and $\alpha\geq 0$, was discussed in [@Da95a], [@Da95b], and in subsequent developments in [@BF03], [@Cl12], [@GMP02], [@GP97]. Moreover, the case where the spectrum is contained in the unit circle or contained in finitely-many smooth arcs was discussed in [@Dy75]. $\diamond$
Some Useful Resolvent Identities
================================
[sA]{}
In this appendix we recall some well-known, yet useful relations for (powers of) resolvents.
We start by recalling the well-known identity (see, e.g., [@We80 p. 178]), $$\begin{aligned}
& (T_2 - z I_{{{\mathcal H}}})^{-1} - (T_1 - z I_{{{\mathcal H}}})^{-1} = (T_2 - z_0 I_{{{\mathcal H}}})(T_2 - z I_{{{\mathcal H}}})^{-1} {\notag}\\
& \quad \times \big[(T_2 - z_0 I_{{{\mathcal H}}})^{-1} - (T_1 - z_0 I_{{{\mathcal H}}})^{-1}\big]
(T_1 - z_0 I_{{{\mathcal H}}})(T_1 - z I_{{{\mathcal H}}})^{-1}, {\label}{A.-1} \\
& \hspace*{6.6cm} z, z_0 \in \rho(T_1) \cap \rho(T_2), {\notag}\end{aligned}$$ where $T_j$, $j=1,2$, are linear operators in ${{\mathcal H}}$ with $\rho(T_1) \cap \rho(T_2) \neq \emptyset$. In addition, if $S$ is self-adjoint in ${{\mathcal H}}$, we recall the elementary estimate, $$\begin{aligned}
\begin{split}
& \big\|(S - z_0 I_{{{\mathcal H}}})(S - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})}
= \big\| I_{{{\mathcal H}}} + (z - z_0) (S - z I_{{{\mathcal H}}})^{-1}\big\|_{{{\mathcal B}}({{\mathcal H}})} {\label}{A.0} \\
& \quad \leq 8^{1/2} \big[|z_0|^2 + |z|^2\big]^{1/2} |{\text{\rm Im}}(z)|^{-1}, \quad z, z_0 \in {{\mathbb{C}}}\backslash {{\mathbb{R}}}.
\end{split} \end{aligned}$$
In addition, for $m\in {{\mathbb{N}}}$, we note (cf. [@Ya92 p. 315]), $$\begin{aligned}
&(T_2 - zI_{{{\mathcal H}}})^{-(m+1)} - (T_1 - zI_{{{\mathcal H}}})^{-(m+1)}{\notag}\\
&\quad = \big[(T_2 - zI_{{{\mathcal H}}})^{-m} - (T_1 - zI_{{{\mathcal H}}})^{-m} \big](T_1 - zI_{{{\mathcal H}}})^{-1}{\label}{A.1}\\
&\qquad + (T_2 - zI_{{{\mathcal H}}})^{-m}\big[(T_2 - zI_{{{\mathcal H}}})^{-1} - (T_1 - zI_{{{\mathcal H}}})^{-1} \big],\quad z\in \rho(T_1) \cap \rho(T_2), {\notag}\end{aligned}$$ and $$\begin{aligned}
&(T_2 - zI_{{{\mathcal H}}})^{-(m+1)} - (T_1 - zI_{{{\mathcal H}}})^{-(m+1)}{\notag}\\
&\quad = (T_2 - zI_{{{\mathcal H}}})^{-1}\big[(T_2 - zI_{{{\mathcal H}}})^{-m} - (T_1 - zI_{{{\mathcal H}}})^{-m} \big]{\label}{A.2}\\
&\qquad + \big[(T_2 - zI_{{{\mathcal H}}})^{-1} - (T_1 - zI_{{{\mathcal H}}})^{-1} \big](T_2 - zI_{{{\mathcal H}}})^{-m}{\notag}\\
&\qquad - \big[(T_2 - zI_{{{\mathcal H}}})^{-1} - (T_1 - zI_{{{\mathcal H}}})^{-1} \big]\big[(T_2 - zI_{{{\mathcal H}}})^{-m}
- (T_1 - zI_{{{\mathcal H}}})^{-m} \big],{\notag}\\
&\hspace*{8.31cm} z\in \rho(T_1) \cap \rho(T_2).{\notag}\end{aligned}$$
Next, by applying Cauchy’s integral formula, $${\label}{A.3}
f^{(k)}(z) = -\frac{k!}{2\pi i} \ointctrclockwise_{\Gamma}d\zeta \, \frac{f(\zeta)}{(\zeta - z)^{k+1}},
\quad z \in \Omega,$$ where $f$ is an analytic function in the open set $\Omega \subset {{\mathbb{C}}}$, and $\Gamma$ is a counterclockwise-oriented contour encompassing the point $z\in \Omega$, to a densely defined, closed linear operator $T$ in ${{\mathcal H}}$ with nonempty resolvent set, one obtains a formula for higher powers of the resolvent of $T$ in terms of a fixed lower power as follows, $$\begin{aligned}
\begin{split}
(T - zI_{{{\mathcal H}}})^{-k} = -\frac{(k-m)!(m-1)!}{2\pi i[(k-1)!]}
\ointctrclockwise_{\Gamma_z}d\zeta\, (\zeta - z)^{m-k-1}(T - \zeta I_{{{\mathcal H}}})^{-m},&{\label}{A.4}\\
z\in \rho(H_0),&
\end{split} \end{aligned}$$ where for each $z\in \rho(T)$, $\Gamma_z$ is any counterclockwise-oriented circular contour centered at $z$ which does not intersect or encompass points of $\sigma(T)$.
The following lemma (cf. [@Ya92 p. 210]) states an elementary, yet useful, fact:
[l3.8]{} Let $S_j$, $j\in \{1,2\}$ be self-adjoint operators in ${{\mathcal H}}$. If $${\label}{3.40}
\big[(S_2 - zI_{{{\mathcal H}}})^{-m} - (S_1 - zI_{{{\mathcal H}}})^{-m}\big]\in {{\mathcal B}}_p({{\mathcal H}}),\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},$$ for some $p\in [1,\infty) \cup \{\infty\}$ and some $m\in {{\mathbb{N}}}$, then $${\label}{3.41}
\big[(S_2 - zI_{{{\mathcal H}}})^{-n} - (S_1 - zI_{{{\mathcal H}}})^{-n}\big]\in {{\mathcal B}}_p({{\mathcal H}}),\quad z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\, n\geq m.$$
It suffices to apply the Cauchy-type formula and note that $$\begin{aligned}
&(S_2 - zI_{{{\mathcal H}}})^{-n} - (S_1 - zI_{{{\mathcal H}}})^{-n} {\notag}\\
&\quad = -\frac{(n-m)!(m-1)!}{2\pi i[(n-1)!]}
\ointctrclockwise_{\Gamma_z}d\zeta\, (\zeta - z)^{m-n-1} \big[(S_2 - \zeta I_{{{\mathcal H}}})^{-m}
- (S_1 - zI_{{{\mathcal H}}})^{-m}\big],{\notag}\\
&\hspace*{8.5cm} z\in {{\mathbb{C}}}\backslash{{\mathbb{R}}},\, n\geq m,{\label}{3.42}\end{aligned}$$ where $\Gamma_z$ is a counterclockwise-oriented circular contour centered at $z$ that does not intersect ${{\mathbb{R}}}$.
[**Acknowledgments.**]{} We are indebted to Alan Carey, Galina Levitina, Denis Potapov, Fedor Sukochev, Yuri Tomilov, and Dmitriy Zanin for helpful discussions, and particularly to Yuri Tomilov for pointing out to us a number of key references in connection with almost analytic extensions.
[99]{} [BF03]{} A. Bátkai, and E. Fačanga, [*The spectral mapping theorem for Davies’ functional calculus*]{}. Rev. Roumaine Math. Pures Appl. [**48**]{}, 365–372 (2003). [Be68]{} Ju. M. Berezanski[ĭ]{}, [*Expansions in Eigenfunctions of Selfadjoint Operators*]{}, Translations of Mathematical Monographs, Vol. 17, Amer. Math. Soc., Providence, R.I. 1968. [Be86]{} Yu. M. Berezanski[ĭ]{}, [*Selfadjoint Operators in Spaces of Functions of Infinitely Many Variables*]{}, Translations of Mathematical Monographs, Vol. 63, Amer. Math. Soc., Providence, R.I. 1986. [BK98]{} Yu. M. Berezansky and A. A. Kalyuzhnyi, [*Harmonic Analysis in Hypercomplex Systems*]{}, Mathematics and its Applications, 434, Kluwer, Dordrecht, 1998. [BK95]{} Y. M. Berezansky and Y. G. Kondtratiev, [*Spectral methods in infinite-dimensional analysis. Vol. 1*]{}, Mathematical Physics and Applied Mathematics, 12/1, Kluwer Academic, Dordrecht, 1995. [BK95a]{} Y. M. Berezansky and Y. G. Kondtratiev, [*Spectral methods in infinite-dimensional analysis. Vol. 2*]{}, Mathematical Physics and Applied Mathematics, 12/2, Kluwer Academic, Dordrecht, 1995. [BSU96]{} Y. M. Berezansky, Z. G. Sheftel, and G. F. Us, [*Functional Analysis. Vol. I*]{}, Operator Theory: Advances and Applications, Vol. 85, Birkhäuser, Basel, 1996. [BSU96a]{} Y. M. Berezansky, Z. G. Sheftel, and G. F. Us, [*Functional Analysis. Vol. II*]{}, Operator Theory: Advances and Applications, Vol. 86, Birkhäuser, Basel, 1996. [BS03]{} M. Sh. Birman and M. Solomyak, [*Double operator integrals in a Hilbert space*]{}, Integral Eq. Oper. Theory [**47**]{}, 131–168 (2003). [CGLPSZ14]{} A. Carey, F. Gesztesy, G. Levitina, D. Potapov, F. Sukochev, and D. Zanin, [*On index theory for non-Fredholm operators: a $(1+1)$-dimensional example*]{}, [[arXiv:arXiv:1509.01356](http://arxiv.org/abs/arXiv:1509.01356)]{}, Math. Nachrichten, to appear. [CGLPSZ14a]{} A. Carey, F. Gesztesy, G. Levitina, D. Potapov, F. Sukochev, and D. Zanin, [*Trace formulas for a $(1+1)$-dimensional model operator*]{}, preprint, 2014. [CGLPSZ14b]{} A. Carey, F. Gesztesy, G. Levitina, D. Potapov, F. Sukochev, and D. Zanin, [*On index theory for non-Fredholm operators: a $(2+1)$-dimensional example*]{}, in preparation. [CGLNPS16]{} A. Carey, F. Gesztesy, G. Levitina, R. Nichols, D. Potapov, and F. Sukochev, [*Double operator integral methods applied to continuity of spectral shift functions*]{}, [[arXiv:1511.07998](http://arxiv.org/abs/1511.07998)]{}. [CGLS14]{} A. Carey, F. Gesztesy, G. Levitina, and F. Sukochev, [*On the index of a non-Fredholm model operator*]{}, [[arXiv:1509.01580](http://arxiv.org/abs/1509.01580)]{}, Operators and Matrices, to appear. [CGPST14]{} A. Carey, F. Gesztesy, D. Potapov, F. Sukochev, and Y. Tomilov, [*On the Witten index in terms of spectral shift functions*]{}, [[arXiv:1404.0740](http://arxiv.org/abs/1404.0740)]{}, J. Analyse Math., to appear. [CCO02]{} G. Carron, T. Coulhon, and E. M. Ouhabaz, [*Gaussian estimates and $L^p$-boundedness of Riesz means*]{}. J. Evol. Equ. [**2**]{}, 299–317 (2002). [Cl12]{} N. S. Claire, [*Spectral mapping theorem for the Davies–Helffer–Sjöstrand functional calculus*]{}, J. Math. Phys. Anal. Geom. [**8**]{}, 221–239 (2012). [Da95a]{} E. B. Davies, [*The functional calculus*]{}, J. London Math. Soc. (2) [**52**]{}, 166–176 (1995). [Da95b]{} E. B. Davies, [*$L^p$ spectral independence and $L^1$ analyticity*]{}, J. London Math. Soc. (2) [**52**]{}, 177–184 (1995). [Da95]{} E. B. Davies, [*Spectral Theory and Differential Operators*]{}, Cambridge University Press, Cambridge, 1995. [Di06]{} M. Dimassi, [*Spectral shift function in the large coupling constant limit*]{}, Ann. H. Poincaré [**7**]{}, 513–525 (2006). [DD14]{} M. Dimassi and A. T. Duong, [*Trace asymptotics formula for the Schrödinger operators with constant magnetic fields*]{}, J. Math. Anal. Appl. [**416**]{}, 427–448, (2014). [DP03]{} M. Dimassi and V. Petkov, [*Spectral shift function and resonances for non-semi-bounded and Stark Hamiltonians*]{}, J. Math. Pures Appl. [**82**]{}, 1303–1342 (2003). [DP10]{} M. Dimassi and V. Petkov, [*Spectral shift function for operators with crossed magnetic and electric fields*]{}, Rev. Math. Phys. [**22**]{}, 355–380 (2010). [DS96]{} M. Dimassi and J. Sjöstrand, [*Trace asymptotics via almost analytic extensions*]{}, in [*Partial Differential Equations and Mathematical Physics*]{}, The Danish-Swedish Analysis Seminar, 1995, L. Hörmander and A. Melin (eds.), Birkhäuser, Basel, 1996, pp. 126–142. [DS99]{} M. Dimassi and J. Sjöstrand, [*Spectral Asymptotics in the Semi-Classical Limit*]{}, London Math. Soc. Lecture Note Series, Vol. 268, Cambridge University Press, Cambridge, 1999. [DZ12]{} M. Dimassi and M. Zerzeri, [*A tme-dependent approach for the study of spectral shift function*]{}, C. R. Acad. Sci. Paris Ser. I [**350**]{}, 375–378 (2012). [Dy75]{} E. M. Dy’nkin, [*An operator calculus based upon the Cauchy–Green formula*]{}, J. Sov. Math. [**4**]{}, 329–334 (1975). [Dy93]{} E. M. Dy’nkin, [*The pseudoanalytic extension*]{}, J. Analyse Math. [**60**]{}, 45–70 (1993). [FHS10]{} R. Froese, D. Hasler, and W. Spitzer, [*On the AC spectrum of one-dimensional random Schrödinger operators with matrix-valued potentials*]{}, Math. Phys. Anal. Geom. [**13**]{}, 219–233 (2010). [FGM08]{} J. Fröhlich, M. Griesemer, and I. M. Sigal, [*Spectral theory for the standard model of non-relativistic QED*]{}, Comm. Math. Phys. [**283**]{}, 613–646 (2008). [GMP02]{} J. E. Galé, P. J. Miana, and T. Pytlik, [*Spectral properties and norm estimates associated to the $C_c^{(k)}$-functional calculus*]{}, J. Operator Theory [**48**]{}, 385–418 (2002). [GP97]{} J. E. Galé and T. Pytlik, [*Functional calculus for infinitesimal generators of holomorphic semigroups*]{}, J. Funct. Anal. [**150**]{}, 307–355 (1997). [Ge92]{} C. Gerard, [*Sharp propagation estimates for $N$-particle systems*]{}, Duke Math. J. [**67**]{}, 483–515 (1992). [Ge08]{} C. Gérard, [*A proof of the abstract limiting absorption principle by energy estimates*]{}, J. Funct. Anal. [**254**]{}, 2707–2724 (2008). [GLMST11]{} F. Gesztesy, Y. Latushkin, K. A. Makarov, F. Sukochev, and Y. Tomilov, [*The index formula and the spectral shift function for relatively trace class perturbations*]{}, Adv. Math. [**227**]{}, 319–420 (2011). [Gr04]{} M. Griesemer, [*Exponential decay and ionization thresholds in non-relativistic quantum electrodynamics*]{}, J. Funct. Anal. [**210**]{}, 321–340 (2004). [Gr73]{} H. R. Grümm, [Two theorems about ${{\mathcal C}}_p$]{}, Rep. Math. Phys. [**4**]{}, 211–215 (1973). [Ho69]{} L. Hörmander, [*Fourier Integral Operators: Lectures at the Nordic Summer School of Mathematics*]{}, 1969. [HS89]{} B. Helffer and J. Sjöstrand, [*Equation de Schrödinger avec champ magnétique et équaton de Harper*]{}, in [*Schrödinger Operators*]{}, H. Holden and A. Jensen (eds.), Lecture Notes in Physics, Vol. 345, Springer, Berlin, 1989, pp. 138–197. [HS90]{} B. Helffer and J. Sjöstrand, [*On diamagnetism and de Haas–van Alphen effect*]{}, Ann. H. Poincaré [**A52**]{}, 303–375 (1990). [JN94]{} A. Jensen and S. Nakamura, [*Mapping properties of functions of Schrödinger operators between $L^p$-spaces and Besov spaces*]{}, in [*Spectral and Scattering Theory and Applications*]{}, K. Yajima (ed.), Adv. Studies in Pure Math., Vol. 23, Math. Soc. Japan, Kinokuniya Company, Tokyo, Japan, 1994, pp. 187–209. [Kh07]{} A. Khochman, [*Resonances and spectral shift function for the semi-classical Dirac operator*]{}, Rev. Math. Phys. [**19**]{}, 1071–1115 (2007). [LPS15]{} G. Levitina, D. Potapov, and F. Sukochev, private communication, January 2015. [MP89]{} M. Martin and M. Putinar, [*Lectures on Hyponormal Operators*]{}, Operator Theory: Advances and Applications, Vol. 39, Birkhäuser, Basel, 1989. [MS75]{} A. Melin and J. Sjöstrand, [*Fourier integral operators with complex-valued phase functions*]{}, in [*Fourier Integral Operators and Partial Differential Equations*]{}, J. Chazarain (ed.), Lecture Notes in Math., Vol. 459, Springer, Berlin, 1975, pp. 120–223. [ORS13]{} S. O’Rourke, D. Renfrew, and A. Soshnikov, [*On fluctuations of matrix entries of regular functions of Wigner matrices with non-identically distributed entries*]{}, J. Theoret. Probab. [**26**]{}, 750–780 (2013). [PRS13]{} A. Pizzo, D. Renfrew, and A. Soshnikov, [*On finite rank deformations of Wigner matrices*]{}. Ann. Inst. Henri Poincaré Probab. Stat. [**49**]{}, 64–94 (2013). [Pu08]{} A. Pushnitski, [*The spectral flow, the Fredholm index, and the spectral shift function*]{}, in [*Spectral Theory of Differential Operators: M. Sh. Birman 80th Anniversary Collection*]{}, T. Suslina and D. Yafaev (eds.), AMS Translations, Ser. 2, Advances in the Mathematical Sciences, Vol. 225, Amer. Math. Soc., Providence, RI, 2008, pp. 141–155. [RS80]{} M. Reed and B. Simon, [*Methods of Modern Mathematical Physics. I: Functional Analysis*]{}, revised and enlarged edition, Academic Press, New York, 1980. [Si05]{} B. Simon, [*Trace Ideals and Their Applications*]{}, Mathematical Surveys and Monographs, Vol. 120, 2nd ed., Amer. Math. Soc., Providence, RI, 2005. [SZ91]{} J. Sjöstrand and M. Zworski, [*Complex scaling and the distribution of scattering poles*]{}, J. Amer. Math. Soc. [**4**]{}, 729–769 (1991). [Sk92]{} E. Skibsted, [*Smoothness of $N$-body scattering amplitudes*]{}, Rev. Math. Phys. [**4**]{}, 619–658 (1992). [We80]{} J. Weidmann, [*Linear Operators in Hilbert Spaces*]{}, Graduate Texts in Mathematics, Vol. 68, Springer, New York, 1980. [Ya92]{} D. R. Yafaev, [*Mathematical Scattering Theory. General Theory*]{}, Amer. Math. Soc., Providence, RI, 1992. [Ya05]{} D. R. Yafaev, [*A trace formula for the Dirac operator*]{}, Bull. London Math. Soc. [**37**]{}, 908–918 (2005).
[^1]: R.N. gratefully acknowledges support from an AMS–Simons Travel Grant.
[^2]: Appeared in [*Methods Funct. Anal. Topology*]{}, [**21**]{}, 151–169 (2015).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Parallel supercomputer-based Monte Carlo applications depend on pseudorandom number generators that produce independent pseudorandom streams across many separate processes. We propose a new scalable class of parallel pseudorandom number generators based on Pohlig–Hellman exponentiation ciphers. The method generates uniformly distributed floating point pseudorandom streams by encrypting simple sequences of integer *messages* into *ciphertexts* by exponentiation modulo prime numbers. The advantages of the method are: the method is trivially parallelizable by parameterization with each pseudorandom number generator derived from an independent prime modulus, the method is fully scalable on massively parallel computing clusters due to the large number of primes available for each implementation, the seeding and initialization of the independent streams is simple, the method requires only a few integer multiply–mod operations per pseudorandom number, the state of each instance is defined by only a few integer values, the period of each instance is different, and the method passes a battery of intrastream and interstream correlation tests using up to $10^{13}$ pseudorandom numbers per test. The 32-bit implementation we propose has millions of possible instances, all with periods greater than $10^{18}$. A 64-bit implementation depends on 128-bit arithmetic, but would have more than $10^{15}$ possible instances and periods greater than $10^{37}$.'
author:
- 'Paul D. Beale'
title: 'A new class of scalable parallel pseudorandom number generators based on Pohlig–Hellman exponentiation ciphers'
---
\[sec:Introduction\]Introduction
================================
We propose a new class of scalable parallel pseudorandom number generators for use in large-scale Monte Carlo and other computer applications that require large numbers of independent streams of pseudorandom numbers. The method we propose is based on Pohlig–Hellman exponentiation ciphers.[@PohligHellman1978; @Schneier1994] The method creates a pseudorandom stream by encrypting a simple sequence of short integer plaintext *messages* $m_k$ into *ciphertexts* $c_k$ using the transformation $$\begin{aligned}
\label{eqn:encrypt2}
c_k&= m_k^e \thinspace\textrm{mod}\thinspace n ,\end{aligned}$$ where each generator instance is based on an independent prime modulus $n$, and an exponent $e$ that is co-prime to $n-1$. Here and throughout, $x=y \thinspace\textrm{mod}\thinspace z$ means $x$ is the remainder of $y$ upon division by $z$, with $0 \leq x < z$. The method is based upon elementary number theory.[@Koshy2002; @Silverman2006; @Koblitz1987] For every prime number $n$, the set of integers $\mathbb{Z}/n\mathbb{Z} = [0 \thinspace . \thinspace . \thinspace n-1]$ forms a finite field that is closed under addition and multiplication modulo $n$. Also, the set of nonzero elements $(\mathbb{Z}/n\mathbb{Z})^* = [1 \thinspace . \thinspace . \thinspace n-1]$ forms a group that is closed under multiplication, and for every element $a$ in $(\mathbb{Z}/n\mathbb{Z})^*$ there exists a unique inverse $a^{-1}$ such that $a a^{-1}\thinspace\textrm{mod}\thinspace n = 1$ . The pseudorandom number generator algorithm described here cycles through a sequence of messages $m_k$ with $k=0,1,2,\ldots$ selected from $\mathbb{Z}/n\mathbb{Z}$ using some simple pattern that uniformly samples the set. The exponentiation step equation then gives a sequence of ciphertexts $c_k$ that is uniformly distributed on $\mathbb{Z}/n\mathbb{Z}$. Uniformly distributed double precision floating point values $R_k$ on the open real interval $(0 \thinspace . \thinspace . \thinspace 1)$ are formed with a floating point division: $R_k=(c_k+1) / (n+1)$.
Most pseudorandom number generators generate the next pseudorandom integer from either the previous pseudorandom integer in the sequence, or by using two or more pseudorandom integers from earlier in the sequence. This method is different in that the pseudorandom sequence arises from the encryption of a sequence of integer messages. In this way, it is similar to cryptographically secure pseudorandom number generators[@Schneier1994; @FergusonSchneierKohno2010] and floating point pseudorandom number generators[@NumericalRecipes1992] based on block ciphers. The quality of the pseudorandom sequence from our method is based on modular exponentiation being a good one-way cryptographic function, meaning that it is computationally difficult to ascertain the message from the ciphertext without knowing both the modulus and exponent.[@Schneier1994; @FergusonSchneierKohno2010; @Koblitz1987; @BlumMichali; @PatelSundaram] The algorithm we propose here produces excellent, long-period pseudorandom sequences even for 32-bit prime moduli. The period of the 32-bit implementation we propose is greater than $10^{18}$, and has millions of possible instances. Sixty-four bit implementations have periods in excess of $10^{37}$, with more than $10^{15}$ possible instances.
Two qualitatively different schemes have been used to create scalable systems of pseudorandom number generators for use on massively parallel supercomputers: stream splitting and parameter splitting.[@BaukeMertens2007] Stream splitting is based on a single pseudorandom number generator with an extremely long period, with parallelization accomplished by subdividing the full period into independent non-overlapping subsequences. This method requires careful seeding to ensure that no two subsequences will overlap for any feasible set of processes. Parameter splitting uses a single algorithm, but produces independent pseudorandom sequences by assigning different parameters to each process. The SPRNG web site[@MascagniSrinivasan2000; @sprng] gives examples of parallel generators of both classes. For example, linear congruential generators of the form $s_k= ( a s_{k-1} + b ) \thinspace\textrm{mod}\thinspace p$, with $p$ a power of two, can be parallelized by parameter splitting.[@lecuyer1999; @MascagniSrinivasan2000; @sprng; @mascagni1998] The parameters are $a$, a well-tested multiplier that ensures a maximal period and $b$, an odd number less than $p$. Each process uses the same values of $p$ and $a$, but the value of $b$ is chosen to be different for each process.[@poweroftwocongruentialweaknesses] Another widely used class of parallel generators are based on the lagged Fibonacci method.[@mascagni1995a; @mascagni1995b; @MascagniSrinivasan2000; @sprng; @MersenneTwister] They are usually of the form $s_k = (s_{k-q} \otimes s_{k-r}) \thinspace\textrm{mod}\thinspace p$, where $p$ is a power of two, $\otimes$ is one of the operations addition, exclusive or, subtraction, or multiplication, and $q<r$ are integer parameters chosen based on primitive polynomials modulo 2.[@Schneier1994; @Koblitz1987; @knuth] Lagged Fibonacci generators have periods of the order of $2^r$, with $r$ typically of the order of several hundred to several thousand. The state of the generator is defined by a table of $r$ integers or $r$ bits, which represent the most recent pseudorandom values in the sequence. Parallel implementation of these algorithms can be accomplished by either stream splitting or parameter splitting. [@MascagniSrinivasan2000; @sprng]
The parallel pseudorandom number generator class we propose here is based on parameter splitting. In our algorithm, every independent process is assigned a different prime modulus $n$, which produces an independent pseudorandom sequence. The sequence of pseudorandom numbers produced using equation are effectively uncorrelated within a single stream and between streams, and the period of every stream is different. The number of independent streams is limited only by the number of prime numbers in the range defined by the implementation. For example, in a 32-bit implementation the prime moduli can be chosen from the set of 98,182,656 primes in the range $[2^{31} . \thinspace . \thinspace\thinspace 2^{32}]$.[@http://oeis.org/A036378]
To demonstrate the relation to encryption and explain some of the useful properties of the method, the message $m$ can be decrypted from the ciphertext $c$ using the decryption exponent $d=e^{-1}\thinspace \textrm{mod}\thinspace (n-1)$, i.e. $ d e = 1+ q (n-1) $ for some $q$.[@PohligHellman1978; @Schneier1994; @Koshy2002; @Silverman2006; @Koblitz1987] The decryption exponent $d$ exists and is unique if $e$ and $n-1$ are co-prime, i.e if $\gcd(e,n-1)=1$. The decryption exponent can be determined quickly using the extended Euclidean algorithm.[@Schneier1994; @Koshy2002; @Silverman2006; @Koblitz1987] The decryption is based on Fermat’s little theorem:[@Koshy2002; @Silverman2006; @Koblitz1987] for any prime n and for all $m$ in $(\mathbb{Z}/n\mathbb{Z})^*$, $m^{n-1}\thinspace\textrm{mod}\thinspace n= 1$. Therefore $$\begin{aligned}
\label{eqn:decrypt}
c^d \thinspace \textrm{mod} \thinspace n = m^{d e} \thinspace \textrm{mod} \thinspace n =
m^{1+q(n-1)} \thinspace \textrm{mod} \thinspace n =
(m (m^{n-1})^q) \thinspace \textrm{mod} \thinspace n =
m \thinspace \textrm{mod} \thinspace n = m .\end{aligned}$$ For any allowed exponent $e$, equation maps $\mathbb{Z}/n\mathbb{Z}$ onto a permutation of the same set. Therefore, any message sequence that uniformly samples $\mathbb{Z}/n\mathbb{Z}$ will produce ciphertexts that uniformly sample $\mathbb{Z}/n\mathbb{Z}$. The Pohlig–Hellman algorithm is closely related to the more widely used RSA public key encryption method[@Schneier1994; @Koshy2002; @Silverman2006; @Koblitz1987; @FergusonSchneierKohno2010; @rsa1979] which uses moduli that are products of two large primes. Using a composite modulus $n=p q$ creates a serous weakness for this application since every message $m$ that is a multiple of $p$ or $q$ gives a ciphertext $c$ that is also a multiple of $p$ or $q$, respectively.
In the Pohlig–Hellman cryptographic scheme, the key consisting of the prime $n$ and exponent $e$ must be known only to the sender and receiver. Otherwise, an eavesdropper can easily determine the decryption exponent. Cryptography theory suggests using safe primes for the moduli, i.e. primes $n$ for which $(n-1)/2$ is also prime.[@PohligHellman1978; @PatelSundaram; @Schneier1994; @SophieGermain] This helps ensure eavesdroppers will find it computationally difficult to take the discrete logarithm to recover the message from the ciphertext.[@PohligHellman1978; @Koshy2002; @Silverman2006; @Koblitz1987] While our empirical tests do not give noticeably better statistics for safe primes than for primes in general, the number of safe primes does not seriously limit the scaling with 32-bit moduli since there are 3,060,794 safe primes in the range $[2^{31} . \thinspace . \thinspace \thinspace 2^{32}]$.[@http://oeis.org/A211395]
The algorithm presented here is similar to a block cipher operating in the counter mode which outputs a sequence of pseudorandom numbers that are ciphertexts resulting from a simple sequence of plaintexts.[@Schneier1994; @FergusonSchneierKohno2010] The Pohlig–Hellman method can be used to generate cryptographically secure pseudorandom sequences, but cryptographic applications require primes with hundreds of digits.[@PohligHellman1978; @Schneier1994; @Koshy2002; @Silverman2006; @Koblitz1987; @FergusonSchneierKohno2010; @BlumMichali; @PatelSundaram] Our application, instead, uses Pohlig–Hellman encryption of 32-bit messages to produce pseudorandom sequences suitable for use in Monte Carlo simulations and other applications. The exponentiations can be accomplished using the method of repeated squaring,[@Schneier1994; @Koshy2002; @Silverman2006; @Koblitz1987] so the number of multiply–mod operations needed to calculate $m^e\thinspace\textrm{mod}\thinspace n$ is less than $2\log_2 e$. As we will show below, the algorithm can be implemented using small exponents that require very few multiply–mod operations per pseudorandom number.[@pohlig-hellansmallexponents] If the modulus $n$ is chosen to be a prime number less than $2^{32}$, each multiply–mod operation can be performed in hardware on a 64-bit processor with one 64-bit multiply, and one 64-bit mod. Our algorithm leads to a new scalable class of fast and effective pseudorandom generators based on parameter splitting. There are several advantages to this pseudorandom number generation method.
- The algorithm is based on elementary number theory and cryptography.
- The method is trivially parallelizable by parameterization, with each instance derived from an independent prime modulus.
- Pseudorandom sequences that result from different prime moduli are independent, and have different periods.
- The method is fast since it requires only a few integer multiply–mod operations per pseudorandom number.
- The method is fully scalable on massively parallel computing clusters due to the millions of available 32-bit primes.
- As we will show below, the period of the generator can be greatly extended by using message skips derived from a prime number linear congruential pseudorandom number generator.
- The seeding and initialization of the independent streams is simple, and it is possible to initialize each process without needing information about the states of any of the other processes. The state of each generator is defined by only seven integers: $m$, $e$, $n$, and $c$, and integer pseudorandom skip parameters $s$, $p$ and $a$ defined below. All of these values can be stored in local memory, and require no global storage. The values $n$, $e$, $p$ and $a$ are fixed in each instance. Only the three values $m$, $s$, and $c$ change during calls to the generator.
- The algorithm allows one to quickly jump far forward or backward in the pseudorandom sequence.
- The method does not require combining generators[@lecuyer1988] or shuffling[@BayesDurham] to remove correlations.
- The algorithm is simple enough to allow the generator to be implemented as an in-line function for efficiency.
- The algorithm can be implemented in parallel on vector computers and GPUs.
- The method passes a battery of intrastream and interstream correlation tests using up to $10^{13}$ pseudorandom numbers per test.
Message skipping algorithms
===========================
The selection of messages to be encrypted for a given modulus $n$ can be accomplished in many different ways, but they can all be expressed in terms of an integer skip sequence with the skips $s_k$ chosen from $\mathbb{Z}/n\mathbb{Z}$:
\[basealgorithm\] $$\begin{aligned}
m_k&= (m_{k-1}+s_k) \thinspace\textrm{mod}\thinspace n,\\
c_k &= m_k^e \thinspace\textrm{mod}\thinspace n .
%R_k &= (c_k+1)/ (n+1) .\end{aligned}$$
Note that if the skip values $s_k$ are chosen uniformly and *randomly* (not pseudorandomly) from $\mathbb{Z}/n\mathbb{Z}$, then the sequence of messages $m_k$ forms a uniform random sequence on the same set. Message $m_k$ is, in effect, a one-time pad encryption of message $m_{k-1}$.[@Schneier1994] Since the set of ciphertexts is one-to-one with the set of messages, the sequence of ciphertexts $c_k$ also forms a uniformly distributed random sequence.
In encryption, it is essential to avoid *cribs*, i.e. messages that result in easily decoded ciphertexts. For example, the messages $m=0,1,n-1$ are cribs for all allowed exponents $e$ since $m^e \thinspace\textrm{mod}\thinspace n =0,1,n-1$, respectively. Calculation efficiency makes it desirable to use small exponents to reduce the number of multiply–mod operations needed to generate each pseudorandom number. RSA-based cryptographic applications often use small exponents such as $e=3,5,17$.[@FergusonSchneierKohno2010] Messages with $m^e < n$ and $(n-m)^e < n$ result in trivially decodable ciphertexts, so exponents $e < \log_2 n$ result in additional cribs. Cribs can be avoided by padding messages to ensure no messages are close to $0$ or $n$. For cryptographic applications, the padding needs to have a random component.[@Schneier1994; @FergusonSchneierKohno2010] For our application, eliminating cribs from the message stream biases the ciphertext distribution due to the elimination of ciphertexts formed from messages with $m^e < n$ and $(n-m)^e < n$, resulting in a slight but systematic under-sampling of small and large values of $c$. Even though the effect of this is small, we choose not to implement an algorithm that does not give a uniform distribution of ciphertexts over the full period. For our purpose, it is not necessary to eliminate cribs, since they would appear in a random message sequence, but rather to prevent correlated sequences of cribs. Our goal is to select a simple skip pattern that ensures a uniform sampling of the set of all messages, is computationally fast, has a long period, allows the use of small encryption exponents, and avoids correlated cribs. We will accomplish this by using a pseudorandom skip sequence that, over the full period of the generator, uniformly samples the messages in $\mathbb{Z}/n\mathbb{Z}$, but we will first examine the properties of pseudorandom ciphertext sequences derived from simpler skip patterns.
- The simplest skip sequence that uniformly samples $\mathbb{Z}/n\mathbb{Z}$ is the unit skip, i.e. $s_k = 1$ for all $k$. Pohlig–Hellman encryption of this sequence is, in effect, a block cipher operating in counter mode.[@Schneier1994; @FergusonSchneierKohno2010] The message sequence is $m_k=(m_0 + k) \thinspace\textrm{mod}\thinspace n$, with the period $P$ of the generator being $P=n$. For 32-bit moduli, one can easily test the full period of the generator. In spite of the cribs near $m=0$ and $m=n-1$, the sequence $c_k = k^e \thinspace\textrm{mod}\thinspace n$ with $0\leqslant k <n $ passes most statistical tests for randomness even for small exponents. Naturally, the full-period sequence produces a perfectly uniform one-dimensional histogram since after $n$ steps every value $c$ will appear once, and only once, in the sequence. Except for the one-dimensional frequency test, and other tests that are affected by the uniformity of the sampling of numbers in the range $\mathbb{Z}/n\mathbb{Z}$, and in spite of the symmetry noted below, the exponentiation cipher with unit skips passes all of our other statistical correlations tests for exponents $e=9$ and $e=17$. By comparison, all linear congruential generators fail *all* $D$-dimensional correlations tests once a substantial fraction of the period has been exhausted, since all $D$-dimensional histograms become uniform as the period of the generator is approached.[@knuth] The exponentiation cipher does not suffer from this; see figures 1 and 2 which display the two-dimensional correlation patterns for a prime number linear congruential generator, and a unit-skip Pohlig–Hellman generator using the same prime modulus. As with most pseudorandom number generators, there is a symmetry in the pseudorandom sequence. The Pohlig–Hellman cipher has the symmetry $$\begin{aligned}
(n-m)^e \thinspace\textrm{mod}\thinspace n=n - m^e \thinspace\textrm{mod}\thinspace n ,\end{aligned}$$ i.e. the ciphertexts derived from messages in $[1\thinspace .\thinspace .\thinspace (n-1)/2]$ are strongly correlated with the messages in $[(n+1)/2\thinspace .\thinspace .\thinspace (n-1)]$, but in reverse order.
- The next simplest skip algorithm is the constant skip: $s_k = b$ with $1<b<n-1$, so $m_k=(m_0 + k b) \thinspace\textrm{mod}\thinspace n$. This message pattern is especially important for understanding the quality of our proposed pseudorandom skip algorithm discussed next. For most values of $b$, the constant skip removes sequential and closely spaced cribs. However, other than breaking up closely spaced cribs, the pattern produced by constant skips is not substantially better than the unit skip pattern since $$\begin{aligned}
\label{constantskipsymmetry}
(kb)^e\thinspace\textrm{mod}\thinspace n=\left( (b^e \thinspace\textrm{mod}\thinspace n)(k^e \thinspace\textrm{mod}\thinspace n)\right) \thinspace\textrm{mod}\thinspace n,\end{aligned}$$ i.e. this is the same as the unit-skip sequence except for a single constant factor $b^e \thinspace\textrm{mod}\thinspace n$. Like the unit skip case, there is a symmetry in the pseudorandom sequence since $((n-k)b)^e \thinspace\textrm{mod}\thinspace n = n - (kb)^e\thinspace\textrm{mod}\thinspace n$. As with the unit skip, the constant skip passes a battery of full-period statistical tests for $e=9$ and $e=17$, except ones affected by the uniformity of the one-dimensional distribution.
- Our recommended skip pattern is a pseudorandom skip produced by a prime number linear congruential pseudorandom number generator:[@knuth; @lecuyer1999; @lecuyer1988] $$\begin{aligned}
\label{congruentialgenerator1}
s_k= a s_{k-1} \thinspace\textrm{mod}\thinspace p= s_0 a^k \thinspace\textrm{mod}\thinspace p ,\end{aligned}$$ with prime modulus $p<n$. The multiplier $a$ is chosen to be a primitive root[@Koshy2002; @Silverman2006; @Koblitz1987] $\textrm{mod}\thinspace p$ that delivers a full period, well-tested pseudorandom sequence.[@knuth; @lecuyer1999; @lecuyer1988] This gives pseudorandom skips $s_k$ in the range $(\mathbb{Z}/p\mathbb{Z})^*$, with the period of the skip sequence being $p-1$. This results in the following pseudorandom ciphertext sequence:
\[pseudoskip\] $$\begin{aligned}
s_k &= s_0 a^k \thinspace\textrm{mod}\thinspace p , \\
m_k &= \left( m_0 + \sum_{j=1}^k{ s_0 a^j \thinspace\textrm{mod}\thinspace p } \right)\thinspace\textrm{mod}\thinspace n ,
\label{pseudoskipb} \\
c_k &= m_k^e \thinspace\textrm{mod}\thinspace n .\end{aligned}$$
Using a pseudorandom skip sequence serves several important purposes:
- A pseudorandom skip effectively eliminates the problem of correlated cribs, allowing the use of small exponents.
- Using a pseudorandom skip extends the period of the generator to $P=n(p-1)$.
- The method provides a uniform sampling of ciphertexts over the full period of the generator. Each message $m$ in $\mathbb{Z}/n\mathbb{Z}$, and hence each ciphertext $c$ in $\mathbb{Z}/n\mathbb{Z}$, will appear exactly $p-1$ times in the sequence.
- The state of each generator is defined by only seven integers: $\{m,s,c,n,e,p,a\}$.
- Implementations using 32-bit primes are fast. With small exponents, each pseudorandom number requires only a few 64-bit multiply–mod operations that can be implemented in hardware on 64-bit processors.
- The implementations suggested below pass a battery of statistical tests with up to $10^{13}$ pseudorandom numbers per test.
First, let’s prove that the period of the generator is $P=n(p-1)$. Since $a$ is a primitive root mod $p$, after $p-1$ pseudorandom skips $s_k$ will have cycled through every value in $(\mathbb{Z}/p\mathbb{Z})^*$, so $s_{k+p-1}=s_k$ and $m_{k+p-1}=\left( m_k + \sum_{s=1}^{p-1}{s} \right) \thinspace\textrm{mod}\thinspace n = ( m_k + p(p-1)/2 ) \thinspace\textrm{mod}\thinspace n$. This means the sequence of messages separated by multiples of $p-1$ steps in the sequence are derived from a constant skip $b=p(p-1)/2 \thinspace\textrm{mod}\thinspace n$. Since $p$ and $(p-1)/2$ are coprime to $n$, the constant skip $b$ is also co-prime to $n$. Therefore, the state of the generator after $k=q(p-1) + r$ steps, with $q=\lfloor k/(p-1)\rfloor$ and $r=k\thinspace\textrm{mod}\thinspace (p-1)$, is given by
\[pseudogenerator1\] $$\begin{aligned}
s_k &= s_0 a^r \thinspace\textrm{mod}\thinspace p , \label{pseudogenerator1a}\\
m_k &= \left( m_0 + q p (p-1)/2 + \sum_{j=1}^r{ \left( s_0 a^j \thinspace\textrm{mod}\thinspace p\right) } \right)\thinspace\textrm{mod}\thinspace n , \label{pseudogenerator1b}\\
c_k &= m_k^e \thinspace\textrm{mod}\thinspace n \label{pseudogenerator1c}.\end{aligned}$$
This demonstrates the mathematical form of the pseudorandom sequence across the full period of the generator. Since $s_{k+n(p-1)}=s_k$ and $m_{k+n(p-1)} = m_k$, and each subsequence of messages of length $p-1$ is different from every other subsequence, the period the generator is $P=n(p-1)$. Over the full period, every ciphertext in $\mathbb{Z}/n\mathbb{Z}$ will appear exactly $p-1$ times. One can use equation to skip $k$ steps forward directly to any point in the pseudorandom sequence, with skips being fast if $k$ is close to a multiple of $p-1$. Backward skips of are accomplished by replacing $a$ with $a^{-1} \thinspace\textrm{mod}\thinspace p$, subtracting rather than adding in equation , and reordering equations and .[@plessthann]
The simple form of the ciphertext sequence allows one to determine the full-period $D$-dimensional correlations pattern for rectangular regions with volume $L_0 L_1^{D-1}$ in $O(p L_0)$ steps, i.e. without needing to exhaust the generator. The $D$-dimensional density of points $(c_k,c_{k+1},\ldots , c_{k+D-1})$ over the full period is $\rho_D = (p-1)/n^{D-1}$ so the average number of points in the volume above is $(p-1)L_0 (L_1/n)^{D-1}$. One can select $L_0$ points $c_k$, with the succeeding points given by $c_{k+1}=(c_k^d \thinspace\textrm{mod}\thinspace n + s)^e \thinspace\textrm{mod}\thinspace n$, $c_{k+2}=(c_k^d \thinspace\textrm{mod}\thinspace n + s + (a s)\thinspace\textrm{mod}\thinspace p)^e \thinspace\textrm{mod}\thinspace n$, etc. where the skips $s$ take on all values in $(\mathbb{Z}/p\mathbb{Z})^*$. A magnified region near the origin of the full-period two-dimensional pattern for safe prime $n=4294967087$, $p=2147483647$, $a=784588716$,[@lecuyer1999] and $e=9$ is shown in figure 3, with the full-period three-dimensional pattern shown in figure 4.
We tested the quality of 32-bit pseudorandom sequences based on pseudorandom skip sequences using a battery of independent statistical tests. We first tested uniform double precision floating point pseudorandom sequences $R_k$ over the period of the skip generator for more than ten-thousand different safe primes between $2^{31}$ and $2^{32}$, and used a single prime number linear congruential pseudorandom number generator recommended by L’Ecuyer:[@lecuyer1999] $p=2^{31}-1=2147483647$ and $a=784588716$. Even using exponents as small as $e=3$, the intrastream pseudorandom sequences pass all of our statistical tests across the period of the skip generator.
We then used equation to test for intrastream correlations among ciphertexts separated by $p-1$ steps in the sequence across the constant skip period $P=n$ using thousands of safe primes between $2^{31}$ and $2^{32}$. This pseudorandom sequence is given by $c_{q(p-1)}= (m_0+ q b)^e \thinspace\textrm{mod}\thinspace n$, with $b=p (p-1)/2 \thinspace\textrm{mod}\thinspace n$ and $q=0,1,\thinspace . \thinspace . n-1$. These constant skip sequences pass all of our the statistical tests for exponents as small as $e=7$ for up to $2^{25}$ pseudorandom numbers, and all but the one-dimensional tests across the period $P=n$. (As noted earlier, the one-dimensional constant skip distribution becomes uniform over the period. These are the only tests that the constant skip sequences appear to fail.) Based on this, we recommend using exponents $e=9$ or $e=17$, which require only four or five multiply–mod operations per exponentiation, respectively.
Since ciphertexts separated by $k<p-1$ steps demonstrate good intrastream statistics because of the pseudorandom skip, and ciphertexts separated by multiples of $p-1$ steps demonstrate good statistics with constant skips $b=p(p-1)/2 \thinspace\textrm{mod}\thinspace n$, one has good reason to believe that the entire pseudorandom sequence should pass a battery of statistical tests until the length of the pseudorandom sequence approaches the full period. To test this, we performed intrastream statistical tests of the pseudorandom 32-bit skip algorithm over as large a fraction of the period and for as many different moduli as possible. We tested sequences of $10^{11}$ pseudorandom numbers using hundreds of different primes, sequences of $10^{12}$ pseudorandom numbers using dozens of different primes, and sequences of $10^{13}$ pseudorandom numbers using a few different primes. This latter test corresponds to several thousand periods of the skip sequence. The method passed every test we applied.
We also tested to ensure that the algorithm displayed lack of correlation between streams. Suppose one has $N_p$ processes each with a different prime modulus $n^{(\alpha)}$ with $\alpha=0,1,\ldots ,N_p-1$, and pseudorandom sequences $R_0^{(\alpha)}, R_1^{(\alpha)}, \ldots$. Our interstream correlations tests draw the pseudorandom numbers from the $N_p$ streams in the order $R_1^{(0)}, R_1^{(1)}, R_1^{(2)}, \ldots , R_1^{(N_p-1)}, R_2^{(0)}, R_2^{(1)}, \ldots $. We tested groupings of $N_p=2$, $32$, $1024$, $32768$ and $1048576$ different streams, using exponents $e= 5, \medspace 9, \medspace \mbox{and} \medspace 17$. We also included tests using $N_p=3,060,794$ safe primes, i.e. all of the safe primes in $[2^{31} \thinspace .\thinspace . \thinspace 2^{32}]$. To test that seeding coincidences do not create spurious correlations, we performed many of the interstream tests by starting every sequence with exactly the same initial values $m_0$ and $s_0$, but avoiding cases in which one of the early messages in the sequence is a crib. The interstream correlations passed every test for up to $10^{13}$ pseudorandom numbers per test.
Tests
=====
We first applied well-established pseudorandom number test suites DIEHARD,[@diehard] NIST[@NIST], and TestU01,[@U01Test] to ensure the generator passes a wide variety of tests. We then applied the following ten additional tests that produce histograms to which one can apply a chi-square test.[@knuth] For each test, we calculated $\chi^2$ and the associated $\mathscr{P}$-value, i.e. the one-sided probability of $\chi^2$ having that value above or below the median. We applied these additional tests to sequences of up to $10^{13}$ pseudorandom numbers per test.
- One-dimensional frequency test:[@knuth] We distributed the sequence of pseudorandom numbers into a one-dimensional histogram with $2^{20}$ bins, and compared the histogram to a uniform Poisson distribution.
- Two-dimensional serial test:[@knuth] We distributed pairs of pseudorandom numbers into a two-dimensional histogram with $2^{20}$ bins, and compared the histogram to a uniform Poisson distribution. This tests for sequential pair correlations in the sequence.
- Three-dimensional serial test:[@knuth] We distributed triplets of pseudorandom numbers into a three-dimensional histogram with $10^6$ bins, and compared the histogram to a uniform Poisson distribution. This tests for sequential triplet correlations in the sequence.
- Four-dimensional serial test:[@knuth] We distributed groups of four pseudorandom numbers into a four-dimensional histogram with $2^{20}$ bins, and compared the histogram to a uniform Poisson distribution. This tests for sequential four-point correlations in the sequence.
- Five dimensional serial test:[@knuth] We distributed groups of five pseudorandom numbers into a five-dimensional histogram with $2^{20}$ bins, and compared the histogram to a uniform Poisson distribution. This tests for sequential five-point correlations in the sequence.
- Poker test:[@knuth] We used groups of five pseudorandom numbers and counted the number of pairs, three-of-a-kind etc. formed from five cards and ten denominations, and compared the resulting histogram to a Poisson distribution derived from the exact probabilities. This tests for a variety of five-point correlations in the sequence.
- Collision test:[@knuth] We distributed $2^{14}$ pseudorandom numbers into $2^{20}$ bins and compared the histogram of the number of collisions against a Poisson distribution derived from the exact probabilities. This tests for long-range correlations in the sequence.
- Runs test:[@knuth] We compared the histogram of the length of runs of 0’s ($R\leq 0.5$) and 1’s ($R>0.5$) to the Poisson distribution derived from the exact probabilities. This tests for short-range correlations of the leading bits.
- Fourier test:[@MascagniSrinivasan2000] We used a fast Fourier transform[@fft] to calculate the Fourier coefficients of sequences of $M=2^{20}$ double precision floating point pseudorandom numbers, $$\begin{aligned}
\hat{x}_k = \frac{1}{\sqrt{M}} \sum_{j=0}^{M-1} { x_j e^{2\pi i j k/M} } ,\end{aligned}$$ where $x_j=R_j - 0.5$. The real and imaginary parts of the Fourier coefficients $\hat{x}_k$ with $k=0,\ldots, M/2 - 1$ should each be gaussian distributed about zero with variance $1/24$. We distributed the real and imaginary parts of the Fourier coefficients into a histogram, and compared the result to a Poisson distribution derived from the exact gaussian distribution. This tests for long-range pair correlations in the sequence.
- Two-dimensional Ising model energy distribution test:[@beale1996; @pathriabeale2011] We performed a Wolff algorithm[@wolff] Monte Carlo simulation at the critical point of the two-dimensional Ising model on a $128\times128$ square lattice, and compared the energy histogram to a Poisson distribution derived from the exact probabilities[@beale1996; @pathriabeale2011] Since the Wolff algorithm is based on stochastically growing fractal critical clusters that can span the system, this tests for long-range correlations in the pseudorandom sequence, and has proven to be effective at identifying weak generators.[@beale1996; @ferrenberglandauwong1992] Assigning an independent generator to each of the 32768 bonds in the lattice tests provides an additional test for exposing interstream correlations.
For several of these tests, we sampled the low-order bits to confirm they do not harbor any hidden correlations. We define passing our tests by there being no $\mathscr{P}<10^{-8}$ events among the tests. Since there were tens of thousands of independent tests, we also counted the number of $\mathscr{P}<10^{-4}$ events in our samples to confirm that the number was consistent with the expected number. Finally, we applied chi-square tests to histograms of the full-period distribution of $(c_k, c_{k+1}, .. , c_{k+D-1})$ sequences in $2^{24}$ $D$-dimensional rectangular volumes of size $L_0 L_1^{D-1}$ with $L_0=2^7$ and $L_1\approx n/2^{24/(D-1)}$, for $D=3,4,5$. The generator passed every test.
Implementation
==============
The algorithm for generating uniformly distributed double precision floating point pseudorandom numbers $R$ on the open interval $(0\thinspace .\thinspace .\thinspace 1)$ is:
\[algorithm1\] $$\begin{aligned}
&s := (a s) \thinspace\textrm{mod}\thinspace p \\
&m := (m + s )\thinspace\textrm{mod}\thinspace n \\
&c := m^e \thinspace\textrm{mod}\thinspace n \label{CodeExponentiationStep}\\
&R := (c+1) / (n+1) \label{CodeFloatingPointStep}\\
&\textrm{RETURN}\medspace R \end{aligned}$$
In a 32-bit implementation, operations (9a)-(9c) of the the algorithm can be implemented using 64-bit unsigned integer arithmetic, which can be executed in hardware on 64-bit processors. This algorithm generates the sequence given by equations . For efficiency, one can precalculate the double precision floating point value $1.0/(n+1)$ and perform a floating point multiplication instead of a floating point divide in step .
For a multiprocessor environment, each of the $N_p$ processes can be assigned an independent prime modulus $n^{(\alpha)}$ using well-established primality tests. The Rabin-Miller test,[@Koblitz1987; @millerprime; @rabinprime] which is the same as Algorithm P in Knuth,[@knuth] provides a simple probabilistic test for primality. Every odd prime $n=1+2^r t$ with $t$ odd satisfies one of the following conditions for every base $g$ in $(\mathbb{Z}/n\mathbb{Z})^*$: either $g^t \thinspace\textrm{mod}\thinspace n = 1 $, or $g^{2^j t} \thinspace\textrm{mod}\thinspace n = n-1$ for some some $j$ in the range $0 \leq j < r$. A composite number $n$ satisfying these criteria is called a strong pseudoprime to base $g$. For any odd composite, the number of bases $g$ for which $n$ is a strong pseudoprime to the base $g$ is less than $ n/4 $, so if the test is applied repeatedly with $M$ randomly chosen bases in $(\mathbb{Z}/n\mathbb{Z})^*$, the probability that a composite will pass every test is less than $4^{-M}$.[@Koblitz1987; @knuth; @millerprime; @rabinprime] Better yet, the Rabin-Miller test can quickly and deterministically identify all primes below $2^{64}$. There are no composite numbers below $2^{64}$ that are strong pseudoprimes to all of the twelve smallest prime bases ($g=2,3,5,7,11,13,17,19,23,29,31,37$).[@PomeranceSelfridgeFlagstaff1980; @http://oeis.org/A014233; @Zhang2001; @JiangDeng2012; @wolframrabinmiller] Therefore, any number less than $2^{64}$ that passes the Rabin-Miller test for all twelve of these bases is prime. Likewise, any number less than $2^{32}$ that passes the Rabin-Miller test for all of the five smallest prime bases ($g = 2, 3, 5, 7, 11$) is prime. For efficiency, one first checks to see if any small primes divide $n$ before applying the Rabin-Miller test.
One can initialize the generators using the system time variable and the process identifier $\alpha$ to construct unique values of $n^{(\alpha)}$, and different starting values $s_0^{(\alpha)}$ and $m_0^{(\alpha)}$ for each of the $N_p$ processes. One might use equation to skip each generator forward relative to some base state such as $m_0=0$ and $s_0=1$ to ensure the initial skips and messages are all unsynchronized. Even if two generators later become synchronized with the same values of $m$ and $s$, the values for the ciphertexts for different moduli will be different due to the encryption step, and the message synchrony is removed after a few skips, even if the skips were to remain synchronized.
The implementation for 32-bit moduli is simple and fast since the multiply–mod operations can be executed in hardware using standard 64-bit unsigned arithmetic on 64-bit processors. The quality of the pseudorandom sequences does not appear to be dependent on the values of the prime moduli, but we recommend using safe primes selected from $[2^{31} \thinspace . \thinspace . \thinspace 2^{32}]$ unless the number of processes exceeds 3,060,794, the number of safe primes in that range. If the implementation requires more than 3,060,794 instances, there are 49,091,941 primes $n$ in $[2^{31}\thinspace . \thinspace . \thinspace 2^{32}]$ with $e=9$ coprime to $n-1$, and 92,045,560 primes in that range with $e=17$ coprime to $n-1$. The number of well-tested 31-bit prime number pseudorandom skip generators is not scalable, but one can use the ones available in the literature[@lecuyer1988; @lecuyer1999] to substantially extend the number of possible instances.[@poweroftwoskips]
The full period of each generator is $P=n(p-1) > 2^{62} \simeq 4.6 \times 10^{18}$. This 32-bit implementation passes all of our intrastream and interstream correlations tests for $e=9$ and $e=17$, for up to $10^{13}$ pseudorandom numbers per prime modulus. The exponent $e=9$ requires only five multiply–mod operations per pseudorandom number, one for the skip and four for the exponentiation, and the exponent $e=17$ requires only six. Consequently, the code is compact and simple enough to be implemented as an in-line function.
If far longer periods or far more instances are needed, one can implement the algorithm using 64-bit primes. There are approximately $3\times 10^{15}$ safe primes in $[2^{63} \thinspace . \thinspace . \thinspace 2^{64}]$. L’Ecuyer[@lecuyer1999] provides suitable pseudorandom skip parameters near $2^{63}$ so the period of each process is $P=n(p-1)>2^{126}\simeq 8.5 \times 10^{37}$. Using 64-bit prime moduli results in a speed penalty with current processors since the multiply–mod operations need to be implemented in using 128-bit unsigned integers. Since a single process is unlikely to exhaust a 63-bit pseudorandom skip, one might consider using smaller encryption exponents for efficiency. Preliminary statistical tests indicate the method works well for 64-bit safe prime moduli even when using $e=3$, the smallest allowed exponent.
Conclusion
==========
We propose a new class of parallel pseudorandom number generators based on Pohlig–Hellman exponentiation ciphers. The method creates pseudorandom streams by encrypting simple sequences of integer messages. The method is fully scalable based on parametrization since each process can be assigned a unique prime modulus. By using pseudorandom skips among messages, one can use small exponents and the period is greatly extended. For 32-bit implementations, only a few 64-bit multiply–mod operations are needed per pseudorandom number. There are millions of possible independent instances, and the period of each instance is is greater than $10^{18}$. We have tested thousands of different pseudorandom streams for intrastream and interstream correlations using up to $10^{13}$ pseudorandom numbers per test, and all pass a battery of statistical tests. A 64-bit implementation would have more than $10^{15}$ possible instances and periods greater than $10^{37}$. Sample C++ code for a 32-bit multi-processor MPI implementation of the Pohlig–Hellman pseudorandom number generator with pseudorandom skip can be found at <http://works.bepress.com/paul_beale/>.
Acknowledgements
================
We thank Matt Glaser, Rudy Horne, Nick Mousouris, Ethan Neil, Robert Blackwell, John Black, and David Grant for helpful discussions. This work utilized the Janus supercomputer, which is supported by the National Science Foundation (award number CNS-0821794) and the University of Colorado Boulder. The Janus supercomputer is a joint effort of the University of Colorado Boulder, the University of Colorado Denver, and the National Center for Atmospheric Research.
[5]{}
S. Pohlig and M. Hellman, IEEE Trans. Inform. Theory **(24)**: 106 (1978).
B. Schneier, *Applied Cryptography* (Wiley, New York, 1994).
T. Koshy, *Elementary Number Theory and Applications* (Academic, San Diego, 2002).
J.H. Silverman, *A Friendly Introduction to Number Theory* (Pearson, New York, 2006).
N. Koblitz, *A Course in Number Theory and Cryptography* (Springer-Verlag, New York, 1987).
N. Ferguson, B. Schneier, T. Kohno, *Cryptography Engineering: Design Principles and Applications*, (Wiley, Indianapolis, 2010).
W.H. Press, S.A. Teukolsky, W.T. Vetterling, and B.P. Flannery, *Numerical Recipes: The Art of Scientific Computing*, (Cambridge, New York, 1992), 2nd ed.
M. Blum and S. Micali, SIAM J. Comput. **13**, 850, (1984).
S. Patel and G.S. Sundaram, CRYPTOÕ98, LNCS **1462**, 304, (1998).
H. Bauke and S. Mertens, Phys. Rev. E **75**, 066701 (2007).
M. Mascagni and A. Srinivasan, ACM Trans. Math. Software **26**, 436 (2000). The Scalable Parallel Random Number Generators Library (SPRNG), <http://www.sprng.org>.
P. L’Ecuyer, Math. Comp. **68**, 249 (1999).
M. Mascagni, Parallel Comput. **24**, 923 (1998). The period of every such generator is $p$, and the low order bits are highly correlated within each stream, as well as between streams.
M. Mascagni, M. L. Robinson, D. V. Pryor and S. A. Cuccaro, [Lec. Notes Statistics]{} **106**, 263 (1995). M. Mascagni, S. A. Cuccaro, D. V. Pryor and M. L. Robinson, [J. Comp. Phys.]{} **119**, 211 (1995). M. Matsumoto and T. Nishimura, ACM Trans. Model. Comput. Sim. **8(1)**, 3 (1998).
D. Knuth, *The Art of Computer Programming*, vol. 2 (Addison-Wesley, Reading, Massachusetts, 1999).
*The On-Line Encyclopedia of Integer Sequences*, <http://oeis.org/A036378>.
R. Rivest, A. Shamir, and L. Adelman, Commun. ACM **21**, 120 (1978).
This makes $(n-1)/2$ a Sophie Germain prime.
*The On-Line Encyclopedia of Integer Sequences*, <http://oeis.org/A211395>; <http://oeis.org/A211397>.
A cryptographically secure implementation of the Pohlig–Hellman cipher would ordinarily be implemented with randomly chosen exponents of the same order as $n$, since that would force an eavesdropper to resort to solving the discrete logarithm problem, even if she knew $n$.[@Koshy2002; @Silverman2006; @Koblitz1987]
P. L’Ecuyer, Commun. ACM **31**, 741 (1988).
C. Bays and S.D. Durham, ACM Trans. Math. Software **2**, 59 (1976). The mathematical form of the message sequence can be used to demonstrate why the method requires $p<n$. If $n=p$, one can evaluate the sum in equation in closed form to give $m_k = (m_0 + s_0 a(a-1)^{-1}(a^k-1))\thinspace\textrm{mod}\thinspace p$. Therefore, using Fermat’s little theorem, $m_{k+p-1}=m_k$ and $s_{k+p-1}=s_k$, so period is reduced to $P=p-1$.
G. Marsaglia, DIEHARD: a battery of tests of randomness (1996); see <http://stat.fsu.edu/pub/diehard/>.
A. Rukhin, J. Soto, J. Nechvatal, M. Smid, E. Barker, S. Leigh, M. Levenson, M. Vangel, D. Banks, A. Heckert, J. Dray, and S. Vo, NIST special publication 800-22, National Institute of Standards and Technology (NIST), Gaithersburg, Maryland, USA, 2001; see <http://csrc.nist.gov/rng/>.
P. L’Ecuyer and R. Simard, ACM Trans. Math. Software **33**, 22 (2007); see <http://www.iro.umontreal.ca/~simardr/testu01/tu01.html>.
J.W. Cooley and J.W. Tukey, Math. Comput. **19**, 297 (1965).
P.D. Beale, [Phys. Rev. Lett.]{} **76**, 78 (1996).
R.K. Pathria and P.D. Beale, *Statistical Mechanics* (Academic, Boston, 2011), 3rd ed.
U. Wolff, [Phys. Rev. Lett.]{} **62**, 361 (1989).
A.M. Ferrenberg, D.P. Landau and Y.J. Wong, [Phys. Rev. Lett.]{} **69**, 3382 (1992).
G.L. Miller, J. Comp. Sys. Sci. **13**, 300 (1976).
M.O. Rabin, J. Number Theory **12**, 128 (1980).
C. Pomerance, J. L. Selfridge and S. S. Wagstaff, Jr., Math. Comp. **35**, 1003 (1980).
*The On-Line Encyclopedia of Integer Sequences* <http://oeis.org/A014233>.
Zhenxiang Zhang, Math. Comp. **70**, 863 (2001). Y. Jiang and Y. Deng, <http://arxiv.org/abs/1207.0063v1>. See <http://mathworld.wolfram.com/Rabin-MillerStrongPseudoprimeTest.html>.
One might consider using power-of-two linear congruential generators for the skip generator since they form a scalable class,[@MascagniSrinivasan2000; @sprng; @lecuyer1999] but we have not investigated the effects the strong correlations in the low-order bits of the skips might have on correlations of the ciphertexts.
![Magnified region near the origin of the two-dimensional full-period $s_{k+1}$ vs. $s_k$ pattern delivered by a prime number linear congruential pseudorandom number generator (equation ) with $p=2^{32}-5=4294967291$ and $a=279470273$.[@lecuyer1999] The axes cover the range $[0\thinspace . \thinspace . \thinspace\thinspace 2^{20}]$, i.e. a linear magnification factor of 4096. The 16 subsquares have area $\Delta x^2=(2^{18})^2$ each. As with all linear congruential generators, every $D$-dimensional full-period pattern forms a perfect lattice,[@knuth] so the number of points in each cell is very close to $N_{cell}=\Delta x^2 / (p-1) \simeq 16$ events in this case.](lecuyer2D4294967291.png){width="5in"}
![Magnified region near the origin of the two-dimensional full-period $c_{k+1}$ vs. $c_k$ pattern for a Pohlig–Hellman exponentiation cipher (equation ) using the same prime as in Fig. 1, $n=2^{32}-5=4294967291$, exponent $e=9$, and unit skip, i.e. $m_k=0, 1,2,\ldots , n-1$. The axes cover the range $[0\thinspace . \thinspace . \thinspace\thinspace 2^{20}]$, i.e. a linear magnification factor of 4096. The 16 subsquares have area $\Delta x^2=(2^{18})^2$ each. Unlike linear congruential generators, the distribution of points in the cells approximate a Poisson distribution with $N_{cell} \approx \Delta x^2 /n \pm \sqrt{\Delta x^2 / n }\simeq 16\pm 4$ in this case. The unit-skip correlated cribs $(0,1)$, $(1,2^9=512)$, $(512,3^9=19683)$, and $(19693,4^9=2^{18}=262144)$ appear near the vertical axis in the lower left square.](pohlighellman32unit_e9A.png){width="5in"}
{width="4.5in"}
![Sequential ciphertext triplets $(c_k,c_{k+1},c_{k+2})$ over the full period of the generator in a rectangular three-dimensional region for a pseudorandom skip with $n=4294967087$, $p=2147483647$, $a=784588716$, $e=9$ and $d=3817748521$. The region shown is a $L_0 L_1^{2}=2^9 \times 2^{16} \times 2^{16}$ rectangular slab closest to the origin with volume $V_3=2^{41}$. The short first dimension is projected onto the front face. Even though the full period is $P=n(p-1)\simeq 4.6\times 10^{18}$, it is feasible to determine the local full-period pattern without exhausting the generator since $c_{k+1}=((c^d_{k}\thinspace\textrm{mod}\thinspace n) +s_{k+1})^e \thinspace\textrm{mod}\thinspace n$ and $c_{k+2}=((c^d_{k}\thinspace\textrm{mod}\thinspace n) +s_{k+1} + (a s_{k+1}) \thinspace\textrm{mod}\thinspace p)^e \thinspace\textrm{mod}\thinspace n$. One selects all ciphertexts $c_{k}$ in $[0 \thinspace .\thinspace . \thinspace L_0-1]$ and all skips $s_{k+1}$ in $(\mathbb{Z}/p\mathbb{Z})^*$ to see which ones give $c_{k+1}$ and $c_{k+2}$ inside the chosen volume. The three-dimensional density of triplets is $\rho_3=(p-1)/n^2$. The result shown here of the occupancies of the sixteen subslabs is consistent with a Poisson distribution with $N_{cell} \approx \rho_3 V_{cell} \pm \sqrt{\rho_3 V_{cell}} \simeq 16\pm 4$. ](fullperiod3D4294967087e9_2_16_3D.png){width="5in"}
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Yaman Umuroglu^\*^, Nicholas J. Fraser^\*^, Giulio Gambardella^\*^, Michaela Blott^\*^,\
Philip Leong^^, Magnus Jahre^^ and Kees Vissers^\*^\
; ;\
bibliography:
- 'references.bib'
title: '[FINN: A Framework for Fast, Scalable Binarized Neural Network Inference]{}'
---
| {
"pile_set_name": "ArXiv"
} |
---
address:
- |
A. Connes: Collège de France\
3, rue d’Ulm\
Paris, F-75005 France
- |
M. Marcolli: Max–Planck Institut für Mathematik\
Vivatsgasse 7\
Bonn, D-53111 Germany
author:
- Alain Connes
- Matilde Marcolli
title: Quantum Fields and Motives
---
Renormalization: particle physics and Hopf algebras
===================================================
The main idea of renormalization is to correct the original Lagrangian of a quantum field theory by an infinite series of counterterms, labelled by the Feynman graphs that encode the combinatorics of the perturbative expansion of the theory. These counterterms have the effect of cancelling the ultraviolet divergences. Thus, in the procedure of perturbative renormalization, one introduces a counterterm $C(\G)$ in the initial Lagrangian for every divergent one particle irreducible (1PI) Feynman diagram $\Gamma$. In the case of a [*renormalizable*]{} theory, all the necessary counterterms $C(\G)$ can be obtained by modifying the numerical parameters that appear in the original Lagrangian. It is possible to modify these parameters and replace them by (divergent) series, since they are not observable, unlike actual physical quantities that have to be finite. One of the fundamental difficulties with any renormalization procedure is a systematic treatment of nested and overlapping divergences in multiloop diagrams.
Dimensional regularization and minimal subtraction {#dimensional-regularization-and-minimal-subtraction .unnumbered}
--------------------------------------------------
One of the most effective renormalization techniques in quantum field theory is dimensional regularization (DimReg). It is widely used in perturbative calculations. It is based on an analytic continuation of Feynman diagrams to complex dimension $d\in \C$, in a neighborhood of the integral dimension $D$ at which UV divergences occur. For the complex dimension $d\to D$, the analytically continued integrals become singular and the expression admits a Laurent series expansion. Thus, within the framework of dimensional regularization, one can implement a renormalization by minimal subtraction, where the singular part of the Laurent series in $z=d-D$ is subtracted at each order in the loop expansion. This renormalization method (DimReg plus minimal subtraction) was developed by ‘t Hooft and Veltman [@tHV], who applied it to one-loop calculations in scalar electrodynamics, discussed the problem of overlapping divergences, the Ward identities, the case of theories with fermions, and anomalies. The method has since been applied widely to perturbative calculations and it quickly became the standard regularization and renormalization method for nonabelian gauge theories and the standard model.
Hopf algebras and the combinatorics of renormalization {#hopf-algebras-and-the-combinatorics-of-renormalization .unnumbered}
------------------------------------------------------
The modern viewpoint on combinatorics, which unfolded in the 70s around the systematic and rigorous restructuring of its foundations advocated by Giancarlo Rota, showed how algebraic structures such as coalgebras, bialgebras, and Hopf algebras govern elaborate combinatorial phenomena ([[*cf.*]{} ]{}[@JoRo], [@Ro], [@Sch]). The reason why such algebraic structures are naturally present lies in the fact that combinatorial objects tend to admit decomposition laws that reduce them to simpler pieces. Such laws are the source of coproduct rules. This principle was illustrated by many examples of incidence Hopf algebras arising from classes of graphs and matroids. The typical situation is families of finite graphs, closed under disjoint union and taking vertex induced subgraphs. These admit a coproduct of the form $$\label{coprodcomb}
\Delta(\Gamma)= \sum_{W\subseteq V} \gamma_W \otimes
\gamma_{V-W},$$ where $V$ is the set of vertices of $\Gamma$ and $\gamma_W$ is the induced subgraph on a set of vertices $W\subseteq V$. Results from Hopf algebras in combinatorics were used, for instance, to study graph coloring problems.
During 1960s and 1970s Quantum Field Theory underwent a season of extraordinary theoretical development. The detailed knowledge theoretical physicists gained on the subject not only made it into something of an art, but refined it into a highly sophisticated instrument, capable of producing theoretical predictions that, to this day, match experiments with unprecedented precision. Renormalization plays a central role in the quantum theory of fields, in as it provides a consistent scheme for extracting from divergent expressions finite values that can be matched to physically observed quantities. Various renormalization schemes can be implemented (though here we will be concerned only with the “dimensional regularization and minimal subtraction” scheme described above). A renormalization scheme produces an extremely elaborate combinatorial recipe that accounts for structuring of subgraphs in a hierarchy of subdivergences and counterterms. Perturbative renormalization hence appears as one of the most elaborate combinatorial recipes imposed on us by nature.
Conceptually, the crucial issue in the combinatorics of perturbative renormalization is a scheme that accounts for subdivergences. This is achieved by a [*forest formula*]{}, which assigns to a graph $\Gamma$ a formal expression where the subdivergences have been dealt with through inductively defined counterterms. Subtraction of the corresponding counterterm from this formal expression finally yields the renormalized value for $\Gamma$. The definition of such formal expressions, as we discuss more in detail below, is related to decomposing a graph by extracting all possible divergent subgraphs $\gamma\subset \Gamma$ and considering corresponding graphs $\Gamma/\gamma$ obtained by collapsing $\gamma\subset \Gamma$ to a single vertex. Such decomposition is more complicated than those derived from incidence relations on graphs in many combinatorial problems, as it is adapted to the specific divergences of the physical theory and has to take into account other data like the distribution of external momenta. Still, one can see a suggestive analogy between the type of decomposition expressed by coproducts and the decomposition $$\label{forest}
\Delta(\Gamma)= \sum_{\gamma \subseteq \Gamma} \gamma \otimes \Gamma/\gamma$$ in a sum over divergent subgraphs, which underlies the combinatorics of the forest formulae. It was the seminal work of Dirk Kreimer [@DK1] in 1997 that paved the way to a conceptual mathematical formulation of perturbative renormalization, precisely by encoding the complicated combinatorics of forest formulae via a coproduct and identifying the Hopf algebra that governed the renormalization procedure.
The extraction of a renormalized value from divergent Feynman integrals was related in [@DK1] and [@CK3] to the antipode in the Hopf algebra. However, the precise formula for the renormalized value given by the BPHZ procedure ([@BP], [@Hepp], [@Zi]) requires a further operation that twists the antipode, which, in this formulation, is not given directly in terms of the Hopf algebra structure. The main conceptual breakthrough in the understanding of the renormalization procedure, that fully reconciles it with the Hopf algebra structure, was then obtained in a later stage of development of the Connes–Kreimer theory of perturbative renormalization, [@CK1], [@CK2], where the BPHZ recursive formulae (see , , below) are described in terms of the Birkhoff factorization of loops. We shall return to this point in Section \[SectBirk\].
Given the state of affairs in combinatorics and in quantum field theory around the late ’70s, it may seem surprising that the pursuit of a conceptual mathematical interpretation of the procedure of perturbative renormalization had to wait, as it did, until the late 1990s. One should keep in mind though that, during the 1970s, mathematicians and physicists were maximally apart. The tendency among physicists was to shift the emphasis heavily towards deriving efficient computational recipes at the expense of conceptual understanding, the latter being often dismissed as a mere exercise of pedantry. This position, though justifiable in developing a theoretical apparatus that could be continuously tested against experiments, had the effect of alienating mathematicians. While quantum mechanics stimulated and in turn benefited from a lot of advancements in modern mathematics (operator algebras, functional analysis), mathematicians shied away from quantum field theory, which they perceived as ill founded, riddled with inevitable divergences, and governed by obscure hands-on recipes. In more recent times, mathematicians and theoretical physicists found a renewed harmony of language, but this happened mostly in the context of string theory. This, however, bypasses many of the crucial problems posed by quantum field theory, by proposing a large restructuring of the foundations of high energy physics, which at present still awaits experimental confirmation. Thus, in particular, the new developments left pretty much untouched the problem of a conceptual understanding of the foundations of quantum field theory. Of course, there were at various times attempts to axiomatize quantum field theory in a way that would be palatable for mathematicians (algebraic and constructive quantum field theory, for instance). Such attempts unfortunately fell short of incorporating the full complexity of quantum field theory, especially with respect to the issue of perturbative renormalization. On the other hand, at present perturbative quantum field theory still remains the most accurate instrument for theoretical predictions in elementary particle physics and this impressive agreement between theory and nature calls for the best possible conceptual understanding of its foundational principles.
Bogoliubov–Parasiuk preparation {#bogoliubovparasiuk-preparation .unnumbered}
-------------------------------
The Bogoliubov–Parasiuk preparation, or BPHZ method (for Bogoliubov–Parasiuk–Hepp–Zimmermann, [@BP], [@Hepp], [@Zi]) accounts for the presence of subdivergences, simultaneously taking care of the problem of the appearance of non-local terms and the organization of subdivergences via an inductive procedure.
The BP preparation of a graph $\Gamma$, whose divergent integral we denote by $U(\Gamma)$, is given by the formal expression $$\label{BPprep}
\overline{R}( \G) = U(\G) + \sum_{\g \sbs \G} C(\g) U( \G / \g),$$ where the sum is over divergent subgraphs. The $C(\g)$ are inductively defined counterterms, obtained (in the minimal subtraction scheme) by taking the pole part (here denoted by $T$) of the Laurent expansion in $z=d-D$ of a divergent expression, $$\label{counter}
C(\G) = -T(\overline{R}( \G)) = -T\left(U(\G) + \sum_{\g \sbs \G}
C(\g) U( \G /
\g)\right).$$ The renormalized value of $\Gamma$ is then given by the formula $$\label{Rgamma}
R(\G) = \overline{R}( \G) +C(\G) =U(\G) +C(\G) + \sum_{\g \sbs \G}
C(\g) U( \G /
\g).$$
Before continuing with the physics, we need to introduce some algebraic notions that will be useful in the rest of the paper.
Hopf algebras and affine group schemes {#hopf-algebras-and-affine-group-schemes .unnumbered}
--------------------------------------
While affine schemes are the geometric manifestation of commutative algebras, affine group schemes are the geometric counterpart of commutative Hopf algebras. The theory of affine group schemes is developed in SGA 3 [@SGA3].
Consider a commutative Hopf algebra $\Hc$ over a field $k$, which we assume here of characteristic zero. Thus, $\Hc$ is a commutative algebra with unit over $k$, endowed with a (not necessarily co-commutative) coproduct $\Delta: \Hc \to
\Hc\otimes_k \Hc$, a counit $\ve: \Hc \to k$, which are $k$-algebra morphisms and an antipode $S: \Hc \to \Hc$ which is a $k$-algebra antihomomorphism. These satisfy the “co-rules” $$\label{corules}
\begin{array}{ll}
(\Delta \otimes id)\Delta = (id\otimes \Delta)\Delta & : \Hc \to
\Hc\otimes_k \Hc \otimes_k \Hc , \\[2mm]
(id\otimes \ve)\Delta =id = (\ve \otimes id)\Delta & : \Hc \to \Hc , \\[2mm]
m (id \otimes S)\Delta = m (S\otimes id) \Delta = 1\,\ve & : \Hc \to \Hc,
\end{array}$$ where we used $m$ to denote multiplication in $\Hc$.
One then lets $G=\,{\rm Spec}\,\Hc$ be the set of prime ideals of the commutative $k$-algebra $\Hc$, with the Zariski topology. The Zariski topology is too coarse to fully recover the “algebra of coordinates” $\Hc$ from the topological space $\Sp(\Hc)$, but one recovers it through the data of the structure sheaf, [[*i.e.*]{} ]{}by considering global sections of the “sheaf of functions” on $\Sp(\Hc)$.
Since $\Hc$ is a commutative $k$-algebra, $G=\Sp(\Hc)$ is an affine scheme over $k$, while the additional structure given by the co-rules endow $G=\Sp(\Hc)$ with a product operation, a unit, and an inverse.
More precisely, one can view such $G$ as a functor that associates to any unital commutative algebra $A$ over $k$ a group $G(A)$, whose elements are the $k$-algebra homomorphisms $$\phi \,: \Hc \to A\,,\quad \phi(x\,y)= \phi(x) \phi(y) \qq
x,y\in \Hc\,, \quad\phi(1)=1\,.$$ The product in $G(A)$ is given as the dual of the coproduct, by $$\label{dualprod}
\phi_1\,\star\,\phi_2\,\,(x)=\,\langle \phi_1\otimes
\phi_2\,,\;\Delta(x)\rangle\,.$$ The inverse and the unit of $G(A)$ are determined by the antipode and the co-unit of $\Hc$. The co-rules imply that these operations define a group structure on $G(A)$. The resulting covariant functor $$A \,\rightarrow G(A)$$ from commutative algebras to groups is representable (in fact by $\Hc$). The functor $G$ obtained in this way is called an [*affine group scheme*]{}. Conversely, any covariant representable functor from the category of commutative algebras over $k$ to groups, is an affine group scheme $G$, represented by a commutative Hopf algebra, uniquely determined up to canonical isomorphism.
Some simple examples of affine group schemes:
- The multiplicative group $G={\mathbb{G}}_m$ is the affine group scheme obtained from the Hopf algebra $\Hc=k[t,t^{-1}]$ with coproduct $\Delta(t)=t\otimes
t$.
- The additive group $G={\mathbb{G}}_a$ corresponds to the Hopf algebra $\Hc=k[t]$ with coproduct $\Delta(t)=t\otimes 1 + 1 \otimes t$.
- The affine group scheme $G=\GL_n$ corresponds to the Hopf algebra $$\Hc=k[x_{i,j},t]_{i,j=1,\ldots,n} / \det(x_{i,j})t-1,$$ with coproduct $\Delta(x_{i,j})= \sum_k x_{i,k}\otimes x_{k,j}$.
The latter example is quite general. In fact, if $\Hc$ is finitely generated as an algebra over $k$, then the corresponding affine group scheme $G$ is a linear algebraic group over $k$, and can be embedded as a Zariski closed subset in some $\GL_n$. Moreover, in the more general case, one can find a collection $\Hc_i\subset \Hc$ of finitely generated algebras over $k$ such that $\Delta(\Hc_i)\subset \Hc_i\otimes \Hc_i$, $S(\Hc_i)\subset \Hc_i$, for all $i$, and such that, for all $i,j$ there exists a $k$ with $\Hc_i \cup \Hc_j \subset \Hc_k$, and $\Hc=\cup_i \Hc_i$. In this case, one obtains linear algebraic groups $G_i=\Sp(\Hc_i)$ such that $$\label{Gprojlim}
G=\varprojlim_i G_i.$$ Thus, in general, an affine group scheme is a projective limit of linear algebraic groups. If the $G_i$ are unipotent, then $G$ is a pro-unipotent affine group scheme.
The Lie algebra ${{\mathfrak{g}}}(k)=\Lie\, G(k)$ is given by the set of linear maps $ L\,:\Hc \to k$ satisfying $$\label{Liescheme}
L(X\,Y)=\, L(X)\,\ve(Y) +\, \ve(X)\, L(Y)\,,\quad \forall X\,,Y
\in \Hc\,,$$ where $\ve$ is the co-unit of $\Hc$, playing the role of the unit in the dual algebra. Equivalently, ${{\mathfrak{g}}}=\hbox{Lie} \ G$ is a covariant functor $$\label{LieGA}
A \,\rightarrow {{\mathfrak{g}}}(A)\,,$$ from commutative $k$-algebras to Lie algebras, where ${{\mathfrak{g}}}(A)$ is the Lie algebra of linear maps $ L\,:\Hc \to A$ satisfying .
Hopf algebra of Feynman graphs and diffeographisms {#hopf-algebra-of-feynman-graphs-and-diffeographisms .unnumbered}
--------------------------------------------------
The Kreimer Hopf algebra of [@DK1] is based on rooted trees, which organize the hierarchy of subdivergences in a given graph. The Hopf algebra depends on the particular physical theory $\cT$ through the use of trees whose vertices are decorated by the divergence free Feynman graphs of the theory ([[*cf.*]{} ]{}[@DK1] [@CK3]). In the work of Connes–Kreimer [@CK1] this Hopf algebra was refined to a Hopf algebra $\Hc(\sT)$, also dependent on the physical theory $\sT$ by construction, which is directly defined in terms of Feynman graphs.
The CK Hopf algebra is the free commutative algebra over $k=\C$ generated by one particle irreducible (1PI) graphs $\Gamma(p_1,\ldots,p_n)$, where $\Gamma$ is not a tree. A graph $\Gamma$ is 1PI if it cannot be disconnected by the removal of a single edge. Here one considers graphs endowed with external momenta $(p_1,\ldots,p_n)$. To account for this external structure one considers distributions $\sigma\in C_c^{-\infty}(E_\Gamma)$ for $$E_\Gamma=\,\left\{ (p_i)_{i=1 , \ldots , N} \ ; \
\sum \, p_i = 0 \right\},$$ and the symmetric algebra $\Hc=\,S(C_c^{-\infty}(\cup E_\Gamma))$, with $\cup E_\Gamma$ the disjoint union.
The coproduct is given by a formula that reflects the BP preparation , namely, it is given on generators by the expression $$\label{CKcoprod}
\Delta(\G) = \G \ot 1 + 1 \ot \G + \sum_{\g \sbs \G} \g_{(i)} \ot \G /
\g_{(i)}.$$ Here the sum is over divergent subgraphs $\gamma \sbs \G$ and $\G
/\g$ denotes the graph obtained by contracting $\g$ to a single vertex. In the notation $\gamma_{(i)}$ accounts for the fact that one has to specify how to assign the external structure to $\gamma$, depending on the type of the corresponding vertex in $\G /
\g_{(i)}$, [[*cf.*]{} ]{}[@CK1].
Up to passing to the Hopf subalgebra constructed on 1PI graphs with fixed external structure, one can reduce to a Hopf algebra $\Hc(\sT)$ that is finite dimensional in each degree, where the degree is defined on 1PI graphs by the loop number. There is an affine group scheme associated to this Hopf algebra $\Hc(\sT)$. This is called the group of [*diffeographisms*]{} $G={\rm
Difg}(\sT)$ of the physical theory. It is a pro-unipotent affine group scheme.
The reason for the terminology lies in the fact that ${\rm
Difg}(\sT)$ has a close relation to the group of formal diffeomorphisms of the complexified coupling constants of the theory. In the simplest case this group is the group ${\rm
Diff}(\C)$ of formal diffeomorphisms of the complex line tangent to the identity. The latter corresponds to the Hopf algebra $\Hc_{\text{diff}}$ whose generators $a_n$ are obtained by writing formal diffeomorphisms as $ \varphi(x) = x + \sum_{n\geq 2}
a_n(\varphi)\, x^n $, and with coproduct $\lgl \D a_n \, , \,
\vp_1 \ot \vp_2 \rgl = a_n (\vp_2 \circ \vp_1)$. A Hopf algebra homomorphism is obtained by writing the effective coupling constant as a formal power series $g_{{\rm eff}}(g) = g +
\sum_{n\geq 2} \alpha_n \, g^n$, where all the coefficients $\alpha_n$ are finite linear combinations of products of graphs, $\alpha_{n} \in \Hc$, for all $n\geq 1$ and mapping $a_n \mapsto \alpha_n$, [[*cf.*]{} ]{}[@CK1].
Birkhoff factorization and renormalization {#SectBirk}
==========================================
Suppose given a complex Lie group $G(\C)$ and a smooth simple curve $C\sbs {{\mathbb P}}^1 (\C)$, with $C^\pm$ the two complementary regions, with $\infty\in C^-$. For a given loop $\g : C \to
G(\C)$, the problem of Birkhoff factorization asks whether there exist holomorphic maps $\g_{\pm} : C_{\pm} \to G(\C)$, such that $$\label{Birk}
\g \, (z) = \g_- (z)^{-1} \, \g_+ (z) \qquad z \in C .$$
This procedure of factorization of Lie group valued loops became well known in algebraic geometry because of its use in the Grothendieck–Birkhoff decomposition [@Gro1] of holomorphic vector bundles on the sphere ${{\mathbb P}}^1(\C)$. In this case, the Lie group is $\GL_n(\C)$ and a weaker form of holds, whereby loops factor as $$\label{BirkL}
\gamma(z) = \gamma_-(z)^{-1}\, \lambda(z)\, \gamma_+(z),$$ where $ \lambda(z)$ is a diagonal matrix with entries $(z^{k_1},z^{k_2},\cdots,z^{k_n})$. The Grothendieck–Birkhoff decomposition hence states that a holomorphic vector bundle on ${{\mathbb P}}^1(\C)$ can be described as $E = L^{k_1} \op \ldots \op L^{k_n}$, where the line bundles $L^{k_i}$ have Chern class $c_1 \, (L^{k_i})
=k_i$. This corresponds to the Birkhoff decomposition when $c_1 \, (L^{k_i}) = 0$.
From a more analytic viewpoint ([[*cf.*]{} ]{}[[*e.g.*]{} ]{}[@Boj]), the Birkhoff factorizations or can be viewed as a (homogeneous) [*transmission problem*]{}, which can be formulated in terms of systems of singular integral equations, with various regularity assumptions. Such transmission problems can be recast in the context of the theory of Fredholm pairs, obtained by considering the spaces of boundary values, on a simple closed curve $C$, of sections of holomorphic vector bundles on ${{\mathbb P}}^1(\C)$.
BPHZ as a Birkhoff factorization {#bphz-as-a-birkhoff-factorization .unnumbered}
--------------------------------
One of the key results of the Connes–Kreimer theory of perturbative renormalization [@CK1] [@CK2] is a reformulation of the BPHZ procedure as a Birkhoff factorization in the pro-unipotent Lie group $G(\C)$ associated to the affine group scheme $G={\rm Difg}(\sT)$.
Unlike the case of $\GL_n$, where the Birkhoff decomposition only holds when $k_i=0$, in the case of interest for renormalization one always has a factorization . This follows from a result of Connes–Kreimer, which we recall in Proposition \[BirkHopf\] below. For the general case where $G$ is the pro-unipotent affine group scheme of a Hopf algebra that is graded in positive degree and connected, the result shows that a factorization of the form always exists. The result, in fact, provides an explicit recursive formula, in Hopf algebra terms, which determines both terms in the factorization.
In this setup, the Lie group $G(\C)$ is the set of complex points of an affine group scheme $G$, whose commutative Hopf algebra $\cH$ is graded in positive degrees $\cH=\cup_k \cH_k$ and connected ([[*i.e.*]{} ]{}the only elements of degree $0$ in $\cH$ are the scalars).
We let $K=\C(\{ z \})$ be the field of Laurent series convergent in some neighborhood of the origin ([[*i.e.*]{} ]{}germs of meromorphic functions at the origin) and ${{\mathcal O}}=\C\{ z \}$ be the ring of convergent power series, and we let $\Qc=\,z^{-1}\,\C([z^{-1}])$, with $\tilde\Qc=\C([z^{-1}])$ the corresponding unital ring. Then a loop $\gamma\,:C\to G$, for $C$ an infinitesimal circle around the origin, is equivalently described by a homomorphism $\phi: \cH
\to K$, [[*i.e.*]{} ]{}by a point in $G(K)$. Because the group structure on $G$ corresponds to the co-rules of the Hopf algebra $\cH$, the product of loops $\gamma(z)=\gamma_1(z)\,\gamma_2(z)$, for $z\in
C$, corresponds to $\phi= \phi_1\star\phi_2$ (dual to the coproduct in $\cH$) and the inverse $ z\mapsto \gamma(z)^{-1}$ to the antipode $\phi \circ S$.
For $z=0 \in C^+$, the condition that the loop $\gamma$ extends to a holomorphic function $\gamma\,:P_1 (\Cb)\backslash\{ 0 \}\to G$ is equivalent to the condition that the homomorphism $\phi$ lies in $G(\tilde\Qc)=\{
\phi\,,\phi({\mathcal H})\subset\;\tilde\Qc \}$, while the condition that $\gamma(0)$ is finite translates in the condition that $\phi$ belongs to $G({{\mathcal O}})=\{ \phi\,, \phi({\mathcal
H})\subset\;{{\mathcal O}}\}$. The normalization condition $\gamma(\infty)=1$ translates algebraically into the condition $\ve_-\circ \phi=\,\ve$, where $\ve_-$ is the augmentation in the ring $\tilde\Qc$ and $\ve$ is the augmentation (co-unit) of $\Hc$. This dictionary shows how interpreting affine group schemes as functors of unital commutative algebras to groups provides a very convenient language in which to reformulate the problem of Birkhoff factorization.
\[BirkHopf\] [*([@CK1])*]{} Let $\cH$ be a Hopf algebra that is graded in positive degree and connected, and $G$ the corresponding affine group scheme. Then any loop $\gamma: C \to G(\C)$ admits a Birkhoff factorization . An explicit recursive formula for the factorization is given, in terms of the corresponding homomorphism $\phi: \cH \to \C(\{ z \})$, by the expressions $$\label{Hbirkhoff1}
\phi_-(X)=-T\left(\phi(X)+\sum\phi_-(X^\prime)
\phi(X^{\prime\prime}) \right)$$ and $$\label{Hbirkhoff2}
\phi_+(X)=\phi(X)+\phi_-(X)+\sum\phi_-(X^\prime)
\phi(X^{\prime\prime}),$$ where $T$ is the projection along ${{\mathcal O}}$ to the augmentation ideal of $\tilde\Qc$ (taking the pole part), and $X'$ and $X''$ denote the terms of lower degree in the coproduct $\Delta(X)= X \otimes 1 + 1 \otimes X + \sum X^\prime
\otimes X^{\prime\prime}$, for $X \in{\mathcal H}$.
Applied to the Hopf algebra $\cH(\sT)$ of Feynman graphs, with $G={\rm
Difg}(\sT)$, the formulae and yield the counterterms and the renormalized values in the BPHZ renormalization procedure.
Mass parameter, counterterms, and the renormalization group {#MassSect .unnumbered}
-----------------------------------------------------------
In DimReg, when analytically continuing the Feynman graphs to complex dimension, in order to preserve the dimensionality of the integrand in physical units, one needs to replace the momentum space integration $d^{D-z}k$ by $\mu^z d^{D-z} k$, where $\mu$ is a mass parameter, so that the resulting quantity has the correct dimensionality of (mass)$^D$. This introduces a dependence on the parameter $\mu$ in the loop $\gamma_\mu(z)$ describing the unrenormalized theory. The behavior of a renormalizable theory under rescaling of the mass parameter $\mu \mapsto e^t\mu$, for $t\in \R$, was analyzed in [@tH].
An important result, which will play a crucial role in our geometric formulation in Section \[GalSect\], is that [*the counterterms do not depend on the mass parameter*]{} $\mu$ ([[*cf.*]{} ]{}[@Collins] §5.8 and §7.1). This result translates in terms of the Birkhoff factorization to the condition that the negative part $\gamma_{\mu^-}(z)$ of the factorization $\g_{\mu} (z) = \g_{\mu^-}
(z)^{-1} \, \g_{\mu^+} (z)$ satisfies $$\frac{\partial}{\partial \mu} \, \g_{\mu^-} (z) = 0 \, . \label{gammamu-}$$
The effect of scaling the mass parameter on the loop $\gamma_\mu(z)$ is instead described by the action of the 1-parameter group of automorphisms generated by the grading by loop number. Namely, if $\theta_t$ denotes the 1-parameter group with infinitesimal generator $\frac{d}{dt}\theta_t|_{t=0}=Y$, where $Y$ is the grading by loop number, we have $$\label{thetagamma}
\gamma_{e^t\mu}(z) =\theta_{tz}(\gamma_\mu(z)) , \ \ \ \ \forall t\in
\R,$$ and for all $z$ in an infinitesimal punctured neighborhood $\Delta^*$ of the origin $z=d-D=0$.
A well known but unpublished result of ‘t Hooft shows that the counterterms in a renormalizable quantum field theory can be reconstructed from the beta function of the theory. In the context of the Connes–Kreimer theory of perturbative renormalization, this can be seen in the following way.
The beta function here is lifted from the space of the coupling constants of the theory to the group of Diffeographisms, namely, it can be regarded as an element in the Lie algebra $\Lie G$ satisfying $$\label{beta}
\b = Y \, {\rm Res} \, \g ,$$ where $Y$ is the grading by loop number, and the residue of $\gamma$ is given by $$\label{resgamma}
{\rm Res}_{z = 0}^{} \g = - \left( \frac{\partial}{\partial u} \,
\g_- \left( \frac{1}{u} \right) \right)_{u=0}.$$ The beta function is the infinitesimal generator $\b= \frac{d}{dt}
{\bf rg}_t
|_{t=0}$ of the renormalization group $$\label{rengroup}
{\bf rg}_t ={\rm lim}_{z \ra 0} \:\g_- (z) \, \t_{t z} (\g_-
(z)^{-1}).$$ Correspondingly, the renormalized value, that is, the finite value $\g_{\mu}^+(0)$ of the Birkhoff decomposition satisfies the equation $$\label{renaction}
\mu \frac{\partial}{\partial \mu}
\,\g^+_{\mu}(0)=\, \b\,\g^+_{\mu}(0)\,.$$
A strong form of the ‘t Hooft relations, deriving the counterterms from the beta function, is given by the following result.
\[thooftrel\] [*([@CK2])*]{} The negative part of the Birkhoff factorization $\gamma_-(z)$ satisfies $$\label{gammaminussum}
\gamma_-(z)^{-1} = 1 + {\displaystyle
\sum_{n=1}^{\ify}} \ \frac{d_n}{z^n},$$ where the coefficients $d_n$ are given by iterated integrals $$\label{dncond}
d_n = \int_{s_1 \geq s_2 \geq \cdots \geq s_n \geq 0} \t_{-s_1}
(\b) \, \t_{-s_2} (\b) \ldots \t_{-s_n} (\b) \, \, ds_1 \cdots
ds_n \, .$$
The result can be formulated ([[*cf.*]{} ]{}[@CK2]) as a scattering formula $$\label{scattering}
\g_- (z) = \lim_{t \ra \ify}
e^{-t \left( \frac{\b}{z} + Z_0 \right)} \, e^{t Z_0},$$ where $Z_0$ is the additional generator of the Lie algebra of $G
\rtimes_\theta {\mathbb{G}}_a$, satisfying $$[Z_0 , X] = Y(X) \qquad \forall \, X \in \Lie\, G \, . \label{LieGstar}$$
This form of the ‘t Hooft relations and the explicit formula in terms of iterated integrals are the starting point for our formulation of perturbative renormalization in terms of the Riemann–Hilbert correspondence and for the relation to motivic Galois theory.
Before continuing with a more detailed account of these topics, we give an introductory tour of some ideas underlying the theory of motives and the Riemann–Hilbert correspondence, that we will need in order to introduce the main result of [@cmln].
The yoga of motives: cohomologies as avatars {#MotSect}
============================================
There are several possible cohomology theories that can be applied to algebraic varieties. Over a field $k$ of characteristic zero one has de Rham cohomology $H_{dR}^\cdot(X)={\mathbb{H}}^\cdot(X,\Omega_X^\cdot)$, defined in terms of sheaves of differential forms, and Betti cohomology $H^\cdot_B(X,\Q)$, which is a version of singular homology for $\sigma
X(\C)$, for an embedding $\sigma: k\hookrightarrow \C$. These are related by the periods isomorphism $$H^i_{dR}(X,k)\otimes_\sigma \C \cong H^i_B (X,\Q)\otimes_\Q \C.$$ Over a perfect field of positive characteristic there is also crystalline cohomology, while in all characteristics one can consider étale cohomology given by finite dimensional $\Q_\ell$-vector spaces $H^i_{et}(\bar X, \Q_\ell)$, where $\bar X$ is obtained by extension of scalars to an algebraic closure $\bar k$, and $\ell\neq
{\rm char}\, k$. In the smooth projective case, these have the expected properties of Poincaré duality, Künneth isomorphisms, etc. Moreover, étale cohomology provides interesting $\ell$-adic representations of $\Gal(\bar k/k)$. There are comparison isomorphisms $$H^i_B(X,\Q)\otimes_\Q \Q_\ell \cong H^i_{et} (\bar X,\Q_\ell).$$
The natural question is then what type of information, such as maps or operations on one cohomology, can be transferred to the other ones. This gave rise to the idea, proposed by Grothendieck, of the existence of a “universal cohomology theory” with realization functors to all the known cohomology theories for algebraic varieties. He called this the theory of [*motives*]{}.
A metaphor [@Gro3] justifying the terminology is provided by music scores, some of which (such as Bach’s “Art of the fugue") are not written for any particular instrument. They are just the motive, which in turn can be realized on different musical instruments. Another powerful metaphor is provided by the notion of avatar in Hindu philosophy, which expresses the idea of a single entity manifesting itself in manifold incarnations (the ten avatars of Vishnu).
We will present here only a very short overview of some ideas and results about motives, following [@De3], [@Man2], [@Se1], and [@Blo], [@dg], [@Gon], [@Le]. We start first by recalling some general algebraic formalism we will need in the following.
Tannakian categories {#tannakian-categories .unnumbered}
--------------------
The basis for a Galois theory of motives lies in a suitable categorical formalism. This was first proposed by Grothendieck, who used the term Galois–Poincaré categories (or rigid tensor categories), and was then developed by Saavedra [@Saa], who introduced the now currently adopted terminology of Tannakian categories, and by Deligne–Milne [@DeMi] ([[*cf.*]{} ]{}also the more recent [@De2]).
It is well known that there are many deep analogies between the theory of coverings of topological spaces and Galois theory. The analogy starts with the observation that, in cases where the covering spaces are defined by algebraic equations, the Galois symmetries of the equation actually correspond to deck transformations of the covering space.
Grothendieck brought this initial simple analogy to far reaching consequences. He developed a common formalism where fundamental groups (of a space, a scheme, or much more generally a topos) and Galois groups both fit naturally. The idea is that, in this very general setting, the group always arises as the group of automorphisms of a fiber functor on a suitable “category of coverings”. The theory of the (pro-finite) fundamental groups is based on the existence of a fiber functor from a certain category $\sC$ of finite étale covers of a connected scheme $S$, with values in finite sets. Then such functor $\omega$ yields an equivalence of categories between $\sC$ and $G$-sets for $G=\Aut(\omega)$ a pro-finite group. This yields a profinite completion of the fundamental group. For $S=\Spec(K)$, it gives Galois theory, thus effectively bringing fundamental groups and Galois groups within the same general formalism.
This is the fundamental idea that guided the development of a motivic Galois theory. The latter appeared as a “linear” version of the general formalism described above, where the fiber functor is a faithful and exact tensor functor with values in vector spaces (instead of finite sets), and the Galois group is the affine group scheme $G=\Aut^\otimes(\omega)$.
More precisely, an abelian category is a category to which the tools of homological algebra apply, that is, a category where the sets of morphisms are abelian groups, there are products and coproducts, kernels and cokernels always exist and satisfy the same basic rules as in the category of modules over a ring. A tensor category over a field $k$ of characteristic zero is a $k$-linear abelian category $\T$ endowed with a tensor functor $\otimes: \T\times \T \to \T$ satisfying associativity and commutativity laws defined by functorial isomorphisms, and with a unit object. Moreover, for each object $X$, there exists a dual $X^\vee$ and maps $ev: X\otimes X^\vee \to 1$ and $\delta: 1 \to X
\otimes X^\vee$, such that the composites $(ev\otimes 1) \circ
(1\otimes \delta)$ and $(1\otimes ev) \circ (\delta \otimes 1)$ are the identity, with an identification $k \simeq \End(1)$.
A [*Tannakian category*]{} $\T$ over $k$ is a tensor category endowed with a fiber functor, namely a functor $\omega$ to finite dimensional vector spaces ${\rm Vect}_K$, for $K$ an extension of $k$, satisfying $\omega(X)\otimes \omega(Y)\simeq \omega(X\otimes
Y)$ compatibly with associativity commutativity and unit. (A more general formulation can be given with values in locally free sheaves over a scheme, see [@De2]). A [*neutral*]{} Tannakian category $\T$ has a ${\rm Vect_k}$-valued fiber functor $\omega$. In this case, the main result is that the fiber functor $\omega$ induces an equivalence of categories between $\T$ and the category ${\rm Rep}_G$ of finite dimensional linear representations of a uniquely determined affine group scheme $G=\Aut^\otimes(\omega)$, given by the automorphisms of the fiber functor.
A $k$-linear abelian category $\T$ is semi-simple if there exists $A\subset Ob(\T)$ such that all objects $X$ in $A$ are simple (namely $\Hom(X,X)\simeq k$), with $\Hom(X,Y)=0$ for $X\neq Y$ in $A$, and such that every object of $\T$ is isomorphic to a direct sum of objects in A. The affine group scheme $G$ of a neutral Tannakian category is pro-reductive if and only if the category is semi-simple.
As an example, one can consider the category of finite dimensional complex linear representations of a group. It is not hard to see what is in this case the structure of neutral Tannakian category, with fiber functor the forgetful functor to complex vector spaces. The affine group scheme determined by this neutral Tannakian category is called the “algebraic hull" of the group. In the case of the group $\Z$, the algebraic hull is an extension of $\hat\Z$, with the corresponding commutative Hopf algebra given by $\Hc=\C
[e(q),t]$, for $q\in \C/\Z$, with the relations $e(q_1+q_2)=e(q_1)e(q_2)$ and the coproduct $\Delta(e(q))=e(q)\otimes e(q)$ and $\Delta(t)=t\otimes 1 + 1
\otimes t$.
The non-neutral case where $\omega$ takes values in ${\rm Vect}_K$ for some extension of $k$, or the more general case of locally free sheaves over a scheme, can also be identified with a category of representations, but now the group $G$ is replaced by a groupoid (Grothendieck’s Galois–Poincaré groupoid). This corresponds to the fact that, even in the original case of fundamental groups of topological spaces, it is more natural to work with the notion of fundamental groupoid, rather than with the base point dependent fundamental group. For our purposes, however, it will be sufficient to work with the more restrictive notion of neutral Tannakian category.
Gauge groups and categories {#gauge-groups-and-categories .unnumbered}
---------------------------
In [@De2], §7, Deligne gives a characterization of Tannakian categories, over a field $k$ of characteristic zero, as tensor categories where the dimensions are positive integers. The dimension of $X\in \T$ is defined in this context as $\Tr(1_X)$, where $\Tr(f) =ev\circ \delta\,(f)$.
This characterization is very close to results developed via different techniques by Doplicher and Roberts in the context of algebraic quantum field theory, [@DR]. Their motivation was to derive the existence of a global compact gauge group, given the local observables of the theory. The group is obtained from a monoidal $C^*$-category where the objects are endomorphisms of certain unital $C^*$-algebras and the arrows are intertwining operators between these endomorphisms. They obtain a characterization of those monoidal $C^*$-categories that are equivalent to the category of finite dimensional continuous unitary representations of a compact group, unique up to isomorphism. Though the context and the techniques employed in the proof are different, the result has a flavor similar to the relation between Tannakian categories and affine group schemes. In their proof, a characterization analogous to the one of [@De2], §7 of the integer dimensions also plays an important role.
Pure and mixed motives {#pure-and-mixed-motives .unnumbered}
----------------------
The first constructions of a category of motives proposed by Grothendieck covers the case of smooth projective varieties. The corresponding motives form a $\Q$-linear abelian category $\cM_{pure}(k)$ of [*pure motives*]{}. There is a contravariant functor assigning a motive to a variety $$\label{hX}
X \mapsto h(X)=\oplus_i h^i(X).$$ If $h^j=0$, for all $j\neq i$, the motive is [*pure of weight $i$*]{}. This way a pure motive can be thought of as a “direct summand of an algebraic variety”. The morphisms $\Hom(X,Y)$ in the category of motives are given by [*correspondences*]{}, namely algebraic cycles in the product $X\times Y$ of codimension equal to the dimension of $X$, modulo a suitable equivalence relation. Different choices of the notion of equivalence for algebraic cycles produce variants of the theory, ranging from the coarsest numerical equivalence to the finest rational equivalence (Chow groups). The objects of the category also include kernels of projectors, namely of idempotents in $\Hom(X,Y)$. Thus, for $p=p^2\in \End(X)$ and $q=q^2\in \End(Y)$, one takes $\Hom((X,p),(Y,q))=q \Hom(X,Y) p$.
One also adds to the objects the Tate motive $\Q(1)$, which is the inverse of $h^2({{\mathbb P}}^1)$. This is a pure motive of weight $-2$. The category is endowed with a tensor product $\otimes$ and a unit $\Q(0)=h(pt)$. The Tate objects $\Q(n)$ satisfy the rule $\Q(n+m)\cong \Q(n)\otimes \Q(m)$.
Grothendieck formulated a set of [*standard conjectures*]{} about pure motives, which are at present still unproven. Assuming the standard conjectures, the category of pure motives is a neutral Tannakian category, with fiber functors given by Betti cohomology (characteristic zero case). Thus, the category of pure motives is equivalent to the category of representations $Rep_G$ of an affine group scheme $G$. This group is called the [*motivic Galois group*]{}. The category of pure motives is conjecturally semi-simple, hence for pure motives $G$ is pro-reductive.
When one considers certain subcategories of the category of motives, one obtains a corresponding Galois group, which is a quotient of the original $G$. For instance, if the subcategory is generated by a single $X$, one obtains a quotient $G_X$, whose identity component is the Mumford–Tate group of $X$. The subcategory of pure Tate motives, generated by $\Q(1)$ has as motivic Galois group the multiplicative group ${\mathbb{G}}_m$.
Some of the first unconditional results about motives were obtained in [@Man2]. In general, a serious technical obstacle in the development of the theory of motives, which accounts for the fact that, decades after its conception, the theory is still largely depending on conjectures, is the fact that not enough is known about algebraic cycles. The situation gets even more complicated when one wishes to consider more general algebraic varieties, which need not be smooth projective. This leads to the notion of [*mixed motives*]{} with $\M_{pure}(k)\subset
\M_{mix}(k)$.
Over a field of characteristic zero (where one has resolution of singularities), one can always write such $X$ as a disjoint union of $X_i - D_i$, where the $X_i$ are smooth projective and the $D_i$ are lower dimensional. Thus, one can assign to $X$ a virtual object in a suitable Grothendieck group of algebraic varieties; however, if one wants a theory that satisfies the main requirements of a category of motives, including the fact of providing a universal cohomology theory (via the Ext functors), the construction of such a category of mixed motives remains a difficult task.
The main properties for a category of mixed motives are that it should be a $\Q$-linear tensor category containing the Tate objects $\Q(n)$ with the usual properties, endowed with a functor $X\mapsto h(X)$ that assigns motives to algebraic varieties, with properties like Künneth isomorphisms. Moreover, the Ext functors in this category of mixed motives define a “motivic cohomology” $$\label{motH}
E^{i,j}_2 = \Ext^i (\Q(0),h^j(X)\otimes \Q(n)) \Rightarrow H^{i+j}_{mot}(X,\Q(n)).$$ One expects also this motivic cohomology to come endowed with Chern classes from algebraic $K$-theory. In fact, if one uses the decomposition $K_n(X)\otimes \Q = \oplus_j K_n(X)^{(j)}$, where the Adams operation $\Psi_k$ acts on $K_n(X)^{(j)}$ as $k^j$, then one expects isomorphisms given by Chern classes $$ch^j: K_n(X)^{(j)} \stackrel{\simeq}{\to}
H_{mot}^{2j-n}(X,\Q(j)).$$ Such motivic cohomology will be universal with respect to all cohomology theories for algebraic varieties satisfying certain natural properties (Bloch–Ogus axioms). Namely, for any such cohomology $H^*(\cdot ,\Gamma(*))$ there will be a natural transformation $H^*_{mot}(\cdot, \Z(*)) \to H^*(\cdot,\Gamma(*))$, compatible with the above isomorphisms. Mixed motives have increasing weight filtrations preserved by the realizations to cohomology theories. More generally, instead of working over a field $k$, one can consider a category $\M_{mix}(S)$ of motives (or “motivic sheaves”) over a scheme $S$. In this case, the functors above are natural in $S$ and to a map of schemes $f: S_1 \to S_2$ there correspond functors $f^*$, $f_*$, $f^{!}$, $f_{!}$, behaving like the corresponding functors of sheaves.
The motivic Galois group for mixed motives will then be an extension of the pro-reductive motivic Galois group of pure motives by a pro-unipotent group. The pro-unipotent property reflects the presence of the weight filtration on mixed motives.
Though, at present, there is not yet a general construction of such a category of mixed motives $\M_{mix}(S)$, there are constructions of a triangulated tensor category $\cD\M(S)$, which has the right properties to be the bounded derived category of the category of mixed motives. The constructions of $\cD\M(S)$ due to Levine [@Le] and Voevodsky [@Vo] are known to be equivalent. In general, given a construction of a triangulated tensor category, one can extract from it an abelian category by considering the [*heart of a $t$-structure*]{}. A caveat with this procedure is that it is not always the case that the given triangulated tensor category is in fact the bounded derived category of the heart of a $t$-structure. The available constructions, in any case, are obtained via this general procedure of $t$-structures developed in [@BBD], which can be summarized as follows. A triangulated category ${\mathcal D}$ is an additive category with an automorphism $T$ and a family of distinguished triangles $X \to Y \to Z \to T(X)$, satisfying suitable axioms (which we do not recall here). We use the notation ${\mathcal D}^{\geq n} = {\mathcal D}^{\geq 0}[-n]$ and ${\mathcal D}^{\leq n} = {\mathcal D}^{\leq 0}[-n]$, with $X[n]=T^n(X)$ and $f[n]=T^n(f)$. A $t$-structure consists of two full subcategories ${\mathcal D}^{\leq 0}$ and ${\mathcal D}^{\geq 0}$ with the properties: ${\mathcal D}^{\leq -1} \subset {\mathcal D}^{\leq 0}$ and ${\mathcal D}^{\geq 1} \subset {\mathcal D}^{\geq 0}$; for all $X\in {\mathcal D}^{\leq 0}$ and all $Y\in {\mathcal D}^{\geq 1}$ one has $\Hom_{\mathcal D} (X,Y)=0$; for all $Y\in {\mathcal D}$ there exists a distinguished triangle as above with $X\in {\mathcal D}^{\leq 0}$ and $Z\in {\mathcal D}^{\geq 1}$. The heart of the t-structure is the full subcategory ${\mathcal D}^0= {\mathcal D}^{\leq 0}\cap
{\mathcal D}^{\geq 0}$. It is an abelian category. This type of construction may be familiar to physicists in the context of mirror symmetry, where continuous families of hearts of $t$-structures play a role in [@Dou].
For our purposes, we will be mostly interested in the full subcategory of Tate motives. The triangulated category of [*mixed Tate motives*]{} $\sD\M\sT(S)$ is then defined as the full triangulated subcategory of $\sD\M(S)$ generated by the Tate objects. It is possible to define on it a $t$-structure whose heart gives a category of mixed Tate motives $\M\sT_{mix}(S)$, provided the Beilinson–Soulé vanishing conjecture holds, namely when $$\label{BSconj}
\Hom^j(\Q(0),\Q(n))=0, \ \ \ \text{ for } n>0, j\leq 0.$$ where $\Hom^j(M,N)=\Hom(M,N[j])$. The conjecture is known to hold in the case of a number field, where one has $$\label{ExtK}
\Ext^1(\Q(0),\Q(n))= K_{2n-1}(k)\otimes \Q$$ and $\Ext^2(\Q(0),\Q(n))=0$. Thus, in this case it is possible to extract from the triangulated tensor category a Tannakian category $\M\sT_{mix}(k)$ of mixed Tate motives, with fiber functor $\omega$ to $\Z$-graded $\Q$-vector spaces, $M
\mapsto \omega(M)=\oplus_n \omega_n(M)$ with $$\label{omegaGr}
\omega_n(M)=\Hom(\Q(n),\Gr_{-2n}^w(M)),$$ where $\Gr_{-2n}^w(M)=W_{-2n}(M)/W_{-2(n+1)}(M)$ is the graded structure associated to the finite increasing weight filtration $W$.
The motivic Galois group of the category $\M\sT_{mix}(k)$ is then an extension $G=U\rtimes {\mathbb{G}}_m$, where the reductive piece is ${\mathbb{G}}_m$ as in the case of pure Tate motives, while $U$ is pro-unipotent. By the results of Goncharov (see [@Gon], [@dg]), it is known that the pro-unipotent affine group scheme $U$ corresponds to a graded Lie algebra ${\rm Lie}\, (U)$ that is free with one generator in each odd degree $n\leq -3$.
A similar construction is possible in the case of the category $\M\sT_{mix}(S)$, where the scheme $S$ is the set of $V$-integers ${{\mathcal O}}_V$ of a number field $k$, for $V$ a set of finite places of $k$. In this case, objects of $\M\sT_{mix}({{\mathcal O}}_V)$ are mixed Tate motives over $k$ that are unramified at each finite place $v\notin V$. For $\M\sT_{mix}({{\mathcal O}}_V)$ we have $$\label{ExtOS}
\Ext^1(\Q(0),\Q(n))=\left\{ \begin{array}{ll} K_{2n-1}(k)\otimes
\Q & n\geq 2 \\[2mm] {{\mathcal O}}_V^*\otimes \Q & n=1 \\[2mm] 0 & n\leq 0.
\end{array} \right.$$ and $\Ext^2(\Q(0),\Q(n))=0$. In fact, the difference between the Ext in $\M\sT_{mix}({{\mathcal O}}_V)$ of and the Ext in $\M\sT_{mix}(k)$ of is the $\Ext^1(\Q(0),\Q(1))$ which is finite dimensional in and infinite dimensional in . The category $\M\sT_{mix}({{\mathcal O}}_V)$ is also a neutral Tannakian category, and the fiber functor determines an equivalence of categories between $\M\sT_{mix}({{\mathcal O}}_V)$ and finite dimensional linear representations of an affine group scheme of the form $U \rtimes {\mathbb{G}}_m$ with $U$ pro-unipotent. The Lie algebra $\Lie(U)$ is freely generated by a set of homogeneous generators in degree $n$ identified with a basis of the dual of $\Ext^1(\Q(0),\Q(n))$ ([[*cf.*]{} ]{}Prop. 2.3 of [@dg]). There is however no [*canonical*]{} identification between $\Lie(U)$ and the free Lie algebra generated by the graded vector space $\oplus
\Ext^1(\Q(0),\Q(n))^\vee$.
We mention the following case, which will be the one most relevant in the context of perturbative renormalization.
\[SNmotives\] [*([@dg], [@Gon])*]{} Consider the scheme $S_N={{\mathcal O}}[1/N]$ for $k=\Q(\zeta_N)$ the cyclotomic field of level $N$ and ${{\mathcal O}}$ its ring of integers. For $N=3$ or $4$, the motivic Galois group of the category $\M\sT_{mix}(S_N)$ is of the form $U \rtimes {\mathbb{G}}_m$, where the Lie algebra $\Lie(U)$ is (noncanonically) isomorphic to the free Lie algebra with one generator $e_n$ in each degree $n\leq -1$.
Hilbert’s XXI problem and the Riemann–Hilbert correspondence
============================================================
Consider an algebraic linear ordinary differential equation, in the form of a system of rank $n$ $$\label{ODE}
\frac{d}{dz} f(z) + A(z) f(z) =0$$ on some open set $U=
\bP^1(\C)\smallsetminus \{ a_1,\ldots a_r \}$, where $A(z)$ is an $n\times n$ matrix of rational functions on $U$. In particular, this includes the case of a linear scalar $n$th order differential equation. The space ${\mathcal S}$ of germs of holomorphic solutions of at a point $z_0\in U$ is an $n$-dimensional complex vector space. Moreover, given any element $\ell \in
\pi_1(U,z_0)$, analytic continuation along a loop representing the homotopy class $\ell$ defines a linear automorphism of ${\mathcal S}$, which only depends on the homotopy class $\ell$. This defines the [*monodromy representation*]{} $\rho:
\pi_1(U,z_0) \to {\rm Aut}({\mathcal S})$ of the differential system . A slightly different formulation requires not the [*Fuchsian condition*]{} ($A(z)$ has simple poles) but the weaker [*regular singular condition*]{} for . The regularity condition at a singular point $a_i \in \bP^1(\C)\smallsetminus U$ is a growth condition on the solutions, namely all solutions in any strict angular sector centered at $a_i$ have at most polynomial growth in $1/|z-a_i|$. The system is regular singular if every $a_i \in \bP^1(\C)\smallsetminus U$ is a regular singular point. The Hilbert 21st problem (or Riemann–Hilbert problem) asks whether any finite dimensional complex linear representation of $\pi_1(U,z_0)$ is the monodromy representation of a differential system with regular (or Fuchsian) singularities at the points of $\bP^1(\C)\smallsetminus U$. A solution to the Hilbert 21st problem in the regular singular case is given by Plemelj’s theorem ([[*cf.*]{} ]{}[@AnBol] §3). The argument first produces a system with the assigned monodromy on $U$, where in principle an analytic solution has no constraint on the behavior at the singularities. Then, one restricts to a [*local problem*]{} in small punctured disks $\Delta^*$ around each of the singularities, for which a system exists with the prescribed behavior of solutions at the origin. The global trivialization of the holomorphic bundle on $U$ determined by the monodromy datum yields the patching of these local problems that produces a global solution with the correct growth condition at the singularities.
From problem to correspondence {#from-problem-to-correspondence .unnumbered}
------------------------------
A modern revival of interest in Fuchsian differential equation, with a new algebraic viewpoint that slowly transformed the original Riemann–Hilbert problem into the broad landscape of the Riemann–Hilbert correspondence, was pioneered in the early 1960s by the influential paper of Yuri Manin [@Man] on Fuchsian modules. This new perspective influenced the work of Deligne [@De1] in 1970, who solved the Riemann–Hilbert problem for regular singular equations on an arbitrary smooth projective variety. In this viewpoint, if $X$ is a smooth projective variety and $U$ is a Zariski open set, with $X\smallsetminus U$ a union of divisors with normal crossing, the data of an algebraic differential system are given by a pair $(M,\nabla)$ of a locally free coherent sheaf on $U$ with a connection $\nabla : M \to
M\otimes \Omega^1_{U/\C}$, while the regular singular condition says that there exists an algebraic extension $(\bar M,
\bar\nabla)$ of the data $(M,\nabla)$ to $X$, where the extended connection $\bar\nabla : \bar M \to \bar M \otimes
\Omega^1_{X/\C}(\log D)$ has log singularities at the divisor $D$. The reconstruction argument for algebraic linear differential systems with regular singularities in terms of their monodromy representation consists then of first producing an analytic solution $(M,\nabla)$ on $U$ with the prescribed monodromy and then restricting to a local problem in punctured polydisks $\Delta^*$ around the singularities, to obtain a local extension of the form $H(z) \prod_j z_j^{B_j}$, where $H\in
\GL_n({{\mathcal O}}_{\Delta^*})$ and the $B_j$ are commuting matrices that give the local monodromy representation $\exp(2\pi i B_j)$ of $\pi_1(\Delta^*)$. An important point of the argument is to show that these local extensions can be patched together. The patching problem does not arise when $\dim U=1$, since in that case the divisor $D$ consists of isolated points. The construction is then completed by showing that the global analytic extension $(\bar M,
\bar \nabla)$ obtained this way on $X$ is equivalent to an algebraic extension.
Starting with the early 1980s, with the work of Mebkhout [@Me1] [@Me2] and of Kashiwara [@Ka1] [@Ka2], and with the development of the theory of perverse sheaves by Beilinson, Bernstein, Deligne, and Gabber [@BBD], the Riemann–Hilbert correspondence was recast in terms of an equivalence of derived categories between regular holonomic $\Dc_X$-module and perverse sheaves. A reason for introducing the language of $\Dc$-modules ([[*cf.*]{} ]{}[[*e.g.*]{} ]{}[@GeMa] §8 or [@LM] for an overview) is that this captures more information in a differential system $(M,\nabla)$, than what was possible with the previous formulations. For instance, the data $(M,\nabla)$ fit into a de Rham type complex. Also, one may want to consider different classes of solutions (smooth, holomorphic, distributional, etc). This type of extra information is taken care of by the formalism of $\Dc$-modules. Namely, a differential equation determines a module $\Mc$ over $\Dc_X$ (differential operators on $X$ with holomorphic coefficients), with solutions to the equation given by $\Hom_{\Dc_X}(\Mc,\Oc_X)$. One can alter the type of solutions by replacing $\Oc_X$ by another module $\Nc$ over $\Dc_X$, and account for the extra structure in the data $(M,\nabla)$ by considering the de Rham complex $\Mc \otimes_{\Oc_X} \Omega_X$. The condition of regular singularities can be extended to modules $\Mc$ subject to another ‘growth’ condition, related to the module structure, compatibly with a natural filtration of $\Dc_X$ (holonomic $\Dc$-modules). Then the equivalence of categories extends to an equivalence of derived categories, between regular holonomic $\Dc_X$-module and perverse sheaves.
With the regular singular hypothesis replaced by the stronger Fuchsian condition, as in Hilbert’s original formulation, counterexamples to the Riemann–Hilbert problem were later found by Bolibruch [@Bol], in the simplest case of $X={{\mathbb P}}^1(\C)$. On the other hand, one can instead relax the regular singular condition and look at classes of differential systems with irregular singularities. It is immediately clear that finite dimensional complex linear representations of the fundamental group no longer suffice to distinguish equations that can have very different analytic behavior at the singularities and equal monodromy. One can see this in a simple example, where all equations of the form $\frac{d}{dz} f(z) + \frac{1}{z^2} P\left( \frac{1}{z} \right)f(z)
=0$ have trivial monodromy, for any polynomial $P$, but they all have inequivalent behavior at the singularity $z=0$.
Thus, one needs a refinement of the fundamental group, whose finite dimensional linear representations are equivalent to (a given class of) irregular differential systems. There are different approaches to the irregular case. Since we are directly interested in the case relevant to perturbative renormalization, we might as well restrict our attention to the one dimensional setting, namely where $\dim U=1$ and $X$ is a compact Riemann surface. In fact, in our case $X={{\mathbb P}}^1(\C)$ will be sufficient, as we will be interested only in the local problem in a punctured disk $\Delta^*$. As we discuss in Section \[GalSect\] below, in physical terms $\Delta^*$ represents the space of complexified dimensions around a given integer dimension $D$ at which the Feynman integrals of the specified theory $\sT$ are divergent.
In this context, the theory that best fits our needs for the application to renormalization was developed by Martinet and Ramis [@MR], where instead of the usual fundamental group one considers representations of a [*wild fundamental group*]{}, which arises from the asymptotic theory of divergent series and differential Galois theory.
Differential Galois theory and the wild fundamental group {#DiffGalSect .unnumbered}
---------------------------------------------------------
We consider a local version of the irregular Riemann–Hilbert correspondence, in a small punctured disk $\Delta^*$ in the complex plane around a singularity $z=0$. We work in the context of differential Galois theory ([[*cf.*]{} ]{}[@vPS], [@vP]). In this setting, one works over a differential field $(K,\delta)$, such that the field of constants $k=\Ker(\delta)$ is an algebraically closed field of characteristic zero. One considers differential systems of the form $\delta f = Af$, for some $A\in {\rm End}(n, K)$.
For $k=\C$, at the formal level we are then working over the differential field of formal complex Laurent series $K=\C((z))=\C[[z]][z^{-1}]$, with differentiation $\delta=z\frac{d}{dz}$, while at the non-formal level one considers the subfield $K=\C(\{z\})$ of convergent Laurent series.
Given a differential system $\delta f = Af$, its [*Picard–Vessiot ring*]{} is a $K$-algebra with a differentiation extending $\delta$. As a differential algebra it is simple and is generated over $K$ by the entries and the inverse determinant of a fundamental matrix for the equation $\delta f = Af$. The [*differential Galois group*]{} of the differential system is given by the automorphisms of the Picard–Vessiot ring commuting with $\delta$.
The formalism of Tannakian categories, that we discussed in Section \[MotSect\] in the context of motives, reappears in the present context and allows for a description of the differential Galois group that fits in the same general picture we recalled regarding motivic Galois groups.
In fact, if we consider the set of all possible such differential systems (differential modules over $K$), these form a neutral Tannakian category, which can therefore be identified with the category of finite dimensional linear representations of a unique affine group scheme over the field $k$.
Similarly to what we discussed in the case of motivic Galois groups, any subcategory $\T$ that inherits the structure of a neutral Tannakian category in turn corresponds to an affine group scheme $G_\T$. This is the universal differential Galois group of the class of differential systems that form the category $\T$. It can be realized as the automorphisms group of the universal Picard–Vessiot ring $R_\T$. The latter is generated over $K$ by the entries and inverse determinants of the fundamental matrices of all the differential systems considered in the category $\T$.
There is therefore a clear analogy between the induced motivic Galois groups of certain subcategories of, say, the category of mixed Tate motives that we discussed in Section \[MotSect\], and the differential Galois group of certain classes of differential systems defining subcategories of the neutral Tannakian category of irregular differential systems over a differential field $K$. Our main result of [@cmln], which we discuss in Section \[GalSect\] below, shows that the theory of perturbative renormalization (in the DimReg and minimal subtraction scheme) identifies a class of differential systems (dictated by physical assumptions), whose differential Galois group is the motivic Galois group of Proposition \[SNmotives\].
The regular–singular case can be seen in this context as follows. The subcategory of regular–singular differential modules over $K=\C((z))$ is a neutral Tannakian category equivalent to $Rep_G$, where the affine group scheme $G$ is the algebraic hull $\bar\Z$ of $\Z$, generated by the formal monodromy $\gamma$. The latter is the automorphism of the universal Picard–Vessiot ring acting by $\gamma \, Z^a = \exp(2\pi i a)\, Z^a$ and $\gamma\, L = L +
2\pi i$ on the generators $\{ Z^a \}_{a\in \C}$ and $L$, which correspond, in angular sectors, to the powers $z^a$ and the function $\log(z)$ ([[*cf.*]{} ]{}[@vPS] §III, [@vP]).
When one allows for an arbitrary degree of irregularity for the differential systems $\delta f =Af$, the universal Picard–Vessiot ring of the formal theory $K=\C((z))$ is generated by elements $\{ Z^a
\}_{a\in \C}$ and $L$ as before, and by additional elements $\{ E(q)
\}_{q\in \Ec}$, where $\Ec= \cup_{\nu\in \N^\times} \Ec_\nu$, for $\Ec_\nu=
z^{-1/\nu} \C [ z^{-1/\nu} ]$. These additional generators correspond, in local sectors, to functions of the form $\exp(\int q\, \frac{dz}{z})$ and satisfy relations $E(q_1+q_2)=E(q_1)E(q_2)$ and $\delta E(q)=q E(q)$.
Correspondingly, the universal differential Galois group $\cG$ is described by an extension ${\mathcal T}\rtimes \bar\Z$, where ${\mathcal T}=\Hom(\Ec, \C^*)$ is the [*Ramis exponential torus*]{}. The algebraic hull $\bar\Z$ generated by the formal monodromy $\gamma$ acts as an automorphism of the universal Picard–Vessiot ring by the same action as above on the $Z^a$ and on $L$, and by $\gamma\, E(q) = E(\gamma q)$ on the additional generators, where the action on $\Ec$ is given by the action $ \gamma : q( z^{-1/\nu}) \mapsto q(
\exp( -2\pi i/\nu) \, z^{-1/\nu})$ of $\Z/\nu\Z$ on $\Ec_\nu$. The exponential torus acts by automorphisms of the universal Picard–Vessiot ring $\tau\,Z^a=Z^a$, $\tau\, L=L$ and $\tau\, E(q)=\tau(q) E(q)$, and the formal monodromy acts on the exponential torus by $(\gamma\tau)(q)=\tau(\gamma q)$.
Thus, at the formal level, the local irregular Riemann–Hilbert correspondence establishes an equivalence of categories between the differential modules over $K=\C((z))$ and finite dimensional linear representations of $G={\mathcal T}\rtimes \bar\Z$. Ramis’ wild fundamental group [@MR] further extends this irregular Riemann–Hilbert correspondence to the non-formal setting. In general, when passing to the non-formal level over convergent Laurent series $K=\C(\{z\})$, the universal differential Galois group acquires additional generators, which depend upon resummation of divergent series and are related to the Stokes phenomenon. However, there are specific classes of differential systems (subcategories of differential modules over $K$), for which the differential Galois group is the same over $\C((z))$ and over $\C(\{z\})$ ([[*cf.*]{} ]{}[[*e.g.*]{} ]{}Proposition 3.40 of [@vPS]). In such cases, the wild fundamental group consists only of the exponential torus and the formal monodromy. This is, in fact, the case in the class of differential systems we obtain from the theory of perturbative renormalization, hence we do not need to discuss here the more complicated case where Stokes phenomena are present, and we simply refer the interested reader to [@MR], [@vPS], and [@vP].
Cartier’s dream of a cosmic Galois group {#GalSect}
========================================
In the section “I have a dream” of [@Cart], Pierre Cartier formulated the hypothesis of the existence of a “cosmic Galois group”, closely related to the Grothendieck–Teichmüller group [@Gro2], underlying the Connes–Kreimer theory of perturbative renormalization, that would relate quantum field theory to the theory of motives and multiple zeta values.
We present in this section the main result of [@cmln], which realizes Cartier’s suggestion, by reformulating the Connes–Kreimer theory of perturbative renormalization in the form of a suitable Riemann–Hilbert correspondence.
Equisingular connections {#equisingular-connections .unnumbered}
------------------------
The first step, in order to pass to this type of geometric formulation, is to identify the loops $\gamma_\mu(z)=\gamma_{\mu,-}(z)^{-1}
\gamma_{\mu,+}(z)$ with solutions of suitable differential equations. The idea of reformulating a Birkhoff factorization problem in terms of a class of differential equations is familiar to the analytic approach to the Riemann–Hilbert problem ([[*cf.*]{} ]{}[@Boj]). In our setting, the key that allows us to pass from the Birkhoff factorization to an appropriate class of differential systems is provided by the ‘t Hooft relations in the form of Proposition \[thooftrel\] and the scattering formula , reformulated more explicitly in terms of iterated integrals.
Here the main tool is the [*time ordered exponential*]{}, formulated mathematically in terms of Chen’s iterated integrals [@Chen1], [@Chen2], also known (in the operator algebra context) as Araki’s expansional [@Araki].
We consider a commutative Hopf algebra $\cH$ that is graded in positive degree and connected, with $G$ the corresponding affine group scheme and ${{\mathfrak{g}}}=\Lie G$. We assume that $\Hc$ is, in each degree, a finite dimensional vector space. Given a ${{\mathfrak{g}}}(\C)$-valued smooth function $\a(t)$ where $t\in[a,b]\subset \R$ is a real parameter, the expansional is defined by the expression $$\label{expansional}
{\bf {\rm T}e^{\int_a^b\,\a(t)\,dt}}=\,1+\, \sum_1^\infty \int_{a\leq
s_1\leq \cdots\leq
s_n\leq b} \,\a(s_1)\cdots\,\a(s_n) \prod ds_j \,,$$ where the products are in the dual algebra $\Hc^*$ and $1\in \Hc^*$ is the unit given by the augmentation $\ve$. When paired with any element $x\in \Hc$, reduces to a finite sum, which defines an element in $G(\C)$.
The fact that, when pairing with elements in $\Hc$ one reduces to an algebraic (polynomial) case plays an important role. In particular, it is related to the fact that, for the class of differential systems we consider, the differential Galois group remains the same in the formal and in the non-formal case, and we need not take into account the possible presence of Stokes’ phenomena.
We are particularly interested in the following property of the expansional: is the value $g(b)$ at $b$ of the unique solution $g(t)\in G(\C)$ with value $g(a)=1$ of the differential equation $$\label{diffexp}
dg(t)=\,g(t)\,\a(t)\,dt\,.$$
More generally, if $(K,\delta)$ is a differential field with $K\supset
\C$ and if $g\in
G(K)$ and $g'=\delta(g)$ is the linear map $\Hc\to K$ defined as $g'(x)=\, \delta(g(x))$ for $x\in \Hc$, then the logarithmic derivative $D(g)$ is defined as the linear map $\Hc\to K$ of the form $D(g)=\,g^{-1}\star\, g'$, with the product dual to the coproduct of $\cH$. It satisfies $$\langle D(g),x\,y\rangle=\,\langle D(g),x\rangle\,\ve(y)
+\,\ve(x)\,\langle D(g),y\rangle \qq x,y \in \Hc\,,$$ hence it gives an element in the Lie algebra $D(g) \in {{\mathfrak{g}}}(K)$. We will work here with the field of convergent Laurent series $K=\C(\{z
\})$.
If we consider over $\Delta^*$ a differential system of the form $$\label{Dfomega}
Df =\omega,$$ where $\omega$ is a flat ${{\mathfrak{g}}}(K)$-valued connection, then the condition of trivial monodromy $$\label{mono}
{\bf {\rm T}e^{\int_0^1\,\g^*\omega}} =1,$$ for $\gamma\in \pi_1(\Delta^*,z_0)$, ensures the existence of a solution. In the expansional form this is given by $$\label{solution}
g(z) =\;{\bf {\rm T}e^{\int_{z_0}^z\,\omega}}\,,$$ independently of the path in $\Delta^*$ from $z_0$ to $z$.
The notion of equivalence relation that we consider for differential systems of the form is the following: two connections $\omega$ and $\omega'$ are equivalent iff they are related by a gauge transformation $h \in G({{\mathcal O}})$, with ${{\mathcal O}}\subset {K}$ the subalgebra of regular functions, $$\label{gaugetransf}
\omega' = Dh + h^{-1} \omega\, h.$$ The behavior of solutions at the singularity is the same for all equivalent connections. When we regard the solutions as $G(\C)$-valued loops, the equivalence of the connections translates to the fact that the loops have the same negative part of the Birkhoff decomposition.
So far we have not taken into account the fact that, in the case of perturbative renormalization, the loop $\gamma_\mu(z)$ that corresponds to the unrenormalized theory depends on the mass parameter $\mu$, as discussed above in Section \[MassSect\]. Because of the presence of this parameter, the geometric reformulation in terms of a class of differential systems takes place, in fact, not just on the 1-dimensional (infinitesimal) punctured disk $\Delta^*$ representing the complexified dimensions of DimReg, but on a principal ${\mathbb{G}}_m(\C)=\C^*$-bundle over $\Delta^*$.
As we discuss below, the fact that the loop $\gamma_\mu(z)$ satisfies the properties and will make it possible to treat this case, which lives naturally over a 2-dimensional space, by applying the same techniques described in Section \[DiffGalSect\] for the irregular Riemann–Hilbert correspondence over the 1-dimensional domain $\Delta^*$.
The conditions and determine a class of differential systems associated to perturbative renormalization. This is given by equivalence classes of flat [*equisingular*]{} $G(\C)$-connections, where $G={\rm Difg}(\sT)$.
Let $\pi\,:B\to \Delta$ be a principal ${\mathbb{G}}_m(\C)=\C^*$-bundle, identified with $\Delta \times \C^*$ by the non-canonical choice of a section $\sigma\,: \Delta \to B$, $\sigma(0)= y_0$. Physically, the latter corresponds to a choice of the Planck constant. Let $P=B\times G(\C)$ be the trival principal $G(\C)$-bundle, and $B^*$ and $P^*$ the restrictions to the punctured disk $\Delta^*$.
We say that the connection $\omega$ on $P^*$ is [*equisingular*]{} if it is ${\mathbb{G}}_m$-invariant and its restrictions to sections of the principal bundle $B$ that agree at $0\in \Delta$ are mutually equivalent, in the sense that they are related by a gauge transformation by a $G(\C)$-valued ${\mathbb{G}}_m$-invariant map $h$ regular in $B$.
The notion of [*equisingularity*]{} is introduced as a geometric reformulation of the properties and . In fact, the property that, when approaching the singular fiber, the type of singularity does not depend on the section along which one restricts the connection but only on the value of the section at $0\in \Delta$ corresponds to the fact that the counterterms are independent of the mass scale, as in .
Thus, we have identified a class of differential systems associated to a physical theory $\sT$, namely the equivalence classes of flat equisingular $G(\C)$-valued connections on $P$, where $G={\rm
Difg}(\sT)$. We can then proceed to investigate the Riemann–Hilbert correspondence underlying this class of differential systems.
The first step consists of writing solutions of and in expansional form through the following result, which we can view as a stronger version of the ‘t Hooft relations.
\[expans\] Let $\g_\mu(z)$ be a family of $G(\C)$-valued loops satisfying the properties and . Then there exists a unique $\beta \in {\rm Lie}\,G(\C)$ and a loop $\g_{\rm reg}(z)$ regular at $z=0$, such that $$\label{solexp}
\g_\mu(z) =\,{\bf {\rm T}e^{-\frac{1}{z}\,
\int^{-z \log\mu}_\infty\,\t_{-t}(\beta)\,dt}}\;
\t_{z\log\mu}(\g_{\rm reg}(z))\,.$$ Conversely, for any $\beta$ and regular loop $\g_{\rm reg}(z)$ the expression gives a solution to and . The Birkhoff decomposition of $\g_\mu(z)$ is of the form $$\label{gammaplusminus}
\begin{array}{ll} \g_{\mu^+}(z)= & {\bf {\rm T}e^{-\frac{1}{z}
\,\int_0^{-z \log\mu}\,\t_{-t}(\beta)\,dt}}\;
\t_{z\log\mu}(\g_{\rm reg}(z))\,, \\[3mm]
\g_-(z) = & \,{\bf {\rm
T}e^{-\frac{1}{z}\,\int_0^\infty\,\t_{-t}(\beta)\,dt}}\,.
\end{array}$$
Using the equivalent geometric formulation in terms of flat equisingular connections, one then obtains the following result.
\[connbeta\] Let $\omega$ be a flat equisingular $G(\C)$-connection. There exists a unique element $\beta \in \Lie\, G(\C)$, such that $\omega$ is equivalent to the flat equisingular connection $D\g$ for $$\label{solexpm}
\g(z,v) =\,{\bf {\rm T}e^{-\frac{1}{z}\,
\int^{v}_0\,u^Y(\beta)\,\frac{du}{u}}}\;
\in G(\C)\; \,,$$ with the integral performed on the straight path $u=t v$, $t\in[0,1]$.
Here a crucial point is the fact that the monodromies with respect to the two generators of $\pi_1(B^*)$ vanish for flat equisingular connections. As we will see in the next section, this fact will be reflected in the form of the affine group scheme associated to the category of equivalence classes of flat equisingular connections (the differential Galois group), which will only contain the part corresponding to the Ramis exponential torus and no contribution from the monodromy. The correspondence of Proposition \[connbeta\] is independent of the choice of the trivialising section $\sigma$ of $B$.
The Riemann–Hilbert correspondence {#the-riemannhilbert-correspondence .unnumbered}
----------------------------------
So far we have been working with an assigned quantum field theory $\sT$ and the corresponding affine group scheme $G={\rm
Difg}(\sT)$. We now pass to considering a universal setting, which encompasses all theories. This is achieved by considering, instead of flat equisingular $G(\C)$-connections, the category of equivalence classes of all [*flat equisingular bundles*]{}. For a specific physical theory, the corresponding category of equivalence classes of flat equisingular $G(\C)$-connections can be recovered from this more general setting by considering the subcategory of those flat equisingular bundles that are finite dimensional linear representations of $G^*=G\rtimes {\mathbb{G}}_m$. This is analogous to what happens when one specializes motivic Galois groups to sucategories of motives, or differential Galois groups to subcategories of differential systems. We describe now in detail the universal setting, with the corresponding group of symmetries and the way it specializes to a given physical theory.
The category of equivalence classes of flat equisingular bundles has as objects $\Theta=(E,\nabla)$ pairs of a finite dimensional $\Z$-graded vector space $E$ and an equisingular flat $W$-connection $\nabla$. To define the latter, we consider the vector bundle $\tilde{E}=B\times E$ with the action of ${\mathbb{G}}_m$ given by the grading and with the weight filtration defined by $W^{-n}(E)=\oplus_{m\geq n} E_m$. A $W$-connection is a connection on the restriction of $\tilde E$ to $B^*$, which is compatible with the weight filtration and induces the trivial connection on the associated graded. The connection $\nabla$ in the data above is a flat $W$-connection that satisfies the equisingular condition, that is, it is ${\mathbb{G}}_m$-invariant and the restrictions to sections $\sigma$ of $B$ with $\sigma(0)=y_0$ are all $W$-equivalent on $B$, where the equivalence relation is realized by an isomorphism of the vector bundles over $B$, compatible with the filtration and identity on the associated graded, that conjugates the connections. We consider the data $\Theta=(E,\nabla)$ as $W$-equivalence classes.
As usual, it is a bit more delicate to define morphisms than objects.
For a linear map $T: E\to E'$, consider the $W$-connections $\nabla_j$, $j=1,2$, on $\tilde{E'}\oplus \tilde{E}$ of the form $$\label{nablaj}
\nabla_1 = \left( \begin{matrix}\nabla' &0 \cr 0 &\nabla \cr
\end{matrix} \right) \ \ \ \text{ and } \ \ \
\nabla_2 = \left( \begin{matrix}\nabla' &T\,\nabla-\,\nabla'\,T
\cr 0 &\nabla \cr \end{matrix} \right),$$ where $\nabla_2$ is the conjugate of $\nabla_1$ by the unipotent matrix $$\left( \begin{matrix}1 &T \cr 0 &1 \cr \end{matrix}\right)\,.$$ Morphisms $T\in{\rm Hom}(\Theta, \Theta')$ in the category of equisingular flat bundles are linear maps $T: E\to E'$ compatible with the grading and such that the connections $\nabla_j$ of are $W$-equivalent on $B$. The condition is independent of the choice of representatives for the connections $\nabla$ and $\nabla'$.
The category $\Ec$ of equisingular flat bundles is a tensor category over $k=\C$, with a fiber functor $\omega: \Ec \to
Vect_\C$ given by $$\label{omegaE}
\omega: \Theta=(E,\nabla) \mapsto E.$$ In fact, one can refine the construction and work over the field $k=\Q$, since the universal singular frame (see below), in which one expresses the connections, has rational coefficients. In this case, the fiber functor $\omega: \Ec_\Q \to
Vect_\Q$ is of the form $\omega = \oplus \,\omega_n$, with $$\omega_n(\Theta)=\,{\rm Hom}(\Q(n), \Gr_{-n}^W(\Theta))\,,$$ where $\Q(n)$ denotes the object in $\Ec_\Q$ given by the class of the pair of the trivial bundle over $B$ with fiber a one-dimensional $\Q$-vector space placed in degree $n$ and the trivial connection.
Let ${{\mathcal}{F}}(1,2,3,\cdots)_{\bullet}$ be the free graded Lie algebra generated by one element $e_{-n}$ in each degree $n\in \Z_{>0}$, and let $$\label{hopfu}
\Hc_u=\; {{\mathcal}{U}}({{\mathcal}{F}}(1,2,3,\cdots)_{\bullet})^\vee$$ be the commutative Hopf algebra obtained by considering the graded dual of the enveloping algebra ${{\mathcal}{U}}({{\mathcal}{F}})$. We can then identify explicitly the affine group scheme associated to the neutral Tannakian category of flat equisingular bundles as follows ([[*cf.*]{} ]{}[@cmln] [@CM2]).
\[EU\] The category $\Ec$ of flat equisingular bundles is a neutral Tannakian category, with fiber functor . It is equivalent to the category ${\rm
Rep}_{U^*}$ of finite dimensional linear representations of the affine group scheme $U^*=U\rtimes {\mathbb{G}}_m$, where $U$ is the pro-unipotent affine group scheme associated to the Hopf algebra $\Hc_u$ of .
The affine group scheme $U^*$ is a motivic Galois group. In fact, by results of Goncharov and Deligne ([@dg], [@Gon], see Proposition \[SNmotives\] above), we have the following identification of the “cosmic Galois group” $U^*$.
\[motgal\] There is a (non-canonical) isomorphism $$\label{MotU}
U^* \cong G_{\cM_T}({{\mathcal{O}}}) \,.$$ of the affine group scheme $U^*$ with the motivic Galois group $G_{\cM_T}({{\mathcal{O}}})$ of the scheme $S_N$ of $N$-cyclotomic integers, for $N=3$ or $N=4$.
The fact that we only have a noncanonical identification suggests that there should be an explicit identification dictated by the form of the iterated integrals that give the expansionals defining the equisingular connections as in Proposition \[expans\]. This should be related to Kontsevich’s formula for multiple zeta values as iterated integrals generalized by Goncharov to multiple polylogarithms ${\rm Li}_{\,k_1,\ldots,k_m}(z_1,z_2,\ldots,z_m)$, in terms of the expansional ${\bf {\rm
T}e^{\int_0^1\,\a(z)\,dz}}$, with the connection $$\alpha(z)dz = \sum_{a\in \mu_m\cup \{0\}} \, \frac{dz}{z-a} \,\, e_a.$$
Notice, moreover, that the group $U^*$, as the differential Galois group in the formal theory of equisingular connections, corresponds to the Ramis exponential torus. In fact, we have no contribution from the monodromy, a fact on which the proof of Proposition \[connbeta\] depends essentially, and we also do not have Stokes phenomena, hence, as far as the differential Galois group is concerned, we can equally work in the formal or in the non-formal setting.
The renormalization group as a Galois group {#the-renormalization-group-as-a-galois-group .unnumbered}
-------------------------------------------
The formulation of Theorem \[EU\] is universal with respect to physical theories. When we consider a particular choice of a renormalizable theory $\sT$, we restrict the category of equisingular flat bundles to the full subcategory of finite dimensional linear representations of $G^*=G\rtimes {\mathbb{G}}_m$, for $G={\rm Difg}(\sT)$. In this case, the Riemann–Hilbert correspondence specializes to a morphism of differential Galois groups, as follows.
\[rhoG\] Let $G$ be a positively graded pro-unipotent affine group scheme. Then there exists a canonical bijection between equivalence classes of flat equisingular $G(\C)$-connections and graded representations $\rho \, : U\,\to G$, of the affine group scheme $U$ in $G$. Compatibility with the grading implies that $\rho$ extends to a homomorphism $\rho^* \, : U^*\,\to G^*$, which is the identity on ${\mathbb{G}}_m$.
This is a reformulation of the result of Proposition \[connbeta\]. In fact, more explicitly, the representation $\rho$ of Proposition \[rhoG\] is obtained as follows. We can write an element $\beta$ in $\Lie\,G$ as an infinite formal sum $$\label{betasum}
\beta=\; \sum_1^\infty\;\beta_n\,,$$ where, for each $n$, $\beta_n$ is homogeneous of degree $n$ for the grading, [[*i.e.*]{} ]{}$Y(\beta_n)=n \beta_n$. Thus, assigning $\beta$ with the action of the grading is the same as giving a collection of homogeneous elements $\beta_n$, with no restriction besides $Y(\beta_n)=n \beta_n$. In particular, there is no condition on their Lie brackets, hence assigning such data is equivalent to giving a homomorphism from the affine group scheme $U$ to $G$, by assigning, at the Lie algebra level, the generator $e_{-n}$ to the component $\beta_n$ of $\beta$.
In particular, the result above means that we can realize the renormalization group as a Galois group. In fact, recall that, for an assigned theory $\sT$, the corresponding $\beta$ that determines the counterterms $\gamma_{-}(z)$ is the infinitesimal generator of the renormalization group . The representation $\rho: U^* \to G^*$ then determines a lifting of the renormalization group ${\bf rg}$ to a canonical 1-parameter subgroup of $U^*$, obtained by considering the element $$\label{esum}
e=\; \sum_1^\infty\;e_{-n}\,,$$ in the Lie algebra $\Lie\,U$. As $U$ is a pro-unipotent affine group scheme, $e$ defines a morphism of affine group schemes $$\label{rgU}
{\bf{rg}}\; :\,{\mathbb{G}}_a \,\to \,U\,,$$ from the additive group ${\mathbb{G}}_a$ to $U$.
Thus, the rest of the affine group scheme $U$ can be throught of as further symmetries that refine the action of the renormalization group on a given physical theory. More precisely, restricting the attention to a generator $e_{-n}$ of the Lie algebra of $U$ corresponds to considering the flow generated by the degree $n$ component of the $\beta$ function with respect to the grading by loop number. Thus, from a physical point of view the Galois group $U$ accounts for a decomposition of the action of the renormalization group in terms of a family of flows restricted to the $n$-loops theory.
Universal singular frame {#universal-singular-frame .unnumbered}
------------------------
The element $e\in \Lie\,U$ defined in determines a “universal singular frame” given by $$\label{univ}
\g_U(z,v) =\,{\bf {\rm T}e^{-\frac{1}{z}\,
\int^{v}_0\,u^Y(e)\,\frac{du}{u}}}\;
\in U\;
\,.$$ This is obtained by applying Proposition \[connbeta\] to the affine group scheme $U$. This can be expressed explicitly in terms of iterated integrals in the form $$\label{framecoeff}
\g_U(z,v) =\,\sum_{n \geq
0}\,\sum_{k_j>0}\,\frac{e_{-k_1}e_{-k_2}\cdots e_{-k_n}}
{k_1\,(k_1+k_2)\cdots (k_1+k_2+\cdots +k_n)}\,v^{\sum
k_j}\,z^{-n},$$ with $e_{-n}$ the generators of $\Lie \,U$. This expansion has rational coefficients. The coefficients are the same as those occurring in in the local index formula of Connes–Moscovici [@cmindex], where the renormalization group idea is used in the case of higher poles in the dimension spectrum.
The Birkhoff factorization in $U$, applied to the universal singular frame, yields universal counterterms that maps under the representation $\rho: U \to {\rm Difg}(\sT)$ to the counterterms of a specific theory $\sT$.
Renormalization and geometry
============================
Quantum mechanics allows for two equivalent formulations of physics at the macroscopic scale, based on coordinate and momentum space, dual to one another by Fourier transform, while gravity, relativistically formulated in terms of the geometry of space-time, appears to privilege coordinates over momenta.
In the quantum theory of fields, at the perturbative level, Feynman integrals are computed in momentum space, using the dimensional regularization scheme. A nice historical and motivational perspective on how this came to be the general “accepted paradigm” in the context of renormalizable perturbation theory can be found in Veltman’s paper [@Vel]. As Veltman suggests, one can assume perturbative field theory as the starting point, defined in terms of Feynman diagrams using dimensional regularization (he refers to this as the “dimensional formulation”). This is very much the approach followed by the Connes–Kreimer theory and by our present work, where such physical data, taken as the given starting point, are reformulated in a more satisfactory conceptual perspective.
It is also possible to follow a different approach and to consider the problem of perturbative renormalization in coordinate space, working geometrically in terms of Fulton–MacPherson compactifications. A mathematical theory of perturbative renormalization under this point of view was developed recently by Kontsevich [@Kon]. It has the advantage of introducing directly geometric objects like algebraic varieties, hence a natural setting for an explicit action of motivic Galois symmetries ([[*cf.*]{} ]{}also [@Kon1]).
As stressed by Veltman [@Vel], space and time do not occur at all in the dimensional formulation, as coordinate space exists solely as Fourier transform of momentum space, which ceases to be defined when momentum space is continued to complex dimension. Notions associated to coordinate space, such as length and time measurements, must be recovered through the gravitational field, with graviton-fermion interactions determined by gauge invariance and Ward identities. Thus, a viewpoint that favors momentum rather than coordinate space is necessarily closer to noncommutative geometry than to classical algebraic geometry. In noncommutative geometry the metric properties of space are assigned not by a local coordinates description of the metric tensor, but through a “dual viewpoint”, spectrally, in terms of the Dirac operator, hence they continue to make sense on spaces that no longer exist classically. This appears to be a promising approach to reconcile space (no longer defined classically) with the dimensional formulation.
It is important to stress, in this respect, that the formulation of Riemannian spin geometry in the setting of noncommutative geometry, in fact, has already built in the possibility of considering a geometric space at dimensions that are complex numbers rather than integers. This is seen through the dimension spectrum, which is the set of points in the complex plane at which a space manifests itself with a nontrivial geometry. There are examples where the dimension spectrum contains points off the real lines ([[*e.g.*]{} ]{}the case of Cantor sets), but here one is rather looking for something like a deformation of the geometry in a small neighborhood of a point of the dimension spectrum, which would reflect dimensional regularization. The possibility of recasting the dimensional formulation in the setting of noncommutative geometry may prove very useful in the problem of extending at a fully quantum level the geometric interpretation of the standard model of elementary particle physics provided by noncommutative geometry ([@CoSM], [@ChCo]).
An important related question, which may be a starting point for such broader program, is to understand the precise relation between the universal singular frame and the local index formula, which in turn may cast some new light on the issue of the relation of the theory of perturbative renormalization illustrated here and noncommutative geometry. Since the local index formula of Connes–Moscovici is closely related to chiral anomalies, a direct comparison with the local index formula will involve a well known problem associated to dimensional regularization in the chiral case, namely the technical issue of how to extend the definition of the product $$\label{gamma5}
\gamma_5= i \gamma^0\gamma^1\gamma^2\gamma^3,$$ of the $\gamma$ matrices, which integer dimension $D=4$ satisfies the Clifford relations $\{ \gamma^\mu,\gamma^\nu \} = 2
g^{\mu\nu}\, I$, with $\Tr(I)=4$, and anticommutativity $\{
\gamma_5, \gamma^\mu \} =0$. The $\gamma^5$ problem, however, is not considered a serious obstacle to the application of dimensional regularization, as there are good methods to address it ([[*cf.*]{} ]{}[@MSR] for a recent discussion of this issue). For instance, the $\gamma^5$ problem is addressed by the Breitenlohner–Maison approach, in which one does not give an explicit expression for the gamma matrices in complex dimension, but just defines them (and the $\gamma_5$ given by ) through their formal properties. In [@DK1], Kreimer described another approach to the problem, in which $\gamma_5$ still anticommutes with $\gamma^\mu$ but the trace is no longer cyclic, an approach that is expected to be equivalent to the one of Breitenlohner–Maison ([[*cf.*]{} ]{}[@DK1], §5).
Finally, we would like to end on a more speculative tone, by mentioning a very different source for the idea of the existence of a deformation of geometry to non-integral complex dimensions. In arithmetic geometry, the Beilinson conjectures relate the values of the first non-vanishing derivative at integer points of the motivic $L$-functions of algebraic varieties to periods, namely numbers obtained by integration of algebraic differential forms on algebraic varieties ([[*cf.*]{} ]{}[[*e.g.*]{} ]{}[@KoZa]). This process of considering the expansion in a neighborhood of an integer point is reminiscent of the procedure of Dim-Reg, where one considers the Feynman integrals in an infinitesimal neighborhood of the integer dimension $D$. Based on this analogy, it becomes extremely suggestive to imagine that the Beilinson conjectures may be related to a “dimensional regularization of algebraic varieties and periods”, and that there may be a geometric interpretation even for the values at non-integer points, in terms of some (noncommutative) geometry in complex dimension.
[99]{}
D.V. Anosov, A.A. Bolibruch, [*The Riemann–Hilbert problem*]{}, Aspects of Mathematics Vol.22, Vieweg, 1994.
H. Araki, [*Expansional in Banach algebras*]{}, Ann. Sci. École Norm. Sup. (4) 6 (1973), 67–84.
A. Beilinson, J. Bernstein, P. Deligne, [*Faisceaux pervers*]{}, Astérisque 100 (1982) 5–171.
S. Bloch, [*Lectures of mixed motives*]{}, in “Algebraic geometry—Santa Cruz 1995”, pp.329–359, Proc. Sympos. Pure Math., 62, Part 1, Amer. Math. Soc., 1997.
N.N. Bogoliubov, O. Parasiuk, [*On the multiplication of the causal function in the quantum theory of fields*]{}, Acta Math. 97, (1957), 227–266.
N.N. Bogoliubov and D.V. Shirkov, [*Introduction to the theory of quantized fields*]{}, 3rd ed., Wiley 1980.
B. Bojarski, [*Connections between complex and global analysis: some analytical and geometrical aspects of the Riemann-Hilbert transmission problem*]{}. in “Complex analysis”, 97–110, Math. Lehrbücher Monogr. II. Abt. Math. Monogr., 61, Akademie-Verlag, 1983.
A.A. Bolibruch, [*The Riemann–Hilbert problem and Fuchsian differential equations on the Riemann sphere*]{}, Proceedings of the International Congress of Mathematicians, Zürich 1994, Birkäuser 1995, 1159–1168.
P. Cartier, [*A mad day’s work: from Grothendieck to Connes and Kontsevich. The evolution of concepts of space and symmetry*]{}, Bull. Amer. Math. Soc. (N.S.) 38 (2001), no. 4, 389–408.
A.H. Chamseddine, A. Connes, [*The spectral action principle*]{}. Comm. Math. Phys. Vol.186 (1997), N.3, 731–750.
K.T. Chen, [*Iterated integrals and exponential homomorphisms*]{}, Proc. London Math. Soc. Vol.3 N.4 (1954), 502–512.
K.T. Chen, [*Iterated integrals of differential forms and loop space homology*]{}. Ann. of Math. (2) 97 (1973), 217–246.
J. Collins, [*Renormalization*]{}, Cambridge Monographs in Math. Physics, Cambridge University Press, 1984.
A. Connes, [*Gravity coupled with matter and the foundation of non-commutative geometry*]{}. Comm. Math. Phys. Vol.182 (1996), N.1, 155–176.
A. Connes, D. Kreimer, [*Hopf algebras, renormalization and noncommutative geometry*]{}, Comm. Math. Phys. Vol. 199 (1998) 203–242.
A. Connes, D. Kreimer, [*Renormalization in quantum field theory and the Riemann-Hilbert problem. I. The Hopf algebra structure of graphs and the main theorem*]{}. Comm. Math. Phys. 210 (2000), no. 1, 249–273.
A. Connes, D. Kreimer, [*Renormalization in quantum field theory and the Riemann-Hilbert problem. II. The $\beta$-function, diffeomorphisms and the renormalization group*]{}. Comm. Math. Phys. 216 (2001), no. 1, 215–241.
A. Connes, M. Marcolli, [*Renormalization and motivic Galois theory*]{}, International Math. Research Notices (2004) N.76, 4073–4092.
A. Connes, M. Marcolli, [*From Physics to Number Theory via Noncommutative Geometry. Part II: Renormalization, the Riemann–Hilbert correspondence, and motivic Galois theory*]{}, preprint hep-th/0411114.
A. Connes, H. Moscovici, [ *The local index formula in noncommutative geometry*]{}, GAFA, Vol. 5 (1995), 174–243.
P. Deligne, [*Equations differentielles à points singuliers réguliers*]{}, Lecture Notes in Mathematics 163, Springer 1970.
P. Deligne, [*Catégories tannakiennes*]{}, in “Grothendieck Festschrift” Vol.2, pp. 111–195, Progress in Mathematics Vol.87, Birkhäuser, 1990.
P. Deligne, [*A quoi servent les motifs?*]{}, in “Motives” (Seattle, WA, 1991), pp.143–161, Proc. Sympos. Pure Math., 55, Part 1, Amer. Math. Soc., 1994.
P. Deligne, A.B. Goncharov, [*Groupes fondamentaux motiviques de Tate mixte*]{}, math.NT/0302267.
P. Deligne, J.S. Milne, [**]{}, in “Hodge cycles, motives, and Shimura varieties", Lecture Notes in Mathematics, 900, pp.101–228. Springer-Verlag, 1982
M. Demazure, A. Grothendieck, et al. [*Séminaire Géometrie Algébrique: Schémas en Groupes*]{}, Lecture Notes in Mathematics, Vol. 151, 152, 153. Springer, 1970.
S. Doplicher, J.E. Roberts, [*A new duality for compact groups*]{}. Invent. Math. 98 (1989) 157–218.
M. Douglas, [*Dirichlet branes, homological mirror symmetry, and stability*]{}, in “Proceedings ICM 2002” Beijing 2002, Vol.III, pp.395–408, Higher Education Press, 2002.
S.I. Gelfand, Yu.I. Manin, [*Homological algebra*]{}, Encyclopaedia of Mathematical Sciences, Vol.38, Springer, 1994.
A. Goncharov, [*Multiple polylogarithms and mixed Tate motives*]{}, math.AG/0103059.
A. Grothendieck, [*Sur la classification des fibrés holomorphes sur la sphère de Riemann*]{}, Amer. J. Math. Vol.79 (1957) 121–138.
A. Grothendieck, [*Esquisse d’un programme*]{}, 1984 manuscript, reproduced in “Geometric Galois actions, 1”, 5–48, Cambridge Univ. Press, 1997.
A. Grothendieck, [*Récoltes et Semailles*]{}, unpublished manuscript, 1985-86.
K. Hepp, [*Proof of the Bogoliubov-Parasiuk theorem on renormalization*]{}, Comm. Math. Phys. 2, (1966), 301–326.
G. ’t Hooft, [*Dimensional regularization and the renormalization group*]{}, Nuclear Physics B, 61 (1973) 455-468.
G. ’t Hooft, M. Veltman, [*Regularization and renormalization of gauge fields*]{} Nuclear Physics B, Vol.44, N.1 (1972), 189–213.
S.A. Joni, G.C. Rota, [*Coalgebras and bialgebras in combinatorics*]{}. Stud. Appl. Math. 61 (1979), no. 2, 93–139.
M. Kashiwara, [*Faisceaux constructibles et systèmes holonômes d’équations aux dérivées partielles linéaires à points singuliers réguliers*]{}, Séminaire Goulaouic-Schwartz, 1979–1980, Exp. No. 19, 7 pp., École Polytech., Palaiseau, 1980.
M. Kashiwara, [*The Riemann-Hilbert problem for holonomic systems*]{}. Publ. Res. Inst. Math. Sci. 20 (1984), no. 2, 319–365.
M. Kontsevich, [*Operads and motives in deformation quantization*]{}, Lett. Math. Phys. 48 (1999), no. 1, 35–72.
M. Kontsevich, private communication.
M. Kontsevich, D. Zagier, [*Periods*]{}, in “Mathematics unlimited—2001 and beyond”, pp. 771–808, Springer, 2001.
D. Kreimer, [*The role of $\gamma_5$ in Dimensional Regularization*]{}, hep-ph/9401354.
D. Kreimer, [*On the Hopf algebra structure of perturbative Quantum Field Theory*]{}, Adv. Theor. Math. Phys. 2 (1998), no. 2, 303–334.
Lê Dũng Tráng, Z. Mebkhout, [*Introduction to linear differential systems*]{}, Proc. Symp. Pure Math. Vol. 40 (1983) Part 2, pp. 31–63, American Mathematical Society 1983.
M. Levine, [*Mixed motives*]{}, Math. Surveys and Monographs, Vol. 57, AMS, 1998.
Yu.I. Manin, [*Moduli fuchsiani*]{}, Ann. Scuola Norm. Sup. Pisa, Vol.19 (1965) N.3, 113–126.
Yu.I. Manin, [*Correspondences, motifs and monoidal transformations*]{}, Mat. Sb. (N.S.) 77 (119) 1968, 475–507.
C.P. Martín, D. Sánchez-Ruiz, [*Action principles, restoration of BRS symmetry and the renormalization group equation for chiral non-Abelian gauge theories in dimensional renormalization with a non-anticommuting $\gamma_5$*]{}, Nucl.Phys. B572 (2000) 387-477.
J. Martinet, J.P. Ramis, [*Elementary acceleration and multisummability, I*]{}, Ann. Inst. Henri Poincaré, Vol.54 (1991) 331–401.
Z. Mebkhout, [*Sur le problème de Hilbert-Riemann*]{} C. R. Acad. Sci. Paris Sér. A-B 290 (1980), no. 9, A415–A417.
Z. Mebkhout, [*Sur le problème de Hilbert-Riemann*]{}. in “Complex analysis, microlocal calculus and relativistic quantum theory” (Proc. Internat. Colloq., Centre Phys., Les Houches, 1979), pp. 90–110, Lecture Notes in Phys., 126, Springer, Berlin-New York, 1980.
M. van der Put, M. Singer, [*Galois theory of linear differential equations*]{}, Springer 2002.
M. van der Put, [*Differential Galois Theory, Universal Rings and Universal groups*]{}, in “Differential Algebra and Related topics”, Editors Li Guo, Phyllis Cassidy, William F. Keigher, William Sitt. World Scientific 2002.
G.C. Rota, [*Hopf algebra methods in combinatorics*]{}, in “Problèmes combinatoires et théorie des graphes” (Colloq. Internat. CNRS, Univ. Orsay, Orsay, 1976), pp. 363–365, Colloq. Internat. CNRS, 260, CNRS, Paris, 1978.
N. Saavedra Rivano, [*Catégories Tannakiennes*]{}. Lecture Notes in Mathematics, Vol. 265. Springer-Verlag, 1972.
W. Schmitt, [*Hopf algebras in enumerative combinatorics*]{}. in“Gian-Carlo Rota on combinatorics”, 345–356, Contemp. Mathematicians, Birkhäuser, 1995.
J.P. Serre, [*Motifs*]{}, in “Journées Arithmétiques, 1989” (Luminy, 1989). Astérisque No. 198-200 (1991), 11, 333–349 (1992).
J.P. Serre, [*Propriétés conjecturales des groupes de Galois motiviques et des representations l-adiques*]{}, Proc. Symp. Pure Math. Vol. 55 (1994) Part I, 377–400.
M. Veltman, [*Perturbation theory and relative space*]{}, Acta Physica Polonica B Vol.25 N.10 (1994) 1399–1411.
V. Voevodsky, [*Triangulated categories of motives over a field*]{} in “Cycles, transfer and motivic homology theories, pp. 188–238, Annals of Mathematical Studies, Vol. 143, Princeton, 2000.
W. Zimmermann, [*Convergence of Bogoliubov’s method of renormalization in momentum space*]{}, Comm. Math. Phys. 15, (1969), 208–234.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Hard exclusive reactions are the tool to learn about generalized parton distributions, which provide a more complete parametrization of the nucleon than the ordinary parton distribution functions. Recent measurements by the HERMES collaboration of the exclusive production of photons, i.e., Deeply-Virtual Compton Scattering, are summarized and compared to model calculations, focusing on the measurements and model comparisons relevant to the extraction of quark orbital angular momentum and on the measurements on heavy nuclei.'
address: |
University of Colorado,\
Department of Physics,\
Boulder, Colorado 80309-0390, USA\
E-mail: [email protected]
author:
- 'F. Ellinghaus[^1]'
- for the HERMES Collaboration
title: 'DVCS at HERMES: Recent Results'
---
Introduction
============
Similar to the case of inclusive and semi-inclusive DIS, where the nucleon structure is described using [**P**]{}arton [**D**]{}istribution [**F**]{}unctions (PDFs), hard exclusive reactions can be expressed in terms of [**G**]{}eneralized [**P**]{}arton [**D**]{}istributions (GPDs)[@Mue94; @Ji97a; @Rad97]. The PDFs and elastic nucleon [**F**]{}orm [**F**]{}actors (FFs) are included in the GPDs as the limiting cases and moments of GPDs, respectively[@Ji97a]. While FFs derived in elastic scattering describe the transverse location of partons inside the nucleon and PDFs describe their longitudinal momentum distribution, GPDs are able to provide information on both at the same time. Thus exclusive reactions are able to give a certain 3–dimensional picture of the nucleon structure [@Bur00; @Bel02b; @Ral02]. In particular, GPDs offer for the first time a possibility to determine the total angular momentum carried by the quarks in the nucleon [@Ji97a].
Below recent HERMES measurements on the hard exclusive electroproduction of real photons ([**D**]{}eeply–[**V**]{}irtual [**C**]{}ompton [**S**]{}cattering, DVCS) are summarized and compared to model calculations. The data has been taken with polarized and unpolarized gas targets using the HERMES spectrometer [@Ack98] at the HERA electron/positron–proton collider at DESY, which offers longitudinally polarized 27.6 GeV electron and positron beams.
GPD $H$ via beam-charge and beam-spin asymmetries
=================================================
DVCS amplitudes can be measured through the interference between the DVCS and [**B**]{}ethe–[**H**]{}eitler (BH) processes, in which the photon is radiated from a parton in the former and from the electron in the latter process. Both processes have an identical final state, i.e., they are indistinguishable, and thus give rise to an interference term $I$. The photon production cross section depends on the Bjorken scaling variable $x_B$, the squared virtual–photon four–momentum $-Q^2$, the squared four-momentum transfer $t$ or the reduced four-momentum transfer $t^\prime = (t-t_{min})$ to the target, and the azimuthal angle $\phi$ defined as the angle between the lepton scattering plane and the (virtual and real) photon production plane. For an unpolarized proton target, and at leading twist, the interference term is given by [@Die97] $$\begin{aligned}
\label {I}
I \propto -C \,
[ \, a \, \cos \phi \, \mathrm{Re} {\cal M}^{1,1}
- b \, P_l \, \sin \phi \, \mathrm{Im} {\cal M}^{1,1}
\, ], \end{aligned}$$ where the lepton beam has longitudinal polarization $P_l$ and charge $C = \pm 1$, and $a$ and $b$ are functions of the ratio of the longitudinal to transverse virtual–photon flux. The real (imaginary) part of the DVCS amplitude $M^{1,1}$ can be accessed by measuring the $\cos \phi$ ($\sin \phi$) dependence of a cross section asymmetry with respect to the charge (spin) of the lepton beam. At HERMES kinematics, the DVCS amplitude $M^{1,1}$ gives access to the GPD $H$. Details and full equations are given in Ref. .
The event selection at HERMES requires events with exactly one photon and one charged track, identified as the scattered lepton, with $Q^2 >$1 GeV$^2$. For data taken prior to 2006 the recoiling proton is not detected and exclusive events are identified by the fact that the missing mass $M_x$ of the reaction $e p \rightarrow e \gamma X$ corresponds to the proton mass. Due to the finite energy resolution the exclusive sample is defined as $-1.5 < M_x < 1.7$ GeV. A recoil detector used during the recent data taking should reduce the underlying background from presently about 15% to less than 1% [@Roberto].
The beam–spin asymmetry (BSA) and the beam–charge asymmetry (BCA) as a function of $\phi$ are calculated as $$A_{LU} (\phi) = \frac {1}{< \left | P_l \right | >} \,
\frac {\overrightarrow N (\phi) - \overleftarrow N (\phi)}
{\overrightarrow N (\phi) + \overleftarrow N (\phi)}, \qquad
A_C(\phi) = \frac{ N^+(\phi) - N^-(\phi)}
{N^+(\phi) + N^-(\phi) },
\label{bsabca}$$ with the normalized yields $\overrightarrow N$ ($\overleftarrow N$) or $N^+$ ($N^-$) using a beam with positive (negative) helicity or a positron (electron) beam, respectively. The BSA (BCA) on the proton as a function of $\phi$ has been extracted at HERMES [@Air01] ([@Air06]), whereby the predominant $\sin \phi$ ($\cos \phi$) dependence expected from Eqn. \[I\] has been observed. The $\cos \phi$ amplitudes of the BCA on hydrogen as a function of $-t$ derived from a fit to the BCA in each $-t$ bin [@Air06] are shown in Fig.\[bca\_t\], and the recent preliminary BSA result on the kinematic dependences of the $\sin \phi$ amplitudes is shown in Fig.\[bsa\_kindep\_vgg\_guzey\].
![ The $\cos \phi$ amplitude of the beam–charge asymmetry [@Air06] on hydrogen as a function of $-t$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with (D) or without (no D) a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups. The variations within these groups are due to the different settings for the $b$ parameters. The calculations in the right panel are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bca_t"}](vgg_c1_t_16_curves.eps "fig:"){width="53.00000%"} ![ The $\cos \phi$ amplitude of the beam–charge asymmetry [@Air06] on hydrogen as a function of $-t$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with (D) or without (no D) a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups. The variations within these groups are due to the different settings for the $b$ parameters. The calculations in the right panel are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bca_t"}](beam_charge_asymmetry_t_newHERMES_Tobias.eps "fig:"){width="46.00000%"}
![ The $\sin \phi$ amplitude of the beam–spin asymmetry on hydrogen from the 1996-2000 data as a function of $-t$, $x_B$ and $Q^2$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with or without a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups, whereby the BSA appears to be insensitive to the D-Term and to the value of $b_v$. The calculations in the right column are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bsa_kindep_vgg_guzey"}](vgg_projection_s1_t_16_curves.eps "fig:"){width="49.50000%"} ![ The $\sin \phi$ amplitude of the beam–spin asymmetry on hydrogen from the 1996-2000 data as a function of $-t$, $x_B$ and $Q^2$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with or without a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups, whereby the BSA appears to be insensitive to the D-Term and to the value of $b_v$. The calculations in the right column are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bsa_kindep_vgg_guzey"}](beam_spin_asymmetry_t_prelim_data.eps "fig:"){width="49.50000%"} ![ The $\sin \phi$ amplitude of the beam–spin asymmetry on hydrogen from the 1996-2000 data as a function of $-t$, $x_B$ and $Q^2$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with or without a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups, whereby the BSA appears to be insensitive to the D-Term and to the value of $b_v$. The calculations in the right column are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bsa_kindep_vgg_guzey"}](vgg_projection_s1_xB_16_curves.eps "fig:"){width="49.50000%"} ![ The $\sin \phi$ amplitude of the beam–spin asymmetry on hydrogen from the 1996-2000 data as a function of $-t$, $x_B$ and $Q^2$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with or without a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups, whereby the BSA appears to be insensitive to the D-Term and to the value of $b_v$. The calculations in the right column are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bsa_kindep_vgg_guzey"}](beam_spin_asymmetry_xB_prelim_data.eps "fig:"){width="49.50000%"} ![ The $\sin \phi$ amplitude of the beam–spin asymmetry on hydrogen from the 1996-2000 data as a function of $-t$, $x_B$ and $Q^2$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with or without a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups, whereby the BSA appears to be insensitive to the D-Term and to the value of $b_v$. The calculations in the right column are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bsa_kindep_vgg_guzey"}](vgg_projection_s1_q2_16_curves.eps "fig:"){width="49.50000%"} ![ The $\sin \phi$ amplitude of the beam–spin asymmetry on hydrogen from the 1996-2000 data as a function of $-t$, $x_B$ and $Q^2$. The error bars show the statistical and systematic uncertainties added in quadrature. The calculations in the left panel are based on a double-distribution GPD model using a factorized (fac.) or a Regge–inspired (Regge) $t$–dependence with or without a D-term contribution. The parameters $b_v$ and $b_s$ are each set to either unity or infinity. Using the resulting 16 sets of model parameters, the calculated asymmetries fall into four main groups, whereby the BSA appears to be insensitive to the D-Term and to the value of $b_v$. The calculations in the right column are based on a dual-parametrization GPD model [@Guz06].[]{data-label="bsa_kindep_vgg_guzey"}](beam_spin_asymmetry_Q2_prelim_data.eps "fig:"){width="49.50000%"}
The model calculations shown in the left panel (left column) of Fig.\[bca\_t\] (Fig.\[bsa\_kindep\_vgg\_guzey\]) are based on a double-distribution GPD model described in Refs. . Since these data are exclusively sensitive to the GPD $H$ as mentioned above, the theoretical calculations shown were derived by only varying the model parameters for the GPD $H$ in the underlying code [@Van01] in order to calculate the asymmetries at the average kinematics of every bin. The model parameters differ with respect to including or neglecting the so–called D–term [@Pol99] in the GPD model and whether the $t$–dependence of the GPD $H$ is calculated in either the simplest ansatz where the $t$–dependence factorizes from the $t$–independent part or in the Regge–motivated ansatz. In addition, the so-called skewness parameters $b_{v}$ and $b_{s}$ in the profile function [@Mus00] have been set to either unity or infinity, the latter value corresponds to a skewness independent ansatz for the GPD. The BCA data appears to disfavor the four parameters sets with the Regge-inspired $t$-dependence and the D–term contribution as well as the factorized one with the D-term which gives the largest asymmetry ($b_v=1$, $b_s=\infty$). For the BSA all models seem to overshoot the absolute size of the $\sin \phi$ amplitude. The calculations based on the dual parametrization GPD model [@Guz06] shown in the right panel (right column) of Fig.\[bca\_t\] (Fig.\[bsa\_kindep\_vgg\_guzey\]) are all in rather good agreement with the data.
GPDs $H$ and $E$ via transverse target-spin asymmetry
=====================================================
Within the next few years HERMES will be able to provide sufficient data to largely constrain the GPD $H$ in the kinematic region of the experiment. It is natural to ask to what extent the GPD $E$ can be accessed, which is the other important GPD necessary in order to determine $J_q$, the total orbital angular momentum of quarks in the nucleon [@Ji97a]. For an unpolarized proton target the contribution from the GPD $E$ is suppressed with respect to $H$, but this is different for transverse target polarization [@Bel02a; @Die03]. Using an unpolarized beam (U) and a transversely (T) polarized target, a $\sin{(\phi-\phi_S)}\cos{\phi}$ modulation in the DVCS transverse target-spin asymmetry (TTSA) gives access to a combination of the GPDs $H$ and $E$ [@Ell05]. Here $\phi_s$ denotes the azimuthal angle of the target polarization vector with respect to the lepton scattering plane.
\[t\] ![The $\sin (\phi-\phi_S) \cos \phi$ amplitude of the transverse target-spin asymmetry as a function of $-t$, $x_B$ and $Q^2$ in comparison to theoretical predictions from Ref. .[]{data-label="trans_proj"}](ttsa-vgg-upper-plot.eps "fig:"){width="90.00000%"}
The results from HERMES data collected on a transversely polarized hydrogen target are shown in Fig. \[trans\_proj\]. They agree with the model calculations [@Ell05] shown in the same figure, which have been calculated for various values of $J_u$. Based on $u$-quark dominance the $d$-quark total angular momentum has been assumed to be zero. Since it was realized that the model calculations are largely insensitive to all model parameters but $J_u$ and $J_d$, it is possible to constrain $J_u$ and $J_d$ even though other model parameters are largely unconstrained [@Ell05]. For example, the model calculations in Fig. \[trans\_proj\] show little variation if calculated for any of the 16 parameter sets which give very different values for the BCA as shown in Fig.\[bca\_t\]. Hence the four parameter sets with the Regge–inspired $t$–dependence and without a contribution of the D–term (solid lines in the left panel of Fig.\[bca\_t\]), have been chosen in order to determine the parameter space allowed for $J_u$ and $J_d$. For different values of $J_u$ and $J_d$ these sets have been compared to the DVCS TTSA data shown in Fig. \[trans\_proj\], leading to a first model dependent constraint on $J_u$ versus $J_d$ shown in Fig.\[jujd\] [@Ye06].
![Model dependent constraint on the quark orbital angular momenta $J_u$ and $J_d$.[]{data-label="jujd"}](vgg-Jq-1.eps){width="70.00000%"}
Further improvement can be expected, taking into account that the GPD $H$ and therefore the available theoretical models will be well constrained by the upcoming HERMES data, and that the data shown here is less than half of the data taken on the transversely polarized hydrogen target. Due to the good agreement between the calculations in the dual-parametrization model and the BSA and BCA results as shown above, the next natural step is to also calculate the allowed parameter space for $J_u$ and $J_d$ within this model. This effort is presently ongoing, but calculations for various values of $J_u$ at a fixed $J_d = 0$ [@Guz06] already indicate that the dual model favors smaller values for $J_u$ than the double-distribution one.
DVCS on Nuclei
==============
Beam-Spin Asymmetries have also been measured on deuterium and various heavier nuclei (He, N, Ne, Kr, Xe). Coherent processes are dominant for small values of $-t$, while at larger values incoherent scattering on the individual protons and neutrons dominates. Based on MC studies a region of small $-t^\prime$ (“coherent enriched”) has been selected for each nucleus in order to have similar average kinematic values for all targets ($\langle -t^\prime \rangle =0.018$ GeV$^2$, $\langle Q^2
\rangle \approx 1.7$ GeV$^2$, $\langle x_B \rangle \approx 0.065$). According to the MC studies the fraction of coherent processes is approximately 82% for all but the lighter targets (D, He). Similarly, an “incoherent enriched” sample has been selected with an average $-t^\prime$ of 0.2 GeV$^2$. The ratio of the BSA on the nuclei divided by the one on the proton at the same average kinematics is shown in the left panel of Fig.\[nuclei\]. A simple fit to a constant for the coherent enriched sample yields a value above unity by two sigma, while the incoherent enriched sample shows asymmetries very similar to the one on the proton. Note that this is in agreement with earlier HERMES preliminary results [@Ell02c] where a value consistent with unity was found when comparing the BSA on deuterium and neon to the one on the proton in the full $t$ range. The BSAs on nuclei are expected to be similar to the one on the proton in the incoherent enriched sample since scattering on the protons inside the nuclei should dominate due to the fact that the BH process on the neutron is suppressed.
![The $\sin \phi$ amplitude of the BSA on D, He, N, Ne, Kr, Xe divided by the one on the proton in two different kinematic regions dominated by either coherent or incoherent processes (left panel). The ’coherent enriched’ result (left panel, upper row) is compared to model calculations [@Guz06b] in the right panel.[]{data-label="nuclei"}](Frac_Hyd_180.eps "fig:"){width="59.50000%"} ![The $\sin \phi$ amplitude of the BSA on D, He, N, Ne, Kr, Xe divided by the one on the proton in two different kinematic regions dominated by either coherent or incoherent processes (left panel). The ’coherent enriched’ result (left panel, upper row) is compared to model calculations [@Guz06b] in the right panel.[]{data-label="nuclei"}](A_LU_A-dependence_siddikov_elli_corrected_errors.ps "fig:"){width="39.50000%"}
The fit value found for the coherent enriched sample is consistent with a very basic prediction of 5/3 for Spin-0 and Spin-1/2 targets, based on the ratio of the involved valence-quark charges squared [@Kir03], as well as with calculations done specifically for neon and krypton [@Guz03]. Also a prediction of R= 1-1.1 for Helium [@Liu05] is in agreement with the measurement. The result of a recent calculation [@Guz06b] is shown in the right panel of Fig.\[nuclei\], whereby one of the three models is disfavored by the data.
Summary
=======
The HERMES DVCS data on beam-charge and beam-spin asymmetries is already able to distinguish between some GPD models. Based on a certain model, the DVCS measurements on transversely polarized hydrogen lead to a first model dependent constraint for the total angular momentum of quarks in the nucleon. This method, together with increased statistics and improved models should allow for a constraint with reasonable statistical and theoretical uncertainties in the future.
[99]{} D. Müller et al., Fortschr. Phys. [**42**]{} (1994) 101.
X. Ji, Phys. Rev. Lett. [**78**]{} (1997) 610, Phys. Rev. [**D55**]{} (1997) 7114.
A.V. Radyushkin, Phys. Lett. **B380** (1996) 417, Phys. Rev. [**D56**]{} (1997) 5524.
M. Burkardt, Phys. Rev. [**D62**]{} (2000) 071503; Erratum–ibid. [**D66**]{} (2002) 119903.
A.V. Belitsky and D. Müller, Nucl. Phys. [**A711**]{} (2002) 118.
J.P. Ralston and B. Pire, Phys. Rev. [**D66**]{} (2002) 111501.
HERMES Coll., K. Ackerstaff et al., Nucl. Instr. and Meth. [**A417**]{} (1998) 230.
M. Diehl et al., Phys. Lett. [**B411**]{} (1997) 193.
A.V. Belitsky, D. Müller and A. Kirchner, Nucl. Phys. [**B629**]{} (2002) 323.
R. Perez-Benito, these proceedings.
HERMES Coll., A. Airapetian et al., Phys. Rev. Lett. [**87**]{} (2001) 182001.
HERMES Coll., A. Airapetian et al., Phys. Rev. **D75** (2007) 011103.
M. Vanderhaeghen, P.A.M. Guichon and M. Guidal, Phys. Rev. [**D60**]{} (1999) 094017.
K. Goeke, M.V. Polyakov and M. Vanderhaeghen, Prog. Part. Nucl. Phys. [**47**]{} (2001) 401.
M. Vanderhaeghen, P.A.M. Guichon and M. Guidal, Computer code for the calculation of DVCS and BH processes, Private Communication, 2001.
M.V. Polyakov and C. Weiss, Phys. Rev. [**D60**]{} (1999) 114017.
I.V. Musatov and A.V. Radyushkin, Phys. Rev. [**D61**]{} (2000) 074027.
V. Guzey and T. Teckentrup, Phys. Rev. **D74** (2006) 054027.
M. Diehl, Phys. Rept. [**388**]{} (2003) 41.
F. Ellinghaus, W.-D. Nowak, A.V. Vinnikov and Z. Ye, Eur. Phys. J. **C46** (2006) 729.
HERMES Coll., Z. Ye, Proceedings of 14th International Workshop on Deep Inelastic Scattering (DIS 2006), Tsukuba, Japan, hep-ex/0606061.
HERMES Coll., F. Ellinghaus et al., Proceedings of the 15th International Spin Physics Symposium (Spin 2002), Upton, New York, AIP Conf. Proc. **675** (2002) 303, hep-ex/0212019.
A. Kirchner and D. Müller, Eur. Phys. J. **C32** (2003) 347.
V. Guzey and M. Strikman, Phys. Rev. **C68** (2003) 015204.
S. Liuti and S.K. Taneja, Phys. Rev. **C72** (2005) 032201.
V. Guzey and M. Siddikov, J. Phys. **G32** (2006) 251.
[^1]: his work is supported in part by the epartment of nergy.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a class of one-dimensional, strictly neutral, Vlasov-Maxwell equilibrium distribution functions for force-free current sheets, with magnetic fields defined in terms of Jacobian elliptic functions, extending the results of @Abraham-Shrauner-2013 to allow for non-uniform density and temperature profiles. To achieve this, we use an approach previously applied to the force-free Harris sheet by @Kolotkov-2015. In one limit of the parameters, we recover the model of @Kolotkov-2015, while another limit gives a linear force-free field. We discuss conditions on the parameters such that the distribution functions are always positive, and give expressions for the pressure, density, temperature and bulk-flow velocities of the equilibrium, discussing differences from previous models. We also present some illustrative plots of the distribution function in velocity space.'
author:
- 'F. Wilson'
- 'T. Neukirch'
- 'O. Allanson'
title: 'Force-free collisionless current sheet models with non-uniform temperature and density profiles'
---
Introduction {#sec: intro}
============
Force-free current sheets, with magnetic fields satisfying $$\begin{aligned}
\nabla\cdot\textbf{B}&=&0\label{ff1}\\
\nabla\times\textbf{B}&=&\mu_0\textbf{j}\label{ff2}\\
\textbf{j}\times\textbf{B}&=&0\label{ff3},\end{aligned}$$ are appropriate for plasma modelling in, e.g., the solar atmosphere and planetary magnetospheres (e.g. Refs. ). Equations (\[ff1\])-(\[ff3\]) imply that the current density is parallel to the magnetic field: $\textbf{j}=\alpha(\textbf{r})\textbf{B}$. The case where $\alpha=0$ defines a potential field, and when $\alpha$ is constant we have a linear force-free field. When $\alpha$ varies with the position $\textbf{r}$, the field is referred to as nonlinear force-free.
Such current sheets as described above can play a crucial role in, e.g, magnetic reconnection processes, for which it is often necessary to consider kinetic length scales (e.g. Ref. ), since many astrophysical plasmas are approximately collisionless. To initialise studies of collisionless reconnection, a Vlasov-Maxwell (VM) equilibrium can be used; since current sheets are strongly localised, they are often well described by one-dimensional (1D) VM equilibrium models. The work by @Wilson-2016 was the first example of a study of collisionless reconnection for which an exact nonlinear force-free equilibrium was used in the initial setup, using a distribution function (DF) found by @Harrison-2009a for the ’force-free Harris’ current sheet, $$\textbf{B}=B_0(\tanh(z/L), \mbox{sech}(z/L), 0).\label{bffhs}$$ Other studies of collisionless reconnection in force-free current sheets have involved the use of approximate force-free equilibria (e.g. Refs. ) or linear force-free equilibria (e.g. Refs. ).
To find VM equilibrium DFs consistent with force-free current sheets involves solving the VM equations in the opposite order from what is usually done; a magnetic field satisfying Equations (\[ff1\])-(\[ff3\]) is specified, and the DFs are then given by the solution of an inverse problem (e.g. Refs. ). As such, finding exact force-free VM equilibria is generally a non-trivial task, and this is reflected in the relatively small number of known solutions. Linear force-free VM equilibria have been discussed in, e.g., Refs. . The first solution for a nonlinear force-free field was found by @Harrison-2009b (see also Ref. ) for the force-free Harris sheet, and these solutions were later extended by @Kolotkov-2015 to allow for non-uniform density and temperature profiles (with respect to the spatial coordinate). A number of other equilibrium DFs have also been found for this field. @Wilson-2011 found DFs with an arbitrary dependence on the particle energy; @Stark-2012 discussed DFs in the relativistic limit; @Allanson-2015 [@Allanson-2016] found DFs in terms of infinite sums over Hermite polynomials, with an arbitrarily low plasma beta (in the previous work on the force-free Harris sheet the plasma beta was constrained to be greater than unity); @Dorville-2015 discussed ’semi-analytic’ DFs for a magnetic field which includes the force-free Harris sheet as a special case.
@Abraham-Shrauner-2013 discussed VM equilibria for a nonlinear force-free magnetic field given in terms of Jacobian elliptic functions. This work can be thought of as a generalisation of some of the previous work, to account for both linear and nonlinear force-free equilibria in one model, since, in one limit of the elliptic modulus, the magnetic field becomes the force-free Harris sheet field, and in another limit it becomes a linear force-free field. The DFs discussed give rise to spatially uniform temperature and density profiles, in a similar way to some of the models mentioned above. In this paper, we will extend this class of DFs to include those consistent with non-uniform temperature and density profiles, using a similar approach used by @Kolotkov-2015 for the force-free Harris sheet. As for Abraham-Shrauner’s DFs, the new DFs we will discuss include both the linear force-free limit and the force-free Harris sheet limit[@Kolotkov-2015].
The paper is laid out as follows; in Section \[sec:vm\_eq\], we outline the background theory of 1D VM equilibria; in Section \[sec: as\], we present an overview of the work by @Abraham-Shrauner-2013; we discuss the extension of this work to include non-uniform temperature and density profiles in Section \[sec:modified\_df\], and the velocity space structure of the new DFs is discussed in Section \[sec:vspace\]; we end with a summary in Section \[sec:summary\].
1D Vlasov-Maxwell equilibria {#sec:vm_eq}
============================
In line with some of the previous work on 1D VM equilibria (e.g. Refs. ), we assume that all quantities depend only on the $z$-coordinate, and that the magnetic field, $\textbf{B}=(B_x,B_y,0)$, can be written as the curl of a vector potential, $\textbf{A}=(A_x, A_y, 0)$. We will not repeat all of the details here, but the result of the above assumptions is that the problem reduces to solving Ampère’s law in the form $$\begin{aligned}
\frac{\mathrm{d}^2A_x}{\mathrm{d}z^2}&=&-\mu_0\frac{\partial P_{zz}}{\partial A_x}\label{ampx}\\
\frac{\mathrm{d}^2A_y}{\mathrm{d}z^2}&=&-\mu_0\frac{\partial P_{zz}}{\partial A_y}\label{ampy},\end{aligned}$$ to find $P_{zz}$, which is the $zz$-component of the pressure tensor, defined by $$P_{zz}(A_x, A_y)=\sum_sm_s\int v_z^2f_s(H_s, p_{xs}, p_{ys})\mathrm{d}^3v,\label{pzz def}$$ where we assume that the DFs can be chosen in such a way that they are compatible with strict neutrality (the scalar potential $\phi=0$) [@Channell-1976]. Note that we only consider $P_{zz}$ since this is the component of the pressure tensor which is important for the force-balance of the 1D equilibrium. The DFs, denoted by $f_s$, are assumed to be functions of the particle energy, $H_s=m_s(v_x^2+v_y^2+v_z^2)/2$, and the $x$- and $y$-components of the canonical momentum, $\textbf{p}=m_s\textbf{v}+q_s\textbf{A}$, since these are known constants of motion for a time-independent system with spatial invariance in the $x$- and $y$-directions. Once Ampère’s law has been solved for $P_{zz}$, the DF can be found by solving Eq. (\[pzz def\]). This is an example of an inverse problem.
Abraham-Shrauner’s model {#sec: as}
========================
In this section we discuss some properties of the the model developed by @Abraham-Shrauner-2013, in order to give context to the discussion we will present in Section \[sec:modified\_df\]. In Abraham-Shrauner’s work, a nonlinear force-free current sheet profile is considered, described by the magnetic field $$\textbf{B}=B_0\Big(\mbox{sn}(z/L), \mbox{cn}(z/L),0\Big),\label{AS field}$$ where $B_0$ is a constant, $L$ is the current sheet half-thickness, and $\mbox{sn}$ and $\mbox{cn}$ are Jacobian elliptic functions[@NIST:DLMF] with the modulus $k$ suppressed (where $0\le k\le 1$). In the limit $k\to0$, $\mbox{sn}(z/L)\to\sin(z/L)$ and $\mbox{cn}(z/L)\to\cos(z/L)$, and so the magnetic field (\[AS field\]) becomes the linear force-free field $\textbf{B}=B_0\left(\sin(z/L), \cos(z/L),0\right)$. In the limit $k\to1$, $\mbox{sn}(z/L)\to\tanh(z/L)$ and $\mbox{cn}(z/L)\to\mbox{sech}(z/L)$, giving the force-free Harris sheet magnetic field (Eq. (\[bffhs\])). The vector potential, $\textbf{A}$, used by @Abraham-Shrauner-2013 is given by $$\begin{aligned}
A_x&=&\frac{B_0L}{k}\left(\mbox{arcsin}\left(k\mbox{sn}(z/L)\right)+\frac{k\pi}{2}\right)\label{ax}\\
A_y&=&\frac{B_0L}{k}\mbox{ln}\left(\frac{k\mbox{cn}(z/L)+\mbox{dn}(z/L)}{1+k}\right)\label{ay},\end{aligned}$$ where $\mbox{dn}$ is also an elliptic function. This can be seen by using standard integrals [@byrd_friedman] and by choosing the integration constants such that, when $k\to1$, ${A_x}\to 2B_0L\mbox{arctan}(e^{z/L})$, $A_y\to-\ln(\cosh(z/L))$ - the vector potential components used in some of the previous work on the force-free Harris sheet (note also that an alternative gauge for $\textbf{A}$ is discussed for the force-free Harris sheet by @Allanson-2016). The current density is given by $$\textbf{j}=\frac{B_0}{\mu_0L}\left(\mbox{sn}(z/L)\mbox{dn}(z/L),\mbox{cn}(z/L)\mbox{dn}(z/L),0\right)=\frac{{\mbox{dn}(z/L)\textbf{B}}}{\mu_0L}\label{asj},$$ and so the force-free parameter $\alpha$ is given by $$\alpha(z)=\frac{\mbox{dn}(z/L)}{\mu_0L}.$$ Note that, in the limit $k\to0$, $\mbox{dn}(z/L)\to1$, and so $\alpha$ is constant (the linear force-free case), but is otherwise a function of position (the nonlinear force-free case).
It is assumed that the pressure has the form $P_{zz}(A_x, A_y)=P_{1}(A_x)+P_{2}(A_y)$; Ampère’s law in the form of equations (\[ampx\]) and (\[ampy\]) can then be solved for $P_{zz}$ in terms of the macroscopic parameters, which gives $$\begin{aligned}
P_{zz}&=&P_{t1}+P_{t2}\nonumber\\
&{}&- \frac{B_0^2}{2\mu_0}\Bigg(\frac{3}{2}+\frac{1}{2k^2}\cos\left(\frac{2kA_x}{B_0L}-k\pi\right)-\frac{1}{4}\left(\frac{1}{k}+1\right)^2\exp\left(\frac{2kA_y}{B_0L}\right)\nonumber\\
&{}&-\frac{1}{4}\left(\frac{1}{k}-1\right)^2\exp\left(-\frac{2kA_y}{B_0L}\right)\Bigg)\label{pzz_macro},\end{aligned}$$ where $P_{t1}$ and $P_{t2}$ are constants. This expression can then be used in Eq. (\[pzz def\]) to determine the DF, which can be written in terms of the constants of motion as $$\begin{aligned}
f_s(H_s, p_{xs}, p_{ys})&=&\frac{n_{0s}e^{-\beta_sH_s}}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\Bigg[a_{0s}-\frac{1}{2k^2}\exp\left(\frac{(1+k^2)u_{ys}^2}{2v_{th,s}^2}\right)\cos\left(k\beta_su_{xs}p_{xs}-k\pi\right)\nonumber\\
&{}&+\frac{1}{4}\left(\frac{1}{k}+1\right)^2\exp\left(\frac{(1-k^2)u_{ys}^2}{2v_{th,s}^2}\right)\exp({k\beta_su_{ys}p_{ys}})\nonumber\\
&{}&+\frac{1}{4}\left(\frac{1}{k}-1\right)^2\exp\left(\frac{(1-k^2)u_{ys}^2}{2v_{th,s}^2}\right)\exp(-{k\beta_su_{ys}p_{ys}})\Bigg],\label{asdf_rewritten}\end{aligned}$$ where $a_{0s}$ is a dimensionless constant, $u_{xs}$ and $u_{ys}$ are constant parameters with the dimension of velocity, $\beta_s=(k_BT_s)^{-1}$ and $v_{th,s}=(\beta_sm_s)^{-1/2}$. In the limit $k\to1$, this DF takes the form of that discussed in Refs. for the force-free Harris sheet. In the opposite limit, i.e. $k\to0$, it takes a general form which is similar to that described in Refs. , but with a shift in $p_{xs}$ and $p_{ys}$ (this corresponds to a regauging of the vector potential).
Note that a number of relations exist between the parameters of the model, to ensure positivity of the DFs, strict neutrality, and consistency between the microscopic and macroscopic descriptions of the equilibrium (see Ref. for further details). Using these relations, the equilibrium density, pressure and temperature can be expressed as $$\begin{aligned}
n&=&n_0\left(a_0+\frac{1}{2}\right)\label{as n}\\
P_{zz}&=&\frac{n_0(\beta_e+\beta_i)}{\beta_e\beta_i}\left(a_0+\frac{1}{2}\right)\\
T&=&\frac{P_{zz}}{n}=\frac{\beta_e+\beta_i}{\beta_e\beta_i}\label{as temp},\end{aligned}$$ where $a_0$ and $n_0$ are constant parameters that are introduced when the strict neutrality condition ($\phi=0$) is imposed. The expressions (\[as n\])-(\[as temp\]) are independent of the elliptic modulus $k$; this can be seen for $P_{zz}$ through the force-balance equation $$\frac{B^2}{2\mu_0}+P_{zz} = P_T,$$ where $P_T$ is the total pressure, since $B^2=\vert\textbf{B}\vert^2=B_0^2$ for the magnetic field (\[AS field\]), which is independent of $k$. Since, in this case, $P_{zz}=(\beta_e+\beta_i)n/(\beta_e\beta_i)$, it follows that the density and temperature will also be independent of $k$. As can be seen from the expressions (\[as n\]) and (\[as temp\]), Abraham-Shrauner’s model has density and temperature profiles that are constant across the current sheet, in a similar way to the models discussed in Refs. . In Section \[sec:modified\_df\], we discuss how the method of @Kolotkov-2015 can be used to extend the model to have spatially non-uniform density and temperature profiles across the current sheet, while still maintaining a constant pressure as is required for a force-free equilibrium (see e.g. Ref. ).
Extension to non-uniform temperature/density case {#sec:modified_df}
=================================================
To extend the model of @Abraham-Shrauner-2013 to have non-uniform temperature and density profiles, we consider a DF of the form $$\begin{aligned}
{f}_{s}&=&\frac{n_{0s}\gamma^{3/2}}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\exp({-\gamma\beta_sH_s})\left(a_{0s}+a_{1s}\cos(\gamma k\beta_su_{xs}p_{xs}-k\pi)\right)\nonumber\\
&{}&+\frac{n_{0s}}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\exp({-\beta_sH_s})\left(b_{0s}+b_{1s}\exp(k\beta_su_{ys}p_{ys})+b_{2s}\exp(-k\beta_su_{ys}p_{ys})\right)\label{modified AS},\end{aligned}$$ (where $\gamma>0$) i.e. a modification of Abraham-Shrauner’s DF. This corresponds to assuming that the $p_{xs}$-dependent population has a different energy dependence than the $p_{ys}$-dependent population, through the factor $\gamma$. We effectively also have two separate constant background populations (through the constants $a_{0s}$ and $b_{0s}$) whose energy dependences differ. These two populations have been included to allow the limit $k\to0$ to exist, and to ensure this we assume that the constants $a_{0s}$ and $b_{0s}$ scale with the elliptic modulus $k$ in the following way; $$\begin{aligned}
a_{0s}&=&\bar{a}_{0s}+\frac{\gamma}{2k^2}\exp\left(\frac{u_{xs}^2}{2v_{th,s}^2}\right)\label{a0s choice}\\
b_{0s}&=&\bar{b}_{0s}-\frac{1}{2k^2}\exp\left(\frac{u_{xs}^2}{2v_{th,s}^2}\right)\label{b0s choice},\end{aligned}$$ for constants $\bar{a}_{0s}$ and $\bar{b}_{0s}$. Note that we have defined the constants in this way so that we have a model that works for all $k$ values between 0 and 1, but for finite small $k$ (or large $u_{xs}/v_{th,s}$), the $k$-dependent parts of $a_{0s}$ and $b_{0s}$ can become very large, which may lead to, e.g., a large maximum density, which may not be physically appropriate. If we were only interested in a particular finite small value of $k$, we could redefine the constants to avoid such issues. By calculating the number density ($n_s=\int f_s\mathrm{d}^3v$) of the modified DF (\[modified AS\]), and imposing the condition $\phi=0$ ($n_i(A_x,A_y)=n_e(A_x,A_y))$, we obtain the neutrality relations (\[modified\_neut1\])-(\[modified\_neut2\]) in Appendix \[sec:neutrality\_app2\]. We can then express $n_s=n$ as $$\begin{aligned}
n(A_x,A_y)&=&n_0[a_0+b_0+a_1\cos\left(\gamma k\beta_su_{xs}q_sA_x-k\pi\right)\nonumber\\
&{}&+b_1\exp\left(k\beta_su_{ys}q_sA_y\right)+b_2\exp\left(-k\beta_su_{ys}q_sA_y\right)]\label{n ax_ay},\end{aligned}$$ and the pressure can be calculated from the DF through Eq. (\[pzz def\]) as $$\begin{aligned}
P_{zz}&=&n_{0}\frac{\beta_e+\beta_i}{\beta_e\beta_i}\Bigg(\frac{a_{0}}{\gamma}+b_0+\frac{a_{1}}{\gamma}\cos\left(\gamma k\beta_s u_{xs}q_sA_x-k\pi\right)\nonumber\\
&{}&+b_{1}\exp\left(k\beta_s u_{ys}q_sA_y\right)+b_{2}\exp\left(-k\beta_s u_{ys}q_sA_y\right)\Bigg).\label{new pzz}\end{aligned}$$ Note the $\gamma^{-1}$ factors appearing in parts of Eq. (\[new pzz\]), meaning that the pressure is no longer simply a multiple of the density as in Abraham-Shrauner’s model. Eq. (\[new pzz\]) for the pressure can be compared with Eq. (\[pzz\_macro\]) to give the relations (\[micromacro1\])-(\[ux uy gamma\]) (see Appendix \[sec:neutrality\_app2\]) between the microscopic and macroscopic parameters. Using these relations, and the neutrality relations in Appendix \[sec:neutrality\_app2\], the modified DF (\[modified AS\]) can then be written as $$\begin{aligned}
f_s&=&\frac{\gamma^{3/2}n_{0s}\exp(-\gamma\beta_sH_s)}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\Bigg(a_{0s}-\frac{\gamma}{2k^2}\exp\left(\frac{(\gamma k^2+1)u_{xs}^2}{2v_{th,s}^2}\right)\cos(\gamma k\beta_su_{xs}p_{xs}-k\pi)\Bigg) \nonumber\\
&{}&+\frac{n_{0s}\exp(-\beta_sH_s)}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\Bigg(\frac{1}{4}\exp\left(\frac{u_{xs}^2-k^2u_{ys}^2}{2v_{th,s}^2}\right)\nonumber\\
&{}&\times\left\{\left(\frac{1}{k}+1\right)^2\exp(k\beta_su_{ys}p_{ys}) + \left(\frac{1}{k}-1\right)^2\exp(-k\beta_su_{ys}p_{ys})\right\} +b_{0s}\Bigg).\label{new df}\end{aligned}$$ Sufficient conditions for the positivity of the DF (\[new df\]) across the whole phase space can be derived by assuming that the functions $$\begin{aligned}
g_{1s}(p_{xs})&=&a_{0s}-\frac{\gamma}{2k^2}\exp\left(\frac{(\gamma k^2+1)u_{xs}^2}{2v_{th,s}^2}\right)\cos(\gamma k\beta_su_{xs}p_{xs}-k\pi)\\
g_{2s}(p_{ys})&=&b_{0s}+\frac{1}{4}\exp\left(\frac{u_{xs}^2-k^2u_{ys}^2}{2v_{th,s}^2}\right)\nonumber\\
&{}&\times\left(\left(\frac{1}{k}+1\right)^2\exp(k\beta_su_{ys}p_{ys}) + \left(\frac{1}{k}-1\right)^2\exp(-k\beta_su_{ys}p_{ys})\right),\end{aligned}$$ are both positive, and are given by $$\begin{aligned}
\bar{a}_0&>&\frac{\gamma}{2k^2}\left[\exp\left(\frac{\gamma k^2u_{xs}^2}{2v_{th,s}^2}\right)-1\right]\label{a0 bar cond}\\
\bar{b}_0&>&\frac{1}{2k^2}\left[1 - (1-k^2)\exp\left(-\frac{k^2u_{ys}^2}{2v_{th,s}^2}\right)\right]\label{b0 bar cond},\end{aligned}$$ where $\bar{a}_0$ and $\bar{b}_0$ are defined in Appendix \[sec:neutrality\_app2\]. Note that these conditions are well defined in the limit $k\to0$. Since $0\le k\le1$, $\gamma>0$ and the exponential term in Eq. (\[a0 bar cond\]) has a minimum value of unity, we see that $\bar{a}_0\ge0$. The new DF (\[new df\]) describes an equilibrium with non-uniform density and temperature profiles; we can show this by writing them as functions of $z$ using Equations (\[ax\]), (\[ay\]), (\[a1\])-(\[a3\]) and the definitions of $\bar{a}_0$ and $\bar{b}_0$, which gives $$\begin{aligned}
n(z)&=&n_0\Bigg[\bar{a}_0+\bar{b}_0+\frac{1}{2}+(\gamma-1)\mbox{sn}^2(z/L)\Bigg]\label{ns}\\
T(z)&=&\frac{P_{zz}}{n}=\frac{\beta_e+\beta_i}{\beta_e\beta_i}\left(\frac{\bar{a}_0}{\gamma}+\bar{b}_0+\frac{1}{2}\right)\left(\bar{a}_0+\bar{b}_0+\frac{1}{2}+\left(\gamma-1\right)\mbox{sn}^2(z/L)\right)^{-1}\label{temp},\end{aligned}$$ where the uniform value of the pressure is given by $$P_{zz}=\frac{n_0(\beta_e+\beta_i)}{\beta_e\beta_i}\left(\frac{\bar{a}_0}{\gamma}+\bar{b}_0+\frac{1}{2}\right)\label{pzz},$$ which is independent of the modulus $k$ (for the same reasons as discussed in Section \[sec: as\]), and is similar to the expression found by @Kolotkov-2015 for the force-free Harris sheet. Note, however, that this time the density depends on $k$, due to the introduction of the $\gamma$ factors in the DF (the pressure can no longer be written as $P_{zz}=(\beta_e+\beta_i)n/(\beta_e\beta_i)$ as it can in the uniform temperature model). It can be seen that, for $\gamma=1$, we recover the constant density/temperature case of @Abraham-Shrauner-2013.
Provided the DF (\[new df\]) is positive over the whole phase space, then the density, pressure and temperature will also be positive everywhere. Note, however, that the opposite is not true, i.e. a positive density and pressure do not imply a positive DF. We ensure that the DF is positive by choosing parameters in such a way that the conditions (\[a0 bar cond\]) and (\[b0 bar cond\]) are satisfied (for both ions and electrons).
Figure \[fig:density\_temp\_ff\] shows profiles of the density and temperature for different values of $\gamma$, with $k=0$ (the linear force-free case). Figure \[fig:density\_temp1\] shows the same quantities with $k=0.5$. They are normalised to have a value of unity at the lower $z$-boundary of each plot, and we have chosen parameters such that the DFs are positive for ions and electrons (note that if we choose $u_{xe}/v_{th,e}$ then this fixes $u_{xi}/v_{th,i}$ through Eq. (\[uxiuxe\]), if we specify the mass ratio and the ratio $\beta_e/\beta_i$). For $\gamma=1.0$ in each figure, we see that both the density and temperature are constant, as in Abraham-Shrauner’s model. For the other values of $\gamma$ shown, the quantities have a periodic structure. In regions where the density is enhanced/depleted (with respect to the constant value for $\gamma=1$), there is a corresponding depletion/enhancement of the temperature, which ensures that the two quantities multiply together to give a constant pressure, as required for the force-free equilibrium. Additionally, in regions where values of $\gamma>1$ lead to an enhancement/depletion of the quantities, the opposite behaviour is seen when $\gamma<1$, i.e. a depletion/enhancement of the quantities. Similar features are seen by @Kolotkov-2015 (which we obtain in the limit $k\to1$), but note that the density and temperature are not periodic in this case, and so, for a particular $\gamma$ value, there is either an enhancement or depletion of the density/temperature (not both).
We will now briefly discuss some other properties of the model. The plasma beta, defined in this case as the ratio of $P_{zz}$ to the magnetic pressure $B_0^2/(2\mu_0)$, is given (using Eq. (\[micromacro1\])) by$$\begin{aligned}
\beta_{pl} &=& \frac{\bar{a}_0}{\gamma}+\bar{b}_0+\frac{1}{2}.\end{aligned}$$ Using the conditions (\[a0 bar cond\]) and (\[b0 bar cond\]) for positivity of the DF, we have that $$\begin{aligned}
\beta_{pl}> \frac{1}{2}+\frac{1}{2k^2}\left[\exp\left(\frac{\gamma k^2u_{xs}^2}{2v_{th,s}^2}\right) - (1-k^2)\exp\left(-\frac{k^2u_{ys}^2}{2v_{th,s}^2}\right)\right].\end{aligned}$$ For $k=0$ and $k=1$, for example, it is straightforward to show that $\beta_{pl}$ must be greater than unity (as in, e.g., the models in Refs. ), since $u_{xs}^2/v_{th,s}^2\ge0$. The bulk-flow velocity components, defined by $$\begin{aligned}
\langle{\textbf{V}_s}\rangle&=&\frac{1}{n_s}\int\textbf{v}f_s\mathrm{d}^3v,\end{aligned}$$ have the form $$\begin{aligned}
\langle{V_{xs}}\rangle&=&\frac{\gamma u_{xs}\mbox{sn}(z/L)\mbox{dn}(z/L)}{\bar{a}_0+\bar{b}_0+1/2+(\gamma-1)\mbox{sn}^2(z/L)}\\
\langle{V_{ys}}\rangle&=&\frac{u_{ys}\mbox{cn}(z/L)\mbox{dn}(z/L)}{\bar{a}_0+\bar{b}_0+1/2+(\gamma-1)\mbox{sn}^2(z/L)}\\
\langle{V_{zs}}\rangle&=&0.\end{aligned}$$ Through these expressions, we see the role played by the parameters $u_{xs}$ and $u_{ys}$, which can also be written in terms of the ratio of the species gyroradius, $r_{g, s}$, to the current sheet half-width, $L$, by using Eq. (\[ux uy gamma\]) (similarly to @Neukirch-2009) as $$\begin{aligned}
\frac{u_{ys}^2}{v_{th,s}^2}&=&\frac{\gamma^2u_{xs}^2}{v_{th,s}^2}=4\frac{r_{g,s}^2}{L^2}.\end{aligned}$$ The current density can be calculated from the bulk flow velocity as $$\begin{aligned}
\textbf{j}&=&\sum_sq_sn_s\langle\textbf{V}_s\rangle,\end{aligned}$$ and has components $$\begin{aligned}
j_x&=&n_{0}e\gamma(u_{xi}-u_{xe})\mbox{sn}(z/L)\mbox{dn}(z/L)\\
j_y&=&{n_0e}(u_{yi}-u_{ye})\mbox{cn}(z/L)\mbox{dn}(z/L)\\
j_z&=&0.\end{aligned}$$ Using Equations (\[micromacro1\]) and (\[asL\]), we can show that these expressions are equivalent to those obtained macroscopically from Ampère’s law (Eq. (\[asj\])).
In the models in e.g. Refs. , the spatial structure of the current density is determined solely by the structure of the bulk flow velocity since the density is constant, in contrast to the classic Harris sheet model [@Harris-1962], where the bulk flow velocity is constant, and it is the spatial dependence of the density that determines the structure of the current density. In this extended model (and also that of @Kolotkov-2015), however, both the bulk-flow velocity and density are spatially dependent, and so the spatial structure of the current density is determined from the product of the two quantities.
Limiting values of $k$
----------------------
In the limit $k\to1$, the number density, temperature, and pressure (Equations (\[ns\])-(\[pzz\])) go to the form discussed by @Kolotkov-2015 for the force-free Harris sheet, and the DF (\[new df\]) becomes the Kolotkov DF (note that our notation is slightly different). In the limit $k\to0$, the field becomes linear force-free, and we get a DF of the form $$\begin{aligned}
f_s&=&\frac{\gamma^{3/2}n_{0}\exp(-\gamma\beta_sH_s)}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\Bigg(\bar{a}_{0}-\frac{\gamma^2u_{xs}^2}{4v_{th,s}^2}+\frac{\gamma}{4}(\gamma\beta_su_{xs}p_{xs}-\pi)^2\Bigg) \nonumber\\
&{}&+\frac{1}{4}\frac{n_{0}\exp(-\beta_sH_s)}{\left(\sqrt{2\pi}v_{th,s}\right)^3}\left(4\bar{b}_{0}-2-\frac{u_{ys}^2}{v_{th,s}^2}+(\beta_su_{ys}p_{ys}+2)^2\right),\label{k to 0}\end{aligned}$$ which is a modified form of the DF obtained in the $k\to0$ limit of the DF (\[asdf\_rewritten\]). The density and temperature have the form given by Equations (\[ns\]) and (\[temp\]) respectively, where $\mbox{sn}(z/L)=\sin(z/L)$.
Velocity space structure of DF {#sec:vspace}
==============================
In this section, we present some illustrative plots of the DF (\[new df\]) to show the effect of changing $\gamma$, i.e. the effect of changing the energy dependence of the different particle populations. In the $v_x$- and $v_y$- directions, it is possible to choose sets of parameters for which there are multiple peaks in the DF, which may have implications for the stability of the equilibrium. @Neukirch-2009 and @Abraham-Shrauner-2013 derive conditions on the parameters in their models such that their DFs will be single-peaked over the whole phase space. Due to the increased complexity of the DFs in terms of energy dependence, however, we have not yet carried out a full analysis of the velocity space structure - this is left for a future investigation.
In the discussion of the plots below, we will refer to cases where the $p_{xs}$ population is ’hotter’/’colder’ than the $p_{ys}$ one. This refers to the $p_{xs}$ population having an energy dependence resulting in a ’narrower’/’wider’ Maxwellian factor in the DF than the $p_{ys}$ one. We note, however, that because the DFs are not purely Maxwellian, the temperature cannot be properly defined in terms of the width of the DF, but the widths of the first and second parts of the DF gives us a qualitative measure of the temperature difference between the different populations. This notion of temperature should not be confused with the definition of the temperature given in Eq. (\[temp\]).
$v_x$-direction
---------------
In Figure \[fig:gamma1\], we plot the electron DF (\[new df\]) in the $v_x$-direction (for $v_y=v_z=0$) with $\gamma=1$ (i.e. the Abraham-Shrauner DF). We have chosen a set of parameters for which, at $z=0$, the DF has a double maximum in $v_x$ (these are the same parameters as in Figure \[fig:density\_temp1\]). We note, however, that it is also possible to choose parameters for which the DF has only a single maximum in $v_x$ over the whole phase space, if required (by increasing the density of the background populations appropriately). In Figure \[fig:gamma1\], and all subsequent figures in this paper, we normalise the DF to have a maximum value of unity.
Our main aim in this section is to investigate the effect of changing $\gamma$ on the velocity space structure of the DF. This is why we have chosen parameters that give a double maximum for $\gamma=1$, since the effect of changing $\gamma$ is illustrated more clearly in such cases. Figure \[fig:gamma\_lt1\] shows plots of the electron DF for various values of $\gamma$ which are less than unity.
For $\gamma=0.92$, the double maximum still exists, but has become more slight; for the smaller values of $\gamma$ shown ($0.2$ and $0.7$), the double maximum has disappeared. In the $v_x$-direction, the second part of the DF (which does not depend on $\gamma$) has the Maxwellian form $g(p_{ys})\exp(-\beta_sH_s)$. For $\gamma<1$, the $p_{xs}$-dependent population and the first background one are ‘hotter‘ than the $p_{ys}$-dependent and second background populations, and so the Maxwellian factor $\exp(-\gamma\beta_sH_s)$ (in the first part of the DF) has a narrower width than the factor in the second part of the DF. The ’narrow’ first part of the DF, including the cosine which can give double maxima in $v_x$, is therefore ‘swamped‘ by the wider second part for decreasing $\gamma$, and we see the behaviour in Figure \[fig:gamma\_lt1\].
Figure \[fig:gamma\_gt1\] shows plots of the electron DF for various values of $\gamma$ which are greater than unity.
We see that the double maximum in the middle becomes more pronounced as $\gamma$ is increased. This is due to the fact that the Maxwellian $\exp(-\gamma\beta_sH_s)$ multiplying the first part of the DF is now wider than the Maxwellian which multiplies the second part (the $p_{xs}$-dependent population and the first background one are now ‘colder‘ than the $p_{ys}$-dependent and second background populations), so the first part dominates and determines the behaviour of the DF. In Figures \[fig:gamma1\] - \[fig:gamma\_gt1\], we have chosen the parameters $\bar{a}_0$ and $\bar{b}_0$ such that the DFs are positive for all values of $\gamma$ we consider. As can be seen from the positivity conditions (\[a0 bar cond\]), the minimum value of $\bar{a}_0$ becomes significantly larger as $\gamma$ is increased (for fixed values of the other parameters). If we were to further increase $\gamma$ then the central ‘dip‘ of the DF would become more pronounced, and the DF would become negative, hence we would need to increase $\bar{a}_0$ (and adjust $\bar{b}_0$ if required).
$v_y$-direction
---------------
In this section we will show some illustrative plots of the electron DF in the $v_y$-direction for various values of $\gamma$. For the parameter set we used in Figures \[fig:gamma1\] - \[fig:gamma\_gt1\], the DFs are single peaked in all cases except for $\gamma=1.3$, where there is a double maximum as illustrated in Figure \[fig:vy\_gamma1p3\_k0pt5\].
From initial investigations, it seems to be difficult to find a set of parameters from which we can illustrate the effect of increasing or decreasing $\gamma$. This may be due to the fact that multiple maxima appear to occur at high values of $u_{xe}/v_{th,e}$, for which we require large values of $\bar{a}_0$ to ensure positivity of the DF - i.e. a large background density. This often results in the DF being single-peaked for smaller values of $\gamma$.
Possible behaviour of the DF in the $v_y$-direction can be explored heuristically by noting that, for given values $v_x$, $v_z$ and $z$, the DF has the general form $$\begin{aligned}
f_s(v_y)&=&C_1\exp\left(-\frac{\gamma v_y^2}{2v_{th,s}^2}\right)+C_2\exp\left(-\frac{v_y^2}{2v_{th,s}^2}\right)\nonumber\\
&{}&+C_3\exp\left(-\frac{(v_y+ku_{ys})^2}{2v_{th,s}^2}\right)+C_4\exp\left(-\frac{(v_y-ku_{ys})^2}{2v_{th,s}^2}\right),\end{aligned}$$ for constants $C_1$-$C_4$, i.e. it consists of two Maxwellian parts with varying widths, and two shifted Maxwellians - one shifted in the positive $v_y$-direction, and the other in the negative $v_y$-direction (by the same amount). Depending on the relative values of $C_1$-$C_4$, therefore, the DF can exhibit different behaviour, some examples of which are given in Figure \[fig:vy\_examples\]. Note that we have taken different values of $\bar{a}_0$ in each plot, to ensure that the DFs are positive in each case.
Summary {#sec:summary}
=======
In this paper, we have presented a class of 1D strictly neutral Vlasov-Maxwell equilibrium DFs for both linear and nonlinear force-free current sheets, with magnetic fields defined in terms of Jacobian elliptic functions, which are an extension of the DFs discussed by @Abraham-Shrauner-2013 to account for non-uniformities in the temperature and density, whilst still maintaining a constant pressure (with respect to the spatial coordinate), as is required for force-balance of the force-free equilibrium. To achieve this, we have used the method of @Kolotkov-2015, which involves modifying the DF of the original case to include temperature differences between the different particle populations in the model, and then ensuring that strict neutrality is satisfied, and that there is consistency between the microscopic and macroscopic parameters of the equilibrium.
The new DF can be regarded as consisting of four particle populations: one depending on $p_{xs}$, one on $p_{ys}$, and two background populations. The $p_{xs}$-dependent and first background population are taken to have the same energy dependence in the DF, as do both the $p_{ys}$-dependent and second background populations. Note that for the limit of vanishing elliptic modulus, $k$, to give continuous DFs and pressure, density and temperature profiles, we require a particular choice of the constants characterising the background populations, but this form can be changed for other $k$ values if desired (it has the ‘drawback‘ of giving a very large maximum density for certain parameter values).
We have derived sufficient conditions on the parameters such that the positivity of the DFs is ensured, and have given explicit expressions for the density, temperature and pressure across the current sheet. Additionally, we have derived the components of the bulk-flow velocity from the DF, to show that the spatial structure of the current density is determined by the product of the spatial structure of the density and bulk-flow velocity, in contrast to the models of, e.g., @Abraham-Shrauner-2013 and @Neukirch-2009, where the current density structure is determined solely by the structure of the bulk-flow velocity, and also in contrast to the Harris sheet case [@Harris-1962], where it is determined solely by the density structure.
We have investigated limiting cases of the elliptic modulus, $k$. For $k\to1$ the magnetic field becomes that of the force-free Harris sheet, and in this limit we recover a DF similar to that found by @Kolotkov-2015 for this magnetic field. In the limit $k\to0$, the magnetic field becomes linear force-free, and in Abraham-Shrauner’s case the DF takes a form which is similar to one discussed in Refs. , but which is shifted in $p_{xs}$ and $p_{ys}$. In our extended model, the $k\to0$ limit simply gives an extension of this shifted DF to include non-uniformity in both the temperature and density. We have also illustrated graphically the effect of changing the temperature difference between the particle populations in the DF. In the $v_x$-direction, we found that making the $p_{xs}$ part ’colder’ than the $p_{ys}$ part can result in rather pronounced double maxima of the DF (due to a cosine term in $v_x$), but when the $p_{xs}$ part is ’hotter’ these maxima are less significant, or the DF becomes single peaked. In the $v_y$-direction, the DF contains two drifting Maxwellians (with the same energy dependence), and two non-drifting Maxwellians (with different energy dependences), and so there is the possibility of double maxima in the DF depending on the relative values of the coefficients of the separate parts.
Double maxima in the DF may lead to velocity space instabilities (e.g. Ref. ). Due to the increased complexity of the model, however, we have not attempted a systematic study of the velocity space structure, i.e. we have not derived conditions on the parameters such that the DF can be multi-peaked for some $z$, such has been done by @Neukirch-2009 and @Abraham-Shrauner-2013. This is left for a future investigation. We note, however, that it will be possible to choose the density of the background populations large enough such that there are only single maxima of the DF over the whole phase space.
Parameter relations {#sec:neutrality_app2}
===================
In Section \[sec:modified\_df\], by imposing the strict neutrality condition $n_e(A_x,A_y)=n_i(A_x,A_y)=n$, we obtain the relations
$$\begin{aligned}
n_{0e}\exp\left(\frac{u_{xe}^2}{2v_{th,e}^2}\right)& = n_{0i}\exp\left(\frac{u_{xi}^2}{2v_{th,i}^2}\right) = n_0\label{modified_neut1}\\
{a}_{0e}\exp\left(-\frac{u_{xe}^2}{2v_{th,e}^2}\right)& = {a}_{0i}\exp\left(-\frac{u_{xi}^2}{2v_{th,i}^2}\right) = {a}_0\label{a0_cond}\\
a_{1e}\exp\left(-\frac{(1+\gamma k^2)u_{xe}^2}{2v_{th,e}^2}\right)& = a_{1i}\exp\left(-\frac{(1+\gamma k^2)u_{xi}^2}{2v_{th,i}^2}\right) = a_1,\\
{b}_{0e}\exp\left(-\frac{u_{xe}^2}{2v_{th,e}^2}\right) & = {b}_{0i}\exp\left(-\frac{u_{xi}^2}{2v_{th,i}^2}\right) = {b}_0.\label{b cond}\\
b_{1e}\exp\left(\frac{k^2u_{ye}^2-u_{xe}^2}{2v_{th,e}^2}\right)& = b_{1i}\exp\left(\frac{k^2u_{yi}^2-u_{xi}^2}{2v_{th,i}^2}\right) = b_1\\
b_{2e}\exp\left(\frac{k^2u_{ye}^2-u_{xe}^2}{2v_{th,e}^2}\right)& = b_{2i}\exp\left(\frac{k^2u_{yi}^2-u_{xi}^2}{2v_{th,i}^2}\right) = b_2\\
\beta_e\vert u_{xe}\vert&=\beta_i\vert u_{xi}\vert\label{uxiuxe}\\
-\beta_eu_{ye}&= \beta_iu_{yi}.\label{modified_neut2}\end{aligned}$$
Using the choices (\[a0s choice\]) and (\[b0s choice\]) for $a_{0s}$ and $b_{0s}$, the conditions (\[a0\_cond\]) and (\[b cond\]) can equivalently be written as $$\begin{aligned}
\bar{a}_{0e}\exp\left(-\frac{u_{xe}^2}{2v_{th,e}^2}\right) &= \bar{a}_{0i}\exp\left(-\frac{u_{xi}^2}{2v_{th,i}^2}\right) = \bar{a}_0\\
\bar{b}_{0e}\exp\left(-\frac{u_{xe}^2}{2v_{th,e}^2}\right) &= \bar{b}_{0i}\exp\left(-\frac{u_{xi}^2}{2v_{th,i}^2}\right) = \bar{b}_0,\end{aligned}$$ where $a_0=\bar{a_0}+\gamma/(2k^2)$, $b_0=\bar{b}_0-1/(2k^2)$.
By calculating two expressions for the pressure $P_{zz}$, in terms of the macroscopic and microscopic parameters of the equilibrium respectively, and comparing these expressions, we obtain the relations
$$\begin{aligned}
n_0\frac{\beta_e+\beta_i}{\beta_e\beta_i}&=&\frac{B_0^2}{2\mu_0}\label{micromacro1}\\
\frac{{a}_0}{\gamma}+{b}_0&=&\frac{P_{t1}+P_{t2}}{B_0^2/2\mu_0}-\frac{3}{2}\label{p mm}\\
\frac{a_1}{\gamma}&=&-\frac{1}{2k^2}\label{a1}\\
b_1&=&\frac{1}{4}\left(\frac{1}{k}+1\right)^2\label{a2}\\
b_2&=&\frac{1}{4}\left(\frac{1}{k}-1\right)^2\label{a3}\\
\frac{2}{B_0L}&=&\gamma\beta_s\vert u_{xs}\vert q_s=\beta_su_{ys}q_s\Rightarrow u_{ys}=\gamma \vert u_{xs}\vert.\label{ux uy gamma}\end{aligned}$$
Similarly to previous work (e.g. Ref. ), we can derive an expression for the current sheet half-width $L$, in terms of the microscopic parameters, as $$L=\left({\frac{2(\beta_e+\beta_i)}{\mu_0e^2\beta_e\beta_in_0(u_{yi}-u_{ye})^2}}\right)^{1/2}.\label{asL}$$
[47]{} \[1\][\#1]{} \[1\][`#1`]{} urlstyle \[1\][doi: \#1]{}
B. [Abraham-Shrauner]{}. . *Physics of Plasmas*, 200 (10):0 102117, October 2013. [doi: ]{}[10.1063/1.4826502]{}.
D. Y. [Kolotkov]{}, I. Y. [Vasko]{}, and V. M. [Nakariakov]{}. . *Physics of Plasmas*, 220 (11):0 112902, November 2015. [doi: ]{}[10.1063/1.4935488]{}.
N. A. [Bobrova]{} and S. I. [Syrovatski[ǐ]{}]{}. . *Soviet Journal of Experimental and Theoretical Physics Letters*, 30:0 535–+, November 1979.
M. G. [Kivelson]{} and K. K. [Khurana]{}. . *Journal of Geophysical Research*, 100:0 23637–23646, December 1995. [doi: ]{}[10.1029/95JA01548]{}.
. *Force-Free Magnetic Fields: Solutions, Topology and Applications*. World Scientific, Singapore, 1996.
E. [Tassi]{}, F. [Pegoraro]{}, and G. [Cicogna]{}. . *Physics of Plasmas*, 150 (9):0 092113–+, September 2008. [doi: ]{}[10.1063/1.2988338]{}.
E. V. [Panov]{}, A. V. [Artemyev]{}, R. [Nakamura]{}, and W. [Baumjohann]{}. . *Journal of Geophysical Research (Space Physics)*, 116:0 A12204, December 2011. [doi: ]{}[10.1029/2011JA016860]{}.
Thomas Wiegelmann and Takashi Sakurai. Solar force-free magnetic fields. *Living Reviews in Solar Physics*, 90 (1):0 5, Sep 2012. ISSN 1614-4961. [doi: ]{}[10.12942/lrsp-2012-5]{}. URL <http://dx.doi.org/10.12942/lrsp-2012-5>.
Eric Priest. *Magnetohydrodynamics of the Sun*. Cambridge University Press, 2014. [doi: ]{}[10.1017/CBO9781139020732]{}.
I. Y. [Vasko]{}, A. V. [Artemyev]{}, A. A. [Petrukovich]{}, and H. V. [Malova]{}. . *Annales Geophysicae*, 32:0 1349–1360, October 2014. [doi: ]{}[10.5194/angeo-32-1349-2014]{}.
L M Zelenyi, A G Frank, A V Artemyev, A A Petrukovich, and R Nakamura. Formation of sub-ion scale filamentary force-free structures in the vicinity of reconnection region. *Plasma Physics and Controlled Fusion*, 580 (5):0 054002, 2016. URL <http://stacks.iop.org/0741-3335/58/i=5/a=054002>.
Cihan Akcay, William Daughton, Vyacheslav S. Lukin, and Yi-Hsin Liu. A two-fluid study of oblique tearing modes in a force-free current sheet. *Physics of Plasmas*, 230 (1):0 012112, 2016. [doi: ]{}[10.1063/1.4940945]{}. URL <http://dx.doi.org/10.1063/1.4940945>.
D. [Burgess]{}, P. W. [Gingell]{}, and L. [Matteini]{}. . *The Astrophysical Journal*, 822:0 38, May 2016. [doi: ]{}[10.3847/0004-637X/822/1/38]{}.
A. V. Artemyev, V. Angelopoulos, J. S. Halekas, A. Runov, L. M. Zelenyi, and J. P. McFadden. Mars’s magnetotail: Nature’s current sheet laboratory. *Journal of Geophysical Research: Space Physics*, pages n/a–n/a, 2017a. ISSN 2169-9402. [doi: ]{}[10.1002/2017JA024078]{}. URL <http://dx.doi.org/10.1002/2017JA024078>. 2017JA024078.
A. V. [Artemyev]{}, V. [Angelopoulos]{}, J. [Liu]{}, and A. [Runov]{}. . *Geophysical Research Letters*, 44:0 5–11, January 2017b. [doi: ]{}[10.1002/2016GL072011]{}.
J. Birn and E. R. Priest, editors. *Reconnection of Magnetic Fields: Magnetohydrodynamics and Collisionless Theory and Observations*. Cambridge University Press, 1 edition, 3 2007. ISBN 9780521854207.
F. [Wilson]{}, T. [Neukirch]{}, M. [Hesse]{}, M. G. [Harrison]{}, and C. R. [Stark]{}. . *Physics of Plasmas*, 230 (3):0 032302, March 2016. [doi: ]{}[10.1063/1.4942939]{}.
M. G. [Harrison]{} and T. [Neukirch]{}. . *Physics of Plasmas*, 160 (2):0 022106–+, February 2009. [doi: ]{}[10.1063/1.3077307]{}.
M. [Hesse]{}, M. [Kuznetsova]{}, K. [Schindler]{}, and J. [Birn]{}. . *Physics of Plasmas*, 120 (10):0 100704, October 2005. [doi: ]{}[10.1063/1.2114350]{}.
Yi-Hsin Liu, W. Daughton, H. Karimabadi, H. Li, and V. Roytershteyn. Bifurcated structure of the electron diffusion region in three-dimensional magnetic reconnection. *Phys. Rev. Lett.*, 110:0 265004, Jun 2013. [doi: ]{}[10.1103/PhysRevLett.110.265004]{}. URL <https://link.aps.org/doi/10.1103/PhysRevLett.110.265004>.
Fan Guo, Hui Li, William Daughton, and Yi-Hsin Liu. Formation of hard power laws in the energetic particle spectra resulting from relativistic magnetic reconnection. *Phys. Rev. Lett.*, 113:0 155005, Oct 2014. [doi: ]{}[10.1103/PhysRevLett.113.155005]{}. URL <https://link.aps.org/doi/10.1103/PhysRevLett.113.155005>.
F. [Guo]{}, Y.-H. [Liu]{}, W. [Daughton]{}, and H. [Li]{}. . *The Astrophysical Journal*, 806:0 167, June 2015. [doi: ]{}[10.1088/0004-637X/806/2/167]{}.
F. [Zhou]{}, C. [Huang]{}, Q. [Lu]{}, J. [Xie]{}, and S. [Wang]{}. . *Physics of Plasmas*, 220 (9):0 092110, September 2015. [doi: ]{}[10.1063/1.4930217]{}.
F. [Guo]{}, X. [Li]{}, H. [Li]{}, W. [Daughton]{}, B. [Zhang]{}, N. [Lloyd-Ronning]{}, Y.-H. [Liu]{}, H. [Zhang]{}, and W. [Deng]{}. . *The Astrophysical Journal Letters*, 818:0 L9, February 2016a. [doi: ]{}[10.3847/2041-8205/818/1/L9]{}.
F. [Guo]{}, H. [Li]{}, W. [Daughton]{}, X. [Li]{}, and Y.-H. [Liu]{}. . *Physics of Plasmas*, 230 (5):0 055708, May 2016b. [doi: ]{}[10.1063/1.4948284]{}.
F. [Fan]{}, C. [Huang]{}, Q. [Lu]{}, J. [Xie]{}, and S. [Wang]{}. . *Physics of Plasmas*, 230 (11):0 112106, November 2016. [doi: ]{}[10.1063/1.4967286]{}.
N. A. [Bobrova]{}, S. V. [Bulanov]{}, J. I. [Sakai]{}, and D. [Sugiyama]{}. . *Physics of Plasmas*, 8:0 759–768, March 2001. [doi: ]{}[10.1063/1.1344196]{}.
K. [Nishimura]{}, S. P. [Gary]{}, H. [Li]{}, and S. A. [Colgate]{}. . *Physics of Plasmas*, 10:0 347–356, February 2003. [doi: ]{}[10.1063/1.1536168]{}.
K. [Bowers]{} and H. [Li]{}. . *Physical Review Letters*, 980 (3):0 035002, January 2007. [doi: ]{}[10.1103/PhysRevLett.98.035002]{}.
W. [Alpers]{}. . *Astrophysics and Space Science*, 5:0 425–437, December 1969. [doi: ]{}[10.1007/BF00652391]{}.
P. J. [Channell]{}. . *Physics of Fluids*, 19:0 1541–1545, October 1976.
F. [Mottez]{}. . *Physics of Plasmas*, 10:0 2501–2508, June 2003. [doi: ]{}[10.1063/1.1573639]{}.
O. [Allanson]{}, T. [Neukirch]{}, S. [Troscheit]{}, and F. [Wilson]{}. . *Journal of Plasma Physics*, 820 (3):0 905820306, June 2016. [doi: ]{}[10.1017/S0022377816000519]{}.
E. [Moratz]{} and E. W. [Richter]{}. . *Zeitschrift Naturforschung Teil A*, 21:0 1963, November 1966.
A. [Sestero]{}. . *Physics of Fluids*, 10:0 193–197, January 1967.
D. [Correa-Restrepo]{} and D. [Pfirsch]{}. . *Physical Review E*, 47:0 545–563, January 1993. [doi: ]{}[10.1103/PhysRevE.47.545]{}.
N. [Attico]{} and F. [Pegoraro]{}. . *Physics of Plasmas*, 6:0 767–770, March 1999. [doi: ]{}[10.1063/1.873315]{}.
M. G. [Harrison]{} and T. [Neukirch]{}. . *Physical Review Letters*, 1020 (13):0 135003, April 2009. [doi: ]{}[10.1103/PhysRevLett.102.135003]{}.
T. [Neukirch]{}, F. [Wilson]{}, and M. G. [Harrison]{}. . *Physics of Plasmas*, 160 (12):0 122102, December 2009. [doi: ]{}[10.1063/1.3268771]{}.
F. [Wilson]{} and T. [Neukirch]{}. . *Physics of Plasmas*, 18:0 082108, August 2011. [doi: ]{}[10.1063/1.3623740]{}.
C. R. [Stark]{} and T. [Neukirch]{}. . *Physics of Plasmas*, 190 (1):0 012115, January 2012. [doi: ]{}[10.1063/1.3677268]{}.
O. [Allanson]{}, T. [Neukirch]{}, F. [Wilson]{}, and S. [Troscheit]{}. . *Physics of Plasmas*, 220 (10):0 102116, October 2015. [doi: ]{}[10.1063/1.4934611]{}.
Nicolas Dorville, Gerard Belmont, Nicolas Aunai, Jeremy Dargent, and Laurence Rezeau. Asymmetric kinetic equilibria: Generalization of the bas model for rotating magnetic profile and non-zero electric field. *Physics of Plasmas*, 220 (9):0 092904, 2015. [doi: ]{}[10.1063/1.4930210]{}. URL <http://dx.doi.org/10.1063/1.4930210>.
. . http://dlmf.nist.gov/, Release 1.0.14 of 2016-12-21. URL <http://dlmf.nist.gov/>. F. W. J. Olver, A. B. [Olde Daalhuis]{}, D. W. Lozier, B. I. Schneider, R. F. Boisvert, C. W. Clark, B. R. Miller and B. V. Saunders, eds.
Paul F. Byrd and Morris David Friedman. *Handbook of Elliptic Integrals for Engineers and Scientists*. Springer Berlin, 2013.
E. G. [Harris]{}. . *Nuovo Cimento*, 23:0 115, 1962.
S. Peter Gary. *Theory of Space Plasma Microinstabilities (Cambridge Atmospheric and Space Science Series)*. Cambridge University Press, 2005. ISBN 0521437482.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The cyanides $M$CN, $M$=Cu, Ag, Au, have experimentally a structure with hexagonally packed, infinite -$M$-CN-$M$-CN- chains. Following our earlier study for AuCN, we now predict that all three $M$CN could have an alternative $M_3$C$_3$N$_3$ sheet structure of comparable energy with the known one. The valence isoelectronic systems $M$C$_2$ versus $M_3$C$_6$, $M$=Be-Ba; Zn-Hg are also studied. Now, the known dicarbides have the CaC$_2$ or MgC$_2$ chain structures. The predicted sheets lie energetically below the chains for $M$ = Zn, Cd, and Hg. All these systems are experimentally unknown. Indeed, they are clearly endothermic, compared to the elements. For some sheet structures the densities of states suggests rather small band gaps and even metallic character. When available, the experimental geometries agree well with the calculated ones for both cyanides and dicarbides.'
author:
- 'P. Zaleski-Ejgierd[^1]'
- 'M. Hakala[^2]'
- 'P. Pyykkö[^3]'
bibliography:
- 'pze-1\_prb.bib'
title: |
**Comparison of chain versus sheet crystal structures for\
cyanides $M$CN ($M$ = Cu-Au) and dicarbides $M$C$_2$\
($M$ = Be-Ba; Zn-Hg). Alternatives to graphene ?**
---
=10000
[^4]
Introduction {#sec:introduction}
============
One of the fundamental goals of computational quantum chemistry is the prediction of new chemical species and the determination of their properties[@Gagliardi:06]. A recent example is gold cyanide, AuCN, a well-known commodity chemical, whose known structure consists of infinite -CN-Au-CN-Au- chains[@Zhdanov:01]. The chains are packed together on a hexagonal grid in such a way, that the Au atoms are in the same plane (see Fig. \[Fig:(A’)(B’)\]: A’). However, Hakala and Pyykkö[@Hakala:06] recently predicted for AuCN an alternative crystal structure, having a closely similar energy but only $\sim{ 70}$% of the density (see Fig. \[Fig:(A’)(B’)\]: B’). The new structure contains triazine-type six-rings of three carbon and three nitrogen atoms, C$_3$N$_3$, which are coupled to each other by linearly coordinated gold(I) atoms forming a two-dimensional sheet structure. The sheets attract each other weakly due to the gold-gold aurophilic interaction. Such a new structure has not yet been observed. Since single-sheet-type 2D atomic crystals, like graphene, are potentially a very important class of materials but much less known than the 3D counterparts[@Novoselov:05], it is interesting to ask whether such a structure could also be found for other similar systems, such as the cyanides involving the other Group 11 (coinage) metals, namely copper and silver. Their known crystalline structure is also chain-type: -CN-$M$-CN-$M$- ($M$ denoting the metal)[@Bowmaker; @Hibble:02; @Hibble:03b].
By applying the valence isoelectronic principle, one can extend the search for such a new sheet-type structures to further cases. Replacing the formula $M$CN of the metal cyanides by the formula $M$C$_2$, with an (ns)$^2$ configuration for the neutral metal atom $M$, leads to the class of crystalline dicarbides. Metallic dicarbides may have interesting electronic properties, such as the superconductivity of YC$_2$ and its layered compounds[@YC2-Henn]. In the present work we have studied metallic elements in Groups 2 (alkaline earth metals $M$=Be, Mg, Ca, Sr, Ba), and 12 (transition metals $M$=Zn, Cd and Hg). Stable crystalline structures are known only for some of these systems (MgC$_2$, CaC$_2$, SrC$_2$, BaC$_2$). No earlier data were found for any Group 12 dicarbides. The previously known crystal structures of CaC$_2$, SrC$_2$, BaC$_2$ from Group 2 consist of infinite -$M$-CC-$M$-CC- chains, but each cation is now equatorially surrounded by four anions, each parallel to the $z$ axis (see Fig. \[Fig:(A)(B)(C)\]: B). Another known Group 2 dicarbide, MgC$_2$, also has infinite chains, but the four equatorial dicarbide ligands around the cation are now perpendicular to the $z$ axis (see Fig. \[Fig:(A)(B)(C)\]: C). These structures have been only very recently clarified by X-ray and neutron diffraction[@MgC2-Karen; @SrC2-Vohn; @BaC2-Vohn].
In the present work we show that sheet-type structures such as the poly(triaurotriazine), one recently reported[@Hakala:06] for AuCN can be also found for the other Group 11 coinage metal cyanides and for the Group 2 and 12 dicarbides. All the proposed sheet structures seem to be entirely new. For Group 12 dicarbides and for BeC$_2$ of Group 2 we also investigate whether they could form stable structures in the tetragonal geometry starting the search from the experimental geometry known for Group 2. We have performed density of states (DOS) analysis for both isolated and packed structures, and discuss the insulator vs. metallic properties of the studied systems. Vibrational analysis has been performed to provide tools for the identification and to analyze the stability of the systems. The geometries and frequencies are compared to the known cyanide and dicarbide results. Finally, the thermodynamical preferences of the predicted dicarbides are estimated by calculating their formation energies with respect to the elemental metals and graphite as starting materials.
Method
======
All calculations were performed within density functional theory (DFT), using the Vienna [*ab-initio*]{} Simulation Package[@VASP:01; @VASP:02; @VASP:03] (VASP). Plane-wave basis sets, Ultrasoft (US) pseudopotentials[@US:01; @US:02] and projector augmented-wave (PAW) potentials[@PAW:01; @PAW:02] with the generalized gradient approximation[@PAW-GGA] (GGA) for the Perdew-Burke-Ernzerhof[@PBE] (PBE) exchange-correlation functional, were employed . When optimizing the lattice parameters and the atomic positions, constant-volume calculations were performed with the symmetry of the unit cell kept fixed. We typically scanned the crystal volume for each system by changing the interchain or intersheet distance in steps of 0.1 Å and letting the system relax. In this work we will report the energy value and the atomic geometry corresponding to the minimum point on the total energy curve for each system.
For the energy cutoff we used a standard 400 eV value. A $\Gamma$-centered 6$\times$6$\times$6 k-point grid was used throughout the work for the lattice optimizations and vibrational analysis. All the chain structures and the majority of the sheet structures studied were found to be insulators, and this choice of the k-point grid was found sufficient for the convergence in total energy and geometry. For the dynamical matrix and vibrational frequencies a shift of $0.025$ Å was applied to the atomic positions. Some sheet structures exhibited the closing of the band gap as the sheets were packed together, and two dicarbides were manifestly metallic already as isolated sheets. The semimetallic vs. metallic character for these systems was inferred by studying the DOS at the Fermi level. For the systems with a small band gap the used k-point grid may lead to a slight overestimation of the total energy ($\sim$0.005 eV) and the sheet-sheet distance ($\sim$0.2 Å), as compared with the calculations using denser k-point grids. We expect a similar behavior also for the two metallic systems. In our tests, the intrachain and intrasheet distances C-N, $M$-C and $M$-N were well converged within 0.2% with respect to increasing the k-point grid above that of 6$\times$6$\times$6. When the dispersion interactions are the dominant mechanism, such as in the case of interchain distances in cyanides, the present DFT-GGA results are not expected to be reliable and should be considered with caution.
The studied geometries are illustrated in Figs. \[Fig:(A’)(B’)\]-\[Fig:(F)(G)\]. We refer to the geometries as ’chains’ (see Fig. \[Fig:(A’)(B’)\]: A’ and Fig. \[Fig:(A)(B)(C)\]: A, B, C), and to the rest as ’sheets’. Furthermore, it turned out to be necessary to consider four different packings of the sheets: ’Simple stacking’, where each 2D ’sheet’ is simply repeated in the perpendicular direction to form the 3D crystal (see Fig. \[Fig:(A’)(B’)\]: B’ and Fig. \[Fig:(D)(E)\]: D); ’Shifted’, where every second 2D sheet is parallelly displaced, so that the benzene-like carbon ring lies on top of the ’triangle’ formed by the metal atoms (see Fig. \[Fig:(D)(E)\]: E); ’Rotated’, as ’simple stacking’, but the carbon ring is rotated inside the sheet with respect to the connecting metals by 30$^\circ$ (see Fig. \[Fig:(F)(G)\]: F); ’Shifted+rotated’, where in addition to the ’rotated’ structure, there is also the displacement parallel to the plane as described above (see Fig. \[Fig:(F)(G)\]: G).
Results and Discussion
======================
Geometries and total energies
-----------------------------
### Cyanides $M$CN
The properties of the studied coinage metal cyanides CuCN, AgCN and AuCN are found to be very similar concerning geometries and total energies. Like in the case of AuCN,[@Hakala:06] , our simulations reproduce, for the two other compounds, the experimentally known geometry within the chain and predict the possible existence of a sheet-type structure. The chain-type structure of AgCN and CuCN is known to differ from that of AuCN due to an even weaker chain-chain interaction. In the case of AuCN the aurophilic interaction keeps the interchain structure to a large extent ordered (6-fold $P6mm$ symmetry about the chain axis), whereas in the former cases there is a considerable degree of disorder, leading to a lower 3-fold symmetry.[@Bowmaker; @Hibble:02; @Hibble:03b] For the present analysis all the chain-type structures were confined to the $P6mm$ symmetry (Fig. \[Fig:(A’)(B’)\]: A’). This is well sufficient for the description of the strong covalent bonds C-N, $M$-C, and $M$-N within the chains. The interchain search geometry for CuCN and AgCN is not the one experimentally suggested but, as the attraction between the chains is due to dispersion, DFT-GGA is not expected to perform correctly for this distance in any case.
For the $M_3$C$_3$N$_3$ sheet structures, a total energy minimum was found for the ’simple stacking’ geometry (see Fig. \[Fig:(A’)(B’)\]: B’), with symmetry $P\overline{6}2m$, whereas no stable structures were found corresponding to the other sheet geometries. The total energies for the chain and sheet structures are compared in Table \[Table:(Cyanides - Energy difference)\]. The energy differences are about 0.2 eV for CuCN and AgCN, but for AuCN there is practically no difference between the two structures. Due to the weak sheet-sheet and chain-chain interactions, the comparison of the isolated structures leads to the same conclusion.
Case CuCN AuCN AuCN
----------------------- ------- ------- ------
3D Sheet vs. 3D Chain -0.22 -0.18 0.00
2D Sheet vs. 1D Chain -0.24 -0.21 0.03
: Total energy difference [*$\Delta$ E*]{} (eV) between the chain and sheet structures for Group 11 metal cyanides (per $M$CN formula unit). Both 3D and isolated cases are reported. Negative value indicates that the chain structure is more stable.
\[Table:(Cyanides - Energy difference)\]
Table \[Table:(Cyanides - bond lengths)\] gives the bond lengths and the lattice constants, along with the experimental values for the known $M$-N, $M$-C and C-N distances. We compare our results with the experimental values from Hibble and co-workers[@Hibble:02; @Hibble:03; @Hibble:03b], who performed a systematic total neutron diffraction study for these structures. Their results differ somewhat from the conventional Bragg scattering studies, which, according to them, do not yield completely satisfactory results. For the chain structures, the experimental intrachain covalent bond lengths are very well described with distances accurate to 2% or better. In particular, the C-N distance is well reproduced. The difference between the $M$-C and $M$-N bond lengths in sheets compared to the same difference in chains, is larger. For sheets, the C-N bond lengths are practically the same for all the cyanides, but larger by about 20 pm compared to the chain structure, reflecting the aromatic six-ring bonding, as contrasted to the -C$\equiv$N- triple bonds.
Because of the well known difficulties in describing dispersion effects by DFT, the interchain and intersheet distances should be considered much more uncertain, which can be seen in the large theory-experiment discrepancy of the lattice parameter $a$ for AuCN, which for this system corresponds to the chain-chain distance. The comparison of the interchain distances with the experimental values is further complicated, because the chains are randomly displaced along the chain axis, and for AgCN and CuCN there are no chemical bonds between the chains[@Hibble:02; @Hibble:03; @Hibble:03b]. As a matter of fact, the latter is reflected in the very weak binding energy between the isolated and packed CuCN and AgCN chains in our calculations. The binding energy for isolated sheets required to form a 3D structure is found to be typically 0.05 eV.
It is interesting to note that room-temperature powder-X-ray pattern for CuCN shows unexplained weak Bragg peaks at distances $d_1=260.5$ pm and $d_2=235.3$ pm (Ref.). These have been attributed to an unknown highly crystalline form $\beta$-CuCN with density 2.97 g cm$^{-3}$ (cf. 3.03 g cm$^{-3}$ for the known $\alpha$-CuCN). Whether such peaks could be correlated with our predicted sheet-type geometry or its modifications remains an open question. The crystal density of the sheet structure is 1.8 g cm$^{-3}$, i.e., $\sim 40$% smaller than that assigned to $\beta$-CuCN. This computed density is, however, strongly influenced by the sheet-sheet distance $c$, which is possibly underestimated due to the lack of dispersion interaction. The present value of $c$ is $\sim 28$% and $\sim 41$% larger than the distances $d_1$ and $d_2$, respectively.
Chains Source $M$-C $M$-N C-N $a$ $c$ $V$
----------------------- ---------- ----------- ----------- ----------- ------------- ------------ -------------------------------------------
CuCN (A’) Calc. 182.0 181.9 117.4 [*407.7*]{} 481.2 [*69.27*]{}
Exp.[^5] 184.6 184.6 117.0 591.2(3) 486.107(3) 49.0467
AgCN (A’) Calc. 201.9 204.0 117.0 [*410.8*]{} 523.9 [*76.56*]{}
Exp.[^6] 206 206 116 590.32 528.29 53.1443
AuCN (A’) Calc. 194.3 198.3 116.8 379.4 511.3 63.74
Exp.[^7] 197.03(5) 197.03(5) 114.99(2) 340.5(4) 509.2(2) 51.1273
Sheet Source $M$-C $M$-N C-N $a$ $c$ $V$
Cu$_3$C$_3$N$_3$ (B’) Calc. 187.6 190.0 136.6 650.7 332.4 81.24
Ag$_3$C$_3$N$_3$ (B’) Calc. 207.4 213.8 136.1 693.3 341.5 94.78
Au$_3$C$_3$N$_3$ (B’) Calc. 199.5 210.4 136.3 682.5 352.4 94.78 \[Table:(Cyanides - bond lengths)\]
### Dicarbides $M$C$_2$ {#Sec:Dicarbides}
The geometries of the metal dicarbides in the known experimental chain structure ($M$ = Mg-Ba in Group 2) are well reproduced by our calculations. In addition, also for the dicarbides we predict the possible existence of sheet structures for both Groups 2 and 12, and for Group 12 we find possible chain structures. Experimentally, the Group 2 dicarbides crystallize in two different tetragonal symmetries, [*viz*]{} I4/mmm for $M$C$_2$ ($M$ = Ca-Ba) and I4$_2$/mnm for MgC$_2$. Table \[Table:(Carbides - group 2 - bond lengths chains)\] contains the calculated bond lengths and lattice parameters for the Group 2 chain dicarbides in comparison with the available experimental data. Table \[Table:(Carbides - group 12 - bond lengths chains)\] contains the predictions for chain structures for the Group 12.
As most of the chain- and sheet-type dicarbide structures were found to have no or low imaginary frequencies, the structures can be considered stable. The frequencies are presented in Sec. \[vibrational\_frequencies\]. Notable exceptions are BeC$_2$ and Be$_3$C$_6$, which exhibited imaginary frequencies of the order of 300$i$ cm$^{-1}$, suggesting that the found minimum energy geometry is only a transition state between phases of other symmetry. We did not try to follow the mode to a lower-symmetry minimum. Some imaginary frequencies of the order of 100$i$ cm$^{-1}$, were found for chain-type MgC$_2$, whose structure is, however, in a very good agreement with the experiment. Smaller imaginary frequencies (about 50$i$ cm$^{-1}$ or smaller) were found for ZnC$_2$, CdC$_2$, Ca$_3$C$_6$, Sr$_3$C$_6$ and Ba$_3$C$_6$. We believe the origin of these imaginary frequencies is related either to the numerical inaccuracies or the anharmonicity of the total energy surface. In fact, small imaginary frequencies have been interpreted in terms of soft mode phonons.[@Sternik:01]
For the known Group 2 dicarbides in the chain structure, the calculated covalent bond lengths are in good agreement with experiment. Our calculations systematically give slightly larger values of the C-C bond length (around 125 pm) compared to the experimental results (around 120 pm), the C-C bond length in acetylene molecule[@Herzberg] (120.4 pm) or the bond length calculated from the carbon triple-bond covalent radius (120 pm)[@Pyykko:05]. A possible reason for the difference could be the vibrational motion of the dicarbide group in the experiments[@MgC2-Karen]. In realistic physical conditions, when thermal energy is applied, C$_2^{2-}$ group can both rotate and vibrate, thus leading to an underestimation of the C-C bond length by experimental measurement. An interesting point to notice is that for ZnC$_2$, CdC$_2$ and HgC$_2$, we predict the possible existence of a chain structure in the I4/mmm geometry similar to the $M$C$_2$ ($M$ = Ca-Ba) structures. The I4$_2$/mnm geometry for ZnC$_2$, CdC$_2$ and HgC$_2$ was also found to be stable, but with slightly higher energies.
The difference in the packing geometry of the chain-type dicarbides compared to the packing geometry of the coinage metal cyanides can be understood from the different relative contributions to the interaction forces. Between the monovalent (Au$^+$)(CN$^-$) chains the dispersion forces predominate and one Au$^+$ is surrounded by six other nominally Au$^+$ ions. For the divalent ($M^{2+}$)(C$_2^{2-}$) chains, on the other hand, the Coulomb forces predominate. Hence in this case the ($M^{2+}$)(C$_2^{2-}$) chains are displaced with respect to each other, so that one $M^{2+}$ ion is surrounded by four C$_2^{2-}$ ions (see Fig. \[Fig:(A)(B)(C)\]: B, C). In support of this, infinite ($M^{2+}$)(C$_2^{2-}$) chains packed into a tetragonal P4mm unit cell (see Fig. \[Fig:(A)(B)(C)\]: A) were found purely repulsive.
--------------------- ------------- ------- --------------- ------- ---------------- ------- --------------- ------- --------------- --
Case BeC$_2$ (B)
Calc. Calc. Exp.[^8] Calc. Exp.[^9] Calc. Exp.[^10] Calc. Exp.[^11]
Lattice const.: $a$ 390.6 396.9 393.42(7)$^a$ 391.3 388.582(4)$^a$ 414.4 411.43(2)$^a$ 440.6 439.43(6)$^a$
$c$ 463.2 492.5 502.1(1) 640.0 640.05(1) 681.8 676.60(4) 724.1 712.5(2)
Volume $V$ 35.01 38.8 38.85 49.0 48.20 58.60 57.26 70.3 68.88
Symmetry I4/mmm
Bond lengths
276.1 280.6 278.2 276.7 274.7 293.0 290.8 311.6 310.7
-C 169.1 217.8 217.4 256.8 255.5(4) 277.7 278.5(8) 298.8 297.0(7)
C-C 124.9 125.8 121.5 126.4 129.7(8) 126.4 120(2) 126.6 118.6(13)
--------------------- ------------- ------- --------------- ------- ---------------- ------- --------------- ------- --------------- --
\[Table:(Carbides - group 2 - bond lengths chains)\]
Case ZnC$_2$ (B) [CdC$_2$]{} (B) [HgC$_2$]{} (B)
-------------------- ----- ------------- ----------------- -----------------
Lattice constants: $a$ 425.1 438.1 510.5
$c$ 505.2 546.9 529.0
Volume $V$ 45.7 52.5 68.9
Symmetry I4/mmm I4/mmm I4/mmm
Bond lengths
300.6 309.8 361.0
-C 190.5 211.2 202.9
C-C 124.2 124.5 123.2
: Calculated lattice parameters $a$ and $c$ (pm), cell volumes $V$ (Å$^3$), symmetries and bond lengths (pm) for Group 12 metal dicarbides in the chain structure. Volume is given with respect to one $M$C$_2$ unit. The capital letter after the compound refers to the crystal structures in Fig. 2.
\[Table:(Carbides - group 12 - bond lengths chains)\]
Case Be$_3$C$_6$ (E) Mg$_3$C$_6$ (G) Ca$_3$C$_6$ (G) Sr$_3$C$_6$ (G) Ba$_3$C$_6$ (G) Zn$_3$C$_6$ (E) Cd$_3$C$_6$ (E) Hg$_3$C$_6$ (E)
--------------------- ------------------- ------------------- ------------------- ------------------- ------------------- ------------------- ------------------- -------------------
Lattice const.: $a$ 390.6 665.0 736.2 782.7 826.2 674.4 715.2 708.2
$c$ 463.2 526.8 543.3 570.0 592.1 711.1 695.1 794.1
Volume $V$ 56.9 67.2 84.9 100.7 116.9 93.4 102.6 115.1
Symmetry $P\overline{6}2m$ $P{6\over m} m m$ $P{6\over m} m m$ $P{6\over m} m m$ $P{6\over m} m m$ $P\overline{6}2m$ $P\overline{6}2m$ $P\overline{6}2m$
Bond lengths
-C 168.9 214.2 246.2 273.2 280.4 194.8 214.9 212.7
C-C 142.9 142.9 143.2 140.0 141.2 142.6 141.9 140.7
\[Table:(Carbides - group 2+12 - bond lengths sheets)\]
In the search of stable 2D infinite metal dicarbide sheets $M_3$C$_6$, the similar hexagonal symmetry D as in the $M_3$C$_3$N$_3$ systems B’ was used as a starting point (initially with a large sheet separation). Here, the benzene-like carbon rings are coupled together from the corners by the metal atoms. The calculations in this geometry yielded minimum energy structures for $M$ = Be, Zn-Hg. For the remaining metals, the isolated 2D sheet spontaneously relaxed to a slightly different geometry within the same hexagonal crystal symmetry. The carbon rings rotated 30$^\circ$ with respect to the initial structure to yield the 2D sheet geometry shown in F and G. In this structure the metal atoms are located in the midpoint of the line connecting the C-C edges of every two benzene-like rings. Considering the 3D stacking of these structures, a ’simple stacking’ (both ’rotated’ and ’nonrotated’ cases) was found purely repulsive. Instead, a new 3D packing had to be introduced, where every second sheet was translated by the vector $\vec{v}= ({1\over 2}a, {1\over 2}a, 0)$. For both the structures (E and H), stable geometries were found.
Fig. \[Fig:(Plot-All)\] summarizes the behavior of total energies for the most stable sheets and chains (the numerical values are tabulated later on in Sec. \[sec:formation\]). The energy difference [*$\Delta E$*]{} is given for each dicarbide per $M$C$_2$ unit; a negative value indicates that the chain structure is preferred. For the experimentally known dicarbides, $M$ = Mg-Ba, the chains are energetically more stable. The energy difference roughly increases following $Z$, being in all the cases less than 1 eV. Interestingly, for the dicarbides not experimentally known, $M$ = Be, Zn, Cd and Hg, the sheet structure is predicted to be the most stable phase, and the energy difference likewise increases roughly following $Z$.
Vibrational frequencies {#vibrational_frequencies}
-----------------------
For the cyanide structures, vibrational analysis was performed for isolated chains and isolated sheets, whereas for dicarbides the analysis was done for packed chains and isolated sheets. The rest of the packed structures were not studied because of the difficulties due to the weak chain-chain (for cyanides) and sheet-sheet (for both cyanides and dicarbides) interactions. The frequencies are given in Figs. \[Fig:vibr\_cyanides\]-\[Fig:vibr\_dicarb12\], and Tables \[Table:(Cyanides - Vibrations)\] and \[Table:(Carbides - Vibrations)\],\[Table:(Carbides - Vibrations)2\] with comparison to available experiments.[@Bowmaker; @MgC2-Karen; @CaC2-Knapp; @SrC2-Vohn; @BaC2-Vohn].
Following the assignment by Bowmaker [*et al.*]{}[@Bowmaker] for the modes in infinite cyanide chains, it can be seen that the simulation reproduces very well both the vibrational and bending normal modes for the known structures. Here the two highest modes are the $\nu$(CN) and $\nu$($M$C/N) stretching modes, and the two lowest ones the $\delta$($M$CN) and $\delta$(N$M$C) bending modes. For the known dicarbides in the chain structure, the highest frequencies correspond to the stretching mode and are very close to the experimental ones.
The sheet structures for both cyanides and dicarbides exhibit systematically smaller maximum frequencies compared to those of the chain structures. The higher eigenfrequencies for sheets correspond to in-plane vibrations. Similarly as the bending modes in chains, the majority of the lower frequencies in sheets corresponds to bending vibrations perpendicular to the plane. For AuCN, the first and second frequency (Table \[Table:(Cyanides - Vibrations)\]) are close to the in-plane ring deformations of an isolated molecule (1395 and 1208cm$^{-1}$) (Ref. ).
We propose that the possible experimental identification of the sheet-type structures for cyanides could be performed within the range of roughly (650-1500) cm$^{-1}$. The majority of the in-plane vibrations of the sheets falls into this wavelength region, where no frequencies corresponding to chains interfere. Similarly, the identification of sheet-type dicarbides could be done in the range (700-1800) cm$^{-1}$, undisturbed by the frequencies of the known chain structures. The frequencies in these ranges correspond mainly to vibrations parallel to the sheets (i.e., parallel to the covalent bonds), and are thus not significantly affected by the DFT deficiencies for dispersion.
------------------ ---------------- ----------- ---------------- ---------- --------------- ----------
Case
Chains Calc. Exp.[^12] Calc. Exp.$^a$ Calc. Exp.$^a$
Stretch (C-N) 2195 2170 2221 2164 2180 2236
Stretch ($M$-CN) 642 591 529 480 571 598
Bend 1 337,336 326 281,277 272 422,421 358
Bend 2 228,221 168 178,173 112 253,251 224
Sheets Calc. Exp. Calc. Exp. Calc. Exp.
1382,1379 - 1392,1386 - 1336,1334 -
1212,1210 - 1221,1216 - 1219,1218 -
1174,1109,1057 - 1174,1087,1039 - 1178,1032,973 -
799,798 - 763,762 - 804,802 -
------------------ ---------------- ----------- ---------------- ---------- --------------- ----------
\[Table:(Cyanides - Vibrations)\]
Chains MgC$_2$ CaC$_2$ SrC$_2$ BaC$_2$ ZnC$_2$ CdC$_2$ HgC$_2$
------------------------- ----------- ----------- ----------- ----------- ----------- ----------- ----------- --
Experimental value[^13] - 1860 1850 1832 - - -
Calculated value 1861,1854 1872,1866 1861,1860 1853,1851 2043,2040 2015,2012 2090,2090
261,260 308,294 245,234 200,190 669,652 563,545 627,621
\[Table:(Carbides - Vibrations)\]
Mg$_3$C$_6$ Ca$_3$C$_6$ Sr$_3$C$_6$ Ba$_3$C$_6$ Zn$_3$C$_6$ Cd$_3$C$_6$ Hg$_3$C$_6$
------------- ------------- ------------- ------------- ------------- ------------- ------------- --
1364,1362 1230,1227 1202,1199 1391,1342 1382,1381 1361,1360 1394,1391
1168 1090,1074 1063,1030 1117,1035 1220 1226 1246
1134,1131 1073 963,901 879,807 1187,1185 1179,1178 1219,1218
966,954 909,828 631 619 1040,1027 1024,947 1050,1001
\[Table:(Carbides - Vibrations)2\]
Band gap and metallicity {#sec:DOS}
------------------------
The band gap values for the studied systems are reported in Table \[Table:(DOS - HOMO/LUMO gap)\]. Since a full band structure analysis was not performed, the values given should be considered as indicative only. In the case of both cyanide and dicarbide chain structures, all the known and predicted 1D (isolated) and 3D structures are found to be insulators. For the packed 3D dicarbide chain structures, the band gap is largest for HgC$_2$ (3.7 eV) and smallest for BaC$_2$ (1.6 eV). The packed chain structures exhibit systematically smaller band gaps than the isolated chains due to the increased overlap of the valence orbitals. The only exception to this picture is BaC$_2$, which seems to show an opposite trend. All the newly predicted chain dicarbides ($M$ = Zn-Hg) have larger gaps than the already known chain dicarbides.
For both cyanide and dicarbide 2D (isolated) and 3D sheet structures the band gaps are clearly smaller than for the chain structures. Packing the sheets to a 3D structure decreases, furthermore, very strongly the band gap value. For dicarbide sheets we identify two candidates that could be purely metallic compounds: Sr$_3$C$_6$ and Ba$_3$C$_6$. Both systems may have a significant amount of charge carriers (we find this for both isolated and 3D sheet structures). The rest of the cyanide and dicarbide compounds are found to be either insulators or semimetals with a slight band overlap between the occupied and unoccupied states. In the 3D packed sheet geometry, Zn$_3$C$_6$ and Hg$_3$C$_6$ have the highest gaps ($\sim 0.7$ eV).
--------- ---------- -------- ---------- ---------
Case
Isolated Packed Isolated Packed
CuCN 3.9 2.5 1.6 0.2
AgCN 4.7 3.1 1.8 0.3
AuCN 4.6 2.3 1.6 0.2
BeC$_2$ 3.6 2.9 0.6 SM[^14]
MgC$_2$ 4.2 2.0 1.5 SM
CaC$_2$ 2.6 1.7 SM SM
SrC$_2$ 2.0 1.7 M[^15] M
BaC$_2$ 1.3 1.6 M M
ZnC$_2$ 3.9 2.4 1.2 0.7
CdC$_2$ 4.0 2.8 0.6 SM
HgC$_2$ 3.8 3.7 0.9 0.7
--------- ---------- -------- ---------- ---------
: HOMO/LUMO band gap (eV) for isolated and packed cyanide and dicarbide structures. The metallic or semimetallic character is indicated when relevant.
\[Table:(DOS - HOMO/LUMO gap)\]
Formation energy of dicarbides $M_3$C$_6$ {#sec:formation}
-----------------------------------------
We have estimated the formation energy of the predicted new dicarbides as the difference between their total energy and the total energy of the corresponding pure elemental metal and graphite: $$E_{\mathrm{F}} = E(M\mathrm{C}_2) - [E(M) + 2E(\mathrm{C})],$$
where $E(M\mathrm{C}_2)$ is the total energy of the chain or sheet structure per $M$C$_2$ unit. $E(M)$ and $E(\mathrm{C})$ are the total energies of one atom in the zero-temperature phase of the bulk crystal and in graphite, respectively. The total energy for graphite was calculated to be -9.18 eV. The results are given in Table \[Table:Formation energy\], where we have also given the atomization energies for the elemental metals, which compare reasonably well with experimental formation enthalpies.
From a thermodynamic consideration, we see that the formation of the dicarbides of Mg and Zn-Hg is clearly endothermic, while CaC$_2$-BaC$_2$ are nearly thermoneutral. Of them, the [*M*]{}C$_2$, [*M*]{} = Mg, Ca, Sr and Ba, exist. As check on the method, we compare the calculated atomization energies for the metals. Instead of graphite, which is a highly stable material, one could think of different benzene-derived substances as possible starting materials in the synthesis of the predicted new sheet-type compounds.
Case $E_{\mathrm{f}}^{\mathrm{sheet}}$ [*$\Delta E$*]{} $E_{\mathrm{at}}^M$ $H_{\mathrm{at}}^M$ (exp.$\footnote{Ref. \onlinecite{CRC-Handbook}}$)
------------- ----------------------------------- ------------------ --------------------- ----------------------------------------------------------------------- -- -- -- -- -- --
Mg$_3$C$_6$ 1.55 -0.36 1.51 1.531
Ca$_3$C$_6$ 0.61 -0.65 1.91 1.847
Sr$_3$C$_6$ 0.92 -0.91 1.61 1.704
Ba$_3$C$_6$ 1.05 -0.85 1.88 1.866
Zn$_3$C$_6$ 1.99 0.67 1.11 1.355
Cd$_3$C$_6$ 2.32 0.65 0.72 1.161
Hg$_3$C$_6$ 1.99 1.21 0.08 0.635
: Formation energy $E_{\mathrm{f}}$ for Group 2 and 12 sheet-type dicarbides and difference to the chains ([*$\Delta E$*]{} = $E^{\mathrm{chain}} - E^{\mathrm{sheet}}$) per $M$C$_2$ unit. The calculated atomization energy $E_{\mathrm{at}}^M$ and the experimental atomization enthalpy $H_{\mathrm{at}}^M$ for the pure element $M$ are also given. All values are in eV.
\[Table:Formation energy\]
Conclusions
===========
We have suggested a number of new solid substances, of which the sheet-type $M_3$C$_3$N$_3$ ($M$ = Cu, Ag, Au) and the sheet-type $M_3$C$_6$ ($M$ = Mg-Ba) may have the best chance of being prepared. The density of the new substances is in general lower than that of the known chain-type structures. For the identification of these new substances we have reported their vibrational frequencies. The majority of the in-plane frequencies of the new compounds are clearly separated from the stretching and bending frequencies of the known chain-type cyanide and dicarbide structures. Evidence is found that the Sr and Ba sheet-type compounds could be metallic both in the 3D crystal and as isolated sheets. All the Group 2 or Group 12 sheet-type structures are endothermic. The least endothermic one is calcium carbide. If any of these compounds can be made, the metallic Sr and Ba ones may find the most interesting applications. The sheet structures of lanthanides might form a further possible class of new compounds. Note that CaC$_2$-structured, insulating or metallic [*Ln*]{}C$_2$ are known for [*Ln*]{} = Y, La, Ce, Tb, Yb, Lu and U.[@Lanthanides]
We belong to the Finnish Center of Excellence in Computational Molecular Science (2006-2011). The grants 110571, 200903, 201291, 205967 and 206102 of The Academy of Finland are also gratefully acknowledged. The work was also supported by the Research Funds of the University of Helsinki.
[^1]: E-mail address: [email protected]
[^2]: E-mail address: [email protected]
[^3]: E-mail address: [email protected]
[^4]: To whom correspondence should be addressed. E-mail: [email protected]
[^5]: Ref.
[^6]: Ref.
[^7]: Ref.
[^8]: Ref. [@MgC2-Karen]
[^9]: Ref. [@CaC2-Knapp]
[^10]: Ref. [@SrC2-Vohn]
[^11]: Ref. [@BaC2-Vohn]
[^12]: Ref. [@Bowmaker]
[^13]: Ref.
[^14]: Semimetallic
[^15]: Metallic
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We present optical and near-infrared photometry, as well as ground-based optical spectra and [*Hubble Space Telescope*]{} ultraviolet spectra, of the Type Ia supernova (SN) 2001ay. At maximum light the Si II and Mg II lines indicated expansion velocities of 14,000 , while Si III and S II showed velocities of 9,000 . There is also evidence for some unburned carbon at 12,000 . SN 2001ay exhibited a decline-rate parameter = 0.68 $\pm$ 0.05 mag; this and the $B$-band photometry at $t
\gtrsim +25$ d past maximum make it the most slowly declining Type Ia SN yet discovered. Three of four super-Chandrasekhar-mass candidates have decline rates almost as slow as this. After correction for Galactic and host-galaxy extinction, SN 2001ay had $M_B = -19.19$ and $M_V = -19.17$ mag at maximum light; thus, it was [*not*]{} overluminous in optical bands. In near-infrared bands it was overluminous only at the 2$\sigma$ level at most. For a rise time of 18 d (explosion to bolometric maximum) the implied $^{56}$Ni yield was (0.58 $\pm$ 0.15)/$\alpha$ M$_{\odot}$, with $\alpha = L_{\rm max}/E_{\rm Ni}$ probably in the range 1.0 to 1.2. The $^{56}$Ni yield is comparable to that of many Type Ia supernovae. The “normal” $^{56}$Ni yield and the typical peak optical brightness suggest that the very broad optical light curve is explained by the trapping of the $\gamma$ rays in the inner regions.
author:
- 'Kevin Krisciunas, Weidong Li, Thomas Matheson, D. Andrew Howell, Maximilian Stritzinger, Greg Aldering, Perry L. Berlind, M. Calkins, Peter Challis, Ryan Chornock, Alexander Conley, Alexei V. Filippenko, Mohan Ganeshalingam, Lisa Germany, Sergio González, Samuel D. Gooding, Eric Hsiao, Daniel Kasen, Robert P. Kirshner, G. H. “Howie” Marion, Cesar Muena, Peter E. Nugent, M. Phelps, Mark M. Phillips, Yulei Qiu, Robert Quimby, K. Rines, Jeffrey M. Silverman, Nicholas B. Suntzeff, Rollin C. Thomas, and Lifan Wang'
title: The Most Slowly Declining Type Ia Supernova 2001ay
---
Introduction
============
@Phi93 first established that Type Ia supernovae (SNe) are [*standardizable*]{} candles at optical wavelengths: there is a correlation between their absolute magnitudes at maximum light and the rate at which these objects fade. This fact allowed Type Ia SNe to be used to determine that the expansion of the Universe is currently accelerating [@Rie_etal98; @Per_etal99]. More recently, we have discovered that in near-infrared (IR) photometric bands Type Ia SNe are [*standard*]{} candles [@Mei00; @Kri_etal04a; @WV_etal08; @Fol_etal10; @Man_etal11]. Except for a fraction of the rapidly declining Type Ia SNe whose prototype is SN 1991bg [@Fil_etal92; @Lei_etal93], which peak in the near-IR after the time of $B$-band maximum, the near-IR absolute magnitudes at maximum light are at most only slightly dependent on the decline-rate parameter [@Kri_etal09; @Fol_etal10].
We do not understand well how to model a Type Ia SN. For much of the previous decade we thought that (most) Type Ia SNe were carbon-oxygen white dwarfs that approach the Chandrasekhar limit (1.4 M$_{\odot}$) owing to mass transfer from a nearby companion [for a review, see, e.g., @Liv00]. Some Type Ia SNe might be mergers of two white dwarfs [@Iben_Tut84]. Over the past few years the double-degenerate scenario has gained prominence among many researchers. Unfortunately, it is not yet possible to determine with a high degree of probability that a given Type Ia SN is a single or double-degenerate event, except perhaps for the few that produced more than a Chandrasekhar mass of fusion products; these would result from double-degenerate mergers. An excellent summary of evidence for different kinds of progenitors of Type Ia SNe is given by @How11.
The spectroscopic classification scheme stipulates that Type I SNe do not exhibit obvious hydrogen in their spectra, and that Type II SNe do show prominent hydrogen [@Min41]; see @Fil97 for a review. Type Ia SNe show Si II in absorption, blueshifted from its rest wavelength at 6355 Å. Two objects have been shown to have strong Si II in absorption and H$\alpha$ in emission: SN 2002ic [@Ham_etal03] and SN 2005gj [@Ald_etal06; @Pri_etal06a]. The interpretation is that these objects are Type Ia SNe interacting with circumstellar material (CSM) or (less likely) with the general interstellar medium (ISM).
SN 2006X was found to exhibit variable Na I D absorption [@Pat_etal07]. Two other examples showing variable Na I absorption are SN 1999cl [@Blo_etal09] and SN 2006dd [@Str_etal10]. Light from the SNe is ionizing the circumstellar medium. This leads to recombination and line variations. It should be pointed out that SNe 1999cl and 2006X have large reddening that may involve multiple scattering as well as normal dust extinction [@Wan05; @Goo08].
It is possible to relate the peak bolometric luminosity of a Type Ia SN and the implied amount of radioactive $^{56}$Ni produced in the explosion. Most events generate between 0.4 and 0.7 M$_{\odot}$ of $^{56}$Ni [@Str_etal06]. The remarkable SN 2003fg [also known as SNLS-03D3bb; @How_etal06] was sufficiently overluminous that it was regarded to have been caused by the explosion of more than 1.4 M$_{\odot}$ of carbon and oxygen. Three additional super-Chandrasekhar-mass candidates have recently been found: SN 2006gz [@Hic_etal07], SN 2007if [@Yua_etal10; @Sca_etal10], and SN 2009dc [@Yam_etal09; @Tan_etal10; @Sil_etal10; @Tau_etal11]. As summarized by @Sca_etal10 [Table 4], these objects can be up to a magnitude more luminous at optical wavelengths than typical Type Ia SNe. Furthermore, three of these four have extremely slowly declining light curves ( $\approx 0.7$ mag). In a double-degenerate system, the more massive white dwarf can tidally disrupt the less massive white dwarf, which has a larger radius. A disk or shell of material is created, retarding the expansion of the more massive white dwarf once it explodes. Another object with a very slow decline rate was SN 2005eq, a target of the Carnegie Supernova Project [@Fol_etal10; @Con_etal10].
In this paper we present optical and near-IR photometry, plus a plethora of spectra, of the equally remarkable SN 2001ay. Preliminary analysis of the light curve and spectra of SN 2001ay is given by @Phi_etal03. Two [*Hubble Space Telescope (HST)*]{} STIS spectra (from 2001 May 2 and May 9) have already been published by @Fol_etal08, in a study of possible luminosity indicators in the ultraviolet (UV) spectra of Type Ia SNe.
Optical and IR Photometry
=========================
SN 2001ay was discovered on 2001 April 18.4 (UT dates are used throughout this paper) by @Swi_Li01. Its position was $\alpha =
14^{\rm h}26^{\rm m}17.0^{\rm s}$, $\delta = +26^\circ 14' 55.8''$ (J2000), some $10.3''$ west and $9.3''$ north of the nucleus of the spiral galaxy IC 4423. Basic information on SN 2001ay is given in Table \[properties\]. Figure \[finder\] shows the host galaxy and SN, and identifies a number of field stars of interest.
A sizable fraction of our optical photometry of SN 2001ay was obtained with the 0.76-m Katzman Automatic Imaging Telescope (KAIT) at Lick Observatory [@Li_etal00; @Fil_etal01]. KAIT images are $6.7'
\times 6.7'$ in size, with a scale of $0.80''$ per pixel. For the first three nights of photometry with the Lick Observatory Nickel 1-m telescope, the CCD camera had a chip with $0.30''$ pixels in $2 \times
2$ readout mode and gave a field size of $5.1' \times 5.1'$. Thereafter, a new chip was installed, having $0.37''$ pixels in $2
\times 2$ readout mode and a field size of $6.3' \times 6.3'$. The two different CCD chips employed with the Nickel telescope had considerably different quantum efficiencies at blue wavelengths.
The $UBVRI$ photometry of some field stars near SN 2001ay is found in Table \[ubvri\_field\]. Calibration of optical photometry was accomplished from observations carried out on five photometric nights (two with the CTIO 0.9-m, two with the Nickel, and one with the CTIO 1.5-m telescopes) using the standard stars of @Lan92. A subset of our photometry (the KAIT data) has already been published by @Gan_etal10 [Table 6], but without K- and S-corrections. Our data presented here were reduced in the [iraf]{} environment. Transformation of the data to the system of @Lan92 was accomplished using equations equivalent to those of @Wan_etal09a. In their Table 1, the reader will find characteristic color terms used for the transformations. Our own direct determinations of these color terms are entirely consistent with their values.
Table \[jh\_field\] gives the $J_s$ and $H$-band photometry of some of the field stars. We give the averages obtained on seven photometric nights using the Las Campanas 1-m Swope telescope. The calibration was accomplished using a stand-alone version of [ daophot]{}, some [fortran]{} programs written by one of us (N.B.S.), and the IR standards of @Per_etal98. Thus, our near-IR photometry of SN 2001ay is on the @Per_etal98 photometric system.
We can compare the photometry of the principal IR secondary standard (“Star 1”) with data from the Two-Micron All Sky Survey [2MASS; @Skr_etal06]. Whereas we obtained $J_s$ = 13.718 $\pm$ 0.014 and $H$ = 13.251 $\pm$ 0.011 mag, 2MASS found $J$ = 13.718 $\pm$ 0.027 and $H$ = 13.214 $\pm$ 0.030 mag. These values are statistically in agreement. For “Star 6” we obtained $J_s$ = 15.153 $\pm$ 0.021 and $H$ = 14.881 $\pm$ 0.008 mag, whereas 2MASS found $J$ = 15.132 $\pm$ 0.045 and $H$ = 14.807 $\pm$ 0.060 mag.
Tables \[ubvri\] and \[jh\] give the optical and IR photometry of SN 2001ay, respectively. Table \[ubvri\] lists point-spread function (PSF) magnitudes from KAIT, the first three Nickel 1-m nights, and one night of the CTIO 1.5-m telescope; the subtracted template images were obtained with KAIT on 2002 July 9 and 10. For the single night of imaging with the CTIO 1.5-m telescope, the $U$-band magnitude was determined from aperture photometry; no $U$-band template was available. For the final night of Nickel 1-m photometry, PSF magnitudes were derived, but without template subtraction. Aperture photometry without template subtraction was conducted for all other optical images.
To compare the photometry of different SNe, one should transform the data to the rest-frame equivalent by subtracting the so-called K-corrections [@Oke_San68; @Ham_etal93; @Kim_etal96; @Hog_etal02; @Nug_etal02]. For each SN one computes the time since maximum brightness, and then obtains a rest-frame timescale by dividing the differential times by $1 + z$, where $z$ is the object’s heliocentric redshift. We have worked out the optical K-corrections using actual spectra of SN 2001ay, adopting the @Bes90 filter profiles as reference; these are given in Table \[opt\_kcorrs\] and shown in Figure \[kcorr\].
To account for differences in SN photometry that result from the use of different telescopes, CCD chips, and filters, we use the method of spectroscopic corrections (S-corrections), as outlined by @Str_etal02 and @Kri_etal03. For optical bands we adopt as reference the filter specifications of @Bes90. [*At minimum*]{} one constructs the effective filter profiles using the laboratory transmission curves of the filters multiplied by an appropriate atmospheric function that includes atmospheric absorption lines; then, one multiplies that result by the quantum efficiency of the CCD chip as a function of wavelength. Telescope and instrument optics also contribute to the effective filter profiles. For the KAIT and Nickel 1-m corrections we have used an atmospheric function appropriate to Lick Observatory (elevation 1290 m). For the CTIO 0.9-m data we have used a different atmospheric function appropriate to Cerro Tololo (elevation 2215 m). In practice, one arbitrarily shifts the effective filter profiles toward longer or shorter wavelengths so that synthetic photometry based on spectra of standard stars gives color terms that match those obtained from actual photometry of photometric standards. We used spectra of 50 stars from the sample of @Str_etal05 to calculate our synthetic magnitudes using an [iraf]{} script written by one of us (N.B.S.). This script was then used to calculate the S-corrections based on spectra of SN 2001ay itself.
In Figure \[scorr\_bv\] we show the $B$- and $V$-band S-corrections for SN 2001ay. One [*adds*]{} these corrections to the photometry. As one can see in Figure \[scorr\_bv\], if an object like SN 2001ay were observed with the CTIO 0.9-m and Nickel 1-m telescopes, there could be photometric discrepancies as large as $\sim 0.06$ mag in the $B$ band, depending on the epoch. In the $V$ band the corresponding differences are much smaller.
Even after the application of the S-corrections, we find that the Nickel 1-m $B$-band data based on aperture magnitudes are systematically 0.09 mag fainter than the KAIT measurements. For the first three nights of Nickel imaging we were able to eliminate these systematic differences by means of template subtraction. The images subsequently obtained with the Nickel 1-m telescope and its newer chip did not lend themselves to image subtraction using the KAIT templates. (Depending on the number of field stars, the angular size of a nearby galaxy in the field, and the seeing, the image rescaling and remapping algorithm can fail.) As a result, we have devised a third set of corrections to the photometry based on images with the new chip in the Nickel camera. Using [addstar]{} within the [daophot]{} package of [iraf]{}, we were able to add an artificial star to the KAIT image-subtraction templates at the pixel location of the SN. This artificial star can be scaled to give a standard magnitude of a desired value such as $B$ = 17.97 and $V$ = 17.14 mag, just about the brightness of the SN on 2001 May 14 (JD 2,452,043.8), in the middle of the run of Nickel 1-m aperture photometry. Artificial stars having the identical PSF magnitude as the fake star at the location of the SN are placed at blank places in the KAIT templates using [addstar]{}. Since the Nickel 1-m aperture photometry was typically derived with a software aperture of radius 10 pixels (px) and a sky annulus ranging from 12 to 20 px, we can then compare the aperture magnitudes of the fake SN with aperture photometry of the fake stars in the blank places of the image. We use a software aperture radius and sky annulus of identical size in [*arc seconds*]{} to that used for the Nickel 1-m photometry. In this way, we can obtain an estimate of the systematic errors of the SN aperture photometry.
We determined that the $B$-band magnitudes from Nickel 1-m aperture photometry of the SN were 0.08 mag fainter than what we would have measured without the presence of the host galaxy. Similarly, in the $V$ band we found that the Nickel 1-m aperture magnitudes were fainter by 0.05 mag. For the two nights of CTIO 0.9-m aperture photometry, when the SN was more than 1 mag brighter, the SN was too faint by 0.03 mag in $B$ and $V$ without this correction. Thus, by applying the S-corrections and these additional corrections, we can reconcile almost all systematic differences in the $B$- and $V$-band photometry obtained with different telescopes, using different cameras, and using different data-reduction methods (i.e., PSF magnitudes with image-subtraction templates vs. aperture magnitudes without subtractions).
Further justification of this method comes from a consideration of the optical photometry from the first Nickel 1-m chip and the photometry from one night with the CTIO 1.5-m telescope. Data from these four nights [*can*]{} be derived using the KAIT host-galaxy templates. In the $B$ band, photometry of the SN using PSF magnitudes and image subtraction was 0.07 to 0.16 mag brighter than aperture photometry using an annulus for the subtraction of the sky level. In the $V$ band, the photometric values of the SN on the first three Nickel 1-m nights were 0.06 mag brighter than values derived from aperture photometry. This result is contrary to typical experience; normally, aperture photometry of a SN is brighter than expected (not fainter) if the SN is located on top of some part of the host galaxy, because the light in the aperture is not entirely due to the SN. That this is not the case here must be due to the relative distributions of host-galaxy light at the SN position and in the sky annulus.
Differences between aperture photometry and PSF photometry are [*usually*]{} larger in redder bands. In fact, similar experiments with adding artificial stars to the KAIT templates of SN 2001ay show that the aperture magnitudes with the Nickel 1-m telescope are too bright by 0.01 mag in $R$ and $I$ for the size of the aperture and annulus used. We never obtained near-IR subtraction templates with the camera used on the Las Campanas Observatory 1-m telescope, and that camera has since been decommissioned. So, we must adopt the available near-IR aperture magnitudes.
The $I$-band photometry is particularly problematic from 15 to 30 d after the time of $B$-band maximum. Consider the effective filter profiles shown in Figure \[i\_filter\]. While the KAIT $I$-band filter very well approximates the Bessell $I$ filter, the filter used on the Nickel 1-m telescope does not. As the SN achieved its reddest optical colors, a significant excess amount of light is let through by the Nickel 1-m $I$-band filter. This leads to positive S-corrections; to correct such photometry to Bessell filter photometry requires making the Nickel 1-m photometry fainter. Our attempts to reconcile up to 0.35 mag differences between the KAIT and Nickel 1-m photometry were not successful. Apparently, the effective filter profile of the Nickel 1-m camera is even more nonstandard than our profile based on laboratory traces of the filter, knowledge of the quantum efficiency as a function of wavelength, and a generic atmospheric function applicable to Lick Observatory. While we list all of our optical photometry in Table \[ubvri\], the Nickel 1-m $I$-band photometry is not included in our plots or used in the analysis.
The $R$- and $I$-band S-corrections are shown in Figure \[scorr\_ri\]. Note that the photometry in Table \[ubvri\] includes the K- and S-corrections, plus the corrections mentioned above for CTIO 0.9-m and Nickel 1-m photometry reduced using aperture magnitudes. Since the K-corrected photometry from one night with the YALO 1-m telescope and four nights with the LCO 2.5-m telescope are in good agreement with photometry obtained with other telescopes, and we have no effective transmission curves for the LCO 2.5-m filters used, we have derived no further corrections for this small fraction of our photometry. The interpolated $BVRI$ K-corrections and S-corrections are listed in Table \[corrections\].
In spite of the rationale outlined above to reconcile as much of the optical photometry as we can, for a derivation of the maximum brightness and decline rate of SN 2001ay we restrict ourselves to the photometry based on PSF magnitudes with image subtraction. After subtracting the derived K-corrections and adding the S-corrections, we scale the time since maximum light to the rest frame by dividing by $1
+ z$. We derive a time of $B$-band maximum of JD 2,452,022.49 (= 2001 April 23.0), with an uncertainty of perhaps $\pm$ 0.8 d. The decline-rate parameter found from our best $B$-band data is = 0.68 $\pm$ 0.05 mag. For comparison, SNe 2005eq had = 0.72 $\pm$ 0.02 mag [@Fol_etal10], though @Con_etal10 give = 0.78 $\pm$ 0.01 mag. SN 2009dc had = 0.72 $\pm$ 0.03 mag [@Sil_etal10]. SNe 2001ay, 2005eq, 2006gz, 2007if, and 2009dc had extremely slow decline rates. Since it is well established that the energy budget of a Type Ia SN is related to the decline rate, we naturally wonder if SNe 2001ay and 2005eq are also super-Chandrasekhar-mass candidates.
The $B$- and $V$-band light curves of SNe 2001ay, 2005eq, and 2009dc are illustrated in Figure \[bv\_comp\]. For the first three weeks after maximum light SNe 2001ay and 2009dc were remarkably similar, but at $t \approx 25$–40 d the light curves diverge. As the peak-to-tail ratio in the $B$- and $V$-band light curves sheds light on the central density and progenitor mass of single-degenerate explosions [@Hoe_etal10], a comparison of the observational features and modeling of very slow decliners may yield similar insights.
In Figure \[01ay\_bv\] we show the unreddened $B-V$ colors of SN 2001ay, the unreddened locus of @Lir95 and @Phi_etal99, and the = 0.83 mag locus of @Pri_etal06b. On the basis of some of the spectra published here, @Bra_etal06 classified SN 2001ay as a “broad-line” Type Ia SN. This is the subtype that @Wan_etal09b suggest is intrinsically redder than normal Type Ia SNe, or occurs in dustier environments, with $R_V \approx 1.7$. However, Figure \[01ay\_bv\] shows that, if anything, SN 2001ay is bluer than other normal Type Ia SNe.
In Figure \[ri\_comp\] we show the corresponding $R$- and $I$-band light curves of SN 2001ay, along with loci derived from the data of SNe 2005eq and 2009dc. The behavior of SN 2001ay in these photometric bands is clearly more like that of SN 2009dc than of SN 2005eq. Note that the $I$-band secondary maximum of SN 2001ay is essentially as bright as the first maximum.
Figure \[jh\_comp\] shows the near-IR photometry of SN 2001ay and the other slow decliners, SNe 2005eq [@Fol_etal10; @Con_etal10] and 2009dc [@Str_etal11]. SN 1999aw [@Stro_etal02] and SN 2001ay were the first objects known to exhibit such flat $H$-band light curves. (Both $H$-band light curves are admittedly somewhat ragged, however.) SN 2005eq was the first to show such early near-IR maxima ($t \leq -7$ d). Interestingly, the $H$-band brightness of SN 2009dc increased steadily over the time frame that the brightness of SN 2001ay was constant.
Spectroscopy
============
We obtained spectra of SN 2001ay on 20 dates using 9 different telescopes; 7 of those nights had spectra with more than one telescope. Spectra were obtained with the 1.5-m telescope of the Fred L. Whipple Observatory (FLWO) on 11 nights. Eight spectra were taken on four nights with the MMT 6.5-m telescope; four with the 2.16-m telescope at the Xinglong Station of the Beijing Astronomical Observatory (BAO); three with the Las Campanas Observatory 2.5-m du Pont telescope; two each with the Lick 3-m Shane reflector (Kast spectrograph) and [*HST*]{}; and one each with the Kitt Peak 2.1-m, Kitt Peak 4-m, and Keck II telescopes. The spectroscopic observations with the FLWO 1.5-m and the MMT are summarized in Table \[sp1\_table\]; we may refer to this as the “CfA set.” A log of other spectroscopic observations is given in Table \[sp2\_table\]. Some, but not all, of the data were obtained at (or close to) the parallactic angle [@Fil82] in order to minimize the effects of atmospheric dispersion.
In Figure \[spectra\] we show a temporal sequence of spectra. Here we have combined the May 2 spectrum from [*HST*]{} with the April 29 spectrum from the KPNO 2.1-m telescope. The rather noisy BAO spectra from May 20, 25, and 30 are not illustrated, but the May 10 spectrum from BAO fills a gap in the temporal coverage. We show additional spectra in Figure \[additional\_spectra\].
Figure \[keck\] includes a portion of our spectrum of SN 2001ay obtained with the Keck Echellette Spectrograph and Imager [ESI; @She_etal02] from 2001 April 22 ($t = -1$ d). We find that this maximum-light spectrum exhibits S II and Si III at $-$9000 km s$^{-1}$. The Si II line (rest wavelength $\lambda_0$ = 6355 Å) and the Mg II line ($\lambda_0$ = 4481 Å) are blueshifted by 14,000 km s$^{-1}$ at this epoch. There is some gas of these species blueshifted as much as 20,000 km s$^{-1}$. As shown in Figure \[spectra\], by $t$ = +17 d the Si II absorption is only blueshifted about 10,000 km s$^{-1}$.
In our Keck spectrum there is a small absorption dip at an observed wavelength of $\lambda \approx 6520$ Å. If this is due to gas in or near the SN or gas in the ISM of the host galaxy, the wavelength of this line in the frame of the host galaxy is 6315 Å. This may be due to C II ($\lambda_0$ = 6580 Å) blueshifted by 12,000 km s$^{-1}$. However, other explanations are possible. An absorption feature at 6520 Å in our Galaxy or in the Earth’s atmosphere could lead to a misidentification, given that telluric lines were not removed from the Keck spectrum. One possible such feature is telluric absorption at 6515 Å due to atmospheric water vapor [@Mat_etal00 Appendix]. We have better evidence of C II absorption in SN 2001ay from the $\lambda_0$ = 4745 Å line, which is seen in the galaxy’s rest frame at 4555 Å, corresponding to an identical blueshift of 12,000 . We see no evidence of the corresponding C II line with $\lambda_0$ = 7234 Å.
One might expect the C II to be concentrated in the outer layers, and therefore at a higher velocity of approach than sulfur or silicon [@Kho_etal93; @Gam_etal03]. But if there is mixing or clumpiness, most species could be observed over a broad range of velocities.
In Figure \[sodium\] we see the two components of Na I D from gas in the Milky Way and gas in the host galaxy of SN 2001ay. The presence of these lines implies some nonzero amount of reddening and extinction. We find equivalent widths of 0.090 $\pm$ 0.018 Å for the Galactic sodium lines and 0.210 $\pm$ 0.018 Å for the host-galaxy sodium lines. Using the @Mun_Zwi97 calibration of the equivalent width of the sodium lines with $B-V$ color excess, we find $E(B-V)_{\rm Gal} = 0.026 \pm 0.006$ mag and $E(B-V)_{\rm host} =
0.072 \pm 0.008$ mag. The @Mun_Zwi97 calibration shows a scatter of $\pm 0.05$ mag for $E(B-V)$, which is a more realistic estimate of the uncertainty of the reddening toward SN 2001ay. The Galactic component may be compared to $E(B-V) = 0.019$ mag obtained by @Sch_etal98.
In Figure \[silicon\] we see the (blueshifted) velocities of Si II $\lambda$4130 and $\lambda$6355, with respective error bars of $\pm 420$ and $\pm 330$ . The $\lambda$6355 line exhibits the larger velocities because its opacity is greater; we are measuring material farther out in the expanding fireball. The high velocities suggest that the outflow of SN 2001ay is essentially unimpeded, which is consistent with it having been a single-degenerate explosion; however, @Mae_etal10 emphasize the importance of the viewing angle on the observed properties of Type Ia SNe. We cannot state with certainty whether SN 2001ay was a single or double-degenerate explosion.
The $\lambda$4130 line gives a velocity gradient of $\dot{v}$ = 226 $\pm$ 32 d$^{-1}$, while the $\lambda$6355 line gives $\dot{v}$ = 171 $\pm$ 35 d$^{-1}$. The Si II velocity gradient is considerably higher than 70 d$^{-1}$, the criterion of @Ben_etal05 to include SN 2001ay with other “high velocity gradient” Type Ia SNe.
Discussion
==========
Reddening
---------
As mentioned above, the Galactic reddening along the line of sight to SN 2001ay is $E(B-V) = 0.026$ mag, and the host-galaxy component is $E(B-V) = 0.072$ mag. Thus, for SN 2001ay, $E(B-V)_{\rm total} \approx
0.098$ mag. Normal Galactic dust is characterized by an average value of $R_V = A_V/E(B-V) = 3.1$ [@CCM89], but dust associated with Type Ia SNe is often characterized by a lower value; see, for example, @Kri_etal07. If we adopt $R_V = 3.1$ for the Galactic component of reddening for SN 2001ay and $R_V = 2.4 \pm 0.2$ for the host-galaxy reddening, it follows that $A_B = 0.35 \pm 0.08$, $A_V =
0.25 \pm 0.06$, $A_R = 0.21 \pm 0.05$, $A_I = 0.15 \pm 0.04$, $A_J =
0.08 \pm 0.02$, and $A_H = 0.04 \pm 0.01$ mag [@Kri_etal06 Table 8].
Absolute Magnitudes at Maximum Light
------------------------------------
@Gar_etal04 give $BVI$ decline-rate relations for what was then the known range of for Type Ia SNe. The slowest decliner used by @Pri_etal06b for their light-curve fitting template algorithm was SN 1999aa, with = 0.83 mag. For such a decline-rate parameter, the implied absolute magnitudes at maximum are $M_B = -19.42$, $M_V = -19.39$, and $M_I = -18.85$ on an $H_0$ = 72 km s$^{-1}$ Mpc$^{-1}$ distance scale. Three of the four super-Chandrasekhar-mass Type Ia SN candidates discussed by @Sca_etal10 are significantly more luminous than this.
In Figure \[slow\_decliners\] we show the absolute $V$-band magnitudes at maximum light of SNe 2001ay, 2003fg, 2005eq, 2006gz, 2007if, and 2009dc, along with the $V$-band decline-rate relation of @Gar_etal04. Figure \[absmag\_ir\] gives the near-IR absolute magnitudes of Type Ia SNe at their respective maxima, including SNe 2001ay and 2005eq. Three of the four super-Chandra candidates are overluminous, while all other very slow decliners have “normal” maximum brightness.
Based on our best optical photometry (using PSF magnitudes and image-subtraction templates), we find apparent magnitudes for SN 2001ay of $B_{\rm max}$ = 16.71, $V_{\rm max}$ = 16.64, $R_{\rm
max}$ = 16.65, and $I_{\rm max}$ = 16.79. These values include the K-corrections and S-corrections. Given the extinction values listed above and the distance modulus of the host galaxy given in Table \[properties\], at maximum light we find $M_B = -19.19 \pm 0.12$, $M_V = -19.17$, $M_R = -19.10$, and $M_I = -18.90$ mag (with uncertainties of $\pm 0.10$ mag); thus, SN 2001ay was [*not*]{} overluminous at optical wavelengths.
Typically, Type Ia SNe peak in the near-IR $\sim 3$ d prior to the time of $B$-band maximum [@Kri_etal04b]. SN 2005eq peaked at least 7 d prior to $B$ maximum. We surmise that SN 2001ay may have been $\sim 0.12$ mag brighter at maximum light than our first $J_s$ measurement, or $J_s$(max) $\approx$ 16.85 mag. In the $H$ band SN 2001ay exhibited a very flat light curve; we adopt $H_{\rm max} = 16.97 \pm 0.08$ mag. The corresponding extinction-corrected absolute magnitudes at IR maximum are $M_{J_s} =
-18.77 \pm 0.10$ and $M_H = -18.62 \pm 0.10$. These values may be compared to the mean values of all but the late-peaking fast decliners from @Kri_etal09, namely $\langle M_J \rangle = -18.61$ and $\langle M_H \rangle = -18.31$ mag. The standard deviations of the distributions of the near-IR absolute magnitudes are about $\pm 0.15$ mag. In the $J_s$ and $H$ bands, SN 2001ay was overluminous by $\sim
1\sigma$ and $2\sigma$, respectively; hence, SN 2001ay was not statistically significantly brighter in the near-IR than other Type Ia SNe.
Bolometric Light Curve and Mass Budget
--------------------------------------
In Figure \[bolom\_lc\] we show the bolometric light curve of SN 2001ay, based on our broad-band photometry. Though Type Ia SNe near maximum light have spectral energy distributions that peak at optical wavelengths, they also emit UV and IR light. One might scale the integrated flux by a factor of $\sim 1.1$ to account for UV and IR light not included in the optical bandpasses [@Str_etal06], but according to @Bli_etal06 this still underestimates the bolometric flux. Adopting distance modulus 35.55 mag ($D \approx$ 129 Mpc; see Table \[properties\]) and a scale factor of 1.15, we obtain a peak derived bolometric luminosity ($4\pi D^2$ times the bolometric flux) of $1.20 \times 10^{43}$ erg s$^{-1}$. Figure \[bolom\_lc\] also shows the bolometric light curves of SNe 2007if and 2009dc.
At maximum light the luminosity produced by radioactive $^{56}$Ni is given by
$$L_{\rm max} \; = \; \alpha \; E_{\rm Ni}~,$$
where $E_{\rm Ni}$ is the energy input from the decay of $^{56}$Ni, evaluated at the time of bolometric maximum. Arnett’s Rule implies that $\alpha$ = 1 [@Arn82], meaning that the gamma-ray deposition matches the bolometric flux at maximum light. However, the value of $\alpha$ can actually range from 0.8 to 1.3, depending on the explosion model. For a delayed detonation model $\alpha$ = 1.2 is appropriate [@Kho_etal93 Fig. 36]; see also @Hoe_Kho96. @How_etal06 adopt $\alpha$ = 1.2.
In Figure \[edep\] we show the radioactive decay energy deposition function fit to the last few points of the bolometric light curve. In this figure we also show the cases of complete trapping of the $\gamma$ rays and complete $\gamma$-ray escape. We adopt a rise time of 18 d from explosion to bolometric maximum, comparable to observational results of @Gar_etal07 for other Type Ia SNe. @Hay_etal10 find an average rise time in the $B$ band of 17.38 d, with a range of 13 to 23 d, while the $B$-band rise time determined by @Gan_etal11 is about 18 d. Figure \[edep\] uses $\alpha$ = 1.0.
The energy deposited by 1 M$_\odot$ of $^{56}$Ni into the explosion of a Type Ia SN is given by @Str_Lei05 as
$$E_{\rm Ni}(1~{\rm M}_{\odot}) = (6.45 \times 10^{43}) e^{-t_R/8.8} +
(1.45 \times 10^{43}) e^{-t_R/111.3},$$
where $t_R$ is the rise time in days from the moment of explosion to the time of bolometric maximum. The $e$-folding times of $^{56}$Ni and $^{56}$Co are 8.8 and 111.3 d, respectively. For $t_R$ = 18 d and $\alpha$ = 1.0, $L_{\rm max}$ = 2.07 $\times
10^{43}$ erg s$^{-1}$ M$_{\odot}^{-1}$; for $\alpha$ = 1.2, $L_{\rm
max}$ = 2.48 $\times 10^{43}$ erg s$^{-1}$ M$_{\odot}^{-1}$.
There was a nondetection of SN 2001ay on April 5.4, which was 17.1 rest-frame days before the time of $B$-band maximum. The upper limit of the brightness on that date was $R \approx 19.5$ mag, or $\sim 3$ mag fainter than the observed $R$-band maximum. Although we have an image of the host galaxy at $t = -17$ d, we would need to reach a much fainter magnitude limit to place a useful constraint on the time of the explosion. The lack of premaximum photometry does not allow us to determine the rise time of SN 2001ay.
To obtain the $^{56}$Ni yield in solar masses, we simply divide the peak luminosity (1.20 $\times$ 10$^{43}$ erg s$^{-1}$) by the coefficient given above for the adopted rise time (e.g., $2.07 \times 10^{43} \times \alpha$ for $t_R$ = 18 d); the result is 0.58/$\alpha$ M$_{\odot}$ of $^{56}$Ni. SN 2001ay certainly did not produce more than a Chandrasekhar mass of $^{56}$Ni.
A 5% uncertainty in the value of $H_0$ leads to a 10% uncertainty in the luminosity calculated from the optical photometry. Given the uncertainties in the parameter $\alpha$ in Equation 1 and uncertainty in the adopted bolometric rise time, the minimum uncertainty in the $^{56}$Ni yield is about $\pm 0.15$ M$_{\odot}$ divided by $\alpha$.
We would like to estimate the mass in the ejecta of SN 2001ay. Following @Jef99 and @Str_etal06, the “fiducial time” ($t_0$) is the time scale for the ejecta to become optically thin:
$$t_0 \; = \; \left( \frac{M_{\rm ej} \kappa q}{8 \pi} \right) ^{\onehalf} \frac{1}{v_e} \; ,$$
= 0 mm
where $M_{\rm ej}$ is the ejecta mass, $\kappa$ is the $\gamma$-ray mean opacity, $q$ is a dimensionless scale factor, and $v_e$ is the $e$-folding velocity of an exponential model’s density profile. @Str_etal06 adopt $\kappa$ = 0.025 cm$^2$ g$^{-1}$ and $v_e$ = 3000 km s$^{-1}$. They also choose $q = 1/3$, meaning that $^{56}$Ni was distributed throughout the ejecta; $q = 1$ for high concentrations of $^{56}$Ni at the center of the ejecta. Using the last three points in our bolometric light curve (from $t$ = 49 to 111 d), adopting $\alpha$ = 1.2, and using a $^{56}$Ni yield of 0.48 M$_{\odot}$ we derive a “fiducial time” of 66.2 $\pm$ 3.0 d. Taking $\alpha$ = 1.0 and a $^{56}$Ni yield of 0.58 M$_{\odot}$ gives $t_0$ = 57.2 $\pm$ 2.4 d. Both of these values are considerably longer than the 22 d to 35 d in Table 1 of @Str_etal06, but not much greater than the 51 d for SN 2007if given in Figure 9 of @Sca_etal10. The SN 2001ay fiducial time depends critically on the final point in the bolometric light curve, a point derived from PSF photometry without template subtraction (hence possibly erroneous). Using $t_0$ = 66.2 d and the values of $q$, $v_e$, and $\kappa$ adopted by @Str_etal06, the implied ejecta mass is 4.4 M$_{\odot}$, which is clearly wrong. For $t_0$ = 57.2 d and the higher $^{56}$Ni yield, the implied ejecta mass is 3.3 M$_{\odot}$. Adopting $q = 1$ instead could cut these down by a factor of three. Choosing $q = 1/3$ but $v_e$ = 1500 km s$^{-1}$ would give 0.8 to 1.1 M$_{\odot}$ for the ejecta mass. The lower $e$-folding velocity is only slightly less than what one derives from the three-dimensional models of @Rop_Hil04. Given the uncertainties of the parameters necessary for Eq. 3 and the lack of high-quality photometry from $\sim 50$ to 100 d after maximum light, we feel that a robust calculation of the ejecta mass of SN 2001ay is beyond the scope of this paper.
= 9 mm
We note that @Tau_etal11 derived an ejecta mass for SN 2009dc of 2.8 M$_{\odot}$, based on the expansion velocity and the timescale around maximum brightness, which depends on the optical opacity. They describe their result as “an utmost challenge for all scenarios that invoke thermonuclear explosions of white dwarfs.”
Spectroscopic Comparison and Modeling
-------------------------------------
@Wag_etal10 have used wavelet analysis to search for spectroscopic correlations of Type Ia SNe. Their results are based on the analysis of a few dozen relatively nearby objects ($z \lesssim 0.04$). SN 2001ay is anomalous in a number of their plots, particularly their Figure 12, showing the correlation of the strength of the 4570 Åemission feature vs. .
In Figure \[3spectra\] we illustrate a comparison of the spectra of SNe 2001ay, 2005eq, and 2009dc. The most obvious difference between SN 2001ay and these other two objects is the much larger blueshift of Si II seen in SN 2001ay prior to $B$ maximum. Also, the C II absorption in SN 2009dc is much stronger. SN 2005eq was apparently the hottest of the three, given the stronger presence of doubly ionized iron and the weakness of singly ionized species. This is more like the classical slow decliner SN 1991T [@Fil97], which is, in fact, how @Con_etal10 [Table 1] classify SN 2005eq.
In Table \[three\] we give a summary of some observational characteristics of SNe 2001ay, 2005eq, and 2009dc. Given the divergence of the light curves of SN 2005eq seen in Figure \[bv\_comp\] compared to the other two objects, the larger decline-rate parameter of SN 2005eq ( = 0.78 mag) from @Con_etal10 is more likely correct than the value of = 0.72 mag from @Fol_etal10, though we have used the latter to plot SN 2005eq in Figure \[slow\_decliners\]. SNe 2001ay and 2009dc have almost the same optical decline rate. SN 2001ay had normal optical peak brightness and high Si II velocity. SN 2009dc was overluminous at optical maximum and had much lower Si II velocity. Their $H$-band light curves are unlike those of any other Type Ia SNe observed thus far, except for SN 1999aw.
To investigate the details of our SN spectra, we use the SN spectrum-synthesis code SYNOW [@Fis_etal97]. Although SYNOW has a simple, parametric approach to creating synthetic spectra, it is a powerful tool to aid line identifications which in turn provide insights into the spectral formation of the objects. To generate a synthetic spectrum one inputs a blackbody temperature ($T_{\rm BB}$), a photospheric velocity ($v_{\rm ph}$), and for each involved ion, an optical depth at a reference line, an excitation temperature ($T_{\rm exc}$), and the maximum velocity of the opacity distribution ($v_{\rm max}$). Moreover, it assumes that the optical depth declines exponentially for velocities above $v_{\rm ph}$ with an $e$-folding scale of $v_{e}$. The strengths of the lines for each ion are determined by oscillator strengths, and the approximation of a Boltzmann distribution of the lower level populations is set by the temperature $T_{\rm exc}$.
In Figure \[synow1\], we present our $-1$ d spectrum of SN 2001ay with a synthetic spectrum generated from SYNOW. This fit has $T_{\rm
BB}$ = 20,500 K and $v_{\rm ph} = 12,000$ km s$^{-1}$. The majority of the observed features are well fit by the synthetic spectrum. The ions used in the fit, as labeled in the figure, are commonly observed in the near-maximum spectra of Type Ia SNe [@Bra_etal05] with the exception of C II. Although the inclusion of C II in the fit produces an absorption feature at $\sim 6300$ Å (due to C II $\lambda$6580) that is too strong compared to the observed spectrum, it nonetheless yields a better fit to the absorption feature at $\sim 4500$ Å (due to C II $\lambda$4745), so we conclude that the inclusion of C II marginally improves the SYNOW fit.
We attempt SYNOW fits to the SN 2001ay spectra at $t \approx +6$ d and 23 d. Figure \[synow2\] shows the result for the $t \approx +6$ d spectrum. The model spectrum has $T_{\rm BB}$ = 14,000 K and $v_{\rm
ph}$ = 11,000 km s$^{-1}$, and has all the regular ions observed in a Type Ia SN. C II is no longer needed, but a relatively strong Na I line is now observed at $\sim 5600$ Å. A SYNOW fit for the $t
\approx 23$ d spectrum (not shown) requires $T_{\rm BB} = 8500$ K and $v_{\rm ph} = 9000$ km s$^{-1}$.
Finally, in Figure \[synapps\] we show a model fit to our Keck spectrum using the SYNAPPS code of @Tho_etal11. The physical assumptions SYNAPPS uses match those of SYNOW [@Fis_etal97], so findings are restricted to identification of features and not quantitative abundances. But where SYNOW is completely interactive, SYNAPPS is automated. This relieves the user from iterative adjustment of a large number of parameters (over 50 variables) to gain fit agreement and assures more exhaustive searching of the parameter space. SYNAPPS can be thought of as the hybridization of a SYNOW-like calculation with a parallel optimization framework, where spectral fit quality serves as the objective function to optimize. A good fit constrains explosion models through interpretive spectral feature identification, with the main result being the detection or exclusion of specific chemical elements. The velocity distribution of detected species within the ejecta can also be constrained.
SYNAPPS indicates that SN 2001ay had a photospheric velocity of 10,800 at $t = -1$ d, which should be robust, as it is derived from 11 ions, namely C II, O I, Mg II, Si II, Si III, S II, Ca II, Fe II, Fe III, Co II, and Ni II. We are confident of the presence of C II.
Conclusions
===========
We have presented the available spectra, as well as optical and near-IR photometry, of SN 2001ay.
While SN 2001ay is the most slowly declining Type Ia SN ever discovered, it was not overluminous in optical bandpasses. In near-IR bands it was overluminous only at the 2$\sigma$ level at most. Unlike other very slow decliners such as SNe 2003fg, 2007if, and 2009dc, which were significantly overluminous, we do not have any evidence that SN 2001ay was a super-Chandrasekhar-mass explosion.
SN 2001ay showed evidence for C II, but it was much weaker than in SN 2009dc. At early times Mg II and Si II were observed in SN 2001ay at high velocity (14,000 and higher). By contrast, spectra of SN 2009dc did not show large velocities for Mg II and Si II. On the basis of a small number of super-Chandrasekhar-mass candidates, it seems that these objects exhibit rather low velocities, possibly the result of retardation due to a shell of material arising from the disruption of the less massive white dwarf in a double-degenerate system.
SN 2001ay produced (0.58 $\pm$ 0.15)/$\alpha$ M$_{\odot}$ of $^{56}$Ni, considerably less than a Chandrasekhar mass. The value of $\alpha$ probably lies in the range 1.0 to 1.2. Naively, one might conclude that SN 2001ay was a single-degenerate explosion, but this is hardly a firm conclusion. The very broad light curve might be explained by the trapping of the $\gamma$ rays in the explosion. An explanation of the extremely slow decline will be discussed in a subsequent paper [@Bar_etal11].
The work presented here is based in part on observations made with the NASA/ESA [*Hubble Space Telescope*]{}, obtained at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555; the Cerro Tololo Inter-American Observatory and the Kitt Peak National Observatory of the National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc. (AURA) under cooperative agreement with the NSF; the MMT Observatory, a joint facility of the Smithsonian Institution and the University of Arizona; the Fred L. Whipple Observatory; the Lick Observatory of the University of California; the Las Campanas Observatory; the Beijing Astronomical Observatory; and the W. M. Keck Observatory, which was generously funded by the W. M. Keck Foundation and is operated as a scientific partnership among the California Institute of Technology, the University of California, and NASA. We thank the staffs at these observatories for their efficient assistance, Don Groom for taking some of the Nickel 1-m images, and Rachel Gibbons, Maryam Modjaz, Isobel Hook, and Saul Perlmutter for other observational assistance. We are grateful to Peter Höflich, Alexei Khokhlov, and Eddie Baron for comments on §4.3.
The supernova research of A.V.F.’s group at U. C. Berkeley is supported by NSF grant AST-0908886 and by the TABASGO Foundation, as well as by NASA through grants AR-11248 and AR-12126 from the Space Telescope Science Institute, which is operated by Associated Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. KAIT and its ongoing operation were made possible by donations from Sun Microsystems, Inc., the Hewlett-Packard Company, AutoScope Corporation, Lick Observatory, the NSF, the University of California, the Sylvia & Jim Katzman Foundation, and the TABASGO Foundation. J.M.S. is grateful to Marc J. Staley for a Graduate Fellowship. K. K., L. W., and N. B. S. are supported in part by NSF grant AST-0709181. Supernova research at Harvard is supported by NSF grant AST-0907903. This work was also supported by the Director, Office of Science, Office of High Energy Physics, of the U.S. Department of Energy under Contract No. DE-AC02-05CH11231.
Aldering, G., et al. 2006, , 650, 510
Arnett, W. D. 1982, , 253, 785
Baron, E., et al. 2011, in preparation
Beers, T. C., Kriessler, J. R., Bird, C. M., & Huchra, J. P. 1995, , 109, 874
Benetti, S., et al. 2005, , 623, 1011
Bessell, M. S. 1990, , 102, 1181
Blinnikov, S. I., Röpke, F. K., Sorokina, E. I., Gieseler, M., Reinecke, M., Travaglio, C., Hillebrandt, W., & Stritzinger, M. 2006, , 453, 229
Blondin, S., Prieto, J. L., Patat, F., Challis, P., Hicken, M., Kirshner, R. P., Matheson, T., & Modjaz, M. 2009, , 693, 207
Branch, D., Baron, E., Hall, N., Melakayil, M., & Parrent, J. 2005, , 117, 545
Branch, D., et al. 2006, , 118, 560
Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, , 345, 245
Contreras, C., et al. 2010, , 139, 519
Filippenko, A. V. 1982, , 94, 715
Filippenko, A. V. 1997, , 35, 309
Filippenko A. V., Li W., Treffers R. R., & Modjaz M. 2001, in Small-Telescope Astronomy on Global Scales, ed. W. P. Chen, C. Lemme, & B. Paczyński (San Francisco: ASP), 121
Filippenko, A. V., et al. 1992, , 104, 1543
Fisher, A., Branch, D., Nugent, P., & Baron, E. 1997, , 481, L89
Folatelli, G., et al. 2010, , 139, 120
Foley, R., Filippenko, A. V., & Jha, S. W. 2008, , 686, 117
Freedman, W., et al. 2001, , 553, 47
Gamezo, V. N., Khokhlov, A. M., Oran, E. S., Chtchelkanova, A. Y., & Rosenberg, R. O. 2003, Science, 299, 77
Ganeshalingam, M., et al. 2010, , 190, 418
Ganeshalingam, M., Li, W., & Filippenko, A. V. 2011, MNRAS, in press
Garg, A., et al. 2007, AJ, 133, 403
Garnavich, P., et al. 2004, , 613, 1120
Goobar, A. 2008, , 686, 103
Hamuy, M., Phillips, M. M., Wells, L. A., & Maza, J. 1993, , 105, 787
Hamuy, M., et al. 2003, Nature, 424, 651
Hayden, B. T., et al. 2010, , 712, 350
Hicken, M., et al. 2007, , 669, L17
Höflich, P., & Khokhlov, A. 1996, , 457, 500
Höflich, P., et al. 2010, , 710, 444
Hogg, D. W., Baldry, I. K., Blanton, M. R., & Eisenstein, D. J. 2002 (arXiv:astro-ph/0210394)
Howell, D. A., et al. 2006, Nature, 443, 308
Howell, D. A. 2011, Nature Communics., in press (arXiv:1011.0441)
Iben, I., Jr., & Tutukov, A. V. 1984, , 54, 335
Jeffrey, D. J. 1999 (arXiv:astro-ph/9907015)
Khokhlov, A., Müller, E., & Höflich, P. 1993, , 270, 223
Kim, A., Goobar, A., & Perlmutter, S. 1996, , 108, 190
Krisciunas, K., Phillips, M. M., & Suntzeff, N. B. 2004a, , 602, L81
Krisciunas, K., Prieto, J. L., Garnavich, P. M., Riley, J.-L. G., Rest, A., Stubbs, C., & McMillan, R. 2006, , 131, 1639
Krisciunas, K., et al. 2003, , 125, 166
Krisciunas, K., et al. 2004b, , 128, 3034
Krisciunas, K., et al. 2007, , 133, 58
Krisciunas, K., et al. 2009, , 138, 1584
Landolt, A. U. 1992, , 104, 340
Leibundgut, B., et al. 1993, , 105, 301
Li, W., et al. 2000, in Cosmic Explosions, ed. S. S. Holt & W. W. Zhang (New York: AIP), 103
Lira, P. 1995, Master’s Thesis, Univ. Chile
Livio, M. 2000, in The Greatest Explosions since the Big Bang: Supernovae and Gamma-Ray Bursts, ed. M. Livio, N. Panagia, & K. Sahu (Baltimore: STScI), 334
Maeda, K., et al. 2010, , 466, 82
Mandel, K. S., Narayan, G., & Kirshner, R. P. 2011, , in press (arXiv:1011.5910)
Matheson, T., Filippenko, A. V., Ho, L. C., Barth, A. J., & Leonard, D. C. 2000, , 120, 1499
Meikle, W. P. S. 2000, , 314, 782
Minkowski, R. 1941, , 53, 224
Munari, U., & Zwitter, T. 1997, , 318, 269
Nugent, P., Kim, A., & Perlmutter, S. 2002, , 114, 803
Oke, J. B., & Sandage, A., 1968, , 154, 21
Patat, F., et al. 2007, Science, 317, 924
Perlmutter, S., et al. 1999, , 517, 565
Persson, S. E., Murphy, D. C., Krzeminski, W., Roth, M., & Rieke, M. J. 1998, , 116, 2475
Phillips, M. M. 1993, , 413, L105
Phillips, M. M., Lira, P., Suntzeff, N. B., Schommer, R. A., Hamuy, M., & Maza, J. 1999, , 118, 1766
Phillips, M. M., et al. 2003, in From Twilight to Highlight, ed. W. Hillebrandt, B. Leibundgut (Berlin: Springer-Verlag), 193
Poznanski, D., Ganeshalingam, M., Silverman, J. M., & Filippenko, A. V. 2011, MNRAS, in press
Prieto, J. L., et al. 2006a (arXiv:0706.4088)
Prieto, J. L., Rest, A., & Suntzeff, N. B. 2006b, , 647, 501
Riess, A. G., et al. 1998, , 116, 1009
Riess, A. G., et al. 2009, , 699, 539
Röpke, F. K., & Hillebrandt, W. 2004, , 420, L1
Scalzo, R. A., et al. 2010, , 713, 1073
Schlegel, D. J., Finkbeiner, D. P., & Davis, M. 1998, , 500, 525
Sheinis, A. I., Bolte, M., Epps, H. W., Kibrick, R. I., Miller, J. S., Radovan, M. V., Bigelow, B. C., & Sutin, B. M., 2002, PASP, 114, 851
Silverman, J. M., Ganeshalingam, M., Li, W. D., Filippenko, A. V., Miller, A. A., & Poznanski, D. 2010, , 410, 585
Skrutskie, M. F., et al. 2006, , 131, 1163
Stritzinger, M., et al. 2002, , 124, 2100
Stritzinger, M., & Leibundgut, B. 2005, , 431, 423
Stritzinger, M., Suntzeff, N. B., Hamuy, M., Challis, P., Demarco, R., Germany, L., & Soderberg, A. M. 2005, , 117 810
Stritzinger, M., Leibundgut, B., Walch, S., & Contardo, G. 2006, , 450, 241
Stritzinger, M., et al. 2010, , 140, 2036
Stritzinger, M., et al. 2011, in press
Strolger, L.-G., et al. 2002, , 124, 2905
Swift, B., & Li, W. D. 2001, 7611
Tanaka, M., et al. 2010, , 714, 1209
Taubenberger, S., et al. 2011, , in press (arXiv:1011.5665)
Thomas, R., et al. 2011, , in press
Wagers, A., Wang, L., & Asztalos, S. 2010, , 711, 711
Wang, L. 2005, , 635, L33
Wang, L., Baade, D., Höflich, P., Wheeler, J. C., Kawabata, K., Khokhlov, A., Nomoto, K., & Patat, F. 2006, , 653, 490
Wang, X., et al. 2009a, , 697, 380
Wang, X., et al. 2009b, , 699, L139
Wood-Vasey, W. M., et al. 2008, ApJ, 689, 377
Yamanaka, M., et al. 2009, , 707, L118
Yuan, F., et al. 2010, , 715, 1338
[ll]{} Host galaxy & IC 4423\
Host-galaxy type & Sbc\
Heliocentric radial velocity & 9067\
CMB-frame radial velocity & 9266\
Distance modulus & $35.55 \pm 0.1$ mag\
$E(B-V)_{\rm Gal}$ & 0.026 $\pm$ 0.006 mag\
$E(B-V)_{\rm host}$ & 0.072 $\pm$ 0.008 mag\
SN $\alpha$(J2000) & $14^{\rm h}26^{\rm m}17^{\rm s}\hspace{-1 mm}.0$\
SN $\delta$(J2000) & $+26\degr 14'55\farcs8$\
Offset from nucleus & 103W 93N\
Julian Date of $B$-band maximum & 2,452,022.49 $\pm 0.8$\
UT Date of $B$-band maximum & 2001 April 23.0\
$\Delta m_{15}(B)$ & 0.68 $\pm$ 0.05 mag\
$M_B$(max) & $-19.19 \pm 0.12$\
$M_V$(max) & $-19.17 \pm 0.10$\
$M_R$(max) & $-19.10 \pm 0.10$\
$M_I$(max) & $-18.90 \pm 0.10$\
$M_J$(max) & $-18.77 \pm 0.10$\
$M_H$(max) & $-18.62 \pm 0.10$\
[rccccc]{} 1 & 17.112 (0.043) & 16.399 (0.019) & 15.480 (0.008) & 14.859 (0.018) & 14.386 (0.015)\
2 & 18.224 (0.047) & 17.262 (0.020) & 16.193 (0.007) & 15.467 (0.019) & 14.921 (0.016)\
3 & 16.705 (0.022) & 16.846 (0.019) & 16.367 (0.008) & 15.981 (0.022) & 15.622 (0.021)\
4 & 17.933 (0.035) & 18.271 (0.030) & 17.864 (0.005) & 17.503 (0.025) & 17.131 (0.025)\
5 & 19.246 (0.174) & 18.729 (0.027) & 17.884 (0.015) & 17.302 (0.024) & 16.796 (0.033)\
6 & 16.813 (0.031) & 16.853 (0.017) & 16.320 (0.007) & 15.917 (0.017) & 15.574 (0.014)\
7 & 20.437 (0.129) & 19.585 (0.015) & 18.313 (0.010) & 17.487 (0.017) & 16.730 (0.010)\
8 & 20.137 (0.230) & 19.987 (0.196) & 18.674 (0.070) & 18.110 (0.100) & 17.467 (0.111)\
9 & 21.270 (0.289) & 19.802 (0.108) & 18.790 (0.059) & 18.009 (0.101) & 17.467 (0.074)\
10 & …& …& …& 20.719 (0.155) & 18.930 (0.061)\
[rcc]{} 1 & 13.718 (0.014) & 13.251 (0.011)\
6 & 15.153 (0.021) & 14.881 (0.008)\
7 & 15.869 (0.007) & 15.281 (0.007)\
8 & 16.566 (0.030) & 15.932 (0.031)\
10 & 17.838 (0.024) & 17.359 (0.029)\
[lccccccc]{} 2020.78 & …& 16.773 (0.043) & 16.708 (0.010) & 16.694 (0.015) & 16.771 (0.017) & 1 & N\
2021.71 & 16.419 (0.034) & 16.731 (0.020) & 16.700 (0.010) & 16.695 (0.006) & 16.802 (0.010) & 1 & N\
2022.90 & …& 16.687 (0.050) & 16.643 (0.020) & 16.635 (0.040) & 16.783 (0.050) & 2 & Y\
2023.88 & …& 16.724 (0.070) & 16.640 (0.020) & 16.661 (0.030) & 16.781 (0.020) & 2 & Y\
2024.89 & …& 16.709 (0.040) & 16.647 (0.020) & 16.626 (0.020) & 16.829 (0.030) & 2 & Y\
2025.89 & …& 16.737 (0.030) & 16.666 (0.020) & 16.692 (0.020) & 16.868 (0.050) & 2 & Y\
2026.86 & …& 16.766 (0.020) & 16.664 (0.020) & 16.659 (0.030) & 16.855 (0.040) & 2 & Y\
2028.89 & …& 16.841 (0.100) & 16.699 (0.080) & …& …& 2 & Y\
2030.88 & …& 16.913 (0.040) & 16.734 (0.020) & 16.775 (0.030) & 16.937 (0.040) & 2 & Y\
2032.88 & …& 17.065 (0.050) & 16.799 (0.030) & 16.860 (0.030) & 16.952 (0.050) & 2 & Y\
2037.88 & …& 17.290 (0.050) & 16.939 (0.030) & 16.948 (0.020) & 17.102 (0.040) & 2 & Y\
2039.86 & …& 17.535 (0.080) & 16.945 (0.050) & 16.963 (0.040) & 17.057 (0.070) & 2 & Y\
2041.80 & …& 17.647 (0.040) & 17.089 (0.050) & …& …& 2 & Y\
2043.85 & …& …& 17.143 (0.020) & 16.968 (0.020) & 17.093 (0.050) & 2 & Y\
2045.86 & …& 17.952 (0.050) & 17.266 (0.030) & 17.006 (0.020) & 17.024 (0.040) & 2 & Y\
2047.84 & …& 18.134 (0.040) & 17.298 (0.040) & 17.015 (0.030) & 16.969 (0.040) & 2 & Y\
2051.81 & …& 18.250 (0.050) & 17.424 (0.030) & 17.123 (0.020) & 16.902 (0.040) & 2 & Y\
2059.81 & …& 18.554 (0.080) & 17.794 (0.040) & 17.385 (0.020) & 17.138 (0.040) & 2 & Y\
2064.83 & …& 18.773 (0.080) & 18.064 (0.070) & 17.629 (0.040) & 17.251 (0.060) & 2 & Y\
2069.79 & …& 19.025 (0.070) & 18.117 (0.060) & 17.813 (0.030) & 17.506 (0.060) & 2 & Y\
2074.74 & …& 19.071 (0.070) & 18.219 (0.070) & 18.008 (0.030) & 17.805 (0.060) & 2 & Y\
2075.81 & …& 18.993 (0.060) & 18.153 (0.060) & 18.099 (0.050) & 17.604 (0.050) & 2 & Y\
2079.75 & …& …& 18.250 (0.070) & 18.180 (0.050) & 17.804 (0.060) & 2 & Y\
2022.99 & …& 16.709 (0.033) & 16.655 (0.013) & 16.693 (0.011) & 16.758 (0.016) & 3 & Y\
2024.91 & …& 16.795 (0.044) & 16.629 (0.025) & 16.686 (0.011) & 16.781 (0.027) & 3 & Y\
2029.81 & …& 16.958 (0.060) & 16.716 (0.010) & 16.781 (0.010) & 16.910 (0.015) & 3 & Y\
2036.86 & …& 17.322 (0.020) & 16.932 (0.010) & 16.901 (0.023) & 17.007 (0.024) & 3 & N\
2037.83 & 17.503 (0.040) & 17.348 (0.019) & 16.990 (0.010) & 16.967 (0.018) & 17.099 (0.020) & 3 & N\
2038.77 & …& 17.488 (0.009) & …& 17.043 (0.010) & 17.079 (0.010) & 3 & N\
2040.79 & 17.655 (0.031) & 17.578 (0.014) & 17.054 (0.010) & 16.964 (0.016) & 16.939 (0.016) & 3 & N\
2041.84 & …& 17.668 (0.034) & 17.003 (0.020) & 16.949 (0.033) & 16.910 (0.034) & 3 & N\
2043.78 & …& 17.784 (0.010) & …& 16.990 (0.010) & 16.845 (0.017) & 3 & N\
2046.82 & …& …& …& 17.107 (0.010) & 16.871 (0.010) & 3 & N\
2050.82 & …& 18.269 (0.013) & …& 16.992 (0.085) & 16.816 (0.010) & 3 & N\
2063.76 & …& …& 18.083 (0.017) & …& 17.225 (0.029) & 3 & N\
2140.70 & …& 20.110 (0.130) & 19.520 (0.100) & 19.390 (0.080) & 19.030 (0.090) & 3 & N\
2026.74 & …& 16.841 (0.062) & 16.673 (0.040) & …& …& 4 & N\
2030.67 & …& 16.953 (0.017) & 16.725 (0.010) & …& …& 5 & N\
2046.62 & …& 18.063 (0.024) & 17.194 (0.013) & …& …& 5 & N\
2047.65 & …& 18.209 (0.047) & 17.247 (0.027) & …& …& 5 & N\
2055.63 & …& …& 17.687 (0.028) & …& …& 5 & N\
2052.65 & 18.633 (0.041) & 18.356 (0.023) & 17.468 (0.012) & 17.152 (0.027) & 16.850 (0.029) & 6 & Y\
[cclcccc]{} Apr 21 & $-$2 & FLWO 1.5-m & $-$0.010 & $-$0.031 & $-$0.120 & $-$0.045\
Apr 24 & +1 & FLWO 1.5-m & 0.000 & $-$0.028 & $-$0.118 & $-$0.068\
Apr 26 & +3 & MMT 6.5-m & $-$0.001 & $-$0.020 & $-$0.090 & $-$0.029\
Apr 27 & +4 & FLWO 1.5-m & 0.018 & $-$0.014 & $-$0.097 & $-$0.015\
Apr 30 & +7 & FLWO 1.5-m & 0.010 & $-$0.017 & $-$0.094 & $-$0.050\
May 02 & +9 & FLWO 1.5-m & 0.023 & $-$0.005 & $-$0.093 & $-$0.065\
May 10 & +17 & BAO 2.16-m & …& $-$0.024: & $-$0.064 & …\
May 16 & +23 & FLWO 1.5-m & 0.080 & 0.047 & $-$0.055 & 0.042\
May 23 & +30 & FLWO 1.5-m & 0.090 & 0.066 & $-$0.057 & 0.030\
May 24 & +31 & MMT 6.5-m & 0.097 & 0.077 & $-$0.039 & $-$0.008\
May 25 & +32 & FLWO 1.5-m & 0.085 & 0.068 & $-$0.061 & 0.025\
May 25 & +32 & MMT 6.5-m & 0.100 & 0.081 & $-$0.040 & 0.022\
May 30 & +37 & FLWO 1.5-m & 0.082 & 0.070 & $-$0.049 & 0.019\
Jun 18 & +56 & FLWO 1.5-m & 0.089 & 0.073 & $-$0.067 & 0.043\
Jun 18 & +56 & MMT 6.5-m & 0.093 & 0.070 & $-$0.056 & 0.059\
[lccccccccc]{} 2020.78 & $-$0.010 & $-$0.031 & $-$0.120 & $-$0.053 & $-$0.008 & $-$0.003 & $-$0.006 & $-$0.002 & 1\
2021.71 & $-$0.006 & $-$0.030 & $-$0.116 & $-$0.052 & $-$0.007 & $-$0.004 & $-$0.005 & $-$0.006 & 1\
2022.90 & $-$0.005 & $-$0.028 & $-$0.113 & $-$0.050 & $-$0.008 & $-$0.005 & 0.022 & $-$0.017 & 2\
2023.88 & $-$0.002 & $-$0.025 & $-$0.110 & $-$0.048 & $-$0.008 & $-$0.005 & 0.021 & $-$0.017 & 2\
2024.89 & 0.004 & $-$0.022 & $-$0.106 & $-$0.046 & $-$0.007 & $-$0.005 & 0.020 & $-$0.017 & 2\
2025.89 & 0.006 & $-$0.021 & $-$0.103 & $-$0.044 & $-$0.007 & $-$0.005 & 0.019 & $-$0.016 & 2\
2026.86 & 0.007 & $-$0.019 & $-$0.100 & $-$0.041 & $-$0.007 & $-$0.005 & 0.019 & $-$0.016 & 2\
2028.89 & 0.013 & $-$0.013 & …& …& $-$0.006 & $-$0.004 & …& …& 2\
2030.88 & 0.022 & $-$0.006 & $-$0.088 & $-$0.033 & $-$0.005 & $-$0.002 & 0.017 & $-$0.016 & 2\
2032.88 & 0.030 & 0.000 & $-$0.083 & $-$0.028 & $-$0.005 & $-$0.001 & 0.017 & $-$0.016 & 2\
2037.88 & 0.045 & 0.016 & $-$0.071 & $-$0.017 & $-$0.005 & 0.005 & 0.017 & $-$0.015 & 2\
2039.86 & 0.050 & 0.022 & $-$0.066 & $-$0.012 & $-$0.005 & 0.007 & 0.017 & $-$0.015 & 2\
2041.80 & 0.057 & 0.029 & …& …& $-$0.006 & 0.008 & …& …& 2\
2043.85 & …& 0.037 & $-$0.059 & $-$0.007 & …& 0.010 & 0.019 & $-$0.014 & 2\
2045.86 & 0.071 & 0.045 & $-$0.056 & 0.002 & $-$0.007 & 0.011 & 0.020 & $-$0.014 & 2\
2047.84 & 0.078 & 0.053 & $-$0.054 & 0.007 & $-$0.008 & 0.011 & 0.021 & $-$0.014 & 2\
2051.81 & 0.089 & 0.068 & $-$0.050 & 0.015 & $-$0.011 & 0.012 & 0.023 & $-$0.013 & 2\
2059.81 & 0.089 & 0.075 & $-$0.048 & 0.031 & $-$0.017 & 0.009 & 0.027 & $-$0.011 & 2\
2064.83 & 0.089 & 0.073 & $-$0.050 & 0.039 & $-$0.018 & 0.007 & 0.029 & $-$0.010 & 2\
2069.79 & 0.087 & 0.067 & $-$0.053 & 0.045 & $-$0.018 & 0.004 & 0.030 & $-$0.009 & 2\
2074.74 & 0.088 & 0.066 & $-$0.058 & 0.048 & $-$0.011 & 0.005 & 0.030 & $-$0.007 & 2\
2075.81 & 0.088 & 0.067 & $-$0.059 & 0.049 & $-$0.009 & 0.005 & 0.030 & $-$0.007 & 2\
2079.75 & …& 0.072 & $-$0.062 & 0.050 & …& 0.010 & 0.028 & $-$0.006 & 2\
2022.99 & $-$0.005 & $-$0.028 & $-$0.113 & $-$0.050 & 0.003 & $-$0.009 & 0.024 & 0.027 & 3\
2024.91 & 0.004 & $-$0.022 & $-$0.106 & $-$0.046 & 0.004 & $-$0.008 & 0.023 & 0.036 & 3\
2029.81 & 0.017 & $-$0.010 & $-$0.091 & $-$0.035 & 0.000 & $-$0.002 & 0.022 & 0.060 & 3\
2036.86 & 0.042 & 0.013 & $-$0.073 & $-$0.019 & $-$0.014 & 0.007 & 0.024 & 0.073 & 3\
2037.83 & 0.045 & 0.016 & $-$0.071 & $-$0.017 & $-$0.016 & 0.008 & 0.024 & 0.076 & 3\
2038.77 & 0.047 & …& $-$0.068 & $-$0.014 & $-$0.018 & …& 0.024 & 0.079 & 3\
2040.79 & 0.053 & 0.026 & $-$0.064 & $-$0.010 & $-$0.022 & 0.011 & 0.024 & 0.082 & 3\
2041.84 & 0.057 & 0.029 & $-$0.063 & $-$0.007 & $-$0.024 & 0.011 & 0.024 & 0.082 & 3\
2043.78 & 0.064 & …& $-$0.059 & $-$0.003 & $-$0.028 & …& 0.024 & 0.082 & 3\
2046.82 & …& …& $-$0.056 & 0.004 & …& …& 0.025 & 0.075 & 3\
2050.82 & 0.087 & …& $-$0.051 & 0.013 & $-$0.037 & …& 0.027 & 0.053 & 3\
2063.76 & …& 0.074 & …& 0.037 & …& $-$0.005 & …& 0.000 & 3\
2026.74 & 0.007 & $-$0.019 & …& …& …& …& …& …& 4\
2030.67 & 0.008 & $-$0.006 & …& …& …& …& …& …& 5\
2046.62 & 0.075 & 0.049 & …& …& …& …& …& …& 5\
2047.65 & 0.078 & 0.053 & …& …& …& …& …& …& 5\
2055.63 & …& 0.075 & …& …& …& …& …& …& 5\
2052.66 & 0.089 & 0.068 & $-$0.050 & 0.017 & 0.018 & 0.006 & 0.000 & $-$ 0.028 & 6\
[cccc]{} 2024.73 & …& 17.045 (0.042) & 1\
2025.67 & 16.977 (0.032) & 16.893 (0.044) & 1\
2026.70 & 17.050 (0.028) & …& 1\
2029.70 & 17.132 (0.022) & 17.000 (0.035) & 1\
2031.68 & 17.136 (0.022) & 16.965 (0.036) & 1\
2032.70 & 17.190 (0.028) & 17.110 (0.027) & 1\
2045.66 & 17.291 (0.024) & 16.873 (0.020) & 2\
2047.65 & 17.509 (0.049) & …& 1\
2049.66 & 17.396 (0.035) & 16.853 (0.039) & 1\
2051.65 & 17.430 (0.034) & 16.894 (0.038) & 1\
2053.67 & 17.337 (0.031) & …& 1\
2055.67 & 17.387 (0.031) & 17.078 (0.056) & 1\
2056.67 & 17.420 (0.031) & 16.998 (0.022) & 1\
2061.65 & …& 16.905 (0.022) & 2\
2068.57 & 17.592 (0.033) & 17.157 (0.028) & 2\
[lccccccccccccl]{} Apr 21.48 & 2452020.98 & FLWO & 3720–7540 & 7.0 & 71.00 & 70.02 & 1.48 & Feige34 & & 1–2 & 3 & 1200 & Calkins\
Apr 24.31 & 2452023.81 & FLWO & 3720–7540 & 7.0 & 0.00 & $-$16.25 & 1.01 & Feige34 & clear & 2 & 3 & 1200 & Berlind\
Apr 26.33 & 2452025.83 & FLWO & 3720–7540 & 7.0 & 45.00 & 43.51 & 1.01 & Feige34 & some clouds & 1–2 & 3 & 1200 & Berlind\
Apr 26.40 & 2452025.90 & MMTO & 3250–8900 & 8.0 & Par.A. & 71.15 & 1.12 & Feige34/BD26 & & & & 900 & Matheson\
Apr 26.41 & 2452025.91 & MMTO & 3250–8900 & 8.0 & Par.A. & 71.52 & 1.15 & Feige34/BD26 & & & & 900 & Matheson\
Apr 27.36 & 2452026.86 & FLWO & 3720–7540 & 7.0 & 64.00 & 65.79 & 1.04 & HZ44 & & 1–2 & 3 & 1200 & Calkins\
Apr 30.36 & 2452029.86 & FLWO & 3720–7540 & 7.0 & 70.00 & 67.86 & 1.05 & Feige34 & clear & 1–2 & 3 & 1200 & Berlind\
May 02.29 & 2452031.79 & FLWO & 3720–7540 & 7.0 & $-$9.00 & $-$19.75 & 1.01 & HZ44 & clear & 1–2 & 3 & 1200 & Berlind\
May 16.26 & 2452045.76 & FLWO & 3720–7540 & 7.0 & $-$33.00 & $-$13.29 & 1.01 & HZ44 & clear & 1–2 & 3 & 1200 & Calkins\
May 23.33 & 2452052.83 & FLWO & 3720–7540 & 7.0 & 69.00 & 71.42 & 1.14 & Feige34 & cirrus & 2 & 3 & 1200 & Rines\
May 24.32 & 2452053.82 & MMTO & 3220–8900 & 8.0 & at Par.A. & 70.92 & 1.11 & BD28/BD26 & clear & & & 1200 & Challis, Phelps\
May 24.33 & 2452053.83 & MMTO & 4950–10000 & 8.0 & Par.A. & 71.51 & 1.16 & BD26 & clear & & & 1200 & Challis, Phelps\
May 25.32 & 2452054.82 & MMTO & 3220–8900 & 8.0 & Par.A. & 71.36 & 1.14 & Feige34/BD26 & clear & & & 1800 & Challis\
May 25.35 & 2452054.85 & MMTO & 4950–10000 & 8.0 & Par.A. & 71.52 & 1.23 & BD26 & clear & & & 1800 & Challis\
May 25.37 & 2452054.87 & FLWO & 3720–7500 & 7.0 & 69.00 & 70.63 & 1.39 & Feige34 & cirrus & 2 & 3 & 1200 & Rines\
May 30.36 & 2452059.86 & FLWO & 3720–7540 & 7.0 & 71.00 & 70.51 & 1.40 & HZ44 & clear & 1–2 & 3 & 1200 & Calkins\
May 30.38 & 2452059.88 & FLWO & 3720–7540 & 7.0 & 71.00 & 69.68 & 1.53 & HZ44 & cirrus & 1–2 & 3 & 1200 & Calkins\
June 18.22 & 2452078.72 & FLWO & 3720–7540 & 7.0 & 69.00 & 66.99 & 1.05 & HZ44 & clouds & 1–2 & 3 & 1200 & Berlind\
June 18.29 & 2452078.79 & MMTO & 3200–8900 & 8.0 & Par.A. & 71.24 & 1.29 & BD28/BD26 & some clouds & & & 900 & Matheson\
June 18.30 & 2452078.80 & MMTO & 3200–8900 & 8.0 & Par.A. & 70.84 & 1.35 & BD28/BD26 & some clouds & & & 900 & Matheson\
[cclcc]{} Apr 22 & $-$1 & Keck II 10-m & 3922–10000 & 1.25\
Apr 29 & +6 & KPNO 2.1-m & 5350–\[9500\] & 4.8\
Apr 30 & +7 & Lick 3-m & 3260–10600 & 5.5/10.5\
May 01 & +8 & LCO 2.5-m & 3500–9199 & 7.0\
May 02 & +9 & [*HST*]{} 2.4-m & 2877–5348 & 3.55 to 4.10\
May 09 & +16 & [*HST*]{} 2.4-m & 2930–5695 & 3.55 to 4.10\
May 10 & +17 & BAO 2.16-m & 4100–9000 & 9.6\
May 16 & +23 & Lick 3-m & 3300–10450 & 5.5/10.5\
May 17 & +24 & LCO 2.5-m & 3500–9350 & 7.0\
May 20 & +27 & BAO 2.16-m & 3450–9000 & 9.6\
May 23 & +30 & KPNO 4.0-m & 3500–5493/5300–11600 & 7.8/13.8\
May 25 & +32 & BAO 2.16-m & 3700–9300 & 9.6\
May 26 & +33 & LCO 2.5-m & 3800–9234 & 7.0\
May 30 & +37 & BAO 2.16-m & 3970–8200 & 4.8\
[lccc]{} & 0.68 (0.05) & 0.72–0.78 & 0.72 (0.03)\
$M_V$(max) & $-$19.17 (0.10) & $-$19.40 (0.08) & $-$20.16 (0.10)\
Si II velocity () & $-$14,340 \[$-$1.5 d\] & $-$10,200 \[$-$5 d\] & $-$8600 \[$-$7 d\]\
C II line(s) & weak & not detectable & very strong\
$H$-band light curve & flat & peaked early & increasing\
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The recent discovery of rotating radio transients and the quasi-periodicity of pulsar activity in the radio pulsar PSR B1931$+$24 has challenged the conventional theory of radio pulsar emission. Here we suggest that these phenomena could be due to the interaction between the neutron star magnetosphere and the surrounding debris disk. The pattern of pulsar emission depends on whether the disk can penetrate the light cylinder and efficiently quench the processes of particle production and acceleration inside the magnetospheric gap. A precessing disk may naturally account for the switch-on/off behavior in PSR B1931$+$24.'
author:
- 'Xiang-Dong Li'
title: On the Nature of Part Time Radio Pulsars
---
Introduction
============
A small group of rotating radio transients (RRATs) were recently reported by @mcl06. These objects are characterized by single, dispersed bursts of radio emission with durations between 2 and 30 ms. The average time intervals between bursts range from 4 minutes to 3 hours, with radio emission typically detectable for $< 1$ s per day. Periodicities in the range $0.4-7$ s for 10 of the 11 sources have been measured, suggesting that they are rotating neutron stars. Three of the sources have measured period derivatives, with one (RRAT J1819$-$1458) having a very high inferred magnetic field of $5\times 10^{13}$ G, if spin-down by magnetic dipole radiation is assumed.
A similar bursting radio source GCRT J1745$-$3009 was detected previously [@hym05], whose notable properties include “flares" approximately 1 Jy in magnitude lasting approximately 10 minutes each and occurring at apparently regular 77 minute intervals. GCRT J1745$-$3009 is located approximately $10\arcmin$ from the Galactic center, and just outside of the shell-type supernova remnant (SNR) G359.1$-$0.5 [@rei84]. If GCRT J1745$-$3009 and the SNR are related, then GCRT J1745$-$3009 would have the age of $\sim 10^5$ yr, comparable to RRAT J1819$-$1458. But the overall age distribution of RRATs is still not clear, although 5 out of 10 have spin periods $P>4$ s.
Despite the small number of RRATs detected, @mcl06 suggest that their ephemeral nature may point to a total Galactic population significantly exceeding that of the regularly pulsing radio pulsars. Investigation of the nature of RRATs will be of great interest on understanding pulsar formation, evolution and radiation mechanisms.
Several possible interpretations for the peculiar properties of RRATs as well as GCRT J1745$-$3009 have been examined. For the latter, models involving a precessing pulsar [@zhu06], binary neutron star [@tur05], and transient white dwarf pulsar [@zha05] have been suggested. More recently, @zha06 discuss two possible interpretations to RRATs: the first model suggests that these objects are pulsars slightly below the radio emission “death line", and become active occasionally when the conditions for pair production and coherent emission are satisfied; the second one invokes a radio emission direction reversal in normal pulsars due to some unknown reasons.
The re-activated dead pulsar model in Zhang et al. (2006) is particularly interesting, although the so-called “death line" is highly uncertain, because it depends not only on the magnetic field configuration, but significantly on the origin of gamma quanta, which are responsible for pair production. The outbursts of radio emission are assumed to be caused by internal magnetic field evolution in the neutron stars - when stronger multipole magnetic fields emerge to the polar cap region of the neutron star, the pair production condition could be satisfied and the neutron star behaves like a radio pulsar. A few lines of implications can be drawn from this idea. First, the large population of RRATs may be either due to higher birthrate of young pulsars than previously thought, or caused by a pile-up effect at large periods. This could be testified by future population synthesis of radio pulsars. Second, since reconnection typically occurs at a fraction of Alfven velocity $v_{\rm A}$, in the case where the instability is driven by the internal field, the growth time of the instability is [@tho95] $$\Delta t\sim \frac{\Delta l}{v_{\rm A}}\sim
10B_{*,12}^{-1}\rho_{15}^{1/2}(\frac{\Delta l}{0.1R_*})\,{\rm s},$$ where $\Delta l$ is the displacement of the field lines, $B_*=10^{12}B_{*,12}$ G, $\rho=10^{15}\rho_{15}$ gcm$^{-1}$, and $R_*$ the surface magnetic field strength, the core density, and the radius of the neutron star, respectively. To be compared to the measured RRAT burst duration of $\sim 10$ ms, one has to assume that the field readjustment should take place in the outer crust of the neutron star, where $\rho_{15}\sim 10^{-6}$.
In this [*Letter*]{} we present an alternative interpretation to RRATs. We propose that some of them may be isolated neutron stars surrounded by a debris disk, which originate either from the supernovae that produced the neutron stars or from the captured interstellar medium. The neutron stars act as a propeller when the disk penetrates inside the light cylinder, and the outflow or wind from the disk may quench the pair production processes in the pulsar magnetosphere with only transient radio emission allowed. The interaction between a processing debris disk with the neutron star magnetic field may also be responsible for the quasi-periodic switch-on/off transition in PSR B1931$+$24 recently reported by @kra06.
RRATs
=====
In the standard model for radio pulsars, the rotationally induced electric field of a rotating, magnetized neutron star pulls the plasma off the surface and eject it beyond the light cylinder to form a relativistic wind [@gol69]. Pulsar emission is then associated with the acceleration of particles to maintain this wind, which appears by the avalanche process of (e$^{\pm}$) pair production [@rud75]. In the polar-cap models, the acceleration and radiation occur near the magnetic poles in the inner magnetosphere, while in the outer-gap models, these processes occur in the outer magnetosphere (see Kaspi, Roberts, & Harding for a review). The flow of the plasma along open field lines results in some plasma void regionsin the vicinity of null-charge surfaces. In such charge-deficient regions (or “gaps") electric field parallel the magnetic field lines $E_{||}\neq 0$ is sustained, electrons and positrons can be accelerated to relativistic energies.
It is interesting to see whether the conditions for pulsar emission will be satisfied if there is a disk surrounding the neutron star. Michel & Dessler (1981, 1983; see also Michel 1988) argued that radio pulsars and X-ray pulsars differ mainly in the fact that the former are surrounded by a supernova fallback disk with negligible accretion, while the latter are surrounded by an accretion disk. Alternatively, the debris disks around isolated neutron stars may result from the captured interstellar medium [e.g. @pop00]. The fallback disk model was recently adopted to account for the observational characteristics of anomalous X-ray pulsars [@yus95; @cha00; @alp01]. Such disks can also influence pulsar braking indices and timing ages. @men01 have explored a disk model for the spin-down of young radio pulsars, in which the neutron star loses rotational energy not only by emitting magnetic dipole radiation, but also by torquing a supernova fallback disk. @mar01 considered a similar model to explain the discrepancies between a pulsar’s true age and its characteristic age. Recent X-ray observations also show that some young pulsars, such as the Crab and Vela pulsars, may have the jet configuration, which suggests the existence of a disk surrounding the neutron star [@bla04 and references therein]. The strongest constraints on the presence of a disk is given by its optical and longer wavelength emission. @perna00 have examined the reprocessing of beamed pulsar emission by the debris disks. They found that, since the reradiated flux gives the dominant contribution at long wavelengths produced in the bulk of the disk, whereas the optical emission is generated in its innermost part, the optical rather the longer wavelength emission would be highly suppressed, if the inner edge of the disk were truncated at a radius larger than the magnetospheric radius. Most recently, @wan06 report the discovery of mid-infrared emission from a cool disk around the isolated young X-ray pulsar 4U 0142$+$61, presenting the first direct evidence for supernova fallback.
In the original picture of @stu71 and @mic81 the debris disk is magnetically coupled with the neutron star with no accretion. This may be true for cold disks with extremely low viscosity. However, it is conventionally thought that the pulsar emission will quenched if the disk wind plasma penetrating into the light cylinder.
Most of the neutron stars with a surrounding disk should have experienced the accretor and propeller regimes [@ill75]. The existence of a disk inside the light cylinder may significantly influence the pulsar radiation processes. Magnetocentrifugally driven outflows from the disks has been discussed by a number of authors during the propeller phase [e.g. @cam90; @kon91; @lov95; @aga00; @ust06]. The disk wind itself may also be strong enough to influence the structure of pulsar winds [@bla04]. It has already been shown that the $\sim 10^{12}$ V potential difference across the magnetospheric gap and the outward-directed electric field required by the Ruderman-Sutherland model for the generation of radio waves will be negated, if the number density of matter at the Alfven surface is greater than $\sim 7.2\times 10^7$ cm$^{-3}$ [@wan83], a condition satisfied by most neutron stars with a debris disk where there is significant wind or outflow from the disk. The density of the outflow plasma, if similar to that in the disk, can be estimated to be $$\rho_{\rm w}=\frac{\dot{M}}{2\pi RHv_{\rm r}m_{\rm H}}\simeq
2.2\times 10^{15}(\frac{\dot{M}}{10^{14}{\rm
gs}^{-1}})(\frac{\alpha}{0.01})^{-1}(\frac{H/R}{0.1})^{-3}(\frac{R}{10^9\,{\rm
cm}})^{-3/2}\,{\rm cm}^{-3},$$ where $\dot{M}$ is the mass inflow rate in the disk, $H$ the half thickness, $v_{\rm r}$ the radial velocity, $R$ the inner radius of the disk, $\alpha$ the viscosity parameter, and $m_{\rm H}$ the proton mass, respectively. It can be much larger than the Goldreich-Julian density $\rho_{\rm GJ}$ for typical values of the adopted parameters, if one assumes that the electron-positron pairs in the gap are close to saturation, $$\rho_{\rm GJ}=\frac{\Omega B}{2\pi ce}=7\times 10^{10}B_{*,
12}P^{-1}(\frac{R_*}{R})^{-3}\,{\rm cm}^{-3},$$ where $P$ is the neutron star spin period. Failure of pulsar emission may also partly result from the fact that coherent radiowaves with a wavelength longer than 75 cm can be absorbed effectively in the wind plasma [@ill75].
The flow in the inner part of the accretion disk is expected to have density fluctuations (“clumps") produced by a variety of mechanisms, such as thermal instability, Kelvin-Helmholtz instability, and magnetoturbulence [see @lam85; @shi87]. The clumpy wind density would be much higher than the averaged value estimated above. They may also leave short, sporadic “transparent" time for the development of particle acceleration in the gap and generation of pulsar emission. If we assume that the typical clump separation is less than the disk height $H_{\rm
in}$ at the inner edge $R_{\rm in}$ of the disk, the duration of successful pulsar emission should be less than $$\tau\sim H_{\rm in}/v_{\rm esc},$$ where $v_{\rm esc}$ is the escape velocity at $R_{\rm in}$. If $R_{\rm c}<R_{\rm in}<R_{\rm lc}$ and $H/R\la 0.1$, we have $11P$ ms $<\tau<1.7P^{3/2}$ s. Here $R_{\rm c}\equiv(GMP^2/4\pi)^2$ is the corotation radius and $R_{\rm lc}\equiv cP/2\pi$ is the light cylinder radius, respectively. As pointed out by Zhang et al. (2006), the dynamical time scale of the inner gap ($\sim h_{\rm
gap}/c\sim 10^{-6}-10^{-4}$ s, where $\sim h_{\rm gap}$ is the height of the gap) is much smaller than the rotation period $P$. So the time scale to develop a pair cascade is much shorter than $\tau$. Its magnitude seems compatible with the burst durations measured so far.
The debris disk may be popular in relatively young neutron stars. Jiang & Li (2005) performed Monte-Carlo simulation of pulsar evolution, assuming that all neutron stars are born with a surrounding supernova fallback disk with the initial masses of the disk ranging from $10^{-6}\,M_{\sun}$ to $10^{-2}\,M_{\sun}$. They found that the emerging proportion of disk-fed neutron stars (i.e. with the disk extending inside the light cylinder) is $\sim
20\%-50\%$ at age of $10^3$ years, and $\sim 10\%-25\%$ at age of $10^4$ years. Obviously these numbers are sensitive to the assumptions for the initial parameters, e.g., the distributions of the initial disk masses, of the neutron star spin periods, magnetic fields, and most importantly, the mechanisms of the propeller spin-down. However, it clearly demonstrates that a considerable fraction of isolated neutron stars could harbor a debris disk with sufficiently long time (@pop00 suggested that a fraction of $0.1\%-0.2\%$ of all isolated neutron stars may be presently in the propeller stage due interaction with the interstellar medium). It was also found that the ratio of the characteristic age $t_{\rm c}=P/2\dot{P}$ and the true age $t$ distributes within a relatively wide range from $\sim 0.1$ to $\sim 10$, indicating that $t_{\rm c}$ and the magnetic field strength estimated from magnetic dipole radiation may considerably deviate from the actual values for these neutron stars. The disk-assisted spin-down may also explain why RRATs have relatively long spin periods compared with normal isolated radio pulsars.
@rey06 recently report the discovery of the X-ray counterpart to RRAT J1819$-$1458. While their data are insufficient for fitting to more detailed neutron star atmosphere models, they suggest that the emission from RRAT J1819$-$1458 is consistent with a cooling neutron star of age $\sim 10^4-10^5$ yr, at a distance $\la 2$ kpc. This seems to be in contradiction with our scenario since thermal, soft radiation is not expected from a propeller, in which nonthermal magnetospheric emission should dominate (e.g. Popov, Turolla, & Possenti 2006). From the work of @can90 and @min97 for supernova fallback, we can roughly estimate the mass inflow rate in the disk as $$\dot{M}\simeq 1.8\times 10^{14}(\frac{M}{1
M_{\odot}})(\frac{\Delta
M}{10^{-5}M_{\odot}})(\frac{\alpha}{0.01}) (\frac{t}{10^5\,{\rm
yr}})^{-1.35}\,{\rm gs}^{-1},$$ where $\Delta M$ is the amount of fallback material. With Eq. (5) the maximum luminosity released by the propeller process is $$L_{\rm prop}=\frac{GM\dot{M}}{R_{\rm m}}\simeq 2.4\times
10^{30}(\frac{M}{1 M_{\odot}})^2(\frac{\Delta
M}{10^{-5}M_{\odot}})(\frac{R_{\rm m}}{0.5R_{\rm
lc}})^{-1}(\frac{\alpha}{0.01}) (\frac{t}{10^5\,{\rm
yr}})^{-1.35}\,{\rm ergs}^{-1},$$ for RRAT J1819$-$1458, which is much smaller than the measured luminosity $\sim 10^{33}\,{\rm ergs}^{-1}$, implying that neutron star cooling could still dominate X-ray emission in this object. But we mention that a neutron star undergoing propeller spindown could be a weak point source of $\gamma$-ray radiation during the (radio-)quiescent state [@wan85].
PSR B1931$+$24
==============
More recently @kra06 report the quasi-periodical pattern in the radio pulsar PSR B1931$+$24: the radio emission switches off in less than 10 seconds after the active phases of $\sim 5-10$ days, and remains undetectable for the next $\sim 25-35$ days when it switches on again. More remarkably, the pulsar rotation slows down $50\%$ faster when it is on than when it is off, indicating an increase in magnetospheric currents when the pulsar switches on. As pointed out by @kra06, the discovery of PSR B1931+24’s behaviour suggests that many more such objects exist in the Galaxy, and the bursting radio source GCRT J1745$-$3009 may turn out to be a short-timescale version of PSR B1931$+$24 and hence to be a radio pulsar.
The $35$ day period is not likely to be attributed to free precession of the neutron star, because no evidence of expected profile changes is found [@kra06]. However, it might be accounted for in our pulsar $+$ debris disk model. Here we suggest that the $35$ day period is the precession period of the debris disk. It is well known that the neutron star receives a kick during the supernova explosion, and it is likely that there is misalignment between the angular momenta of the fallback disk and the neutron star, leading to free precession of the disk [@kat73; @rob74]. The disk precession can also be induced by the radiation or magnetic torques generated from the neutron star [e.g. @pet77; @pri96; @lai99]. There is extensive evidence for a warped, precessing disk in X-ray binaries and active galactic nuclei [e.g. @ogi01 and references therein].
The quasi-periodicity in PSR B1931$+$24 may be explained in the following picture. We assume that the inner edge of the debris disk is close to the pulsar’s light cylinder. It is known that the horizontal distance of the disk from the spin axis of the neutron star always changes during the precession. As soon as the disk penetrates inside the light cylinder, the propeller process commences along with outflows from the disk, particle acceleration processes in the magnetospheric gap are then quenched and the coherent radio emission cuts off. The neutron star slows down only by magnetic dipole radiation. The pulsar radiation switches on when the disk moves outside the light cylinder. In this case both magnetic dipole radiation and pulsar wind brake the neutron star, so that the pulsar slows down faster than during the off phase. This scenario also suggests that PSR B1931$+$24 may appear as a RRAT during the off phase. The recent detection of transient pulsed radio emission from the anomalous X-ray pulsar XTE J1810$-$197 [@cam06] could be an example of this transition.
Agapitou, V. & Papaloizou, J. C. B. 2000, MNRAS, 317, 273 Alpar, M. A. 2001, ApJ, 554, 1245 Blackman, E. G. & Perna, R. 2004, ApJ, 601, L71 Camenzind, M. 1990, Rev. Mod. Astron., 3, 234 Camilo, F. et al. 2006, astro-ph/06605429 Cannizzo, J. K., Lee, H. M., & Goodman, J., 1990, ApJ, 351, 38 Chattterjee, P., Hernquist, L., & Narayan, R., 2000, ApJ, 534, 373 Goldreich, P. & Julian, W. H. 1969, ApJ, 157, 869 Hyman, S. D., Lazio, T. J. W., Kassim, N. E., Ray, P. S., Markwardt, C. B. & Yusef-Zadeh, F. 2005, Nat, 434, 50 Illarionov, A. F. & Sunyave, R. A., 1975, A&A, 39, 185 Jiang, Z.-B. & Li, X.-D. 2005, ChJA&A, 5, 487 Kaspi, V., Roberts, M. S. E., & Harding, A. K. 2006, in Compact Stellar X-ray Source, eds. M. van der Klis, W. H. G. Lewin (Cambridge University Press), in press Katz, J. I. 1973, Nat, 246, 87 Königl, A. 1991, ApJ, 370, L39 Kramer, M., Lyne, A. G. L, O’Brien, J. T., Jordan, C. A., & Lorimer, D. R. 2006, Sci, in press Lai, D. 1999, ApJ, 524, 1030 Lamb, F. K. Shibazaki, N., Alpar, A., & Shaham, J. 1985, Nat, 317, 681 Lovelace, R. V. E., Romanova, M. M., & Bisnovatyi-Kogan, G. S. 1995, MNRAS, 275, 244 Marsden, D., Lingenfelter, R. E., & Rothschild, R. E. 2001, ApJ, 547, L45 McLaughlin, M. A., Lyne, A. G., Lorimer, D. R., Kramer, M., Faulkner, A. J. 2006, Nat, 439, 817 Menou, K., Perna, R., & Hernquist, L. 2001, ApJ, 554, L63 Michel, F. C. 1988, Nat, 333, 644 Michel, F. C. & Dessler, A. J. 1981, ApJ, 251, 654 Michel, F. C. & Dessler, A. J., 1983, Nat, 303, 48 Mineshige, S., Nomura, H., Hirose, M., Nomoto, K., & Suzuki, T. 1997, ApJ, 489, 227 Ogilvie, G. I. & Dubus, G. 2001, , 320, 485
Perna, R. & Hernquist, L. 2000, ApJ, 544, L57 Petterson, J. A. 1977, ApJ, 218, 783 Popov, S. B., Colpi, M., Treves, A., Turolla, R., Lipunov, V. M., & Prokhorov, M.E., 2000, ApJ, 530, 896 Popov, S. B., Turolla, R., & Possenti, A. 2006, MNRAS, in press (astro-ph/0603258) Pringle, J. E. 1996, MNRAS, 281, 357 Reich, W. & Fürst, E. 1984, A&AS, 57, 165 Reynolds, S. P. et al. 2006, ApJ, 639, L71 Roberts, W. J. 1974, ApJ, 187, 575 Ruderman, M. & Sutherland, P. G. 1975, ApJ, 196, 51 Shibazaki, N., & Lamb, F. K. 1987, ApJ, 318, 767 Sturrock, P. A. 1971, ApJ, 164, 529 Thompson, C. & Duncan, R. C. 1995, , 275, 265 Turolla, R., Possenti, A., & Treves, A. 2005, ApJ, 628, L49 Ustyugova, G. V., Koldoba, A. V., Romanova, M. M., & Lovelace, R. V. E. 2006, ApJ, in press Wang, Y.-M. & Robertson, J. A. 1985, A&A, 151, 361 Wang, Z., Chakrabarty, D., & Kaplan, D. L. 2006, Nat, 440, 772 Wang, Z.-R. 1983, In High energy astrophysics and cosmology (A85-19326 07-90). Beijing/New York, Science Press/Gordon and Breach Science Publishers, S.A., p. 270 Yusifov, I. M., Alpar, M. A., Gok, F., & Huseyinov O. H., 1995, In The Lives of the Neutron Stars, M. A. Alpar, Ü. Kiziloglu, J. van Paradijs, eds. (NATO ASI Ser. C, 450; Dordrecht: Kluwer), 201 Zhang, B. & Gil, J. 2005, ApJ, 631, L143 Zhang, B., Gil, J., & Dyks, J. 2006, ApJ, submitted (astro-ph/0601063) Zhu, W. W. & Xu, R. X. 2006, , 365, L16
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
A theoretical technique which allows to include contributions from non-orthogonality of the electron states in the leads connected to a tunneling junction is derived. The theory is applied to a single barrier tunneling structure and a simple expression for the time-dependent tunneling current is derived showing explicit dependence of the overlap. The overlap proves to be necessary for a better quantitative description of the tunneling current, and our theory reproduces experimental results substantially better compared to standard approaches.\
PACS numbers: 72.10.Bg, 72.15.-v, 73.23.-b, 73.40.Gk, 73.63.-bk\
address: |
$^1$ Condensed Matter Theory group, Uppsala University, Box 530, 751 21 Uppsala, Sweden\
$^2$ Department of Physics, Linköping University, 581 83 Linköping, Sweden\
$^3$ Kirensky Institute of Physics, RAS, 660036 Krasnoyarsk, Russian Federation
author:
- 'J. Fransson$^1$, O. Eriksson$^1$ and I. Sandalov$^{1,2,3}$'
title: 'Effects of non-orthogonality in the time-dependent current through tunnel junctions'
---
24.0cm
\#1 \#1 \#1[|\#1]{} \#1[\#1|]{} \#1\#2 \#1[[d]{}\#1]{} \#1[[tr]{}\#1]{} \#1\#2[[diag]{}{\#1}\_[\#2]{}]{} \#1[\_[\#1]{}]{} \#1[\_[\#1]{}]{} \#1[a\_[\#1]{}]{} \#1[a\_[\#1]{}\^]{} c\#1[c\_[\#1]{}]{} \#1[c\_[\#1]{}\^]{} \#1[c\_[\#1]{}]{} \#1[c\_[\#1]{}\^]{} \#1[d\_[\#1]{}]{} \#1[d\_[\#1]{}\^]{} ł\#1[l\_[\#1]{}]{} \#1[r\_[\#1]{}]{} \#1[l\_[\#1]{}\^]{} \#1[r\_[\#1]{}\^]{} \#1[n\_[\#1]{}]{} \#1[n\_[\#1]{}]{}
Achievements in nano-materials science is expected to have importance in many scientific fields, including information technology, quantum computing and fuel cells. In particular, tunneling phenomena have been under focus recently, both in magnetic heterostructures as well as for quantum dot systems. The purpose of this paper is to develop an improved description of this phenomenon for general tunnel junctions, with possible application to the aforementioned scientific questions.
To focus the discussion, we mention that conductance measurements on extremely small metal-insulator-metal (MIM) junctions were carried out by Vullers *et al.*[@vullers2000] showing a non-linear conductance as a function of the bias voltage for low temperatures. The same behaviour has been reported for MIM double junctions[@nakayama1999] and Ti/TiO$_x$ tunneling barrier systems[@irmer1997; @matsumoto1996; @haraichi1997]. The non-linearity in the current-voltage $(J-V)$ characteristics appears for source-drain bias voltages larger than the spacing of the quasi one dimensional subbands since different numbers of subbands become available for transport in the forward and reverse directions [@glazman1989]. In the study by Simmons [@simmons1963] the current was found to depend non-linearly on the voltage, roughly as $V+\gamma V^3$.
Many theoretical studies of transport in nanostructures with tunneling barriers rely on the transfer Hamiltonian[@bardeen1961; @payne1986; @larkin1987; @meir1992; @jauho1994] which contains serious inconsistencies [@svidsinskii1982]. The principle of the transfer Hamiltonian is a division of the system into subsystems. This is motivated by the fact that the physical properties of the subsystems may be different and, hence, require different descriptions. Another motivation is that one is directly offered the possibility to generalize the approach to any number of tunneling barriers in the system. The transfer (tunneling) between the subsystems arises due to an overlap of the wave functions in the region of the barrier whereas the electron operators of the different subsystems are assumed to be anti-commuting. Qualitatively this may be motivated since the leakage of a wave function in one subsystem into the other is exponentially small. The $J-V$ characteristics given in this picture also shows a non-linear structure for large bias voltages. Quantitatively, though, the assumption of anti-commuting operators creates serious errors in the calculations of the current. This becomes particularly evident in the equilibrium situation displayed in Table \[table-equilibrium\], in which the four lowest states of a particle in a one dimensional hard walled box with a scattering potential are given. The energy levels are, as expected, reproduced within the non-orthogonal representation (NOR) with much higher accuracy than in the orthogonal representation (OR). Attempts that go beyond the transfer Hamiltonian have been made, *e.g.* by expanding the non-orthogonal states into a new Hilbert space [@emberly1998]. However, the proven success and physical transparency of the transfer Hamiltonian approach makes it desirable to extend its applicability to more general situations where the overlap is large, without making use of perturbation theory. This can indeed be achieved, which we demonstrate in this paper.
In order to overcome the inconsistencies with the transfer Hamiltonian formalism, we develop a theoretical approach for time-dependent transport through tunneling systems in which the overlap between the subsystems give explicit contribution to the current. Technically, we will express the properties of the original system in terms of the operators constructed of the wave functions of each subsystem. The resulting model structurally resembles the transfer Hamiltonian, although the physical interpretation is different. We have chosen the single barrier system simply to show the features of our approach. The main result of this paper is the Eqn. (\[eq-JIcurrent\]) for the time-dependent tunneling current through a single barrier. This expression is applied to a MIM junction in order to analyze the effect of overlap on the current. To our knowledge there does not exist any derivation nor analysis of time-dependent transport in tunneling junctions where the non-orthogonality is *not* disregarded.
Let us now proceed starting with the one particle Hamiltonian $$H=\frac{p^2}{2}+V,$$ where $V$ is any potential describing a system of two leads with an insulating layer in between. We introduce the two potentials $V_{\alpha},\ \alpha=L,R$, for the left $(L)$ and the right $(R)$ subsystem, respectively[@bardeen1961; @payne1986; @fran1999]. For instance, the left potential can be written as $V_L=V(x)\theta(-x+a_L)+V(a_L)\theta(x-a_L)$, where $\theta(x)$ is the Heaviside function and $a_L$ is a turning point for the left subsystem. In each subsystem there are orthonormal eigenstates $\{\phi_k,\leade{k}\}_{k\sigma\in\alpha}$ from which the corresponding field operator $\psi_{\alpha}(t,x)=\sum_{k\sigma\in\alpha}\c{k}(t)\phi_k(x)$ is constructed. Here $t$ is time and $x=(\bfr,\sigma)$ is a vector of the spatial coordinate $\bfr$ and the spin $\sigma$. Suppose that $\psi$ is the field operator of the system formed by the potential $V$. Then, this operator can be expanded in terms of $\psi_{\alpha}$ by the trivial identity $\psi(t,x)=\sum_{\alpha}\psi_{\alpha}(t,x)+[\psi(t,x)-\sum_{\alpha}\psi_{\alpha}(t,x)]$. Following reference[@sandalov1990] we project $\psi$ onto the subsystem $\alpha$ by $$\tilde{c}_{k\sigma}(t)=\int\phi_k^*(x)\psi(t,x)\dmu{x},$$ $k\in\alpha$, interpreted as the annihilation of a particle in the state $\phi_k$ with spin projection $\sigma$. Creation $\tilde{c}^{\dagger}_{k\sigma}$ of a particle in the state $\phi_k$ is defined similarly. These projections are possible to use directly for a second quantized form of the Hamiltonian. However, such an expansion gives an inconvenient expression of the Hamiltonian with the overlap matrix appearing explicitly. Thus, in order to proceed further, we define the operators $$\begin{array}{rcl}
\c{k}(t)&=&\sum_{k'}{{\cal O}^{-1}_{{k}{k'}}}\tilde{c}_{k'\sigma}(t),
\\
\cdagger{k}(t)&=&\sum_{k'}\left({{\cal O}^{-1}_{{k}{k'}}}\right)^*
\tilde{c}^{\dagger}_{k'\sigma}(t),
\end{array}
\label{eq-annihilator}$$ where $k'$ runs over all states in $L\cup R$ and ${{\cal O}^{-1}_{{k}{k'}}}$ is the element $kk'$ of the inverse of the overlap matrix of the wave functions $\phi_k,\ \phi_{k'}$ given by ${{\cal O}_{{k}{k'}}}=\inner{\phi_k}{\phi_{k'}}={{\cal O}_{{k'}{k}}}^*$. By a limitation to the case of spin conservation we can omit the spin indices in the overlap integral. The expectation value of the Hamiltonian in these operators is $${{\cal H}}=\int\psi^{\dagger}H\psi\dmu{x}=
{{\cal H}}_L+{{\cal H}}_R+{{\cal H}}_T,
\label{eq-hamiltonian1}$$ where we have defined ${{\cal H}}_{\alpha}=\int\psi_{\alpha}^{\dagger}H\psi_{\alpha}\dmu{x}$ and ${{\cal H}}_T=\sum_{\alpha\alpha'}(\int\psi_{\alpha}^{\dagger}H\psi_{\alpha'}\dmu{x}+H.c.)$. Here, we have neglected all expectation values that contain $\psi-\sum_{\alpha}\psi_{\alpha}$. Furthermore, we note that from the identity $V=V_{\alpha}+[V-V_{\alpha}],\
\alpha=L,R$, we find that the Hamiltonian of the lead ${{\cal H}}_{\alpha}=\sum_{k\sigma\in\alpha}\leade{k}\cdagger{k}\c{k}+\sum_{kk'\in\alpha}{\langle{\phi_k}|}(V-V_{\alpha}){|{\phi_{k'}}\rangle}\cdagger{k}\c{k'},\
\alpha=L,R$. The last term is a sum of terms proportional to the integral of $\phi_k^*\phi_{k'}$ over $(a_R,\infty)$ or $(-\infty,a_L)$ when $\alpha=L$ or $\alpha=R$, respectively, in which domains the wave functions are exponentially small. Thus, this term is negligible and we arrive at the appealing form of the Hamiltonian $$\begin{aligned}
{{\cal H}}= \sum_{p\sigma\in L}\leade{p}\cdagger{p}\c{p}+
\sum_{q\sigma\in R}\leade{q}\cdagger{q}\c{q}
+\sum_{pq\sigma}(v_{pq\sigma}\cdagger{p}\c{q}+H.c.),
\label{eq-hamiltonian}\end{aligned}$$ where $v_{pq\sigma}={\langle{\phi_p}|}H{|{\phi_q}\rangle}$ is the mixing matrix element. The structure of the Hamiltonian (\[eq-hamiltonian\]) very much resembles the usual transfer Hamiltonian. Nevertheless, the meaning of the electron operators $\cdagger{k},\c{k}$ is altered, now carrying information of the full system rather than just of its subsystem. This fact is legible from the anti-commutation relation ${\{{\c{k}},{\cdagger{k'}}\}}={{\cal O}^{-1}_{{k}{k'}}}$. Indeed, when ${{\cal O}^{-1}_{{k}{k'}}}\rightarrow\delta_{kk'}$ we recover the transfer Hamiltonian with the usual interpretation of the operators $\c{k}$. In this sense we conclude that the Eq. (\[eq-hamiltonian\]) generalizes the conventional transfer Hamiltonian.
The expression in Eqn. (\[eq-hamiltonian\]) is derived for the system in equilibrium. It is straight forward applicable to the non-equilibrium case by letting $\leade{k}\rightarrow\leade{k}(t)$ and $v_{pq\sigma}\rightarrow v_{pq\sigma}(t)$. For definiteness we derive an expression for the current flowing through the barrier from the left to right. The tunneling current through the barrier separating the leads is expressed as the rate of change of the number of particles on, say, the left side of the junction ${\langle{N_L(t)}\rangle}=\sum_{p\sigma}{\langle{n_{p\sigma}(t)}\rangle}$, where ${\langle{n_{p\sigma}(t)}\rangle}={\langle{\cdagger{p}(t)\c{p}(t)}\rangle}$. The time development of ${\langle{n_{p\sigma}}\rangle}$ is given by the Heisenberg equation of motion yielding the tunneling current for each spin projection $\sigma$ $$\begin{aligned}
J_{\sigma}(t)&=&
2e\Im\sum_{pq}\biggl(
V^*_{pq\sigma}(t){\langle{\cdagger{q}(t)\c{p}(t)}\rangle}+
v^*_{pq\sigma}(t){{\cal O}^{-1}_{{p}{q}}}{\langle{\cdagger{p}(t)\c{p}(t)}\rangle}
\biggr)=
\nonumber\\
&=&-2e\Re\sum_{pq}\biggl(
V^*_{pq\sigma}(t)F^<_{pq\sigma}(t,t)-
v^*_{pq\sigma}(t){{\cal O}^{-1}_{{p}{q}}}g^<_{p\sigma}(t,t)\biggr),
\label{eq-dynamicnumber}\end{aligned}$$ with the coefficients $V_{pq\sigma}=v_{pq\sigma}+{{\cal O}^{-1}_{{p}{q}}}\leade{q}$ describing the tunneling. In Eq. (\[eq-dynamicnumber\]) we have identified the correlation function ${\langle{\cdagger{q}\c{p}}\rangle}$ with the *lesser* Green function $F^<_{pq\sigma}(t,t)=i{\langle{\cdagger{q}(t)\c{p}(t)}\rangle}$. This propagator is calculated within the non-equilibrium technique of Kadanoff and Baym [@kada62] for the Green function $F_{pq\sigma}(t,t')={(-i)\langle {\rm T}{\c{p}(t)}{\cdagger{q}(t')}\rangle}$. From the equation of motion for $F_{pq\sigma}(t,t')$ we obtain $$\begin{aligned}
F_{pq\sigma}(t,t')&=&g_{p\sigma}(t,t'){{\cal O}^{-1}_{{p}{q}}}+
\int_0^{-i\beta}g_{p\sigma}(t,t_1)
V_{pq\sigma}(t_1)g_{q\sigma}(t_1,t')\dmu{t_1},
\label{eq-transferGF}\end{aligned}$$ where $g_{k\sigma}=F_{kk\sigma}$ is the conduction electron Green function satisfying the equation $(i{\partial/\partial t}-\leade{k})g_{k\sigma}(t,t')=\delta(t-t')$. The contour integration in Eq. (\[eq-transferGF\]) is brought to real time integration by the Langreth analytical continuation rules [@langreth1976], thus $$\begin{aligned}
F_{pq\sigma}^<(t,t')=g^<_{p\sigma}(t,t'){{\cal O}^{-1}_{{p}{q}}}+
\int_{-\infty}^{\infty}V_{pq\sigma}(t_1)
\biggl[g_{p\sigma}^r(t,t_1)g_{q\sigma}^<(t_1,t')+
g_{p\sigma}^<(t,t_1)g_{q\sigma}^a(t_1,t')\biggr]\dmu{t_1}.\end{aligned}$$ The lesser, retarded and advanced expressions of the conduction electron GF are $$\begin{aligned}
g^<_{k\sigma}(t,t')&=&if_{\alpha}(\leade{k})
e^{-i\int_{t'}^t\leade{k}(t_1)\dmu{t_1}},\\
g^{r,a}_{k\sigma}(t,t')&=&
\mp i\theta(\pm t\mp t')e^{-i\int_{t'}^t\leade{k}(t_1)\dmu{t_1}},\end{aligned}$$ respectively, where $f_{\alpha}(x)$ is the Fermi-Dirac distribution function. Before we continue the derivation we rewrite the electron operators in terms of current states, *i.e.* $\cdagger{k}(t)=\cdagger{k}\exp{[i\mu_{\alpha}(t)]}$, and $\c{k}(t)=\c{k}\exp{[-i\mu_{\alpha}(t)]}$. This will explicitly show the applied voltage dependence $V(t)$ of the current, since $\mu_L(t)-\mu_R(t)=eV(t)$. Replacing the summation over $p$ and $q$ in Eq. (\[eq-dynamicnumber\]) by energy integration in terms of the density of states $\rho_{\sigma}(\dote{\alpha})$ and noting that $\Re[(V_{pq\sigma}^*-v_{pq\sigma}^*){{\cal O}^{-1}_{{p}{q}}}g^<_{p\sigma}]=0$, the time-dependent tunneling current becomes $$\begin{aligned}
J_{\sigma}(t)&=&-2e\Re\int V_{LR\sigma}^*(t)
\rho_{\sigma}(\dote{L})\rho_{\sigma}(\dote{R})
\nonumber\\
&& \times\int_{-\infty}^t
V_{LR\sigma}(t_1)[f(\dote{R})-f(\dote{L})]
e^{-i\int_{t_1}^t(eV(t_2)+(\dote{L}-\dote{R}))\dmu{t_2}}
\dmu{t_1}\dmu{\dote{L}}\dmu{\dote{R}}.
\label{eq-JIcurrent}\end{aligned}$$ The mixing and the overlap are here replaced by the functions $V_{LR\sigma}(t)\equiv V_{\sigma}(\dote{L},\dote{R},t)$ and ${{\cal O}^{-1}_{{L}{R}}}\equiv{{\cal O}^{-1}_{{}{}}}(\dote{L},\dote{R})$, respectively, satisfying $V_{\sigma}(\leade{p},\leade{q},t)=V_{pq\sigma}(t)$ and ${{\cal O}^{-1}_{{}{}}}(\leade{p},\leade{q})={{\cal O}^{-1}_{{p}{q}}}$. The formula (\[eq-JIcurrent\]) reproduces results based on the transfer Hamiltonian in the limit of orthogonal subsystems, *i.e.* when ${{\cal O}^{-1}_{{k}{k'}}}\rightarrow\delta_{kk'}$. It is important to note the fact that the tunneling coefficient $V_{LR\sigma}=v_{LR\sigma}+{{\cal O}^{-1}_{{L}{R}}}\leade{R}$ in our formulation, explicitly depends on the energies of the electrons involved in the conduction process.
When $V(t)=V$ and a stationary current is established through the barrier the Eq. (\[eq-JIcurrent\]) reduces to $$\begin{aligned}
J_{\sigma}=2e\frac{\pi}{4W^2}\int_{-W}^{W}
|V_{LR\sigma}|^2[f(\dote{}-eV)-f(\dote{})]
\dmu{\dote{}}.
\label{eq-JIstationarycurrent}\end{aligned}$$ This expression is given by assuming a constant density of states $\rho_{\sigma}(\dote{\alpha})=1/2W$, where $2W$ is the conduction band width, and slowly varying mixing and overlap so that their respective values can be taken at the chemical potential, which are reasonable conditions for MIM-junctions. In order to compare our theory with a realistic example we show in Fig. \[fig-JIstationarycurrent\] the experimental $J-V$ characteristics from Ref. [@haraichi1997] (solid-dotted line) together with that of Eqn. (\[eq-JIstationarycurrent\]) in both the non-orthogonal (solid line) and orthogonal (dashed line) representations. We have also included the corresponding result given by Simmons formula (dotted line) [@simmons1963]. Note that Simmons formula and the orthogonal representation correspond to the standard methods used to calculate transport. From the figure, it stands clear that inclusion of the overlap contributes significantly to the behaviour of the $J-V$ characteristics and the quantitative agreement is remarkably improved. The increase in the agreement with the experiments lies not only in the low voltage regime but also in that the current rises rapidly at a certain threshold voltage, which influences the time-dependent current. For a $6\ \%$ increase in the barrier width our calculation (bold dash-dotted line) agrees exactly with the experimental results for positive voltages. The remaining discrepancy from the experimental curve, *e.g.* the observed asymmetry, is believed to stem from the lack of electron interactions in our model, for example charging effects. Moreover, in the simple calculations presented here we have merely computed the wave functions $\phi_{\mu_L}$ and $\phi_{\mu_R}$, normalized to a unit probability flow [@landau1977] at their asymptotic distances from the barrier $x\rightarrow-\infty$ and $x\rightarrow\infty$. For simplicity we have used a rectangular potential barrier leading.
In conclusion, we have developed a simple and transparent theoretical approach for time-dependent tunneling current through nanostructures which has a far wider applicability compared to standard methods. The ability of dividing the system into several subsystems, which then can be treated individually, is preserved without loss of accuracy when the inclusion of the overlap of the subsystems is allowed and all attractive features of the transfer Hamiltonian approach can be kept. The non-orthogonality is reflected in the non-zero anti-commutation relations of the electron operators of different subsystems. A formula for the time-dependent tunneling current through a single barrier structure, Eqn. (\[eq-JIcurrent\]), has been derived, which shows the necessity of including the overlap for a substantially better quantitative agreement with experiments. We also note that the formalism simply generalizes to the case of a two, or multiple, barrier structure. In particular, the region between the barriers can be interacting, for example a quantum dot. Then, a generalization to any number of contact leads is straight forward.
J.F. wants to thank U. Lundin for helpful and encouraging discussions. Support from the Göran Gustafsson foundation, the Swedish national science foundation (NFR and TRF) and the Swedish foundation for strategic research (SSF) are acknowledged.
[5]{} R.J.M. Vullers *et al.*, Appl. Phys. Lett. [**76**]{}, 1947 (2000). M. Nakayama *et al.*, Jap. J. Appl. Phys. [**38**]{}, 7151 (1999). B. Irmer *et al.*, Appl. Phys. Lett. [**71**]{}, 1733 (1997). K. Mastumoto *et al.*, Appl. Phys. Lett. [**68**]{}, 34 (1996). S. Haraichi *et al.*, J. Vac. Sci. Technol. B [**15**]{}, 1406 (1997). L.I. Glazman and A.V. Khaetskii, Europhys. Lett. [**9**]{}, 263 (1989). J.G. Simmons, J. Appl. Phys. [**34**]{}, 1793 (1963), *ibid.* [**34**]{}, 1828 (1963). J. Bardeen, Phys. Rev. Lett. [**6**]{}, 57 (1961). M.C. Payne, J. Phys. C [**19**]{}, 1145 (1986). A.I. Larkin and K.A. Matveev, Sov. Phys. JETP [**66**]{}, 580 (1987). Y. Meir and N.S. Wingreen, Phys. Rev. Lett. [**68**]{}, 2512 (1992). A-P Jauho *et al.*, Phys. Rev. B [**50**]{}, 5528 (1994). A.V. Svidisinskii, *Space-non-homogeneous Problems in Superconductivity* (Nauka, Moscow, 1982) *in russian*. E. Emberly and G. Kirczenow, Phys. Rev. Lett. [**81**]{}, 5205 (1998). J. Fransson, O. Eriksson, B. Johansson and I. Sandalov, Physica B [**272**]{}, 28 (1999). I. Sandalov and V.I. Filatjev, Physica B [**162**]{}, 139 (1990). L.P. Kadanoff and G. Baym, *Quantum Statistical Mechanics* (Benjamin, New York, 1962). D.C. Langreth, in *Linear and nonlinear electron transport in solids*, Vol. 17 of *Nato Advanced Study Institute, Series B: Physics*, edited by J.T. Devreese and V.E. van Doren (Plenum, New York, 1976). The experimental data of the barrier’s width and height given by Haraichi [*et al.* ]{}[@haraichi1997] are calculated from $\beta(V+\gamma V^3)$ , which is only valid for low and intermediate voltages. In the Fig \[fig-JIstationarycurrent\], we have used the more precise $J_0[(\phi-eV/2)\exp{(-2s\sqrt{2\phi-eV})}-(\phi+eV/2)\exp{(-2s\sqrt{2\phi+eV})}]$. The two formulae for the current are found in Ref. [@simmons1963]. L.D. Landau and E.M. Lifshitz *Quantum Mechanics* 3 ed. (Pergamon Press 1977).
exact NOR OR
-- --------- --------- --------- --
20.265 20.266 18.866
27.781 27.862 27.342
83.868 83.592 79.383
111.088 113.793 107.176
: The four lowest energy levels of a 37 nm long hard walled box with a 5.3 nm wide and 178 meV high scattering potential located in the middle of the box. The energies (meV) are computed exact, with the overlap matrix taken into account (NOR) and ignored (OR).
\[table-equilibrium\]
![The $J-V$ characteristics of a 1.46 nm wide and 1.85 eV high MIM junction (the height measured from the equilibrium chemical potential) \[19\]. The experimental results by Haraichi *et al.* \[5\] (solid-dotted) is compared with the computations within the NOR (solid), NOR with a 6 $\%$ increase of the width (dash-dotted), OR (dashed) and Simmons formula (dotted) \[7\]. The equilibrium chemical potential is 1.75 eV and the conduction band width is $2W=40$ eV.[]{data-label="fig-JIstationarycurrent"}](Fig.eps){width="8cm"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We investigate the time evolution of the mass distribution of pre-stellar cores (PSCs) and their transition to the initial stellar mass function (IMF) in the central parts of a molecular cloud (MC) under the assumption that the coalescence of cores is important. Our aim is to explain the observed shallow IMF in dense stellar clusters such as the Arches cluster. The initial distributions of PSCs at various distances from the MC center are those of gravitationally unstable cores resulting from the gravo-turbulent fragmentation of the MC. As time evolves, there is a competition between the PSCs rates of coalescence and collapse. Whenever the local rate of collapse is larger than the rate of coalescence in a given mass bin, cores are collapsed into stars. With appropriate parameters, we find that the coalescence-collapse model reproduces very well all the observed characteristics of the Arches stellar cluster IMF; Namely, the slopes at high and low mass ends and the peculiar bump observed at $\sim 5-6$ $M_{\odot}$. Our results suggest that today’s IMF of the Arches cluster is very similar to the primordial one and is prior to the dynamical effects of mass segregation becoming important.'
author:
- |
Sami Dib$^{1}$[^1], Jongsoo Kim$^{1}$, and Mohsen Shadmehri$^{2,3}$\
$^{1}$Korea Astronomy and Space Science Institute, 61-1, Hwaam-dong, Yuseong-gu, Daejeon 305-348, Korea\
$^{2}$School of Mathematical Sciences, Dublin City University, Glasnevin, Dublin 9, Ireland\
$^{3}$Department of Physics, School of Science, Ferdowsi University, Mashad, Iran
date: 'Accepted XXX. Received XXX'
title: The origin of the Arches stellar cluster mass function
---
\[firstpage\]
galaxies: star clusters - Galaxy: centre - Turbulence - ISM: clouds - open clusters and associations:individual: Arches
MOTIVATION
==========
Understanding the origin of the initial stellar mass function (IMF) remains one of the most challenging issues in modern astrophysics. When averaged over the total volume of galaxies or whole stellar clusters, the IMF is observed to follow a nearly uniform behavior which consists in an increased number of stars counted when going from the most massive stars up to $\sim 0.5$ $M_{\odot}$, followed by a shallower increase between $\sim 0.5$ and $\sim 0.1$ $M_{\odot}$ and a decline in the number of stars at masses $\lesssim 0.1$ $M_\odot$. This standard IMF has been described, with continuous refinements, by several analytical functions (e.g., Salpeter 1955; Miller-Scalo 1979; Kroupa 2002; Chabrier 2003). Yet, deviations from the standard IMF at low and high mass ends have been reported in many observations (see review in Elmegreen 2004). At high mass, the IMF is observed to be generally top-heavy in dense cluster cores such as in the Arches cluster (e.g., Stolte et al. 2005; Kim et al. 2006) and stars appear to be, preferentially located in the central parts of the clusters (e.g., Hillenbrand & Hartmann 1998; Figer et al. 1999; Stolte 2002; Gouliermis et al. 2004). Star-bursts regions are also observed to possess a top-heavy IMF, either in the form of a shallow slope at high mass (e.g., Einsenhauer et al. 1998; Sternberg 1998) or by having a value of the high mass-low mass turnover of a few to several $M_\odot$ which is substantially larger than that of the standard IMF (e.g., Rieke 1993). The IMF of dense clusters seems also to be truncated at the very high mass end (e.g., Stolte 2005).
The mass truncation can be attributed to the short lifetimes of the most massive stars. Ideas that have been proposed to explain the shallowness of the slope at the high mass end include a) a model based on the coalescence of pre-stellar cores (PSCs) and their subsequent gravitational collapse to produce stars (e.g., Nakano 1966; Silk & Takahashi 1979; Elmegreen & Shadmehri 2003; Elmegreen 2004; Shadmehri 2004), b) the mass segregation of stars in the cluster (e.g., Vesperini & Heggie 1997; Kroupa 2002; Mouri & Taniguchi 2002), and c) A renewed episode of gas accretion by the cluster under favorable conditions, which leads to the formation of a new generation of massive stars (e.g., Lin & Murray 2007). This latter idea is somehow inconsistent with the fact that a cluster such as the Arches cluster is overall very young (i.e, age $\sim 2 \pm 1$ Myrs) and may apply only to older clusters. Concerning mass segregation, whereas there is little doubt that the enhancement in the numbers of massive stars in the inner parts of the cluster by dynamical processes will lead to a shallower IMF, this does not constitute a direct proof that the primordial IMF of stars in those regions was not shallower than a Salpeter IMF initially. The latter is commonly used as an initial input for the stellar distribution functions at all cluster radii in N-Body models (e.g., Portegies Zwart et al. 2007). Furthermore, the IMF of the Arches cluster is characterized by a peculiar bump at $\sim 6$ $M_{\odot}$ which is not, to date, well reproduced by the effect of mass segregation in N-body simulations (e.g., Kim et al. 2006, Portegies Zwart et al. 2007).
In this letter, we propose a coalescence model in which the local initial PSCs populations are those resulting from the local gravo-turbulent fragmentation of the protocluster cloud. We follow the time evolution of the mass function of PSCs and the transition to the IMF under the assumption that the coalescence of PSCs is important. This is very likely to be the case for the PSCs located in the central parts of the protocluster cloud.
THE COALESCENCE MODEL {#model}
=====================
We consider PSCs (e.g., André et al. 2000) embedded in an isothermal MC (at a temperature of $T=10$ K), at different locations $r$ from the cloud’s center. We assume that both the PSCs and the MC are axisymmetric (PSCs are initially spherical but are likely to quickly flatten as time evolves). The radial density profile of the MC is given by:
$$\rho_{c}(r)= \frac{\rho_{c0}}{1+(r/R_{c0})^{2}},
\label{eq1}$$
where $\rho_{c0}$ and $R_{c0}$ are the cloud’s central density and core radius, respectively. The central density, $\rho_{c0}$, is given by:
$$\rho_{c0}= \frac{M_{c}}{4 \pi R_{c0}^{3} [(R_{c}/R_{c0})-arctan(R_{c}/R_{c0})]},
\label{eq2}$$
where $M_{c}$ is the mass of the cloud and $R_{c}$ its radius. The density profiles of PSCs are assumed to follow the formula given by Whitworth & Ward-Thompson (2001):
$$\rho_{p}(r_{p})= \frac{\rho_{p0}}{[1+(r_{p}/R_{p0})^{2}]^{2}},
\label{eq3}$$
where $\rho_{p0}$ and $R_{p0}$ are the central density and core radius of the PSC, respectively. The radius $R_{p}$ of the the PSC depends both on its mass and on its position within the MC. The dependence of $R_{p}$ on $r$ requires that the density at the edges of the PSC equals the ambient cloud density, i.e., $\rho_{p}(R_{p})=\rho_{c}(r)$. This would result in smaller radii for PSCs of a given mass when they are located in their inner parts of the cloud. The density contrast between the edge of the PSC and its center is given by
$${\cal C}(r) = \frac {\rho_{p0}}{\rho_{c} (r)}=\frac {\rho_{p0}} {\rho_{c0}} \left(1+ \frac{r^{2}}{R^{2}_{c0}} \right).
\label{eq4}$$
Depending on its position $r$ in the cloud, the radius of the PSC of mass $M_{p}$, $R_{p}$, can be calculated as being $R_{p}(r,M_{p})=a(r) R_{p0} (r,M_{p})$, where:
$$R_{p0}(r,M_{p})= \left(\frac{M_{p}}{2 \pi \rho_{p0}} \right)^{1/3} \left(arctan[a(r)]-\frac{a(r)}{1+a(r)^{2}} \right)^{-1/3},
\label{eq5}$$
and with $a(r)=({\cal C}(r)^{1/2}-1)^{1/2}$. With our set of parameters, the quantity ${\cal C}^{1/2}-1$ is always guaranteed to be positive. The value $R_p(r,M)$ can be considered as being the radius of the PSC at the moment of its formation. However, the radius of the PSC will decrease as time advances due to gravitational contraction. The PSC contracts on a timescale, $t_{cont,p}$ which is equal to a few times its free fall timescale, and can be parametrized as:
$$t_{cont,p}(r,M)= \nu ~ t_{ff}(r,M)= \nu \left( \frac {3 \pi} {32~G \bar{\rho_{p}} (r,M)} \right)^{1/2},
\label{eq6}$$
where $\nu \ge 1$ and $\bar{\rho_{p}}$ is the radially averaged density of the PSC of mass $M_{p}$, located at position $r$ in the cloud, and which is calculated as being:
$$\bar{\rho_{p}}(r,M_{p})=\frac{1}{R_{p}(r,M_{p})}~\int^{R_{p}(r,M_{p})}_{0}\frac{\rho_{p0}}{[1+(r_{p}/R_{p0})^{2}]^{2}} dr_{p},
\label{eq7}$$
Thus, the time evolution of the radius of the PSC can be described by the following equation:
$$R_{p}(r,M,t)=R_p(r,M)~e^{-(t/t_{cont,p})}.
\label{eq8}$$
Once the instantaneous radius of a PSC of mass $M_{p}$, located at position $r$ form the cloud’s center is defined, it becomes possible to calculate its cross section for collision with PSCs of different masses. The cross section for the collision of a PSC of mass $M_{i}$ and radius $R_{i}$ with another of mass $M_{j}$ and radius $R_{j}$ and which accounts for the effect of gravitational focusing is given by:
$$\begin{aligned}
\sigma(M_{i},M_{j},r,t) = \pi \left[R_{p,i}(r,M_{i},t)+R_{p,j}(r,M_{j},t)\right]^{2} \nonumber \\
\times \left[ 1+\frac{2G (M_{i}+M_{j})} {2 v^{2} (R_{p,i}(r,M_{i},t)+R_{p,j}(r,M_{j},t))} \right].
\label{eq9}\end{aligned}$$
Elmegreen & Shadmehri (2003) and Shadmehri (2004) assumed that the collision velocity between PSCs is equal to the virialized velocity dispersion inside the MC. This might be a plausible hypothesis if MCs were indeed the dissipative structures of turbulence in the interstellar medium. It is however unlikely to be the case. Numerical simulations (e.g., Dib et al. 2007) show that clumps and cores in MCs are not in virial equilibrium. In this work, we assume that the relative collision velocity between the PSCs follows the local gas dynamics at their position in the cloud (this remains a simplification as in reality PSCs motions can be decoupled from that of the local ambient gas) according to a Larson type relation $v(r)=v_{0} r^{\alpha}$ (Larson 1981; $v_{0}=1.1$ km s$^{-1}$), with a lower limit being the local thermal sound speed, which is uniform across the isothermal MC.
INITIAL CONDITIONS
==================
As initial conditions for the PSCs mass distribution at different cloud radii, we adopt distributions that are the result of the gravo-turbulent fragmentation of the cloud, following the formulation given in Padoan et al. (1997) and Padoan & Nordlund (2002). In these models, the probability function of the density field is well represented by a log-normal function:
$$P(ln~x) d~ln~x=\frac{1}{\sqrt{2 \pi \sigma_{d}}} exp \left[-\frac{1}{2} \left(\frac{ln~x-\bar{ln~x}}{\sigma_{d}} \right)^{2} \right] d~ln~x,
\label{eq10}$$
where $x$ is the number density normalized by the average number density, $x=n/\bar{n}$. The standard deviation of the density distribution $\sigma_{d}$ and the mean value $\bar {ln~x}$ are functions of the thermal rms Mach number, $\cal M$: $\bar{ln x}=-\sigma^{2}_{d}/2$ and $\sigma^{2}_{d}=ln(1+{\cal M}^{2} \gamma^{2})$. Padoan & Nordlund (2002) suggest a value of $\gamma \sim 0.5$. A second step in this approach is to determine the mass distribution of dense cores. Padoan & Nordlund (2002) showed that by making the assumptions that: (a) the power spectrum of turbulence is a power law and, (b) the typical size of a dense core scales as the thickness of the postschock gas layer, the cores mass spectrum is given by:
$$N(M)~d~log~M \propto M^{-3/(4-\beta)} d~log~M,
\label{eq11}$$
where $\beta$ is the exponent of the kinetic energy power spectrum, $E_{k} \propto k^{-\beta}$, and is related to the exponent $\alpha$ of the size-velocity dispersion relation in the cloud with $\beta=2 \alpha+1$. However, Eq. \[eq11\] can not be directly used to estimate the number of cores that are prone to star formation. It must be multiplied by the local distribution of Jeans masses. At constant temperature, this distribution can be written as:
$$P(M_{J})~dM_{J}=\frac{2~M_{J0}^{2}}{\sqrt{2 \pi \sigma^{2}_{d}}} M^{-3}_{J} exp \left[-\frac{1}{2} \left(\frac{ln~M_{J}-A}{\sigma_{d}} \right)^{2} \right] dM_{J},
\label{eq12}$$
where $M_{J0}$ is the Jeans mass at the mean density $\bar{n}$. Thus, Eq. \[eq11\] becomes, locally:
$$N (r,M)~d~log~M =f_{0}(r)~M^{-3/(4-\beta)} \left[\int^{m}_{0} P(M_{J}) dM_{J}\right]d~log~M.
\label{eq13}$$
The local normalization coefficient $f_{0}(r)$ is obtained by requiring that $\int^{M_{max}}_{M_{min}} N (r,M)~dM=1$ in the shell of width $dr$ located at distance $r$ from the cloud’s center. Then, the local distribution of cores at time $t=0$, $N(r,M,0)$, is obtained by multiplying the local normalized function $N(r,M)$ by the local rate of fragmentation such that:
$$N(r,M,0)=\frac{\epsilon_{c}(r) \rho_{c}(r)} {<M>(r)~t_{cont,p} (r,M)} N(r,M),
\label{eq14}$$
where $<M>$ is the average core mass in the local distribution and is calculated by $<M>=\int_{M_{min}}^{M_{max}} M~N(r,M)~dM$ and $\epsilon_{c}$ is a parameter smaller than unity which describes the local mass fraction of gas that is present in the dense PSCs. In principle, $\epsilon_{c}$ will have a radial and probably outwardly decreasing dependence. For simplicity we shall assume $\epsilon_{c}$ to be a constant independent of radius. As our comparisons with the observations will be focused on the inner parts of the protocluster cloud which will be transformed into a stellar cluster (i.e., the Arches cluster), it is likely that these regions will be characterized by a uniform mass fraction of the dense gas.
Fig. \[fig1\] displays the local mass spectrum of Jeans unstable PSCs in rings of width 0.025 pc, obtained with Eq. \[eq14\], located at different distances from the cloud’s center (top), as well as the cumulative number of PSCs in each mass bin in regions of the protocluster cloud located between \[$0,R_{c0}$\], \[$0,2~R_{c0}$\], and \[$R_{c0},2~R_{c0}$\].
FROM THE PRE-STELLAR CORES MASS FUNCTION TO THE PRIMORDIAL IMF
==============================================================
{height="12cm" width="17.5cm"}
With the initial conditions described in §. 3, we follow the time evolution of the PSCs mass spectrum by solving the following integro-differential equation of $N(r,M,t)$:
$$\begin{aligned}
\frac{dN(r,M,t)}{dt}=0.5\times \eta(r) \times \nonumber \\
\int^{\Delta M}_{M_{min}}~N(r,m,t)~N(r,M-m,t)~\sigma(m,M-m,r,t)~v(r)~dm \nonumber \\
-\eta (r) N(r,M,t) \int^{M_{max}}_{M_{min}} N(r,m,t) \sigma(m,M,r,t) v(r)~dm
\label{eq15}\end{aligned}$$
where the first and second terms in the right hand side of Eq. \[eq15\] correspond to the rate of creation and destruction of a PSC of mass $M$, at location $r$, respectively (Nakano 1966; Shadmehri 2004). In Eq. \[eq15\], $\Delta M=M-M_{min}$, and $\eta (r)$ is a coefficient which represents the coalescence efficiency, with $\eta \leq 1 $. This efficiency can be the result of various physical processes which can affect the coalescence of PSCs, such as if the merger of cores occurs preferentially parallel or perpendicular to the local magnetic field lines, and is likely to have a radial dependence. For simplicity, we shall assume that $\eta$ is independent of position. In order to evaluate the transition from PSCs to stars, we compare, at each timestep, the local coalescence timescale to the local contraction timescale for PSCs of a given mass. The local coalescence timescale is $t_{coal}(r,M)=1/w_{coal}(r,M)$ where $w_{coal}$ is the coalescence rate (Elmegreen & Shadmehri 2003):
$$w_{coal}(r,M)=\frac{2^{1/2} v(r)}{V_{shell}(r)} \sum_{j=1}^{mbin} (R_{i}+R_{j})^{2} \left[1+\frac{2 G (M_{i}+M_{j})}{2 v^{2} (R_{i}+R_{j})} \right],
\label{eq16}$$
where $mbin$ is the number of mass bins, and $V_{shell}$ is the volume of the shell of width $dr$ located at distance $r$ from the MC’s center. The contraction timescale is given by Eq. \[eq8\]. Whenever the local contraction timescale is shorter than the local coalescence timescale, PSCs are collapsed into stars. When a PSC collapses to form a star, we assume that a fraction of its mass is re-injected into the protocluster cloud in the form of an outflow. We account for this mass loss in a purely phenomenological way by assuming that the mass of a star which is formed out of a PSC of mass $M_{p}$ is given by M$_{\star}$=$\psi$ M$_{p}$, where $\psi \le 1$. Matzner & McKee (2000) showed that $\psi$ can vary between $0.25-0.7$ for stars in the mass range $0.5-2$ $M_{\odot}$. There is no evidence so far, for or against, whether this result holds at higher masses. However, the similarity between the IMF and the dense cores mass function observed by Alves et al. (2007) in the Pipe Nebula might be an indication of a constant $\psi$ across the mass spectrum (i.e., in their case it is $\psi \sim 1/3$). Here also, we shall assume that a similar fraction of the mass of a PSC is lost in the outflow independent of its mass.
The algorithm was tested by performing runs with $\eta=0$ (i.e., no-coalescence) and $\eta=0.001$ (i.e., inefficient coalescence) and with the other parameters fixed at $M_{c}=5 \times 10^{5}$ $M_{\odot}$, $R_{c}=5$ pc, $R_{c0}=0.2$ pc, $\rho_{p0}=10^{7}$ cm$^{-3}$, $\epsilon=0.5$, $\alpha=0.37$, $\nu=10$, and $\psi=0.58$. As expected, for $\eta=0$, the resulting stellar mass spectrum after the PSCs collapse into stars is similar to the initial cumulative PSCs spectrum, and only slightly modified for $\eta=0.001$. Models were performed with permutations over the parameters $\eta$ and $\nu$ fixing the other parameters to the above stated values. It should be stressed at this stage that our semi-analytical modeling is not aimed at recovering the initial characteristics of the Arches protocluster cloud, but rather at showing whether or not, the Archer cluster IMF can be generated by the coalescence of PSCs and their subsequent collapse into stars.
Fig. \[fig2\] displays the time evolution of the cumulative PSCs populations in the region \[$R_{c0}-2~R_{c0}$\]=\[0.2-0.4\] pc, which corresponds to the annulus between $\sim$ 1-2 core radii of the Arches cluster for a model with $\eta=0.5$ and $\nu=10$. In the initial stages, the most massive PSCs, who have a larger cross section, coalesce faster than the less massive ones, essentially by capturing the numerous intermediate mass PSCs and causing a rapid flattening of the spectrum at the high mass end. By $t \sim 0.07~t_{ff,c}$ ($t_{ff,c}=(3 \pi/32 G \bar{\rho_{c}})^{1/2} \sim 3 \times 10^{4}$ yr is the MC free fall timescale), a first generation of the smallest PSCs collapses to form stars. As time advances, more massive stars are formed in the shell (massive cores collapse later because of their lower average density) and in parallel the PSCs population decreases. By $t \sim 0.1~t_{ff,c}$ the intermediate mass PSCs which constitutes the largest mass reservoir for coalescence collapse into stars. At this time, the turnover in the PSCs mass spectrum is located at $\sim 8-10$ $M_{\odot}$. Since the reservoir of intermediate mass objects is depleted, the remaining massive PSCs coalesce at a slower pace before they collapse. By $t \sim 0.25~t_{ff,c}$, all PSCs of different masses in the shell have collapsed and formed stars. Because of mass loss, the stellar IMF is shifted to lower masses (bump shifted to $\sim 5-6$ $M_{\odot}$). In summary, the resulting IMF is not very different from the PSCs mass spectrum after the initial and rapid stage of strong coalescence until $t \sim 0.01~t_{ff,c}$. This is due to the fact that low and intermediate mass PSCs collapse at early stages, thus depleting the reservoir of objects with which the massive PSCs can continue to coalesce, in addition to their own contraction. Both effects reduce the massive PSCs ulterior merger rate. Overall, the stellar mass spectrum is formed very quickly, on a timescale which is of the order of the contraction timescale of the most massive cores i.e., $\sim 0.25~t_{ff,c}$.
In Fig. \[fig2\], over-plotted to our result is the cumulative mass spectrum of the Arches cluster in the annulus of \[0.2-0.4\] pc (Kim et al. 2006). The coalescence-collapse model agrees better with the observations than models based on mass segregation by dynamical friction. In particular, the bump at $\sim 5-6$ $M_{\odot}$ is reproduced. A fit to the stellar spectrum yields slopes of $-2.04 \pm 0.02$ and $-1.72 \pm 0.01$ in the mass ranges of \[$1-3$\] $M_{\odot}$ and $\ge 15$ $M_{\odot}$, respectively, in very good agreement with observational values.
We also performed additional runs where the maximum mass in the PSC spectrum was set to 250 $M_{\odot}$ (instead of 100 $M_{\odot}$). In this case, the resulting slope of the IMF in the low and high mass regimes are shallower than the Salpeter IMF, yet shallower than those of the Arches IMF. The reason is that PSCs with masses larger than $100$ $M_{\odot}$ will form quickly from the coalescence of lower mass ones, and the number of PSCs of masses $\gtrsim 100$ $M_{\odot}$ will grow at an even faster pace as their cross sections are very large. The mismatch in this case with the Arches IMF might be an indication that PSCs with masses $\ge 100$ $M_\odot$, if they form, might undergo a certain amount of sub-fragmentation.
SUMMARY
=======
In this work, we use semi-analytical modeling to study the evolution of the pre-stellar cores (PSCs) mass spectrum and its transition to the stellar initial mass function (IMF) at different locations in a molecular cloud (MC) under the assumption that the coalescence of PSCs is important. The aim is to reproduce the observed IMF in the inner regions of dense stellar clusters such as the Arches cluster (Kim et al. 2006). The initial conditions for the local populations of PSCs are those of Jeans unstable cores resulting from the gravo-turbulent fragmentation of the MC. PSCs of a given mass are transformed into stars whenever their local rate of contraction is higher than their rate of coalescence. With appropriate, yet very realistic parameters, we are able to reproduce all of the observed characteristics of the IMF of the Arches cluster. Namely, the slopes at the high and low mass ends, and the peculiar bump observed at $\sim 5-6$ $M_{\odot}$. Our results suggest that today’s IMF of the Arches cluster is primordial. This might be a common property of young and dense stellar clusters (e.g., Chen et al. 2007). Another consequence of the coalescence-collapse model is that it might help explain the formation of intermediate-mass black holes ($M_{BH} \gtrsim 100$ $M_{\odot}$) in the central regions of dense stellar clusters, either by the direct gravitational collapse of massive PSCs or by the runaway collisions of massive stars (e.g., Bonnell et al. 1998; Freitag et al. 2006) which would be fostered if the primordial IMF is top-heavy.
Acknowledgments {#acknowledgments .unnumbered}
===============
We thank the anonymous referee for valuable comments and Dongsu Ryu, Christopher Matzner, Carl Jakob Walcher, Sungsoo Kim, and Zhi-Yun Li for useful discussions.
Alves, J., Lombardi, M., & Lada, C. J. 2007, A&A, 462, L17 André, P., Ward-Thompson, D., & Barsony, M. 2000, in Protostars and Planets IV, ed. V. Mannings, A. P. Boss, S. S. Russel (Tuscon:University of Arizona Press), 59 Bonnell, I. A., Bate, M. R., & Zinnecker, H. 1998, MNRAS, 298, 93 Chabrier, G. 2003, PASP, 115, 763 Chen, L., de Grijs, R., & Zhao, J. L. 2007, AJ, accepted, (arXiv:0706.2723) Dib, S., Kim, J., Vázquez-Semadeni, E., Burkert, A., & Shadmehri, M. 2007, ApJ, 661, 262 Eisenhauer, F., Quirrenbach, A., Zinnecker, H., & Genzel, R. 1998, ApJ, 498, 278 Elmegreen, B. G., & Shadmehri, M. 2003, MNRAS, 338, 817 Elmegreen, B. G. 2004, MNRAS, 354, 367 Figer, D. F., McLean, I. S., & Morris, M. 1999, ApJ, 514, 202 Freitag, M., Atakan Gürkan, M., & Rasio, F. 2006, MNRAS, 368, 141 Gouliermis, D., Keller, S. C., Kontizas, M., Kontizas, E., & Bellas-Velidis, I. 2004, A&A, 416, 137 Hillenbrand, L. A., & Hartmann, L. W. 1998, ApJ, 492, 540 Kim, S. S., Figer, D. F., Kudritzki, R. P., & Najarro, F. 2006, ApJ, 653, L113 Kroupa, P. 2002, Science, 295, 82 Larson, R. B. 1981, MNRAS, 194, 809 Lin, D. N. C., & Murray, S. D. 2007, ApJ, accepted, (astro-ph/0703807) Matzner, C. D., & McKee, C. F. 2000, ApJ, 545, 364 Miller, G. E., & Scalo, J. M. 1979, ApJS, 41, 513 Mouri, H., & Taniguchi, Y. 2002, ApJ, 580, 844 Nakano, T. 1966, Prog. Theo. Phys., 36, 515 Padoan, P., Nordlund, Å., & Jones, B. J. T. 1997, MNRAS, 288, 145 Padoan, P., & Nordlund, Å. 2002, ApJ, 576, 870 Portegies Zwart, S., Gaburov, E., Chen, H.-C., & Atakan Gürkan, M. 2007, MNRAS, accepted, (astro-ph/0702693) Rieke, G. H., Loken, K., Rieke, M. J., & Tamblyn, P. 1993, ApJ, 412, 99 Salpeter, E. E. 1955, ApJ, 121, 161 Shadmehri, M. 2004, MNRAS, 354, 373 Silk, J., & Takahashi, T. 1979, ApJ, 229, 242 Sternberg, A. 1998, ApJ, 506, 721 Stolte, A., Grebel, E. K., Brandner, W., & Figer, D. F. 2002, A&A, 394, 459 Stolte, A., Brandner, W., Grebel, E. K., Lenzen, R., & Lagrange, A.-M. 2005, ApJ, 628, L113 Vesperini, E., & Heggie, D. C. 1997, MNRAS, 289, 898 Whitworth, A. P., & Ward-Thompson, D. 2001, ApJ, 547, 317
\[lastpage\]
[^1]: E-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose a source capable of on-demand emission of single electrons with a wave packet of controllable shape and phase. The source consists of a hybrid quantum system, relying on currently experimentally accessible components. We analyze in detail the emission of single electron time-bin qubits, which we characterize using the well known electronic Hong–Ou–Mandel (HOM) interferometry scheme. Specifically, we show that, by controlling the phase difference of two time-bin qubits, the Pauli peak, the electronic analogue of the well known optical HOM dip, can be continuously removed. The proposed source constitutes a promising approach for scalable solid-state architectures for quantum operations using electrons and possibly for an interface for photon to electron time-bin qubit conversion.'
author:
- 'J. R. '
- 'M. '
title: 'An On-Demand Single-Electron Time-Bin Qubit Source'
---
*Introduction.—*
One main ingredient in quantum information processes is the quantum bit or qubit [@MANielsen_1st_ed2000]. Like its classical counterpart, the qubit consists of two different states, but contrary to the classical bit, the qubit can be in a superposition of the two states. An example of such is the time-bin qubit of the form ${\left|\psi\right>}=\alpha{\left|0\right>}+\beta e^{i\varphi}{\left|1\right>}$, where $\alpha$, $\beta$, and $\varphi$ are real-valued, $\alpha^2+\beta^2=1$, and ${\left|0\right>}$ and ${\left|1\right>}$ describe two time-bins of a propagating state in which quantum information is encoded. In this Letter we propose a method for creating and characterizing an on-demand single-electron time-bin qubit (SETBQ) with tunable phase difference, $\varphi$. With the progress in hybrid circuit quantum electrodynamics (QED) in mind [@wallraff_nature04a; @schoelkopf_nature08a; @franceschi_naturenano10a; @you_nature11a; @delbecq_prl11a; @frey_prl12a; @frey_prb12a; @petersson_nature12a; @basset_prb13a; @delbecq_ncomm13a; @van_loo_science13a; @toida_prl13a; @viennot_arxiv13a; @deng_arxiv13a; @liu_arxiv14a], we take inspiration from a quantum optics scheme for manipulating the shape and, importantly, also phase of a single-photon wave packet envelope [@law_jmo97a; @keller_njp04a; @vasilev_njp10a] and transfer it to the electronic realm.
In quantum optics, a single-photon envelope can be shaped using a $\Lambda$-type three-level system, such as an atom or a quantum dot (QD), with one transition resonantly driven by a time-dependent driving field, while the other transition is coupled to a cavity [@law_jmo97a; @keller_njp04a]. By controlling the temporal dependence of the phase and amplitude of the driving field, a single photon is emitted into the cavity with an envelope having a time-dependent phase and amplitude. This scheme was initially proposed and demonstrated for three-level quantum emitters in free space [@law_jmo97a; @kuhn_prl02a; @keller_nature04a; @keller_njp04a; @vasilev_njp10a; @nisbet_jones_njp11a], for which the creation and characterization of high quality single-photon time-bin qubits has been achieved [@vasilev_njp10a; @nisbet_jones_njp13a]. Recently, the shaping of single-photon envelopes has also been demonstrated with solid-state QDs at optical frequencies [@matthisen_naturecom13a] and at microwave frequencies [@pechal_arxiv13a; @wenner_arxiv13a].
![Color online. A schematic representation of the setup. A time-dependent driving field, ${\Omega}(t)$, in the microwave transmission line (left orange region) is coupled to QD1, a two-level quantum dot with ground level ${\left|g\right>}$ (blue line) and excited level ${\left|e\right>}$ (red line). QD1 is in turn tunnel coupled, with amplitude $\tau$, to QD2, a quantum dot with one level ${\left|x\right>}$ (green line) from which the electron may escape, with rate $\Gamma$, into the electronic waveguide (right gray region). The dashed rectangle shows the components of the single-electron time-bin qubit source.[]{data-label="fig:fig1_setup"}](fig1_setup.eps){width="45.00000%"}
Within quantum electronics, several types of single-electron sources (SES) are realized experimentally, for a review see Ref. [@pekola_rmp2013a], while electron emission using an optically driven double quantum dot has been proposed theoretically [@stafford_prl96a], for a review see Ref. [@platero_pr04a], and recently also using circuit QED [@van_den_berg_arxiv14a]. However, for an electronic flying qubit, controlling the electron phase is crucial. The earliest efforts to electrically control an electron phase in solid-state circuits required strong magnetic fields [@washburn_prl87a; @de_vegvar_prb89a; @van_oudenaarden_nature98a]. The most recent demonstration combines an Aharonov-Bohm ring with a two-channel wire to generate solid-state flying qubits [@yamamoto_nnano12a]. The later setup is limited by backscattering at the ring-wire interface, resulting in a visibility of less than $1\%$. Here we propose a different approach, which allows to efficiently produce a highly controllable time-bin single-electron qubit. We suggest to employ a hybrid quantum system, see Fig. \[fig:fig1\_setup\], where a microwave driving field is used to raise an electron from the ground state of a double quantum dot to the excited state above the Fermi level of a nearby electronic waveguide. The essential ingredient of our proposal is a controlled temporal dependence of the amplitude and phase of the driving field, that allows to produce the time-dependent amplitude and phase of an emitted electron. The experimental toolbox for such time-dependent manipulation of the microwave driving field was recently demonstrated for the amplitude [@pechal_arxiv13a; @wenner_arxiv13a] and phase [@hoi_prl13a; @abdo_prl13a; @tancredi_apl13a].
To demonstrate coherence properties of single particles one may use one of several characterization schemes, *e.g.* Hanbury Brown–Twiss [@hanbury_brown_nature56a; @hanbury_brown_nature56b] (HBT), Mach–Zehnder [@zehnder_zfi81a; @mach_zfi82a] (MZ), and Hong–Ou–Mandel [@hong_prl87a] (HOM) interferometry. These schemes were originally developed for optics, but have within the last two decades also been realized in electronics [@henny_science99a; @oliver_science99a; @ji_nature03a; @samuelsson_prl04a; @neder_prl06a; @neder_nature07a; @samuelsson_prl09a; @bocquillon_prl12a; @dubois_nature_13a; @bocquillon_science13a; @dubois_nature_13a]. The coherence properties of electrons emitted on-demand [@feve_science07a; @dubois_nature_13a] were recently characterized via electronic Hong-Ou-Mandel interferometry [@bocquillon_science13a; @dubois_nature_13a]. These experiments demonstrate the possibility to achieve correlations between electrons emitted by independent sources and, for instance, open the opportunity to generate time-bin entangled electron pairs as suggested in Ref. [@splettstoesser_prl09a]. Other recent proposals for such states utilize the helical edge states of a quantum spin Hall insulator [@inhofer_prb13a; @hofer_prb13a].
In the following we first describe the setup and model of a source capable of creating SETBQs and then analyze its characterization using HOM interferometry.
*Setup.—*
We consider a two-level quantum dot, QD1, tunnel coupled to a single-level quantum dot, QD2, which is coupled to an electronic waveguide, *i.e.* a ballistic conductor or an edge state, see Fig. \[fig:fig1\_setup\]. A microwave transmission line in the vicinity of QD1 allows the ground state, ${\left|g\right>}$, and exited state, ${\left|e\right>}$, to be coupled via a classical time-dependent microwave field, ${\Omega}(t)$. An electron in ${\left|e\right>}$ may tunnel with amplitude $\tau$ into the state ${\left|x\right>}$ of QD2 from which it can escape with rate $\Gamma$ into the electron waveguide at an energy state $\Delta \epsilon$ above the Fermi energy, $\epsilon_F$. The state ${\left|x\right>}$ of QD2 ensures that the electrons below the Fermi level of the electronic waveguide do not couple to QD1. While the analysis does not refer to a specific experimental setup, the scheme could be realized in a variety of systems such as by discrete levels of a carbon nanotube, where coupling to fermionic leads and a microwave circuit cavity has been realized [@delbecq_prl11a; @delbecq_ncomm13a] or gate defined QDs coupled to microwave transmission lines as *e.g.* investigated in Refs. [@frey_prl12a; @frey_prb12a; @basset_prb13a].
*Model.—*
The Hamiltonian describing a quantum dot interacting with a microwave transmission line is well known [@childress_pra04a; @blais_pra04a; @bergenfeldt_prb12a]. We assume an infinite on-site Coulomb interaction, such that the source is at most occupied by one electron, consider low temperature, such that no electrons leak from the electronic waveguide into the source, and treat the escape rate by a dissipative Lindblad term of state ${\left|x\right>}$. Thereby, in the interaction picture, the coherent evolution of a single electron in the source is described by the effective non-Hermitian Hamiltonian, ($\hbar=1$) $$\begin{aligned}
{\mathcal{H}}=({\Omega}(t){\hat{a}^{\dagger}}_e{\hat{a}}_g+\tau{\hat{a}^{\dagger}}_e{\hat{a}}_x+H.c.)-i\Gamma{\hat{a}^{\dagger}}_x{\hat{a}}_x,\label{eqn:H'}\end{aligned}$$ where the operators ${\hat{a}}_i$ and ${\hat{a}^{\dagger}}_i$ annihilate and create an electron in state $i$. For the time being we analyze the idealized situation and disregard decoherence effects from relaxation and dephasing for clarity, while we return to these important effects later.
In the basis of ${\left|g\right>}$, ${\left|e\right>}$, and ${\left|x\right>}$, this gives the Schrödinger equation $$\begin{aligned}
\frac{{\partial}}{{\partial}t}
\left[
\begin{array}{c}
a_{g}\\
a_{e}\\
a_{x}
\end{array}
\right]
=
\left[
\begin{array}{ccc}
0 & -i{\Omega}^*(t) & 0 \\
-i{\Omega}(t) & 0 & -i\tau \\
0 & -i\tau^* & -\Gamma
\end{array}
\right]
\left[
\begin{array}{c}
a_{g}\\
a_{e}\\
a_{x}
\end{array}
\right].\label{eqn:schr}\end{aligned}$$ The amplitudes $a_i$ of states $i=g, e, x$ are related to the density matrix elements $\rho_{ij}$ through $\rho_{ij}=a_ia_j^*$.
Equation is identical to the effective Hamiltonian of the quantum optical analog for shaping a single photon envelope [@law_jmo97a; @vasilev_njp10a] described in the introduction. Specifically, our single-electron source is related to the driven $\Lambda$-type 3-level atom single-photon source, by mapping the microwave drive, ${\Omega}(t)$, the tunneling amplitude, $\tau$, and the escape rate, $\Gamma$, respectively to the optical drive, the atom-cavity coupling, and the cavity leakage. We may thus draw a parallel to the approach for photon shaping to design the single-electron wave packet.
*Creating Single-Electron Time-Bin Qubits.—*
The electron that coherently escapes the source, propagates in the $z$-direction along the electronic waveguide with a single electron wave packet of the form $\Psi(t,z)=\psi(t,z)e^{i(\epsilon_F+\Delta\epsilon)t-ik z}$. Here, the envelope $\psi(t,z)$ is determined by the rate $\Gamma$ at which the population of the state ${\left|x\right>}$ decays, *i.e.* at $z_S$, the position of the source, $\psi$ is related to the density matrix element $\rho_{xx}(t)$ through $|\psi(t,z_S)|^2= 2\Gamma\rho_{xx}(t)$, and thereby, $a_{x}(t)=\psi(t,z_S)/\sqrt{2\Gamma}$. In the following we omit the explicit dependence of $z$ for simplicity. Similar to the optical analog [@vasilev_njp10a], by solving Eq. we can derive an equation, which determines the driving field, ${\Omega}(t)$, to be imposed in order to emit a specified single-electron envelope into an electronic waveguide. Furthermore, for time-bin qubits we find that the time-dependent phase of $\Omega(t)$ directly transfers to the phase of the envelope.
From the derivation of ${\Omega}(t)$, the shape of the envelope is limited by $$\begin{aligned}
|\psi(t)|^2\leq2\Gamma, \quad \mbox{and} \quad|{\partial}_t\psi(t)+\Gamma\psi(t)|^2\leq2\Gamma|\tau|^2.\label{eqn:limits}\end{aligned}$$ These inequalities physically signify that the temporal variation of the envelope cannot be faster than the internal timescales of the source. Specifically, the first inequality results in a limit to the width, $T$, of the envelope as a result of the finite escape rate, $\Gamma$. The second inequality shows that the temporal variation of the envelope is limited by the tunneling rate between the two dots in combination with the escape rate. Moreover, as it is known from the optical analog [@vasilev_njp10a], the coupling between the driving field and a QD depends on the occupancy of the QD. Thus, since the occupancy of QD1 decreases during the emission process, the amplitude of the driving field has to be increasingly strong to fully emit the single electron. Therefore, in practice, the electron is partially emitted. To account for this effect we introduce the efficiency of the source, $\eta$, and represent $\psi = \sqrt{\eta}\psi_0$, with the efficiency $0< \eta < 1$ and the ideal envelope $\psi_0$ having unit time integral. These features are best illustrated by an example.
We consider the emission of a time-binned single electron with an envelope $\psi_0(t)=\psi_{sp}(t,0)+\psi_{sp}(t,T/2)e^{i\varphi}$ where $\psi_{sp}(t,t')$ is a semi-pulse given $$\begin{aligned}
\psi_{sp}(t,t')=\frac{4}{\sqrt{5T}}\sin^3\left(\frac{2\pi t}{T}\right), \qquad t'\leq t \leq \frac{T}{2}+t'. \label{eqn:TBQ}\end{aligned}$$ This gives a time-bin qubit of the form ${\left|\psi\right>}=\frac{1}{\sqrt{2}}({\left|10\right>}+e^{i\varphi}{\left|01\right>})$, where *e.g.* ${\left|10\right>}$ represents an electron in the first semi-pulse. The envelope is shown in the inset of Fig. \[fig:fig2\_drive\]. For $\tau=2\Gamma$ and $T=100/G$ the temporal shape of the needed field amplitude, $|{\Omega}(t)|$, is shown in Fig. \[fig:fig2\_drive\] for different $\eta$. The temporal shape of the first part of ${\Omega}(t)$ only changes slightly when increasing $\eta$, while the second part on the other hand has to be increasingly skewed due to the reduced occupancy of QD1.
![Color online. The calculated temporal shape of the drive $|{\Omega}(t)|/\Gamma$ in order to obtain the single-electron TBQ, Eq. , for different $\eta$. The parameters used for the source are $\tau=2\Gamma$ and $T=100/\Gamma$. Inset: The squared envelope of an emitted electron normalized in maximum to unity.[]{data-label="fig:fig2_drive"}](fig2_drive.eps){width="45.00000%"}
We have thus presented a source capable of emitting time-bin qubits. We notice that, even though we have focused on such single electron envelopes, the scheme is not limited to these. In fact any single electron envelope shape is allowed, only restricted by the internal timescales of the source through inequalities Eq. .
*Characterization.—*
Having described a source for creating SETBQs, we next analyze a possible way of characterizing it, specifically using the HOM scheme. The HOM scheme, shown in the inset of Fig. \[fig:fig3\_characterization\_phi\], consists of a beam-splitter (BS) having two input ports, 1 and 2, and two output ports, 3 and 4. Electrons are emitted into the input ports followed by the zero-frequency current correlations measurement of the output ports, *i.e.* [@blanter_pr00a] $$\begin{aligned}
S_{34}&=\iint{\mathrm{d}}t{\mathrm{d}}t' [{\left<{\hat{n}}_3(t){\hat{n}}_4(t')\right>}-{\left<{\hat{n}}_3(t)\right>}{\left<{\hat{n}}_4(t')\right>}],\end{aligned}$$ where ${\hat{n}}_{\alpha}(t)$ is the current flux operator of channel $\alpha$. If only one of the sources, S1 or S2, is active we get the HBT correlations, $S^{HBT}_{S1}$ or $S^{HBT}_{S2}$, which are identical for identical sources, $S^{HBT}_{S1}=S^{HBT}_{S2}=S^{HBT}$. If both sources are active, the HOM correlations, $S^{HOM}$, are measured and one may define the HOM correlations normalized with respect to the HBT correlations, $\mathcal{S}_{34}=S^{HOM}_{34}/(2S^{HBT})$, to get a quantity which is independent of the transmission and reflection probabilities of the BS and, furthermore, to reduce the effect of temperature [@bocquillon_science13a]. Assuming that none of the energy components of the envelopes overlap with the Fermi sea, we have that [@jonckheere_prb12a] $$\begin{aligned}
\mathcal{S}_{34}&=1-\left|\int{\mathrm{d}}t \;\psi_{1}^*(t)\psi_{2}(t)\right|^2. \label{eqn:S_{34}}\end{aligned}$$ In Fig. \[fig:fig3\_characterization\_phi\] we show $\mathcal{S}_{34}$ calculated for two sources, S1 and S2, emitting single electrons with envelopes $\psi_1$ and $\psi_2$ of the form in Eq. and partial phases $\varphi_1$ and $\varphi_2$, respectively, with $\delta\varphi=\varphi_2-\varphi_1$. The sources S1 and S2 are synchronized such that emitted electrons arrive at the beam splitter with a time delay $\delta t$.
![The HOM correlations vs phase difference, $\delta\varphi$, and time difference, $\delta t$. Parameters used are $\tau=2\Gamma$, a desired envelope of width $T=100/\Gamma$, and source efficiency $\eta=0.99$. For $\Omega(t)$, see the solid black line in Fig. \[fig:fig2\_drive\]. Inset: Sketch of the HOM setup. The two sources S1 and S2 emit single electrons, which collide at the beam splitter BS and the current correlations, $\mathcal{S}_{34}$, between the arms 3 and 4 are measured.\[fig:fig3\_characterization\_phi\]](fig3_HOM.eps){width="45.00000%"}
First, for $\delta t=0$, we have $\mathcal{S}_{34}=1-\eta^2\cos^2(\delta\varphi/2)$ and thus for $\delta\varphi=0$, corresponding to two identical incident single-electron envelopes, the current correlation is $\mathcal{S}_{34}=1-\eta^2$. For unit efficiency, $\eta=1$, each electron incoming from the port $1$ encounters an electron incoming from the port $2$. At $\delta t =0$, the perfect overlap of the envelopes results in $\mathcal{S}_{34} = 0$, *i.e.* the electrons are scattered into different output ports $3$ and $4$. This anti-bunching is known as the Pauli peak and reflects the fermionic nature such that two electrons cannot occupy the same state at the same time [@blanter_pr00a; @olkhovskaya_prl2008a]. The Pauli peak can be seen in Fig. \[fig:fig3\_characterization\_phi\] at $\delta\varphi =0$ close to $\delta t=0$.
As $\delta\varphi$ changes from zero to $\pi$ we observe that, interestingly, the Pauli peak at $\delta t=0$ disappears. That is, even though the two electrons arrive at exactly the same time, such that $|\psi_1(t)|=|\psi_2(t)|$, they seemingly do not obey the Pauli principle. This peculiar observation is caused by the two envelopes at $\delta\varphi=\pi$ being orthogonal to each other and thus do not constitute the same states, *i.e.* their overlap is zero and electrons can be scattered to the same output port. At equal arrival time, the difference between $\mathcal{S}_{34}$ for $\delta\varphi=0$ and $\delta\varphi=\pi$ shows the difference between the two states ${\left|\psi\right>}=\frac{1}{\sqrt{2}}({\left|10\right>}\pm{\left|01\right>})$ and thus $\mathcal{S}_{34}$ constitutes the source visibility.
Lastly, by changing the time-delay, $\delta t$, two phase-independent smaller peaks are seen at $\delta t=\pm T/2$. These correspond to the overlap of the front pulse of $\psi_1(t)$ with the tail pulse of $\psi_2(t)$ and vice versa. We have thus shown that, by varying the time-delay, $\delta t$, and phase difference, $\delta\varphi$, between the two SETBQ, we are able to characterize our proposed source using the HOM scheme.
*Decoherence Effects.—*
Until now we have neglected the important issues of the relaxation of QD1 and dephasing due to the tunneling between QD1 and QD2. Since we are interested in the coherent evolution of the electron, we follow Ref. [@vasilev_njp10a] and describe the decoherence by a dissipation of the coherent electron to derive an effective Hamiltonian as used in the quantum-jump approach to dissipative systems known from quantum optics, see *e.g.* Refs. [@zoller_pra87a; @HCarmichael_1st_1993; @plenio_rmp98a]. This is done by including a relaxation rate $\gamma_{r}$ of state ${\left|e\right>}$ and tunneling dephasing rate $\gamma_{\phi}$ of states ${\left|e\right>}$ and ${\left|x\right>}$ as dissipations in the effective Hamiltonian, Eq. , governing the coherent electron. This implicitly assumes that a photon emitted into the microwave transmission line due to the relaxation of QD1, does not reexcite the QD. We may then again derive an equation, which determines ${\Omega}(t)$ for the coherent emission of a specified single electron envelope into the electronic waveguide. From the conservation of charge one finds that the decoherence terms lead to the physical limit on the maximal efficiency, similar to the optical analog [@vasilev_njp10a], $$\begin{aligned}
\eta_{\max} = \frac{1}{1+\frac{(\gamma_{r}+\gamma_{\phi})}{\Gamma|\tau|^2}\left[\Gamma^2+\int_0^{T}{\mathrm{d}}t'|{\partial}_{t'}\psi_0(t')|^2\right]+\frac{\gamma_{\phi}}{\Gamma}},\end{aligned}$$ *i.e.* the efficiency has to be in the interval $0\leq\eta\leq\eta_{\max}$. The case $\eta=\eta_{\max}<1$ signifies that the decoherence results in the emitted electron being in a statistical mixture of being coherently emitted with probability $\eta$ and incoherently emitted with probability $1-\eta$. To give an estimate of $\eta_{\max}$ we take the currently achievable experimental values $\tau/(2\pi)\sim15.5$GHz, $\gamma_{r}/(2\pi)\sim100$MHz, and $\gamma_{\phi}/(2\pi)\sim 1.5$GHz from Ref. [@basset_prb13a] and $\Gamma \sim \tau/2$ from [@frey_prb12a] and $T=100/G$ as used in Figs. \[fig:fig2\_drive\] and \[fig:fig3\_characterization\_phi\] giving $\eta_{\max}\sim0.8$ for the SETBQ, Eq. . This suggests that, with current technology, our proposed source has the possibility of a significantly increased efficiency compared to state-of-the-art flying qubit sources [@yamamoto_nnano12a].
*Reloading the Source.—*
Lastly, let us describe three possible methods for reloading the source for gate defined QDs. First, a laser pulse could excite electrons from the buffer layer to the state ${\left|g\right>}$ as experimentally demonstrated in Ref. [@fujita_prl13a] and thereby deterministically load the source with a new electron on demand. A second method is to tune the voltage gates to lower the energy level of state ${\left|x\right>}$ below the Fermi sea of the electronic waveguide to the level of ${\left|g\right>}$. This would permit an electron (only one due to the Coulomb energy) to stochastically flow from the electronic waveguide into ${\left|x\right>}$ and then tunnel back and forth between ${\left|g\right>}$ and ${\left|x\right>}$. Since it is below the Fermi sea it would not decay back into the electronic waveguide. Then if ${\left|x\right>}$ is slowly raised, the electron would end in ${\left|g\right>}$. A third option is to couple QD1 very weakly to an electron reservoir. If the time scale of the electron tunneling from the reservoir to QD1 is much longer than the electron emission time of the source, then there would only be a minimal risk of having more than one electron in the emitted electron time-bin.
*Summary and Outlook.—*
We proposed a single-electron source relying on a hybrid quantum system. With our scheme one can design an electron envelope thereby providing control over the phase and the amplitude of a single electron. In particular, we analyzed the emission of a single-electron time-bit qubit, and showed how to characterize it using a well-known interferometric technique. Specifically, we showed that in Hong–Ou–Mandel interferometry the Pauli peak can be continuously removed by controlling the phase difference of two time-bin qubits. Our analysis showed that, with experimentally relevant parameters, the source efficiency is expected to be close to unity. This opens the possibility of using single electrons for quantum operations in scalable solid-state architectures. Furthermore, with the recent demonstration of creating single microwave photons with controlled envelopes [@pechal_arxiv13a; @wenner_arxiv13a] our proposed scheme constitutes a possible photon-electron interface for photon-to-electron time-bin qubit conversion with microwave coupling in the vacuum Rabi regime.
*Acknowledgment.—*
We are most grateful to Markus Büttiker who raised the problem addressed here and participated in the initial stage of this work. Markus sadly passed away before the project was finalized. We thank D. Dasenbrook, P. P. Hofer, C. Flindt, P. Samuelsson, and M. Wubs for comments to the manuscript. JRO acknowledges financial support from the Danish Council of Independent Research. Research in Geneva is supported by the Swiss NSF.
[66]{}ifxundefined \[1\][ ifx[\#1]{} ]{}ifnum \[1\][ \#1firstoftwo secondoftwo ]{}ifx \[1\][ \#1firstoftwo secondoftwo ]{}““\#1””@noop \[0\][secondoftwo]{}sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{}@startlink\[1\]@endlink\[0\]@bib@innerbibempty @noop [**]{}, ed. (, , ) @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} @noop [ ()]{} @noop [ ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} @noop [ ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [**]{} (, , ) @noop [****, ()]{} @noop [****, ()]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recent studies have raised doubts about the occurrence of $r$ modes in Newtonian stars with a large degree of differential rotation. To assess the validity of this conjecture we have solved the eigenvalue problem for Rossby-Haurwitz waves (the analogues of $r$ waves on a thin-shell) in the presence of differential rotation. The results obtained indicate that the eigenvalue problem is never singular and that, at least for the case of a thin-shell, the analogues of $r$ modes can be found for arbitrarily large degrees of differential rotation. This work clarifies the puzzling results obtained in calculations of differentially rotating axi-symmetric Newtonian stars.'
address:
- '$^1$SISSA, International School for Advanced Studies, Via Beirut 2, 34014 Trieste, Italy'
- '$^2$INFN, Department of Physics, University of Trieste, Via Valerio, 2 34127 Trieste, Italy'
author:
- 'Luciano Rezzolla$^{1,2}$ and Shin’ichirou Yoshida$^1$'
---
Introduction
============
The investigation of $r$ modes in rotating stars has seen a renewed interest in recent years since they were shown to be unstable to the emission of gravitational waves (Andersson 1998, Friedman and Morsink 1998). In the last three years a number of papers have outlined the basic properties of the $r$-mode oscillations, their role in triggering and feeding the instability, and the importance of this instability in the emission of gravitational waves as well as in the slowing down of rapidly rotating, hot neutron stars (see Andersson and Kokkotas 2001 and Friedman and Lockitch 2001 for some recent reviews). Recent studies, on the other hand, have focussed on more intricate but also more realistic aspects of the instability. Such aspects include the interaction of the instability with magnetic fields (Spruit 1999, Rezzolla et al. 2000), nonlinear and secular effects (Rezzolla et al. 2001a, 2001b; Stergioulas and Font 2001; Lindblom et al. 2001), the presence of differential rotation in Newtonian stars (Karino et al. 2001, hereafter KYE), of a continuous frequency spectrum in the slow-rotation approximation of uniformly rotating relativistic stars (Kojima and Hosonuma 1999) and the lack of discrete, physically plausible mode solutions within the continuous part of the spectrum (Ruoff and Kokkotas 2001). We here pay attention to one of these more “subtle” aspects and focus on the role played by differential rotation in Newtonian stars. In particular, we address the question of whether a large degree of differential rotation in slowly rotating stars turns the eigenvalue problem for $r$ modes into a singular eigenvalue problem, thus preventing the existence of $r$ modes. This is an important question because of the possible observation of a singular problem in $r$-mode oscillations that has emerged from recent calculations of differentially rotating axi-symmetric Newtonian stars (KYE). More intriguingly, similar evidence is converging also from different approaches such as those looking at purely axial $r$ modes in rotating non-isentropic relativistic stars in the slow-rotation approximation (Yoshida 2001, Yoshida and Futamase 2001; Ruoff and Kokkotas, 2001a, 2001b) although a regularization may be possible in some cases (Lockitch and Andersson 2001).
Part of the subtlety introduced by differential rotation in the eigenvalue problem is due to the fact that, above a certain degree of differential rotation, the eigenvalue equations become very stiff as a result of large radial and polar gradients of the rotational angular frequency. Under these circumstances, the numerical solution requires increasingly high accuracy and computational costs. To circumvent this problem, we have here resorted to a simpler model based on a differentially rotating, thin shell. A thin-shell model has been adopted also by Levin and Ushomirsky (2001) to investigate the effect of electromagnetic radiation-reaction for $r$ modes on a uniformly rotating shell. This approach replaces the set of partial differential equations of a multidimensional star model with an ordinary differential equation which can be solved to much higher accuracy and with modest computational costs. Although much simpler to solve, the $r$-mode eigenvalue problem for a thin shell incorporates many of the mathematical properties of the corresponding eigenvalue problems for multidimensional Newtonian stars or for slowly-rotating relativistic stars. As will become clear in the following Sections, adopting a thin-shell model has been very valuable for gaining insight into the behaviour of $r$ modes in differentially rotating fluids.
Basic Equations {#be}
===============
The thin-shell model adopted here is based on the following simplifying assumptions: [*(i)*]{} the background star is slowly rotating and we will omit terms in the hydrodynamical equations of ${\cal O}(\Omega^2)$; [*(ii)*]{} the fluid is incompressible and inviscid; [*(iii)*]{} the shell is spherical with radius $R$ for every rate of differential rotation; [*(iv)*]{} the fluid motion is constrained on the shell, that is, we have no radial component of velocity.
In the inertial (nonrotating) frame the fluid velocity $\vec{v}$ is the composition of the unperturbed velocity of the differentially rotating shell $\vec{v}_{_0}$, where, in spherical polar coordinates, $v^i_{_0}=(0,0,\Omega(\theta) R \sin\theta)$, with the velocity perturbation $\delta \vec {u}$ which has only tangential components, i.e. $\delta u^i =
(0,u^{\theta},u^{\phi})$. As a result, the components of the fluid velocity can be written as $$v^i = (0, u^{\theta}, \Omega R\sin\theta + u^{\phi}),$$ In the simple model considered here, the fluid motion in the shell is fully described by the Euler and continuity equations. The latter, in particular, is trivially satisfied by the background motion and assumes the following form in its perturbed part $$\partial_{\theta}(u^{\theta}\sin\theta)
+ \partial_{\phi} u^{\phi} = 0 \ .
\label{continuity}$$
As customary with this type of problem, it is here convenient to introduce the fluid vorticity $\vec{\xi} \equiv\frac{1}{2}\nabla\times\vec{v}$, whose radial component has the form $$\xi^r = \Omega\cos\theta + \frac{1}{2}\frac{d\Omega}{d\theta}
\sin\theta +\frac{1}{2r\sin\theta}
\left[\partial_{\theta}(u^{\phi}\sin\theta)
- \partial_{\phi} u^{\theta}
\right] \ .$$ It is a simple exercise to show that the Euler equation can then be cast into a conservation equation for the vorticity $\partial_t\vec{\xi} + \nabla\times
(\vec{\xi}\times\vec{v}) = 0$, whose radial component is $$\left( \partial_t + \frac{v^{\theta}}{r}
\partial_{\theta} + \frac{v^{\phi}}{r\sin\theta}
\partial_{\phi}\right)\xi^r = 0 \ .
\label{xi-r}$$
Because we are interested in the harmonic modes, we assume that the tangential components of the perturbation velocity can be written as $u^{\theta}, u^{\phi} \sim {\rm exp}(-i\sigma t +
im\phi)$, so that the linearized version of equation (\[xi-r\]) is $$\fl -i(\sigma-m\Omega)
\left[\frac{d}{d\theta}(u^{\phi}\sin\theta)
-imu^{\theta}\right]
+ 2u^{\theta}\sin\theta \frac{d}{d\theta}
\left(\Omega\cos\theta + \frac{1}{2}
\frac{d\Omega}{d\theta}\sin\theta\right) = 0\ .$$
We can now use the continuity equation (\[continuity\]) to rewrite the radial component of the vorticity conservation equation as $$\label{eigen_eq}
\fl \frac{d}{d\mu}\left[(1-\mu^2)\frac{d\chi}{d\mu}\right]
- \frac{m^2}{1-\mu^2}\chi
- \frac{2m\Omega}{\sigma-m\Omega}
\left[
1 + \left(\frac{2\mu}{\Omega}\right)\frac{d\Omega}{d\mu}
- \left(\frac{1-\mu^2}{2\Omega}\right)
\frac{d^2\Omega}{d\mu^2}\right]\chi = 0\ ,$$ where we have introduced the new coordinate $\mu \equiv
\cos\theta$ and where $\chi \equiv u^{\theta}
\sin\theta$. Together with regular boundary conditions at the northern and southern poles ($\mu=\pm 1$), equation (\[eigen\_eq\]) accounts for the eigenvalue problem of the fluid normal modes on the shell. In the case of a uniformly rotating shell, ${d\Omega}/{d\mu}=0={d^2\Omega}/{d\mu^2}$, and equation (\[eigen\_eq\]), first derived by Haurwitz in 1940 (see also Stewartson and Rickard, 1969), reduces to Legendre’s equation whose regular solution is $\chi=P_{^l}^{_m}(\mu)$, with $P_{^l}^{_m}(\mu)$ being the associated Legendre functions. Note that in the case of uniform rotation the eigenfrequencies obey the well known dispersion relation for $r$ modes $$\label{dispersion}
\sigma = m\Omega - \frac{2m\Omega}{l(l+1)}
\ ,$$ and this justifies calling these modes Rossby-Haurwitz waves. For simplicity, however, hereafter we will refer to them as $r$ modes.
The solution of equation (\[eigen\_eq\]) in the case of a differentially rotating shell depends on the law of differential rotation chosen and on its first and second polar derivatives. The choice of a law of differential rotation is, in this sense, somewhat arbitrary but we have here followed previous work on the subject (Eriguchi and Müller 1985, KYE) and modeled the differential rotation through either a “$j={\rm
const.}$” law $$\label{diff_rot_j}
\Omega = \frac{A^2_j + 1}{A^2_j + 1 - \mu^2}
\Omega_{_{\rm E}} \ ,$$ or a “$v={\rm const.}$” law[^1] $$\label{diff_rot_v}
\Omega = \frac{A_v + 1}{A_v + \sqrt{1 - \mu^2}}
\Omega_{_{\rm E}}\ .$$ In both cases $\Omega_{_{\rm E}}$ is the angular velocity at the equator ($\mu=0$) and the parameter $A_{j,v} > 0$ accounts for the degree of differential rotation so that uniform rotation (i.e. ${\Omega}/{\Omega_{_{\rm E}}}=
{\rm const.}$) is reached for $A_{j,v} \to \infty$. Note that the use of a law of differential rotation is physically plausible as long as such a law does not violate Rayleigh’s stability criterion for rotating inviscid fluids: $d\left(\varpi^2
\Omega\right)^2/{d\varpi} > 0$, where $\varpi$ is the cylindrical radial coordinate. It is straightforward to check that both expressions (\[diff\_rot\_j\]) and (\[diff\_rot\_v\]) satisfy Rayleigh’s criterion for all values of $A_{j,v} > 0$.
The solution of equation (\[eigen\_eq\]) is considerably simpler than the solution of the corresponding set of partial differential equations describing normal modes of a differentially rotating axi-symmetric star and discussed by KYE. Nevertheless, most of the mathematical properties that are found in the solution of the system of partial differential equations can already be found in (\[eigen\_eq\]). A particularly important property of equation (\[eigen\_eq\]), which has been encountered also when dealing with $r$ modes of differentially and rapidly rotating stars (KYE), is that it may become a singular eigenvalue problem at the angular position $\mu_s$ for which $$\label{corot}
{\sigma-m\Omega(\mu_s)}= 0 = \omega_{\rm ph} -
\Omega(\mu_s) \ ,$$ where $\omega_{\rm ph} \equiv \sigma/m$ is the phase velocity of the mode. The condition (\[corot\]), which can be interpreted as the appearance of a [*“corotation point”*]{} (i.e. a point on the shell at which the perturbation pattern rotates at the same angular velocity as the background shell) has been interpreted by KYE as the cause impeding the numerical solution of the eigenvalue problem for a sufficiently large degree of differential rotation[^2]. Interestingly, equation (\[eigen\_eq\]) offers close analogies also with the corresponding equation obtained in full General Relativity for a slowly and uniformly rotating relativistic star (Yoshida 2001, Yoshida and Futamase 2001; Ruoff and Kokkotas, 2001a, 2001b). The analogy is brought about by the fact that the corrections due to the relativistic dragging of inertial frames are mathematically similar to the corrections due to differential rotation and introduce a similar coefficient which could become singular for certain rates of rotation[^3].
Strategy of the Numerical Solution
==================================
The numerical procedure adopted in the solution of equation (\[eigen\_eq\]) is straightforward and needs particular care only if a singular point should appear during the solution. In general, we can numerically integrate equation (\[eigen\_eq\]) as a two point boundary value problem with boundary conditions at both poles, $\mu=\pm 1$, being given by the requirement that the solution $\chi$ is regular there. The two solutions found are then matched at an arbitrary point $\mu_{_{\rm
M}}$ in the domain $(-1,1)$ following the standard procedure in the solution of an eigenvalue problem. In particular, for a trial value of the eigenfrequencies we look for a zero of the Wronskian evaluated at $\mu_{_{\rm
M}}$, with different zeros representing sequences of different mode numbers. Once a zero is found, the eigenfrequency is used in equation (\[eigen\_eq\]) to calculate the corresponding eigenfunction. This procedure is repeated for different degrees of differential rotation so that a sequence is built describing the eigenfrequencies of the rotating shell from uniform rotation up to extreme differential rotation. The results of these calculations are discussed in detail in the following Section.
Numerical Results
=================
This Section briefly presents the results of the numerical solution of equation (\[eigen\_eq\]) for the two laws of differential rotation discussed above.
$j={\rm const.}$ differential rotation
--------------------------------------
We show in figure \[fig1\] the values of the phase velocity $\omega_{\rm ph}$ normalized to the value of the angular velocity at different latitudes on the shell. (This normalization is particularly convenient as it allows one to detect immediately whether corotation, corresponding to $\omega_{\rm ph}/\Omega(\mu) \to$ 1, takes place or not.) As expected, in the limit of uniform rotation, the curves converge to $2/3$, the value given by the dispersion relation (\[dispersion\]) for an $m=l=2$, $r$ mode. For progressively smaller values of $A_j$, on the other hand, the curves split in response to the different angular velocities at different angular positions on the shell. Surprisingly, corotation is reached only at the equator (i.e. $\mu=0$) and then only in the limit of $A\to 0$, which is not physically interesting.
This can be clearly seen in figure \[fig2\], where we show the deviation away from unity of the ratio $\omega_{\rm ph}/\Omega(\mu=0)$. For both curves, which correspond in the limit of uniform rotation to an $m=l=2$ and an $m=2$, $l=3$ mode respectively, the behaviour for small values of the parameter $A$ is well fitted by a power law $\omega_{\rm ph}/\Omega(\mu=0) \sim K A^n$, where $n\simeq 3.76$ for $m=2$, ’$l=2$’, and $n\simeq
53.4$ for $m=2$, ’$l=3$’ ($K$ is here a positive constant). As a result, the deviation away from unity tends to zero only in the limit of $A\to 0$ and we therefore conclude that [*no corotation*]{} appears in these modes in the range of validity of a $j={\rm
const.}$ law of differential rotation.
The small inset in figure \[fig1\] shows a comparison between the ratio $\omega_{\rm
ph}/\Omega(\mu=0)$ obtained with the present shell approach and the corresponding quantity (indicated with filled squares) obtained by KYE with their numerical code for a slowly rotating stellar model with axis ratio 0.95. There is a rather good agreement, at least for the range of differential rotation rates in which the mode calculation was possible. Such an agreement brings confidence about the relevance of the results obtained with the shell model also for multidimensional star models.
When looking at the results of KYE, it becomes apparent how the behaviour of the eigenfrequencies might have suggested the appearance of a corotation point. As mentioned in the Introduction, when the degree of differential rotation increases past a certain threshold, very large radial and polar gradients appear in the equations for the eigenvalue problem. The solution of very stiff equations using finite difference techniques is a very difficult task and it is therefore not surprising that KYE were not able to find convergent solutions to the eigenvalue problem for $A\lesssim 0.3$. By making use only of an ordinary differential equation, the shell approach bypasses this difficulty and provides an accurate solution for any value of $A$.
$v={\rm const.}$ differential rotation
--------------------------------------
In the case of the $v={\rm const.}$ law, the absence of a corotation point is even more evident. In figure \[fig3\] the eigenfrequency is plotted as in figure \[fig1\]. There, we clearly see that the ratio $\omega_{\rm ph}/\Omega(\mu)$ is below unity everywhere on the shell. The comparison with the results by KYE, shown in the small inset of figure \[fig3\], is less good than the one seen for a $j={\rm const.}$ differential rotation law but the overall behaviour is rather similar.
In addition to the differential rotation laws (\[diff\_rot\_j\]) and (\[diff\_rot\_v\]) we have also investigated a quadratic differential rotation law of the type $\Omega/\Omega_{_{\rm P}} = 1 + (\mu^2-1)/{B}$, where $B>0$ and uniform rotation is reached for $B \to
\infty$. Also in this case, no evidence for corotation was found for the values of the parameter $B$ satisfying the Rayleigh stability criterion (i.e $B>2$).
Conclusion
==========
We have used a thin-shell model to investigate the behaviour of Rossby-Haurwitz waves (the analogues of $r$ waves on thin shells) in the presence of differential rotation. Our simplified approach replaces the set of partial differential equations for a multidimensional stellar model with a single ordinary differential equation that can be solved to arbitrary accuracy. The numerical solution of the eigenvalue problem has shown that there is no evidence of corotation, even for asymptotically large values of differential rotation. As a result, the eigenvalue problem for Rossby-Haurwitz waves never becomes singular, as had instead been suggested by recent calculations of differentially rotating axi-symmetric Newtonian stars. We have found that, for this simplified model, $r$ modes can in principle coexist with arbitrarily large rates of differential rotation, the eigenfrequencies and eigenfunctions being suitably modified in response to the degree of differential rotation.
The relevance of the results found here for more realistic stellar models is not easy to assess, although the equations solved here show many of the mathematical features of the corresponding equations for Newtonian differentially rotating stars, or for slowly rotating relativistic stars.
It is pleasure to acknowledge fruitful discussions with M. Abramowicz, N. Andersson, S. Bonazzola, Y. Eriguchi, K. Kokkotas, J. Miller, J. Ruoff and N. Stergioulas. Financial support for this research has been provided by the MURST and by the EU Programme “Improving the Human Research Potential and the Socio-Economic Knowledge Base" (Research Training Network Contract HPRN-CT-2000-00137).
References {#references .unnumbered}
==========
Andersson N. 1998 [*Astrophys. J.*]{} [**502**]{} 708
Andersson N. and Kokkotas K.D. 2001 to appear in [*Int. J. Mod. Phys.*]{} [gr-qc/0010102]{}
Eriguchi Y. and Müller E. 1985 [ *Astron. Astrophys.*]{} [**146**]{} 260
Friedman J.L. and Morsink S.M. 1998 [ *Astrophys. J.*]{} [**502**]{} 714
Friedman J.L. and Lockitch K.H. 2001, Proceedings of the IX Marcel Grossman Meeting, World Scientific, eds. V. Gurzadyan, R. Jantzen, R. Ruffini; [ gr-qc/0102114]{}
Haurwitz B, 1940 [*J. Mar. Res.*]{} [**3**]{}, 254.
Karino K., Yoshida S. and Eriguchi Y. 2001 [ *Phys. Rev. D*]{} [**64**]{} 024003
Kojima K. and Hosonuma 1999 [*Astrophys. J.*]{} [**520**]{} 788
Levin Y. and Ushomirsky G. 2001 [ *Mon. Not. R. Astr. Soc.*]{} [**322**]{} 515
Lindblom L., Tohline J.E. and Vallisneri M. 1152
Lockitch K. H. and Andersson N. 2001, [*preprint*]{} [gr-qc/0106088]{}
Rezzolla L., Lamb F.K. and Shapiro S.L. 2000 [*Astrophys. J.*]{} [**531**]{} L139
Rezzolla L., Lamb F.K., Markovic D. and Shapiro S.L. 2001a [*Phys. Rev. D*]{} in press
2001b [*Phys. Rev. D*]{} in press
Ruoff J. and Kokkotas K.D. 2001a, [*preprint*]{} [gr-qc/0101105]{}
Ruoff J. and Kokkotas K.D. 2001b, [*preprint*]{} [gr-qc/0106073]{}
Spruit H.C. 1999 [*Astron. Astrophys.*]{} [**341**]{} L1
Stergioulas N. and Font J.A. 2001 1148
Stewartson K. and Rickard J.A. 1969 [ *J. Fluid. Mech.*]{}, [**35**]{} 759
Yoshida S. 2001 to appear in [*Astrophys. J.*]{} [gr-qc/0101115]{}
Yoshida S. and Futamase T., [*preprint*]{} [gr-qc/0106076]{}
[^1]: These differential laws take their names from the fact that through the variation of the parameter $A_{j,v}$ they represent families of rotation laws spanning the range between uniform rotation (for $A_{j,v}\to\infty$) and differential rotation with constant specific angular momentum or constant linear velocity (for $A_{j,v}\to
0$), respectively.
[^2]: KYE also noted that the difficulty in finding a convergent numerical solution was dependent on the rate of rotation of the background models, with rapidly rotating models providing solutions for comparably smaller values of $A_{j,v}$.
[^3]: Note that although equation (\[eigen\_eq\]) can in principle be singular, a global solution can still be found if series expansion techniques, such as the Frobenius method, are employed in the vicinity of the singular point $\mu_s$.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Markus Mugrauer, R. Neuhäuser, T. Mazeh, E. Guenther M. Fernández$^{,}$'
- 'C. Broeg'
date: 'Received; accepted; published online'
title: 'A Search for wide visual companions of exoplanet host stars. The Calar Alto Survey'
---
Introduction
============
More than 100 stars are already known to host exoplanets. Detected by radial velocity searches, theses planets orbit their host stars on relative close orbits with semi-major axes smaller 5AU. Some of these exoplanets were found in binary systems, in which the exoplanet host star is mostly the more massive primary star. First statistical differences between planets orbiting single stars and planets located in binary systems were already reported by Zucker & Mazeh (2002) and Eggenberger et al. (2004), using samples of 9 and 15 binaries, respectively. In particular, planets with orbital periods shorter than 40 days exhibit a difference in their mass-period and eccentricity-period distribution. Mugrauer et al. (2005b) updated and extended the sample of binaries among the exoplanet host stars (21 binaries) and confirmed the reported statistical differences for short period planets. These statistical differences between planets orbiting single stars and planets residing in a binary systems might be an implication of the host star multiplicity.
Today only four of the binary systems with exoplanets are known to have small projected separations of the order of 20AU (HD188753, $\gamma$Cep, HD41004 and Gl86). Due to the proximity of the stars in these four stellar systems, they are a challenge for planet formation theories (core accretion and gravitational collapse scenario). In these systems the size of a protoplanetary disk around the exoplanet host star is truncated by its companion star to only a few AU and does not extend beyond the ice line (Pichardo et al. 2005). Furthermore, objects revolving around the exoplanet host star farther outside are perturbed by the secondary star and are not longterm stable, i.e. finally these objects will be ejected from the system or collide with one of the two stars (Holman & Wiegert 1999). Therefore, planets orbiting a component of a close binary system should be formed and reside only in the adjacency of their parent star. On the other hand, close to the star, within the ice line, the core accretion (lack of solid material to form the planet core in a solar minimum mass nebula) as well as the gravitational collapse scenario (high disk temperature dumps gravitational instability) cannot explain the formation of gas giant planets (Jang-Condell 2005).
Nevertheless, there are ways around this problem. The secondary star could excite density waves in the disk, increasing the surface density in some parts of the disk, leading to planet formation via gravitational instability. Or, the planet bearing disk might be different from the disk of our own solar system. If disks of the same mass differ only in their angular momentum, such that in smaller disks more mass is closer in, planets indeed might be formed close to the star within the iceline (Hatzes & Wuchterl 2005).
The problem to form planets in close binaries by both the gravitational instability and the core accretion scenario clearly demonstrates that planets detected in these systems are most intriguing objects. They provide the possibility to study the effect of stellar multiplicity on the planet formation, the longterm stability and evolution of planetary orbits and yield ancillary conditions for the formation of Jovian planets.
So far only few programs have searched for additional visual companions of exoplanet host stars. Adaptive optics (AO) imaging search campaigns reported several new close companions during the last years. Patience et al. (2002) detected companions close to HD114762 and $\tau$Boo. Furthermore Els et al. (2000) reported a faint companion, separated from the exoplanet host star Gl86 by $\sim$2arcsec. They concluded from infrared photometry that this companion is a late L or early T dwarf. Queloz et al. (2000) reported a longterm linear trend in the radial velocity of Gl86 and according to Jahreiß (2001) Gl86 turned out to be a highly significant $\Delta\mu$ binary after combining Hipparcos measurements with ground based astrometric catalogues. Both results indicating that Gl86 has a companion of stellar mass. Mugrauer & Neuhäuser (2005a) could finally prove that the companion, firstly detected by Els et al. (2000), is indeed a white dwarf, the first white dwarf found close to an exoplanet host star.
Despite being successful, AO imaging surveys cannot find wide companions with separations up to a few thousand AU because of their small field of view (typically only a few tens of arcsec). Lowrance et al. (2002) presented a new wide (750AU) low-mass stellar companion ($\sim$0.2M$\rm_{\sun}$) which was detected in the digitized plates of the Palomar Obervatory Sky Survey, a first example that there could be many more of these objects, all with separations larger than 100AU. The whole sample of exoplanet host stars was not systematically surveyed so far for these objects, i.e. the multiplicity of exoplanet host stars might be much higher than derived from AO surveys alone. Therefore, we have started a search program for wide companions of exoplanet host stars and several new binaries were already detected (see Mugrauer et al. 2004 a,b and 2005 b).
Most of the exoplanet host stars have large proper motions ($\mu$$\sim$200mas/yr), well known due to precise measurements of the *HIPPARCOS* satellite (Perryman et al. 1997). Therefore, real companions can be identified as co-moving objects by comparing images taken with several years of epoch difference. Photometry and spectroscopy can then confirm the companionship — the measured photometry of the companion must be consistent with an object of the given spectral type at the distance and age of the exoplanet host star. To be sensitive to low-mass faint substellar companions, we observed all targets in the near-infrared (H band at 1.6$\mu$m), as substellar companions are several magnitudes brighter compared to the visible spectral range. Furthermore, the contrast between the hot primary and a substellar companion is smaller in the infrared. Hence, close companions separated from their primary star by only a few arcsec, can be detected.
At the beginning of 2002 we started our multiplicity study of exoplanet hosts stars in the northern sky with a first imaging campaign carried out at the Calar Alto observatory (Spain). We selected as targets all exoplanet host stars, published before 2002 and which are observable with small airmasses (AM $\le$ 1.5) from Calar Alto (37$^{\circ}$ latitude), i.e. in total 44 exoplanet host stars. The first observations were obtained in February 2002 followed by a further observing run in July 2002. The third and final run was scheduled for September 2002 but clouded out and no data could be taken. In total we have observed 18 exoplanet host stars. Most of the remaining 26 stars were observed in the meantime with either UKIRT[^1] on Hawaii and/or NTT[^2] on La Silla.
The Calar Alto Survey - Results
===============================
Our direct imaging campaign was carried out with the 2.2m telescope at the Calar Alto observatory, using the near-infrared imager MAGIC. This camera is equipped with a 256x256 HgCdTe-detector with a pixelscale of 640mas in its high resolution mode, i.e. 164x164arcsec field of view.
All MAGIC images were astrometrically calibrated using the 2MASS[^3] Point source catalogue (Cutri et al. 2003), yielding the detector pixelscale and the offset in the position angle (PA). The results of the astrometric calibration for both MAGIC runs are summarized in Table\[tab1\].
------- --------------- ----------------
epoch pixelscale offset PA
\[mas/pixel\] \[$^\circ$\]
02/02 640.2$\pm$3.8 0.01$\pm$0.19
07/02 639.1$\pm$3.8 -0.21$\pm$0.26
------- --------------- ----------------
: Astrometric calibration of all observing runs.[]{data-label="tab1"}
Because most of the observed exoplanet host stars are nearby relatively bright stars the integration time had to be reduced to only a few tenth of a second, in order to avoid saturation effects of the detector. Many of these short exposures were averaged to one frame with a total integration time of $\sim$3 minutes. The telescope was then moved by a few arcsec and the integration procedure was repeated. With this jitter observing technique the bright infrared sky background could effectively be subtracted from the images. To correct for the individual pixel sensitivity all frames were flatfielded using sky flat frames which were taken at the beginning of the night in twilight. Background subtraction, flatfielding, image registration, shifting and final averaging of all images were achieved with the data reduction package ESO eclipse [^4] (Devillard 2001).
The achieved observational data of all exoplanet host stars observed with MAGIC in the two observing runs carried out in January and July 2002 are summarized in Table\[tab2\]. We list the total integration time per target (*time*) and the average seeing measured in each MAGIC image (*seeing*). In general, after 30 min of integration a detection limit (S/N=3) of H=18mag was reached. With the given distances of the exoplanet host stars we can convert the detection limits to the absolute magnitudes of the faintest detectable companions. The mass of these companions can be approximated with theoretical models, using mass-magnitude relations therein. With Baraffe et al. (2003) models and an average system age of 5Gyr we estimate to be sensitive to substellar companions down to $\sim$60M$\rm_{Jup}$. Table\[tab2\] shows the derived mass limit of any detectable companions as well as the inner and outer detection radii (*r$\rm_{in}$*, *r$\rm_{out}$*) of stellar companions in all MAGIC images. The inner limit depends on the brightness of the primary and on the seeing. The outer radius is only limited by the MAGIC field of view. In average additional stellar companions can be detected around all targets in a range of separations between about 270AU and 2500AU.
As a typical example, we plot in Fig.[\[limit\]]{} the achieved detection limit as a function of separation to the exoplanet host star HD52265. Stellar companions can be detected from 5.5arcsec (154AU) up to 67.8arcsec (1904AU). A limiting magnitude of 17.9mag is reached beyond 14arcsec (393AU). This is about $\sim$1.5mag deeper than the 2MASS detection limit, i.e. all objects detected in 2MASS are also visible in our MAGIC images. However close and faint companions around all these stars are not accessible for our wide field imaging. For example with AO imaging Patience et al. (2002) could detected a close companion of HD114762 (H$\sim$13.4mag) which is located only 3.3arcsec north-east of the exoplanet host star. This object is not visible in our MAGIC image as expected from the derived detection limit.
Because real companions of the exoplanet host stars are co-moving to their parent stars, i.e. both objects form a common proper motion pair, they can be distinguished from unrelated slow or non-moving background stars. By comparing all our MAGIC images with images from 2MASS and POSS[^5] I and II, the companionship of all detected objects in the MAGIC images can be checked. Among the 18 observed exoplanet host stars, no further so far unknown co-moving companions could be detected around the target stars.
However our sample contains four systems which are listed as binaries in the Washington Visual Double Star (WDS) catalog (Worley 1997). The MAGIC images of these four systems are shown in Fig.\[pics\]. The measured separations and position angles are summarized in Table\[tab3\] together with the distances of the exoplanet host stars. Thereby the distances of HD80606, 55Cnc and HD46375 can be derived from Hipparcos parallaxes (Perryman et al. 1997). BD$-$10$^{\circ}$3166 is not listed in the Hipparcos catalogue therefore no accurate parallax is available for this star. To estimate the distance of BD$-$10$^{\circ}$3166 we use its spectral type, K0V, derived by Butler et al. (2000), as well as photometric data in different filters (B=10.727$\pm$0.081mag, V=10.000$\pm$0.044mag[^6], J=8.611$\pm$0.032mag, H=8.300$\pm$0.040mag, and K$\rm{_{S}}$=8.124$\pm$0.026mag[^7]). The apparent photometry of BD$-$10$^{\circ}$3166 is consistent with a K0V dwarf at a distance of 67$\pm$3pc.
The proper motions of both components of the four observed WDS binaries are already given by the Hipparcos, the USNO-B1.0 (Monet et al. 2003) and the UCAC2 catalogues (Zacharias et al. 2004). Only the secondary of HD46375 is not listed in any of these catalogues. HD46375 is not well resolved in the POSS images but the companion is clearly separated from the bright primary star in the 2MASS image. By comparing this image with our MAGIC image we can derive the proper motion of HD46375B with precision which is limited mainly by the 2MASS astrometric accuracy. We summarize the proper motions of the four WDS binaries in Table\[tab4\].
The 2MASS point source catalogue provides accurate near infrared photometry of the secondaries and Kharchenko (2001) lists V band magnitudes for HD80606B and BD$-$10$^{\circ}$3166B. The V band magnitudes of 55CncB and HD46375B are listed in the WDS catalogue but no photometric uncertainties are given there (see Table\[photodata\]).
Discussion
==========
We observed 18 exoplanet host stars with MAGIC at Calar Alto. New co-moving companions could not be detected but 4 binary systems were observed which are already listed in the Washington Visual Double star catalogue (WDS). The proper motions of the primary and the secondary components of these binary systems are summarized in Table\[tab4\]. Both components of the WDS binaries HD80606, HD46375, and 55Cnc share a common proper motion. Therefore the companionship of these systems is confirmed by astrometry.
The WDS binary BD$-$10$^{\circ}$3166 consists of two high proper motion stars which exhibit proper motions being similar in right ascension (Ra) but significantly differ in Declination (Dec). According to the astrometric UCAC2 catalogue (see Table\[tab4\]) the motion of the secondary relative to the primary is $-$12$\pm$4.7mas/yr in Ra but $-$86.8$\pm$4.6mas/yr in Dec, i.e. a total relative motion of 87.6$\pm$4.6mas/yr. Comparing our MAGIC image with the 2MASS and the POSS-I and POSS-II images yields a similar result. We derive a motion of the secondary relative to the primary component of $-$13$\pm$7mas/yr in Ra and $-$79$\pm$7mas/yr in Dec. Therefore we can conclude that this WDS binary is clearly not a common proper motion pair.
In a next step we can test the companionship of the four WDS binaries with photometry. We derive the absolute magnitudes of the four secondaries using their apparent magnitudes. Thereby we always assume that the companions are located at the distances of the exoplanet host stars, as it is expected for real companions. For all secondaries accurate 2MASS near infrared photometry is available and Kharchenko (2001) provides V band magnitudes of HD80606B and BD$-$10$^{\circ}$3166B. The V band magnitudes of 55CncB and HD46375B (photometric uncertainties are not available) are listed in the WDS catalogue (see Table\[photodata\]). We plot all secondaries in a J-K$\rm{_{S}}$, M$_{\rm H}$ diagram (upper panel of Fig.\[photoir\]), together with comparison dwarfs from the Hipparcos catalogue, the Nearby Stars catalogue (Gliese & Jahreiß 1995), and cool dwarfs from Cruz et al. (2003). Only comparison objects with accurate colors ($\rm\sigma(J-K_{S})<0.05$) and accurate absolute magnitudes ($\rm\sigma(M_{\rm H})<0.25$) are plotted. Furthermore we plot all secondaries in a V-K$\rm{_{S}}$, M$_{\rm K\rm{_{S}}}$ diagram (see bottom panel of Fig.\[photoir\]), using again the same Hipparcos comparison dwarfs as in the infrared color-magnitude diagram, as well as M dwarfs from Leggett et al. (1992).
According to the intrinsic colors for dwarfs and giants published by Tokunaga (2000) the V-K$\rm{_{S}}$ and J-K$\rm{_{S}}$ colors of BD$-$10$^{\circ}$3166B are both fully consistent with a M4 to M5 dwarf. Nevertheless if we assume that this object is indeed located at the distance of the exoplanet host star the derived absolute K$\rm{_{S}}$ band magnitude is about 3.5 mag brighter than comparison dwarfs with the same V-K$\rm{_{S}}$ color (see the bottom panel of Fig.\[photoir\]), i.e. the distance of this object is overestimated by a factor of 5. The apparent infrared and V band photometry of BD$-$10$^{\circ}$3166B is consistent with a M4 to M5 dwarf located at a distance of about 13pc. Therefore we can conclude that the WDS binary BD$-$10$^{\circ}$3166 is only a visual pair of stars, consisting of a K0 dwarf in the background at a distance of 67pc and an unrelated foreground M dwarf at a distance of about 13pc.
As it is illustrated in the color-magnitude diagrams shown in Fig.\[photoir\] the photometry of the secondaries of the common proper motion pairs HD80606, HD46375 and 55Cnc is consistent with dwarfs located at the distances of the exoplanet host stars, hence the companionship of the these three WDS binaries is confirmed by both astrometry and photometry.
The masses of the three secondaries HD80606B, 55CncB, and HD46375B can be derived by converting their absolute infrared magnitudes to masses, using the evolutionary Baraffe et al. (1998) models. Thereby we always assume a system age of 5Gyr, which is a good estimate for most of the exoplanet host stars. It is important to mention that, the age uncertainty of the binary systems in the order of a few Gyr is not important here because for system ages between 1 and 10Gyr the infrared magnitudes of stars with masses below one solar mass are not strongly age dependant. The derived companion masses range from $\rm{0.9\,M_{\odot}}$ to $\rm{0.27\,M_{\odot}}$, with mass ratios (M$\rm_{c}/M\rm_{p}$) between 0.3 to 0.87.
Finally with the measured angular separations and known distances of the binary systems we can derive their projected separations. Because neither the orientation of the binary orbit to the line of sight nor the position of both stars on their orbits around their common barycenter is known, we always use the observed projected separation also as estimate of the binary semi-mayor axis. The separation of the three binaries range from 350 up to 1200AU.
According to Holman & Wiegert (1998) objects, e.g. additional companions, located in a binary system are only longterm stable if their semi-major axes do not exceed a critical value, the critical semi-major axis ($\rm{a_{c}}$), which depends on the binary semi-major axis, its eccentricity as well as its mass-ratio. The critical semi-major axes of possible additional companions in the three binary systems are listed in Table\[tab5\]. They range from about 400AU for an assumed circular binary orbit down to only 10AU for an eccentric binary system, respectively. Our observations didn’t reveal any further, wide co-moving companions in these systems, as expected from the Holman & Wiegert (1998) stability criteria.
Eggenberger et al. (2004) have recently compared the statistical characteristics of planets in binary systems with those orbiting single stars. They found that the distribution of the masses of binary-star planets with periods shorter than 40 days is approximately flat, whereas single-star planets all exhibit masses less than $\rm{2M_{Jup}}$. Furthermore, Eggenberger et al. (2004) found that all known close binary-star planets have almost circular orbits (e$<$0.05), while eccentric orbits are only detected among single-star planets.
However, these statistical differences are based only on a small number of known binary-star planets, and therefore the significance of these differences is still not clear. Note that the whole sample of the exoplanet host stars has not been systematically surveyed so far for close or wide companions. Only systematic search programs for companions can clarify the multiplicity status of the stars in the sample.
In our study we have already shown that there are indeed several exoplanet host stars considered as single stars in the published statistical analyses which emerge as binary systems (see e.g. Mugrauer et al. 2005b). A further example of these former unknown or unconfirmed binary systems among the exoplanet host stars is HD46375, whose binary nature was confirmed here for the first time with astrometry as well as photometry. The exoplanet in that binary system was detected by Marcy et al. (2000). It is a hot Jupiter (Msin(i)=0.249$\rm{M_{Jup}}$) which revolves its parent star in only 3.024 days on an almost circular (e=0.04) orbit, typical for such a short period binary-star planet.
The Calar Alto imaging survey was just the beginning of our multiplicity study of the exoplanet host stars, using a 2m class telescope. Only a relative small number of exoplanet host stars was observed. We expanded our search for additional wide companions of exoplanet host stars, using larger mirrors to be sensitive to fainter companions. On the northern sky we use the UKIRT on Hawaii and southern targets are observed with the NTT on La Silla, and since begin of 2005 also with the VLT on Paranal. The results of these surveys finally will clarify the multiplicity status of most of the exoplanet host stars and will verify the significance of the reported statistical differences between single-star and binary-star planets.
Baraffe, I., Chabrier, G., Allard, F., Hauschildt, P.H: 1998, A&A 337, 403 Baraffe, I., Chabrier, G., Barman, T.S., Allard, F., Hauschildt, P.H.: 2003, A&A 402, 701 Butler, R.P., Vogt, S.S., Marcy, G.W., Fischer, Debra, A., Henry, G.W., Apps, K.: 2000, ApJ 545, 504 Cruz, K.L., Reid, I.N., Liebert, J., Kirkpatrick, J.D., Lowrance, P.J.: 2003, AJ 126, 2421 Cutri, R.M., Skrutskie, M.F., van Dyk, S., Beichman, C.A., Carpenter, J.M., Chester, T., Cambresy, L., Evans, T., et al.: 2003, yCat 2246, 0 Devillard, N.: 2001, ASPC 238, 525 Els, S.G., Sterzik, M.F., Marchis, F., Pantin, E., Endl, M., Kürster, M.: 2001, A&A 370, 1 Eggenberger, A., Udry, S., Mayor, M.: 2004, A&A 417, 353 Gliese, W., Jahreiß, H.: 1995, yCat 5070, 0 Hatzes, A.P., Wuchterl, G.: Nature 436, 182 Holman, M.J., Wiegert, P.A.: 1999, AJ 117, 621 Jahreiß, H.: 2001, AGM 18, 110 Jang-Condell, H.: 2005, submitted to ApJ, astro-ph/0507356 Kharchenko, N.V.: 2001, KFNT 17, 409 Leggett, S.K.: 1992, ApJS 82, 351 Lowrance, P.J., Kirkpatrick, J.D., Beichman, C.A.: 2002, ApJ 572, 79 Marcy, G.W., Butler, R.P., Vogt, S.S.: 2000, ApJ 536, 43 Mugrauer, M., Neuhäuser, R., Mazeh, T., Guenther, E., Fernández, M.: 2004a, AN 325, 718 Mugrauer, M., Neuhäuser, R., Mazeh, T., Alves, J., Guenther, E.: 2004b, A&A 425, 249 Mugrauer, M., Neuhäuser, R.: 2005a, MNRAS 361, 15 Mugrauer, M., Neuhäuser, R., Seifahrt, A., Mazeh, T., Guenther, E.: 2005b, A&A 440, 1051 Monet, D.G., Levine, S.E., Canzian, B., Ables, H.D., Bird, A.R., Dahn, C.C., Guetter, H.H., Harris, H.C.: 2003, AJ 125, 984 Patience, J., White, R.J., Ghez, A.M., McCabe, C., McLean, I.S., Larkin, J.E., Prato, L., Kim, S.S., et al.: 2002, ApJ 581, 654 Perryman, M., Lindegren, L., Kovalevsky, J., Hoeg, E., Bastian, U., Bernacca, P.L., Crézé, M., Donati, F., et al.: 1997, A&A 323, 49 Pichardo, B., Sparke, L.S., Aguilar, L.A.: 2005, MNRAS 359, 521 Queloz, D., Mayor, M., Weber, L., Blécha, A., Burnet, M, Confino, B., Naef, D., Pepe, F., et al.: 2000, A&A 354, 99 Santos, N.C., Israelian, G., Mayor, M.: 2004, A&A 415, 1153 Tokunaga, A.T.: 2000, Allen’s Astrophysical Quantities, 4th edition, ed. A.N. Cox, Springer-Verlag, NY, p. 143 Worley, C.E., Douglass, G.G.: 1997, A&A 125, 523 Zacharias, N., Urban, S.E., Zacharias, M.I., Wycoff, G.L., Hall, D.M., Monet, D.G., Rafferty, T.J.: 2004, AJ 127, 3043 Zucker, S., Mazeh, T.: 2002, ApJ 568, 113
[^1]: United Kingdom Infrared Telescope
[^2]: New Technology Telescope
[^3]: 2 Micron All Sky Survey
[^4]: ECLIPSE: ESO C Library for an Image Processing Software Environment
[^5]: Palomar All Sky Survey
[^6]: B,V magnitudes from Kharchenko (2001)
[^7]: J,H,K$\rm{_{S}}$ from the 2MASS point source catalogue
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
The HERMES Collaboration\
\
A. Airapetian$^{30}$, N. Akopov$^{30}$, I. Akushevich$^{7}$, M. Amarian$^{23,25,30}$, J. Arrington$^{2}$, E.C. Aschenauer$^{7,13,23}$, H. Avakian$^{11,a}$, R. Avakian$^{30}$, A. Avetissian$^{30}$, E. Avetissian$^{30}$, P. Bailey$^{15}$, B. Bains$^{15}$, C. Baumgarten$^{21}$, M. Beckmann$^{12}$, S. Belostotski$^{24}$, S. Bernreuther$^{9}$, N. Bianchi$^{11}$, H. Böttcher$^{7}$, A. Borissov$^{6,14,19}$, M. Bouwhuis$^{15}$, J. Brack$^{5}$, S. Brauksiepe$^{12}$, B. Braun$^{9,21}$, W. Brückner$^{14}$, A. Brüll$^{14,18}$, P. Budz$^{9}$, H.J. Bulten$^{17,23,29}$, G.P. Capitani$^{11}$, P. Carter$^{4}$, P. Chumney$^{22}$, E. Cisbani$^{25}$, G.R. Court$^{16}$, P.F. Dalpiaz$^{10}$, R. De Leo$^{3}$, L. De Nardo$^{1}$, E. De Sanctis$^{11}$, D. De Schepper$^{2,18}$, E. Devitsin$^{20}$, P.K.A. de Witt Huberts$^{23}$, P. Di Nezza$^{11}$, V. Djordjadze$^{7}$, M. Düren$^{9}$, A. Dvoredsky$^{4}$, G. Elbakian$^{30}$, J. Ely$^{5}$, A. Fantoni$^{11}$, A. Fechtchenko$^{8}$, M. Ferro-Luzzi$^{23}$, K. Fiedler$^{9}$, B.W. Filippone$^{4}$, H. Fischer$^{12}$, B. Fox$^{5}$, J. Franz$^{12}$, S. Frullani$^{25}$, Y. Gärber$^{7}$, F. Garibaldi$^{25}$, E. Garutti$^{10,23}$, G. Gavrilov$^{24}$, V. Gharibyan$^{30}$, A. Golendukhin$^{6,21,30}$, G. Graw$^{21}$, O. Grebeniouk$^{24}$, P.W. Green$^{1,27}$, L.G. Greeniaus$^{1,27}$, A. Gute$^{9}$, W. Haeberli$^{17}$, M. Hartig$^{27}$, D. Hasch$^{7,11}$, D. Heesbeen$^{23}$, F.H. Heinsius$^{12}$, M. Henoch$^{9}$, R. Hertenberger$^{21}$, W.H.A. Hesselink$^{23,29}$, P. Hoffmann-Rothe$^{23}$, G. Hofman$^{5}$, Y. Holler$^{6}$, R.J. Holt$^{15}$, B. Hommez$^{13}$, W. Hoprich$^{14}$, G. Iarygin$^{8}$, H. Ihssen$^{6,23}$, M. Iodice$^{25}$, A. Izotov$^{24}$, H.E. Jackson$^{2}$, A. Jgoun$^{24}$, R. Kaiser$^{7,26,27}$, J. Kanesaka$^{28}$, E. Kinney$^{5}$, A. Kisselev$^{24}$, P. Kitching$^{1}$, H. Kobayashi$^{28}$, N. Koch$^{9}$, K. Königsmann$^{12}$, M. Kolstein$^{23}$, H. Kolster$^{21,23}$, V. Korotkov$^{7}$, E. Kotik$^{1}$, V. Kozlov$^{20}$, V.G. Krivokhijine$^{8}$, G. Kyle$^{22}$, L. Lagamba$^{3}$, A. Laziev$^{23,29}$, P. Lenisa$^{10}$, T. Lindemann$^{6}$, W. Lorenzon$^{19}$, N.C.R. Makins$^{2,15}$, J.W. Martin$^{18}$, H. Marukyan$^{30}$, F. Masoli$^{10}$, M. McAndrew$^{16}$, K. McIlhany$^{4,18}$, R.D. McKeown$^{4}$, F. Meissner$^{7}$, F. Menden$^{12,27}$, A. Metz$^{21}$, N. Meyners$^{6}$, O. Mikloukho$^{24}$, C.A. Miller$^{1,27}$, R. Milner$^{18}$, V. Mitsyn$^{8}$, V. Muccifora$^{11}$, R. Mussa$^{10}$, A. Nagaitsev$^{8}$, E. Nappi$^{3}$, Y. Naryshkin$^{24}$, A. Nass$^{9}$, W.-D. Nowak$^{7}$, T.G. O’Neill$^{2}$, R. Openshaw$^{27}$, J. Ouyang$^{27}$, B.R. Owen$^{15}$, S.F. Pate$^{18,22,b}$, S. Potashov$^{20}$, D.H. Potterveld$^{2}$, G. Rakness$^{5}$, R. Redwine$^{18}$, D. Reggiani$^{10}$, A.R. Reolon$^{11}$, R. Ristinen$^{5}$, K. Rith$^{9}$, D. Robinson$^{15}$, M. Ruh$^{12}$, D. Ryckbosch$^{13}$, Y. Sakemi$^{28}$, I. Savin$^{8}$, C. Scarlett$^{19}$, C. Schill$^{12}$, F. Schmidt$^{9}$, M. Schmitt$^{9}$, G. Schnell$^{22}$, K.P. Schüler$^{6}$, A. Schwind$^{7}$, J. Seibert$^{12}$, T.-A. Shibata$^{28}$, T. Shin$^{}$, V. Shutov$^{8}$, C. Simani$^{10,23,29}$, A. Simon$^{12,22}$, K. Sinram$^{6}$, E. Steffens$^{9}$, J.J.M. Steijger$^{23}$, J. Stewart$^{16,27}$, U. Stösslein$^{7}$, K. Suetsugu$^{28}$, M. Sutter$^{18}$, H. Tallini$^{}$, S. Taroian$^{30}$, A. Terkulov$^{20}$, S. Tessarin$^{10}$, E. Thomas$^{11}$, B. Tipton$^{18,4}$, M. Tytgat$^{13}$, G.M. Urciuoli$^{25}$, J.F.J. van den Brand$^{23,29}$, G. van der Steenhoven$^{23}$, R. van de Vyver$^{13}$, J.J. van Hunen$^{23}$, M.C. Vetterli$^{26,27}$, V. Vikhrov$^{24}$, M.G. Vincter$^{27,1}$, J. Visser$^{23}$, E. Volk$^{14}$, C. Weiskopf$^{9}$, J. Wendland$^{26,27}$, J. Wilbert$^{9}$, T. Wise$^{17}$, K. Woller$^{6}$, S. Yoneyama$^{28}$, H. Zohrabian$^{30}$,
date: 'Received: April 25, 2000; Revised: July 3, 2000'
title: 'Exclusive Leptoproduction of $\rho^0$ Mesons from Hydrogen at Intermediate Virtual Photon Energies'
---
Introduction {#sec:intro}
============
This paper presents cross section measurements for exclusive diffractive production of $\rho^0$(770) vector mesons in positron scattering on a $^1$H target. The production of vector mesons by real or virtual photons is of considerable interest, as the corresponding cross section is closely related to other observables in lepton scattering. For example, with help of the recently introduced Off-Forward Parton Distributions (OFPDs), one can relate elastic nucleon form factors, deep-inelastic scattering (DIS) structure functions, virtual Compton scattering cross sections, and vector meson production cross sections [@Rady; @Ji; @Collins; @FS96; @MvdH; @MvdH2; @Mank1; @Mank2]. The OFPDs represent a generalization of the parton distributions measured, for instance, in inclusive DIS experiments.
In Refs. [@MvdH; @MvdH2] the longitudinal part of the $\rho^0$ virtual-photoproduction cross section is calculated in the OFPD framework. At large values of the photon-nucleon invariant mass ($W >$ 10 GeV) the calculated cross section is dominated by a two-gluon exchange mechanism, which has been treated using the perturbative approach of Ref. [@FS96]. These calculations reproduce existing data. However, uncertainties arise due to the size of higher-order and higher-twist contributions [@FS96; @Mank1; @Mank2; @Beli99], which are larger at small values of $Q^2$ (the negative square of the four-momentum of the virtual photon). Below 10 GeV the calculated cross section is dominated (in leading twist) by a handbag diagram, in which the virtual photon is absorbed by a valence quark in the nucleon [@MvdH; @MvdH2]. Following Ref. [@MvdH] this mechanism is called quark exchange. However, few data are available between 4 and 10 GeV. Several data sets exist for $W <$ 4 GeV, but in this domain the OFPD calculations do not apply, as the reaction receives contributions from other reaction channels in this domain.
The relation between the vector meson leptoproduction cross section and the structure function $F_2^{\mathrm p}(x)$ of the proton (with $x$ the Bjorken scaling variable) is apparent in a model calculation by Haakman et al. [@ref4] based on Reggeon field theory. In this model the $x$-dependence of $F_2^{\mathrm p}(x)$ is related to the $W$-dependence of the cross section for vector meson leptoproduction. The model gives a good description of $\gamma^* p \rightarrow \rho^0 p$ cross sections for $W$ ranging from 10 to 180 GeV. Between 4 and 10 GeV the calculated shape of the $W$-dependence changes significantly with $Q^2$. The model has not yet been properly assessed in this domain due to lack of data.
The two examples above illustrate the need for additional exclusive $\rho^0$ leptoproduction data in the $W$-range between 4 and 10 GeV. Such data are reported in the present paper. Exclusive cross sections have been measured for $W$ values between 4.0 and 6.0 GeV, and $Q^2$ values between 0.7 and 5.0 GeV$^2$. The kinematical variables used to describe $\rho^0$ leptoproduction are introduced in the next section. Experimental details are provided in section 3, while the data analysis is discussed in section 4. The experimental results are compared to various model calculations in section 5, and the paper is summarized in the last section.
Kinematics {#sec:kinem}
==========
In the present leptoproduction measurement both the scattered lepton and two oppositely charged hadrons are observed. The $\rho^0$ leptoproduction events are identified by requiring that the reconstructed invariant mass of the two hadrons is close to the mass of the $\rho^0$ meson, i.e. 0.77 GeV. The kinematic variables that characterize such measurements are defined here in the laboratory frame. The variables describing the kinematics of the virtual photon include its energy $\nu$, its fractional energy $y = \nu / E_0$ (with $E_0$ the incident lepton energy), and its four-momentum $q$. The latter quantity is related to the previously introduced variable $Q^2$ through $Q^2 = -q^2 > 0$. The Bjorken scaling variable $x$ is related to $\nu$ and $Q^2$ by $x = Q^2/(2 M \nu)$, and the photon-nucleon invariant mass $W$ is given by $W^2 = M^2 + 2 M \nu - Q^2$. The virtual-photon polarization parameter is represented by $\epsilon$.
The $\rho^0$ meson is characterized by its four-momentum $v$ and $\theta_{\pi\pi}$, which represents the angle between the two pions into which the $\rho^0$ meson decays.
Combining some of the lepton and meson variables, one can introduce $t = (q - v)^2 < 0$, the square of the four-momentum exchange between the virtual photon and the target, and $\Delta E = (P_{\mathrm Y}^2 - M^2)/2M$, which is a measure of the missing energy (where $P_{\mathrm Y} = P + q - v$ represents the 4-momentum of the unobserved final state $\mathrm Y$ with $P = (M,0)$ that of the target nucleon).
Instead of $-t$, the above-threshold momentum transfer $-t' = -t + t_0$ is often used, which is approximately equal to $p_t^2$ – the square of the transverse momentum of the $\rho^0$ meson with respect to the direction $\vec{q}$. In this expression $-t_0$ represents the minimum value of $-t$ for fixed values of $\nu$, $Q^2$ and $P_{\mathrm Y}^2$.
Experiment {#sec:data}
==========
The data were collected during the 1996 and 1997 running periods of the HERMES experiment [@specpaper] at DESY using a 27.5 GeV longitudinally polarized positron beam with a $^1$H gas target in the HERA storage ring. Part of the data set was collected with longitudinally polarized targets. Since the polarization degrees of freedom were not exploited in the present analysis, the average over both target polarization states is taken.
The HERMES polarized proton target [@ABStarget] is formed by injecting a nuclear-polarized beam of atomic hydrogen from an atomic beam source into a tubular open-ended storage cell inside the positron ring. The cell provides a 40 cm long target of pure atomic species with an areal density of approximately 7 $\times$ 10$^{13}$ atoms/cm$^2$. With unpolarized targets of molecular hydrogen, areal densities of about 10$^{15}$ atoms/cm$^2$ were obtained. The storage cell was shielded from synchrotron radiation by two sets of collimators, one of which is moveable. No particles were observed to originate from scattering events in the walls of the storage cell.
During the course of one fill (typically 8 hours long), the positron current in the ring decreased from typically 30 – 40 mA at injection to $\sim 10$ mA, at which point the ring was emptied. The data presented in this paper correspond to an integrated luminosity of 108 pb$^{-1}$.
The HERMES spectrometer is described in detail elsewhere [@specpaper]. It is a forward spectrometer in which both the scattered positron and produced hadrons are detected within an angular acceptance $\pm$ 170 mrad horizontally, and $\pm$ (40 – 140) mrad vertically. The scattered-positron trigger was formed from a coincidence between a pair of scintillator hodoscope planes and a lead-glass calorimeter. The trigger required an energy of more than 3.5 GeV deposited in the calorimeter. (For part of the running the trigger threshold was reduced to 1.5 GeV, for which a correction was applied in the data analysis.) Positron identification was accomplished using the calorimeter, the preshower counter consisting of the second hodoscope preceded by a lead sheet, a transition-radiation detector, and a threshold gas $\breve {\mathrm C}$erenkov counter. This system provided positron identification with an average efficiency of 99% and a hadron contamination of less than 1%.
Data analysis
=============
Only those events were selected that contained a scattered positron and exactly two hadrons with opposite charge. (A more detailed description of the analysis is given in Refs. [@Machiel; @belz].) A number of geometric requirements were imposed on the particle tracks to ensure that they were well contained within the acceptance of the spectrometer. It was also required that the tracks originated from along the beam line within $\pm$ 18 cm of the centre of the target. In addition, several constraints were imposed on the kinematic variables. The size of the radiative corrections was limited by requiring $y \le$ 0.85. Because the $W$-acceptance of the HERMES spectrometer for $\rho^0$ production is sharply reduced both below 4 GeV and above 6 GeV, cross sections for only two $W$-bins (4–5 and 5–6 GeV) have been extracted from the data.
The $\rho^0$ vector mesons were identified by requiring 0.6 GeV $< M_{\pi \pi} <$ 1 GeV, with $M_{\pi \pi}$ the invariant mass of the pair of detected hadrons, assuming that they are pions. It has been verified that this requirement also removes the $\phi \rightarrow \mathrm{K^{+} K^{-}}$ background, by confirming that the $\phi$ events, which occur at $M_{\mathrm K \mathrm K}$ $\approx$ $M_{\phi}$, appear in the $M_{\pi \pi}$ spectrum at $M_{\pi \pi} <$ 0.6 GeV. Here $M_{\mathrm K \mathrm K}$ is the invariant mass calculated assuming that the two hadrons are kaons, and $M_{\phi}$ = 1.019 GeV is the mass of the $\phi$ meson.
![ Missing energy spectra for leptoproduction of $\rho^0$ mesons off $^1$H at an incident energy of 27.5 GeV. The upper (lower) panel corresponds to the $W$-range between 4 and 5 (5 and 6) GeV. The data have been selected requiring 0.6 GeV $< M_{\pi\pi} <$ 1 GeV and $-t' <$ 0.4 GeV$^2$. The solid histograms represent the results of a Monte Carlo simulation [@LEPTO] of the background, which has been scaled to the data for $\Delta E >$ 3.0 GeV. The dotted histogram includes a simulation [@belz] of exclusive $\rho^0$-production as well. As the Monte Carlo simulations do not include inelastic nucleon excitations or internal radiative effects for the exclusive channel, the $\Delta E$-region between 0.5 and 2.5 GeV cannot be properly described. []{data-label="fig:deltaE"}](figdE5.epsi){width="47.00000%"}
In exclusive $\rho^0$ electroproduction, i.e. $e N \rightarrow e' \rho^0 N$, the part of the final state that is unobserved at HERMES consists of a nucleon recoiling without excitation. Such events are selected by requiring that the missing energy $\Delta E$ is approximately zero. In this domain the $\Delta E$ spectrum, as displayed in Fig. \[fig:deltaE\] for the two $W$-bins, shows a clear peak near $\Delta E \approx$ 0. The exclusive events were selected by requiring $\Delta E$ $<$ 0.4 GeV.
Using the Monte Carlo simulation for exclusive $\rho^0$ production described in section 4.2, it has been evaluated that on average 18% of the exclusive events fall outside the imposed $\Delta E$ window. This effect has been accounted for (bin by bin) in the acceptance correction.
The requirement on $\Delta E$ is relatively tight in order to suppress inelastic contributions involving nucleon excitations. As no Monte Carlo simulation is available for such double diffractive processes, their remaining contribution is estimated from other data in section 4.3.
Background subtraction
----------------------
There is a background contribution under the exclusive $\rho^0$ peak that is caused by hadrons from DIS fragmentation processes. Part of this background is removed by excluding the region $-t'$ $>$ 0.4 GeV$^2$, where the background dominates the $\rho^0$ yield. The remaining background is estimated and subtracted using the methods described below.
The background estimate is based on the LEPTO Monte Carlo program [@LEPTO]. This simulation includes hadrons resulting from fragmentation processes in deep-inelastic scattering (using the Lund fragmentation code), but not from diffractive $\rho^0$ production. The $\Delta E$ spectrum was generated independently for each ($Q^2$,$W$)-bin, and normalized to the experimental data in the $\Delta E$ region above 3.0 GeV (see Fig. \[fig:deltaE\]). Thus normalized, the background was subtracted from the data in the $\Delta E$ domain that was used for the determination of the cross section. Depending on kinematics the background contribution ranges from (3 $\pm$ 1)% to (9 $\pm$ 4)%, where the uncertainties include the effect of varying the $\Delta E$-ranges used in evaluating the background contribution. The lower limit of the $\Delta E$-range used to normalize the background was varied from 3 to 5 GeV, and the upper limit used in selecting the exclusive events was varied from 0.3 to 0.6 GeV.
The background subtraction procedure was verified by also using an alternative method. In this case a $\Delta E$ spectrum was extracted from the data by requiring $-t'$ values between 0.7 and 5 GeV$^2$, i.e. well beyond the region used to determine the diffractive cross section. A similar normalization procedure is used, i.e. the background spectrum is normalized such as to reproduce the same number of events for $\Delta E >$ 3 GeV as in the spectrum obtained with the standard $-t'$ requirement. The tail of the normalized background spectrum for $\Delta E <$ 0.4 GeV was taken to represent the actual background contribution to the $\rho^0$ mass peak. The cross sections obtained after this background subtraction method were found to be consistent with those obtained using the Monte-Carlo based method within the systematic errors listed below in Table \[tab:cross\].
Acceptance correction
---------------------
The data were corrected for the finite acceptance and inefficiencies of the HERMES spectrometer for $\rho^0$ production. The correction was evaluated using a Monte Carlo simulation based on the Vector Meson Dominance (VMD) model. Details are given in Ref. [@belz]. For each kinematic variable ($\nu$, $Q^2$, $W$, $M_{\pi\pi}$, $\cos{\theta_{\pi\pi}}$ and $-t'$) good agreement between the measured distribution and the simulation was obtained.
The acceptance correction is large because the HERMES spectrometer has a relatively small acceptance for $\gamma^* p \rightarrow
\rho^0 p$ events. The acceptance for $\rho^0$ mesons (given an observed scattered positron) ranges from only 5.9% at low $Q^2$ to 18% at high $Q^2$. As the corresponding correction factors are large, it is mandatory to study the dependence of the acceptance correction on the details of the Monte Carlo calculation. For that purpose, the entire analysis was redone varying the assumptions for the angular distributions of the production and decay of the $\rho^0$-meson. The original Monte Carlo simulation, which assumed s-Channel Helicity Conservation (SCHC), was replaced by one that produces uniform angular distributions for all relevant angles. In addition, the $Q^2$-dependence of the VMD propagator (as given by Eq. \[eq:res-fittingf\] in section 5.1) was varied by changing the exponent from 2 to 2.5, and an alternative event generator DIPSI [@dipsi] was used. The sensitivity of the Monte Carlo yield to variations of the angular distributions and changes of the assumed $Q^2$-dependence leads to typical systematic uncertainties of 10% and 6%, respectively. The effect of the event generator itself amounts to typically 8%. Including all variations, the uncertainty of the cross sections due to the acceptance correction ranges from 12% to 17%, depending on the kinematics.
Cross section determination
---------------------------
![ Invariant mass spectrum of the two oppositely charged hadrons assuming that they are pions. Backgrounds have been subtracted, the contribution from $\phi$ decays has been removed by requiring that $M_{\mathrm K \mathrm K}$ $>$ 1.04 GeV, and the acceptance correction has been applied. The $\rho^0$ mass peak at 770 MeV is described by a Breit-Wigner function (solid line), of which the parameters were fitted to the data. []{data-label="fig:Mpipi"}](figMpipi.epsi){width="47.00000%"}
The $M_{\pi\pi}$ mass spectrum is shown in Fig. \[fig:Mpipi\]. In this spectrum backgrounds are subtracted, the acceptance correction is applied and all kinematic requirements mentioned above are imposed. The $\rho^0$ mass peak was fitted using a Breit-Wigner (p-wave) function and a constant background term representing the non-resonant physics background. The Breit-Wigner function was multiplied by the factor $(M_{\rho}/M_{\pi\pi})^{n_s}$ in which $n_s$ represents the skewing parameter. This procedure is identical to the one described in Ref. [@e665], and was first proposed by Ross and Stodolsky [@RS]. The resulting fit parameters ($M_{\rho^0}$ = 0.776 $\pm$ 0.003 GeV, $\Gamma_{\rho^0}$ = 0.147 $\pm$ 0.009 GeV, and $n_s$ = 2.2 $\pm$ 0.4) are in good agreement with the values published by E665 [@e665]: $M_{\rho^0}$ = 0.777 $\pm$ 0.002 GeV, $\Gamma_{\rho^0}$ = 0.146 $\pm$ 0.003 GeV, and $n_s$ = 2.7 $\pm$ 0.6 (for the same $Q^2$-region). Both experiments find a background term that is consistent with zero, and are in fair agreement with the PDG-values [@booklet]: $M_{\rho^0}$ = 0.7700 $\pm$ 0.0008 GeV, $\Gamma_{\rho^0}$ = 0.1507 $\pm$ 0.0011 GeV. If the effect of the $\omega$ decay is also considered in the fit, $\omega$ mass and width parameters are found that are also consistent with Ref. [@booklet]. The contribution of this interference averages out to zero in the considered mass interval. The absolute contribution of $\omega$-production to the apparent $\rho^0$ yield is estimated to be less than 1%, and has therefore been neglected. It should be noted that the $\rho^0$ production cross sections presented in this paper correspond to the data in the mass region 0.6 GeV $< M_{\pi\pi} <$ 1.0 GeV with a correction (of 14 $\pm$ 2%) for the tails of the skewed Breit-Wigner distribution [@RS] outside this mass window. This correction factor has been evaluated relative to the mass range between $2 M_{\pi}$ and $(M_{\rho^0} + 5 \Gamma_{\rho^0})$.
Examples of the $-t'$ distribution of the data can be found in Refs. [@belz; @ref7]. The $-t'$ distribution is fit to a falling exponential, and the fit is used to correct the measured cross section for the excluded region $-t' >$ 0.4 GeV$^2$. The slope parameter $b$ from the fit is (6.82 $\pm$ 0.15) GeV$^{-2}$, which is consistent with previously published values.
The $\rho^0$ production data were normalized using the inclusive deep-inelastic scattering (DIS) yield derived from the same data sample. By comparing the DIS yield to the LEPTO Monte-Carlo yield based on the world data [@ref6] evaluated at the appropriate $Q^2$ and $x$ values, the required normalization factors were determined. Virtual photoproduction cross sections, $\sigma_{\gamma^* p \rightarrow
\rho^0 p}$, were obtained from the leptoproduction cross section after applying the photon flux factor $\Gamma_T$ (using the Hand convention as outlined in Ref. [@e665], for instance): $$\sigma_{\gamma^* p \rightarrow \rho^0 p} = { \frac{ 1 } {\Gamma_T}
\frac{d \sigma_{e p \rightarrow e' \rho^0 p}} {d \nu dQ^2} }
\label{eq:Xsect}$$ In double-diffractive (DD) processes the target nucleon is broken up. In a recent study of target dissociation in $\rho^0$ production at center-of-mass energies between 60 and 180 GeV, the cross section ratio of double-diffractive to single-diffractive $\rho^0$ production was measured to be 0.65 $\pm$ 0.17 [@H1]. Accounting for the shape of the baryonic spectrum (taken from $\mathrm p \mathrm p$ diffractive scattering [@Akimov]) and the resolution of the present experiment, the DD contribution to the diffractive $\rho^0$ production cross section at HERMES is found to be (4 $\pm$ 2)%, for which the data were corrected. The uncertainty in the DD contribution originates from a rough estimate of the acceptance for double diffractive events relative to that for single diffractive $\rho^0$ events, and the uncertainty of the data of Ref. [@H1]. Taking into account other estimates of the ratio of double diffractive to single diffractive $\rho^0$ production also at lower $W$-values [@Holt], which are smaller in general, a total systematic uncertainty of about 3% (relative to the measured cross section) has been assigned to the DD contribution. The small size of the DD contribution is related to the good energy resolution of the HERMES experiment ($\sigma_{\Delta E}
\approx$ 0.25 GeV – see Fig. \[fig:deltaE\]).
The final cross sections are obtained by applying a radiative correction to the experimentally determined $\rho^0$ production cross section. The internal radiative correction has been evaluated separately for each ($Q^2$,$W$) bin, and typically amounts to 18% [@radcor]. External radiative effects (caused by detector materials) are included in the acceptance Monte Carlo.
Results
=======
In Table \[tab:cross\] the results for the $\rho^0$ virtual-photoproduction cross section are given for each $Q^2$ and $W$ bin. The systematic uncertainties are dominated by those from the acceptance correction factors, which amount to 17%, 11%, 14% and 17% in the bins centered at $Q^2$ values of 0.83, 1.3, 2.3 and 4.0 GeV$^2$, respectively. Another important contribution to the systematic uncertainty stems from from the uncertainty in the reconstruction and data selection efficiency, which totals about 9%. This contribution has been estimated by varying the large number of requirements used to select data and reconstruct valid events. The quoted 9% also includes uncertainties in tracking effeciency, kaon contamination, radiative corrections and DD-contribution. Remaining contributions to the total systematic uncertainty are the absolute normalization (6%) and the background subtraction (4%). The combined systematic uncertainty on the $\rho^0$ cross sections ranges from 16 to 21%, depending on the kinematics.
\[bt\]
$\langle Q^2 \rangle$ \[GeV$^2$\] $\langle W \rangle$ = 4.6 GeV $\langle W \rangle$ = 5.4 GeV
----------------------------------- ------------------------------- -------------------------------
0.83 2.46 $\pm$ 0.13 $\pm$ 0.51 2.04 $\pm$ 0.10 $\pm$ 0.43
1.3 0.92 $\pm$ 0.03 $\pm$ 0.15 1.00 $\pm$ 0.04 $\pm$ 0.16
2.3 0.43 $\pm$ 0.02 $\pm$ 0.08 0.41 $\pm$ 0.02 $\pm$ 0.07
4.0 0.16 $\pm$ 0.01 $\pm$ 0.03 0.10 $\pm$ 0.01 $\pm$ 0.02
: The measured virtual-photoproduction cross sections $\sigma_{\gamma^* p \rightarrow \rho^0 p}$ for exclusive $\rho^0$ production on $^1$H (in $\mu$b) corrected for radiative effects. Both the statistical (first) and systematic (second) uncertainties are listed. \[tab:cross\]
\[t\]
![\[fig:crossw1\] The virtual-photoproduction cross section for exclusive $\rho^0$ production versus $W$ for the indicated values of $\langle Q^2 \rangle$. The data collected in the present experiment are represented by solid circles. The open squares are from Ref. [@Cassel], the open triangles from Ref. [@Joos], the open stars from Ref. [@Papa], the crosses from Ref. [@Sham], the open circles from Ref. [@e665], and the open diamonds from Ref. [@nmc]. Previous data have been scaled to the presently used $Q^2$-bins using Eq. (2). The error bars include both the statistical and systematic uncertainties (added in quadrature). ](figW-all2.epsi){width="47.00000%"}
In Fig. \[fig:crossw1\] the virtual-photoproduction cross section is plotted versus $W$, and compared to existing measurements [@e665; @Joos; @Papa; @Cassel; @Sham; @nmc] at nearby values of $W$. As the various data sets have been measured at different average $Q^2$ values, most data sets were rescaled to the average $Q^2$ values of the HERMES data using the $Q^2$-dependence of the VMD model, which is in agreement with the present data as will be shown below (see Fig. \[fig:crossqq2\] and Eq. (\[eq:res-fittingf\]) with $m$ = 2). It is noted that the various data sets have also been obtained at different average $\epsilon$-values, for which no correction has been applied. The HERMES data are seen to fill the gap which previously existed between the steeply declining data at low $W$, and the data collected at higher $W$ values, which have a much flatter $W$-dependence. With a few exceptions, there is a fair consistency among the world’s data on exclusive $\rho^0$ production. The first exception concerns the highest $W$ point (at 3.5 GeV) of the Cornell data [@Cassel] at $Q^2$ = 0.83 GeV$^2$, which deviates by about 2$\sigma$–3$\sigma$ from a smooth interpolation of all other data in that $W$-region. Secondly, there is a difference of about a factor of two between the NMC [@nmc] and E665 data [@e665] at $W \approx$ 15 GeV in the two highest $Q^2$-bins. The latter discrepancy has been reported before [@e665; @zeus]. Part of the discrepancy might be related to a (model-dependent) subtraction of the non-resonant contribution to the $\rho^0$-peak that was applied by NMC [@nmc] in contrast to E665 [@e665] and the present experiment.
It should be noted that it is the total cross section for the process $\rm{ep} \rightarrow \rm{eh^+h^-p}$ with $M_{\pi \pi} \approx M_{\rho^0}$ that is displayed in fig. \[fig:crossw1\] for all data sets. This total cross section is believed to receive contributions from exclusive $\rho^0$ production through Reggeon exchange, as well as from reaction channels such as those involving nucleon resonances at low $W$, as reported in Refs. [@Joos; @Papa; @Cassel]. In virtual-photoproduction it is usually assumed that the latter contributions are negligible beyond $W \approx$ 4 GeV. The data of Fig. \[fig:crossw1\] indicate that these additional processes have a decreasing contribution to $\rho^0$ production up to about 4–5 GeV, where the steep decrease of the cross section changes into an almost flat $W$-dependence. Henceforth, we restrict the comparison with existing calculations to data collected at $W >$ 4 GeV.
$Q^2$-dependence
----------------
The HERMES data are displayed in Fig. \[fig:crossqq2\] as a function of $Q^2$ for the two $W$ bins. For the purpose of extrapolating to $Q^2$ = 0, the $Q^2$-dependence of the data has been parameterized using the following functional form, inspired by the VMD model [@e665; @Bauer] $$\sigma(Q^{2}) = \sigma_0 \cdot
\left( \frac{ M_{\rho}^2 }{ Q^2 + M_{\rho}^2 } \right)^m \cdot
( 1 + \epsilon R(Q^2)).
\label{eq:res-fittingf}$$ The VMD-model predicts that $m$ = 2 and that the ratio of longitudinal to transverse $\rho^0$ photoproduction cross sections is given by $R = { \xi^2 Q^2 } / { M_{\rho}^2 }$. A good description of the data is obtained by fixing $m$ to its VMD value, and treating $\sigma_0$ and $\xi^2$ as free parameters. Here $\sigma_0$ represents the $\rho^0$ production cross section for real ($Q^2$ = 0) photons. The results of the fit (solid curves in Fig. \[fig:crossqq2\]) are listed in table \[tab:photocross\].
\[t\]
![ The virtual-photoproduction cross section for $\rho^0$ production on $^1$H versus $Q^2$ for (upper panel) and (lower panel). The error bars include both the statistical and systematic uncertainties, added in quadrature. The solid, dashed and dotted curves represent fits using the values of $\sigma_0$ and $\xi^2$, $\sigma_0$ and $m$, or only $m$ as free parameters, respectively. \[fig:crossqq2\] ](figQ2-3.epsi){width="47.00000%"}
\[bt\]
Variables $\langle W \rangle$ = 4.6 GeV $\langle W \rangle$ = 5.4 GeV
----------------------- ------------------------------- -------------------------------
$\sigma_0$ \[$\mu$b\] 10.7 $\pm$ 1.8 13.1 $\pm$ 1.7
$\xi^2$ -0.023 $\pm$ 0.049 -0.13 $\pm$ 0.03
$\sigma_0$ \[$\mu$b\] 11.2 $\pm$ 2.9 19.7 $\pm$ 4.9
$m$ 2.42 $\pm$ 0.18 2.83 $\pm$ 0.18
$\sigma_0$ \[$\mu$b\] 11.5 (fixed) 11.0 (fixed)
$m$ 2.44 $\pm$ 0.05 2.46 $\pm$ 0.06
: The fitted photoproduction cross sections for $\rho^0$ production on $^1$H. In the top row the results are shown with the variables $\sigma_0$ and $\xi^2$ of Eq. (2) as free parameters, the middle row lists the results obtained with $\sigma_0$ and $m$ left free, and the bottom row shows the fitted values of $m$ if a parameterization of the world data is used to constrain $\sigma_0$ The listed errors include both the statistical and systematic uncertainties, added in quadrature. In all cases the $\chi^2$/d.o.f. values of the fits are close to or less than unity. \[tab:photocross\]
As the ratio $R$ must be positive, the negative fit values of $\xi^2$ indicate a deficiency of the VMD-model. Hence, a different approach was also used, i.e. the exponent $m$ of the propagator in Eq. (\[eq:res-fittingf\]) was used as a free parameter, while the value of $R(Q^2)$ was fixed using information extracted from measurements of the $\rho^0$ decay angular distribution [@belz]: $$R(W,Q^2) = c_0(W) \cdot (Q^2 / M_{\rho}^2)^{c_1}.
\label{eq:res-r_q2}$$ The parameterization of $R$ expressed by Eq. (\[eq:res-r\_q2\]) has been obtained from a fit of the world data on $R$, yielding $c_1$ = 0.61 $\pm$ 0.04, and $c_0(W)$ = 0.33 $\pm$ 0.03 (0.48 $\pm$ 0.03) for 4 GeV $< W <$ 7 GeV ($W >$ 7 GeV). The resulting fits (dashed curves in Fig. \[fig:crossqq2\]) also give a good account of the $Q^2$-dependence of the $\rho^0$ production cross section. The fitted parameters are listed in the middle row of table \[tab:photocross\]. The values of $\sigma_0$ are somewhat different from the results of the previous method, but are consistent as the errors have increased considerably.
It is concluded that an increase of the exponent $m$ can avoid the negative values of $\xi^2$ found in the previous fits. This result quantitatively confirms similar conclusions obtained by the E665 collaboration ($m$ = 2.51 $\pm$ 0.07 at $W \approx$ 15 GeV [@e665]).
The fitted values of the extrapolated photoproduction cross section $\sigma_0$ are in agreement with existing real-photon data [@desy-rho], but carry larger error bars particularly if the exponent and the cross section are treated as free parameters. Since the VMD cross section scales with $Q^{-4}$ while a perturbative QCD description predicts [@Brod] that the longitudinal part of the cross section scales with $Q^{-6}$, it is of particular interest to study the value of the exponent of the propagator in the transitional domain probed by the present data. By fixing the value of the photoproduction cross section $\sigma_0$ using a parameterization of the existing photoproduction data [@desy-rho], a one-parameter fit of the $Q^2$-dependence of the cross section has been carried out, yielding fairly precise values for the exponent $m$ as can be seen from the bottom row of table \[tab:photocross\]. The fitted value is remarkably close to 2.5, which is in agreement with results obtained by E665 [@e665] (2.51 $\pm$ 0.07) at an average $W$ value of 17 GeV, but somewhat above the result obtained by H1 [@H196] (2.24 $\pm$ 0.09) at $W$ = 75 GeV.
\[t\]
![\[fig:crossw2\] The virtual-photoproduction cross section for $\rho^0$ production versus $W$ at average $Q^2$ values of 0.83, 1.3, 2.3 and 4.0 GeV$^2$ (from top to bottom). The coding of the data symbols is the same as in Fig. 2. The solid (and dashed) lines represent the results of calculations by L. Haakman [*et al.*]{} [@ref4]. For the solid (dashed) lines the HERMES and E665 (NMC) data were used to normalize the curves. Note that both the data and the calculations have been multiplied by the factors indicated on the left-hand side of the figure for plotting purposes only. ](figHAAK4.epsi){width="47.00000%"}
$W$-dependence
--------------
In Fig. \[fig:crossw2\] the $\rho^0$ virtual-photoproduction cross sections are shown as a function of $W$ for the four different $Q^2$ bins. As was argued above, only the data for $W$ $>$ 4 GeV are being considered. The data are compared to the calculations of Haakman [*et al.*]{} [@ref4], which relate the $W$-dependence of the vector-meson production cross section to the proton structure function $F_2^{\mathrm p}(x,Q^2)$. In order to understand how the $x$-dependence of $F_2^{\mathrm p}(x,Q^2)$ influences the $W$-dependence of the $\rho^0$ production cross section, it is useful to realize that $x$ and $W$ are related by $W^2 = Q^2 \frac{1-x}{x} + M^2$, thus showing that the two variables are essentially inversely proportional at large values of $Q^2$ and small values of $x$. The relation between the $\rho^0$ production cross section and the structure function is apparent in the expression for the $t$-dependence of the cross section given in Ref. [@ref4]: $${ \frac{d \sigma}{dt} } = f_{\rho} (Q^2) F_2^{\mathrm p}(x,Q^2)^2
\exp[\Lambda(W)t].
\label{eq:haakm}$$ In this expression the scaling function $f_{\rho}(Q^2)$ is not determined by the model, and has been obtained from a fit to the experimental data displayed in Fig. \[fig:crossw2\]. For $F_2^{\mathrm p}(x,Q^2)$ the parameterization of Capella [*et al.*]{} [@Cap] in terms of Pomeron and Reggeon exchanges has been taken. The slope parameter $\Lambda$ is given by $$\Lambda(W) = 2 [ R_{sc}^2 + 2 \alpha_P^{\prime} \ln(W/W_0) ]
\label{eq:haakm2}$$ with the radius parameter $R_{sc}^2 = R_{\rm p}^2 + R_{V}^2(Q^2)$ expressing the combined size of the scattering objects, and $W_0^2$ given by $M_{\rho^0}^2 + Q^2$. Expressions for $\alpha_P^{\prime}$, the slope of the effective Pomeron trajectory, and $R_{V}^2(Q^2)$ are given in Ref. [@ref4]. Thus evaluated, the value of $\Lambda$ turns out to be consistent with the value of the slope parameter derived from a fit of the $-t'$ distributions [@ref4].
In order to compare the $W$-dependence of the model of Ref. [@ref4] to that of the data, the calculations have been normalized to the HERMES and E665 data, for each $Q^2$ bin separately. The results are displayed in Fig. \[fig:crossw2\] for $Q^2$ bins centered at 0.83, 1.3, 2.3 and 4.0 GeV$^2$, respectively. A good description of these data is obtained, thus confirming the findings of Ref. [@ref4] but now at lower values of $W$. It should be noted, however, that the data collected at $W$ $<$ 4 GeV – as displayed in Fig. \[fig:crossw1\] – show a steep rise with decreasing $W$ that is not reproduced by the calculations of Ref. [@ref4]. As was mentioned above, this is presumably caused by additional reaction processes that are not contained in the theoretical framework of Ref. [@ref4]. Without explicit calculations that include these effects, it cannot be excluded that there is a finite contribution due to such processes above $W$ = 4 GeV, especially at the lowest $Q^2$ values.
The NMC data (also shown in Fig. \[fig:crossw2\]) can be compared to the calculations of Ref. [@ref4] as well. This leads to differently normalized curves (dashed lines in Fig. \[fig:crossw2\]). If the NMC normalization is adopted, the curves fall below the E665 and HERMES data. Within the constraints of the present model calculation, the E665 normalization is therefore preferred. However, on the basis of the HERMES data alone no such distinction can be made.
The longitudinal cross section
------------------------------
The results of recent OFPD calculations, such as those described in Refs. [@MvdH; @MvdH2; @Mank1; @Mank2], only concern the longitudinal component $\sigma_L$ of the $\rho^0$ virtual-photoproduction cross section, because the factorization theorem [@Collins] applies only to the longitudinal case. Hence, values of $\sigma_L$ were extracted from the data.
The longitudinal cross section $\sigma_L$ is related to the total $\rho^0$ production cross section $\sigma_{\gamma^* p \rightarrow \rho^0 p}$: $$\sigma_L ={{R} \over {1 + \epsilon R}} \sigma_{\gamma^* p \rightarrow \rho^0 p}.
\label{eq:LT}$$ Assuming SCHC the ratio $R = \sigma_L / \sigma_T$ can be derived from the longitudinal fraction $r_{00}^{04}$ of $\rho^0$ mesons, which can be extracted from their polar decay angular distribution $W(\cos\Theta)$. In Ref. [@belz] such an analysis is presented, which has provided us with the parameterization of $R$ shown in Eq. (\[eq:res-r\_q2\]). This has been used to evaluate $\sigma_L$ for each of the cross sections listed in Table \[tab:cross\]. (Note that $\epsilon$ can be determined from the kinematics of each event.) The results are listed in Table \[tab:longcross\] and, for $Q^2 >$ 2 GeV$^2$, displayed in Fig. \[fig:crossw3\]. It may be noted that the values of $\sigma_L$ increase by about 30% if the quoted parameterization of $R$ for $W$ $>$ 7 GeV is used.
\[t\]
![\[fig:crossw3\] The longitudinal component of the virtual-photoproduction cross section for $\rho^0$ production versus $W$ at average $Q^2$ values of 2.3 (left) and 4.0 GeV$^2$ (right). The solid lines represent the results of the calculations of Refs. [@MvdH; @MvdH2]. The dashed (dotted) curves represent the quark (two-gluon) exchange contributions within these calculations. ](figOFPD4b.epsi){width="47.00000%"}
The data are compared to the model calculations described in Ref. [@MvdH], but with the improvements described in Ref. [@MvdH2]: higher twist effects are included in a phenomenological fashion, the $Q^2$-dependence of the strong coupling constant $\alpha_s$ was accounted for, and the sea quarks are included using the MRST98 parton distributions [@MRST98]. As can be seen from Fig. 17 of Ref. [@MvdH2] the higher-twist contributions, which were included through the dependence on the intrinsic transverse momentum $k_{\perp}$, reduce the longitudinal cross section by typically a factor of 5 at $Q^2 \approx$ 2 GeV$^2$. As the (phenomenological) higher twist corrections at lower $Q^2$ values are even larger, the comparison between the OFPD calculations of Ref. [@MvdH; @MvdH2] and the data has not been extended to the data below $Q^2 \approx$ 2 GeV$^2$. Moreover, the OFPD calculations cannot be used for data collected at $W$ $<$ 4 GeV, as no other reaction channels were included.
\[bt\]
$<Q^2>$
--------- -------------- ----------------------- -------------- -----------------------
$<\epsilon>$ $\sigma_L$ \[$\mu$b\] $<\epsilon>$ $\sigma_L$ \[$\mu$b\]
0.83 0.87 0.77 $\pm$ 0.16 0.74 0.67 $\pm$ 0.14
1.3 0.87 0.35 $\pm$ 0.07 0.73 0.41 $\pm$ 0.08
2.3 0.86 0.21 $\pm$ 0.04 0.71 0.21 $\pm$ 0.04
4.0 0.83 0.09 $\pm$ 0.02 0.69 0.06 $\pm$ 0.01
: Longitudinal cross sections $\sigma_L$ for $\rho^0$ virtual photoproduction on $^1$H in $\mu$b. The listed uncertainties include both the total error on the measured $\rho^0$ virtual-photoproduction cross sections and the error on the parameterization of $R$ for $W$ $<$ 7 GeV, which was used in the evaluation of Eq. (\[eq:LT\]). The average values of $\epsilon$ for each bin are listed as well.\[tab:longcross\]
The OFPD calculations are compared in Fig. \[fig:crossw3\] to the longitudinal cross sections measured by E665 [@e665] and the present experiment. The calculated contributions due to quark exchange (dashed curves) and two-gluon exchange (dotted curves) are also shown separately. In the present kinematic domain the quark contributions dominate, and only at $Q^2$ = 4.0 GeV$^2$ and $W >$ 10 GeV the gluon contribution starts to contribute significantly. A fairly good agreement between the calculations and the data is obtained, given the existing theoretical uncertainties related to the size of the higher-twist contributions, and the relatively low $Q^2$ values involved. The calculated rise of the cross section at $W \approx$ 5 GeV, which is associated with the contribution due to the exchange of valence quarks, is not inconsistent with the new HERMES data.
Summary {#sec:sum}
=======
In summary, cross sections have been presented for exclusive diffractive $\rho^0$ virtual photoproduction in the $W$-domain between 4 and 6 GeV. The $Q^2$-dependence of the cross section is well described by the propagator of the Vector Meson Dominance model, although an increase of the exponent from its original value of 2 to about 2.5 is needed when consistency with existing data for $R = \sigma_L / \sigma_T$ is required. By extrapolating the $ \gamma^* p \rightarrow \rho^0 p$ cross section to $Q^2$ = 0, photoproduction cross sections are found which are in fair agreement with previously published values of about 11 $\mu$b. The $W$-dependence of the present and (most) existing data in the $W$-domain between 4 and 25 GeV is well described by a model linking the $x$-dependence of the proton structure function $F_2^{\mathrm p}(x,Q^2)$ to the vector meson production cross section. The longitudinal component of the cross section has been compared to calculations based on the Off-Forward Parton Distribution framework. A fairly good agreement with the data is found.
We gratefully acknowledge the DESY Directorate for its support and the DESY staff and the staffs of the collaborating institutions. We particularly appreciate the efforts of the HERA machine group in providing high beam polarization. Additional support for this work was provided by the Deutscher Akademischer Austauschdienst (DAAD) and INTAS, HCM, and TMR network contributions from the European Community.
[99]{}
A.V. Radyushkin, Phys. Lett. B [**380**]{} (1996) 417; Phys. Rev. D [**56**]{} (1997) 5524.
X. Ji, Phys. Rev. Lett. [**78**]{} (1997) 610; Phys. Rev. D [**55**]{} (1997) 7114.
J.C. Collins, L. Frankfurt and M. Strikman, Phys. Rev. D [**56**]{} (1997) 2982.
L. Frankfurt, W. Koepf and M. Strikman, Phys. Rev. D [**54**]{} (1996) 3194.
M. Vanderhaeghen, P.A.M. Guichon and M. Guidal, Phys. Rev. Lett. [**80**]{} (1998) 5064.
M. Vanderhaeghen, P.A.M. Guichon and M. Guidal, Phys. Rev. D [**60**]{} (1999) 094017.
L. Mankiewicz, G. Piller and T. Weigl, Eur. Phys. J. C [**5**]{} (1998) 119.
L. Mankiewicz, G. Piller and A. Radyushkin, Eur. Phys. J. C [**10**]{} (1999) 307; and L. Mankiewicz, G. Piller and T. Weigl, Phys. Rev. D [**59**]{} (1999) 017501.
A.V. Belitsky, D. Müller, L. Niedermeier and A. Schäfer, hep-ph/9908337.
L.P.A. Haakman, A. Kaidalov and J.H. Koch, Phys. Lett. B [**365**]{} (1996) 411.
HERMES Collaboration, K. Ackerstaff [*et al.*]{}, Nucl. Instr. and Meth. A [**417**]{} (1998) 230.
J. Stewart, Proc. of the Workshop “Polarized gas targets and polarized beams,” eds. R.J. Holt and M.A. Miller, Urbana-Champaign, USA, AIP Conf. Proc. 421 (1997) 69.
M. Kolstein, PhD Thesis, Vrije Universiteit, Amsterdam (1998), HERMES Report 98-046.
HERMES Collaboration, K. Ackerstaff [*et al.*]{}, submitted to Eur. Phys. J., DESY preprint 99-199 (1999), hep-ex/0002016.
LEPTO: G. Ingelmann, A. Edin and J. Rathsman, Comp. Phys. Comm. [**101**]{} (1997) 108.
M. Arneodo, L.Lamberti and M. Ryskin, DESY preprint 96-149 (1996).
E665 Collaboration, M.R. Adams [*et al.*]{}, Z. Phys. C [**74**]{} (1997) 237.
M. Ross and L. Stodolsky, Phys. Rev. [**149**]{} (1966) 1172.
C. Caso [*et al.*]{} (Particle Data Group), Eur. Phys. J. C [**3**]{} (1998) 1.
HERMES Collaboration, K. Ackerstaff [*et al.*]{}, Phys. Rev. Lett. [**82**]{} (1999) 3025.
NMC Collaboration, M. Arneodo [*et al.*]{}, Phys Lett. B [**364**]{} (1995) 107; NMC Collaboration, P. Amaudruz [*et al.*]{}, Phys Lett. B [**295**]{} (1992) 159.
H1 Collaboration, C. Adloff [*et al.*]{}, Z. Phys. C [**75**]{} (1997) 607.
Y. Akimov [*et al.*]{}, Phys. Rev. D [**14**]{} (1976) 3148.
H. Holtmann [*et al.*]{}, Z. Phys. C [**69**]{} (1996) 297.
I. Akushevich, Eur. Phys. J. C [**8**]{} (1999) 457; and I. Akushevich, A. Ilychev, N. Shumeiko, A. Soroko and A. Tolkachev, Comp. Phys. Comm. [**104**]{} (1997) 201.
P. Joos [*et al.*]{}, Nucl. Phys. B [**113**]{} (1976) 53.
C. del Papa [*et al.*]{}, Phys. Rev. D [**19**]{} (1979) 1303.
D.G. Cassel [*et al.*]{}, Phys. Rev. D [**24**]{} (1981) 2787.
W.D. Shambroom [*et al.*]{}, Phys. Rev. D [**26**]{} (1982) 1.
NMC Collaboration, M. Arneodo [*et al.*]{}, Nucl. Phys. B [**429**]{} (1994) 503.
ZEUS Collaboration, J. Breitweg [*et al.*]{}, Eur. Phys. J. C [**6**]{} (1999) 603.
T.H. Bauer [*et al.*]{}, Rev. Mod. Phys. [**50**]{} (1978) 261.
ZEUS Collaboration, J. Breitweg [*et al.*]{}, Eur. Phys. J. C [**2**]{} (1998) 247.
S.J. Brodsky [*et al.*]{}, Phys. Rev. D [**50**]{} (1994) 3134.
H1 Collaboration, C. Adolff [*et al.*]{}, Eur. Phys. J. [**C13**]{} (2000) 371.
A. Capella, A. Kaidalov, C. Merino and J. Tran Tanh Van, Phys. Lett. B [**337**]{} (1994) 358.
A. D. Martin, R. G. Roberts, W. J. Stirling, and R. S. Thorne (MRST 98), Eur. Phys. J. C [**4**]{} (1998) 463.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The [[*FIRST*]{}]{} survey, begun over twenty years ago, provides the definitive high-resolution map of the radio sky. This VLA survey reaches a detection sensitivity of 1 mJy at 20 cm over a final footprint of 10,575 deg$^2$ that is largely coincident with the Sloan Digital Sky Survey area. Both the images and a catalog containing 946,432 sources are available through the [[*FIRST*]{}]{} web site (<http://sundog.stsci.edu>). We record here the authoritative survey history, including hardware and software changes that affect the catalog’s reliability and completeness. In particular, we use recent observations taken with the JVLA to test various aspects of the survey data (astrometry, CLEAN bias, and the flux density scale). We describe a new, sophisticated algorithm for flagging potential sidelobes in this snapshot survey, and show that fewer than 10% of the catalogued objects are likely sidelobes, and that these are heavily concentrated at low flux densities and in the vicinity of bright sources, as expected. We also report a comparison of the survey with the NRAO VLA Sky Survey (NVSS), as well as a match of the [[*FIRST*]{}]{} catalog to the SDSS and 2MASS sky surveys. The NVSS match shows very good consistency in flux density scale and astrometry between the two surveys. The matches with 2MASS and SDSS indicate a systematic $\sim10-20$ mas astrometric error with respect to the optical reference frame in all VLA data that has disappeared with the advent of the JVLA. We demonstrate strikingly different behavior between the radio matches to stellar objects and to galaxies in the optical and IR surveys reflecting the different radio populations present over the flux density range 1–1000 mJy. As the radio flux density declines, stellar counterparts (quasars) get redder and fainter, while galaxies get brighter and have colors that initially redden but then turn bluer near the [[*FIRST*]{}]{} detection limit.
Implications for future radio sky surveys are also briefly discussed. In particular, we show that for radio source identification at faint optical magnitudes, high angular resolution observations are essential, and cannot be sacrificed in exchange for high signal-to-noise data. The value of a JVLA survey as a complement to SKA precursor surveys is briefly discussed.
author:
- 'David J. Helfand, Richard L. White and Robert H. Becker'
title: 'The Last of [[*FIRST*]{}]{}: The Final Catalog and Source Identifications'
---
Introduction {#intro}
============
Faint Images of the Radio Sky at Twenty-centimeters were collected over a period of eighteen years at the (now, Jansky) Very Large Array. The original proposal to use the world’s premiere radio telescope for the relatively mundane task of surveying the sky was submitted in August of 1990. Pilot observations for the [[*FIRST*]{}]{}survey began in April of 1993, fifty years to the month after Grote Reber’s original radio sky survey [@reber1944]. The final observations were completed in the Spring of 2011 in an expansion of the survey to cover the SDSS3 sky survey area. Over this extended period, a number of VLA hardware changes were implemented, not least being the transformation of the entire array into the JVLA. In addition, a number of changes to the data reduction software and processing algorithms also took place over the course of the survey. Furthermore, the lower-resolution NVSS survey was completed during this interval, as were sky surveys at infrared (2MASS) and optical (SDSS) wavelengths.
In this paper, we describe the final products of the [[*FIRST*]{}]{}survey. The [[*FIRST*]{}]{} images and catalogs are distributed through the [[*FIRST*]{}]{} web site[^1], the Mikulski Archive for Space Telescopes[^2] and other archival sites.
A history of the hardware and software changes during the project is provided to alert users to possible systematics in the survey images and catalog (§\[history\]); we also note various rare anomalies in the VLA system that have been discovered as a consequence of examining over 75,000 snapshot images. In some of this work, we make use of matches of the [[*FIRST*]{}]{} catalog to other all-sky surveys, the science from which is reported later in the paper. We assess the astrometric accuracy of the [[*FIRST*]{}]{} catalog, which reveals some small systematic offsets in the positions (§\[astrometry\]). We go on to describe the algorithm developed to assign to each source a probability that it is a spurious catalog entry resulting from a sidelobe of a bright source elsewhere in the field (§\[sidelobes\]). Various tests of this algorithm using radio and optical catalog matching provide an assessment of the algorithm’s reliability.
{width="\linewidth"}
We compare the [[*FIRST*]{}]{} survey catalog with the results of three other major sky surveys that cover most or all of the same $> 10,000$ deg$^2$ of sky: the NVSS [@condon1998], the 2MASS survey [@skrutskie2006], and the SDSS [@york2000] plus its more recent extensions [@ahn2014]. In particular, we use the NVSS match (§\[NVSS\]) to quantify the accuracy of the [[*FIRST*]{}]{} flux density scale and the degree to which faint sources are missed as a consequence of the increased noise level near bright sources, as well as the incompleteness of [[*FIRST*]{}]{}for very extended objects. The 2MASS (§\[2MASS\]) and SDSS (§\[SDSS\]) matches provide information on the astrometric accuracy of the respective surveys, as well as insight into the classes of sources that populate the radio sky at milliJansky flux densities. We conclude with a commentary (§\[conclusions\]) on the survey’s utility and on the lessons learned from this undertaking that can be used to inform the next generation of radio sky surveys. In particular, we examine critically the effect of survey resolution on optical/IR identification programs for radio sources.
Readers uninterested in the detailed technical history and/or the subtle catalog biases useful only for those using [[*FIRST*]{}]{} in large, statistical surveys, should skip to the science results, which begin in §\[NVSS\].
[lcccc]{} 1995 Jan 06 & 26,892 & 300 & 0 & 300\
1995 Oct 16 & 138,665 & 1559 & 0 & 1559\
1996 May 28 & 138,665 & 1559 & 0 & 1559\
1997 Feb 27 & 236,040 & 2576 & 0 & 2576\
1997 Apr 24 & 268,047 & 2576 & 350 & 2926\
1998 Feb 04 & 437,429 & 4153 & 611 & 4764\
1999 Jul 21 & 549,707 & 5448 & 611 & 6060\
2000 Jul 05 & 722,354 & 7377 & 611 & 7988\
2001 Oct 15 & 771,076 & 7954 & 611 & 8565\
2003 Apr 11 & 811,117 & 8422 & 611 & 9033\
2008 Jul 16 & 816,331 & 8444 & 611 & 9055\
2012 Feb 16 & 946,464 & 8444 & 2191& 10635\
2013 Jun 05 & 971,268 & 8444 & 2191& 10635\
2014 Mar 04 & 946,432 & 8444 & 2131& 10575\
2014 Dec 17 & 946,432 & 8444 & 2131& 10575
The Survey History {#history}
==================
The VLA[^3] pilot observations in 1993 aided us in designing the pointing grid for the survey (also adopted by the NVSS) and in developing the basic data reduction algorithms for turning snapshot visibilities into final survey images. These basic attributes of the project are described in detail in [@becker1995] and the catalog, constructed as detailed in [@white1997]; those papers should remain the primary references when making use of [[*FIRST*]{}]{} results. Here, for the record, we document the technical history of the project and describe all factors that might affect the character and integrity of the data products.
In Figure \[coverage\], we show the time allocated and the sky coverage achieved over the 18 years of survey observations. A total of eleven observing sessions in the B-configuration led to the accumulation of just over 4000 hours (5.5 months) of observing time that resulted in 10,575 deg$^2$ of sky coverage (8444 deg$^2$ in the north Galactic cap and 2131 deg$^2$ in the southern cap). A coverage map, color-coded by observation epoch, is displayed in the inset to Figure 1; the catalog-release points are colored to match the relevant coverage areas.
As originally envisioned, all data were released to the public archive on the day they were taken, and all images were fully reduced and put on a public website before a new set of observations commenced. More than a dozen catalog releases were issued; they are located in Figure 1 by date and sky coverage (right axis). A summary of the catalog releases is given in Table \[table-catalog-releases\]. The final 2014 December 17 release contains 946,432 entries. A history of catalog format changes and error corrections can be found online[^4].
Hardware and observing mode changes {#hardware}
-----------------------------------
The only major hardware change over the period of the survey was occasioned by the transition of the array from its original configuration to the JVLA. During the 2009 observing session, roughly half the array had the original 20 cm receivers, while half had been changed to the JVLA receivers. The center frequencies of the two IFs were held constant, but the shapes of the receiver bandpasses were different. In the 2011 session, the JVLA transition was complete. We observed at a different center frequency (1335 MHz vs. two narrow bands centered at 1365 MHz and 1435 MHz) with a 128 MHz total bandwidth comprised of 64 2-MHz channels (vs. $2 \times 7$ 3-MHz channels for the remainder of the survey); data taken in a second 128-MHz band centered at 1730 MHz, to be used in deriving source spectral indices, will be reported elsewhere. All the changes in 2011 were motivated by a combination of the changing requirements of the JVLA receiver/correlator system, plus the desire to reduce the observing time and to extract additional science using the new capabilities of the JVLA.
While the pilot observations used a 3-second integration time, we quickly adopted 5-second integrations as standard. For the 2011 JVLA observations, however, the wide bandwidth required a reduction of the integration time to 1 second. These changes affect the data analysis but do not lead to appreciable changes in the survey data products. The on-target dwell time was 165-seconds per field until 2011, at which time the wider bandwidth available allowed us to reach our 1 mJy sensitivity while reducing the observing time to 1 minute per field. The wider bandwidth and different frequency do have some effects on the data (discussed further below). The shorter integration time also required a change in the pointing pattern, since the time spent per field was then much less than the sidereal rate at which the sky passed overhead. A traveling-salesman algorithm using simulated annealing [@kirkpatrick1983] was developed to minimize slew time between fields.
During both 2009 and 2011, the Sun was located in the sky region to be observed, requiring us to schedule around a zone of avoidance centered on the apparent solar position. A 7-degree avoidance radius was used in 2009; the zone was expanded to 10 degrees in 2011 owing to increased solar activity. This, coupled with non-optimal scheduling of our time allocations, led to many of the observations being taken farther from the zenith than was desirable. The scheduling challenges combined with the loss of a significant quantity of data to interference also resulted in some holes in the sky coverage and led to the disconnected island of coverage that is visible in Figure \[coverage\] near $\delta=15^\circ$ in the south Galactic cap.
![Distribution of hour angles for the [[*FIRST*]{}]{} observations. The hour angle is zero for fields observed as they pass the meridian. The black line shows the distribution for the entire survey. The red line shows the distribution for the 2011 JVLA observations, normalized to the same area as the overall distribution. The latter distribution differs owing to shorter 1-minute integrations and the need to avoid pointing toward the Sun. \[fig-hour-angles\]](f2.eps){width="\linewidth"}
Figure \[fig-hour-angles\] shows how the observations were distributed with respect to the meridian. Ideally the fields would all have been observed as they passed the meridian (zero hour angle), but the real distribution is more complex. 90% of the observations for the entire survey were acquired within 1.4 hours of the meridian. The distribution is noticeably spiky and asymmetrical because the observing time blocks allocated were often non-optimal, requiring that we observe fields off the meridian. The distribution for the short 1-minute JVLA observations, shown by the red line, is more symmetrical (the result of our traveling salesman algorithm) but also has a very extended tail at large hour angles (due to the need to avoid the Sun.) All these effects conspire to produce point-spread functions (PSFs) and sidelobe patterns that vary slightly from year-to-year. For most purposes, however, the PSF can be treated as if it were uniform. Over the northern sky the PSF is a circular Gaussian with FWHM 5.4. South of declination ${+}4^\circ33\arcmin21\arcsec$ the beam becomes elliptical, $6.4\arcsec\times5.4\arcsec$, with the major axis running north-south. South of ${-}2^\circ30\arcmin25\arcsec$ the elliptical beam size increases further to $6.8\arcsec\times5.4\arcsec$.
In the final 2014Dec17 version of the catalog, the JVLA images are not co-added with older VLA images to avoid problems resulting from the different frequencies and noise properties of the two datasets. That leads to small gaps in sky coverage at the boundaries between the JVLA and VLA regions, but has the advantage that it cleanly separates sources from the old and new configurations. All sources in the final catalog that have field names ending with ‘W’ come from the JVLA data.
VLA anomalies
-------------
Unsurprisingly, with 4000 hours of observing time to reduce and a million sources to catalog, subtle and/or rare effects, not noticeable in ordinary VLA programs, can appear. We summarize several such anomalies here.
**Misassignment of data blocks:** Searches of the [[*FIRST*]{}]{}database for source variability led to the discovery of a bug in the online VLA software that appends to the current observation a few integrations from the previous observation. This was revealed when investigating the (ultimately spurious) detection of variability in a bright source [@galyam2006; @ofek2010]. All $uv$ datasets were subsequently checked for this error and 190 cases of discontinuities in $uvw$ values within a single scan were identified [@thyagarajan2011]. These data sets were edited and the images re-made. Note that all versions of the catalog prior to 2012 may contain a few spurious sources as a consequence of this error.
**Misassignment of array configuration:** In collecting the global observation records to include in this paper, we noticed that the VLA archive sometimes records incorrect array configuration labels for some observations. All observations for the [[*FIRST*]{}]{}survey were, in fact, taken in the B configuration. The VLA archive also is currently missing entries for $\sim1$% of the [[*FIRST*]{}]{} fields.
**Image stretch and rotation:** By comparing the positions of sources observed in multiple pointings, we discovered that there exist both a stretch and a rotation of VLA images, which we infer are related to small clock errors at the VLA and small changes in bandpass shape [see footnote 6 of @white1997]. The astrometric errors introduced are very small ($< 0.1\arcsec$) for most sources, as the errors tend to cancel out in the co-added images. Only for sources at the very edge of the survey coverage area can the errors rise to $\sim 0.3\arcsec$. For the first four epochs of observation, we solved for the stretch and rotation corrections using multiply observed sources. Since the parameters were quite stable from epoch to epoch, we then fixed the correction and applied it to each epoch’s images through 2003. The last two epochs have not had these small corrections applied. Thus, the positions of sources at the edge of the coverage in the south Galactic cap should be assigned slightly larger astrometric uncertainties.
Data reduction changes {#data-reduction}
----------------------
Computing power and storage capacity have increased dramatically over the life of this project (Fig. \[fig-disk-prices\]). In addition, the AIPS software package has evolved, as have the scripts we run to process the [[*FIRST*]{}]{}data. Finally, the recent upgrade to the JVLA hardware has required changes in the processing pipeline. We briefly note here all significant changes to the data processing.
![Price of disk storage (dollars per terabyte, using constant 2012 dollars) over the duration of the [[*FIRST*]{}]{} survey. Storage prices have dropped by a factor of roughly 300,000 since the [[*FIRST*]{}]{} survey proposal was submitted in 1990. Computer processing and memory have also become dramatically less expensive over the lifetime of the survey. \[fig-disk-prices\]](f3.eps){width="\linewidth"}
**Data flagging:** Prior to 2011, the data were collected in two sets of seven 3-MHz channels, and if interference was strong in one channel, all seven channels of data for that integration period (5 seconds) were deleted. With the advent of the broad-bandwidth JVLA correlator, identified interference in one 2-MHz channel leads to the deletion of that channel and the two adjacent channels only.
**Self-calibration:** The self-calibration process [@hogbom1974] requires iteratively CLEANing images and using the flux models from those images to improve the antenna phase (and sometimes amplitude) calibrations. During the initial years of the [[*FIRST*]{}]{} survey, the computing required to construct a map was daunting. To render the computing tractable, we adopted an approach of making an initial low-resolution map to find bright sources in the image and then using only a set of small maps centered on those source positions for the self-calibration iteration [@becker1995]. Following the six-year hiatus in 2009, we simplified the process for new data by repeatedly making full-field images during the self-calibration iteration as well as for producing the final map.
**Wide-field bright-source mapping:** Owing to processing speed limitations, our original approach for the final map was to augment a $2048\times2048$ map covering the central primary beam with small satellite maps placed at the positions of nearby bright sources (from single-dish catalogs) that lie within $10\arcdeg$ of each field center [for details see @becker1995]. Beginning in 2009, we simply made $4096 \times 4096$ images and dispensed with special treatment for far off-axis sources. With 1.8 arcsec pixels, these images span 2.0 degrees, which is four times larger than the 20 cm FWHM primary beam diameter. Note that because these images are snapshot observations, there is no need for any special treatment of 3-D sky effects in the processing. The 3-D distortions are removed in post-processing by warping the map as described in [@becker1995].
Also, in the first section of the survey, each field center was shifted by up to 0.9 so that the brightest source in the field fell at the precise center of a pixel. We never found any evidence that this significantly affected image quality and, since it complicated the image processing and the final products (which otherwise had a predictable pixel grid), we abandoned this procedure in processing the final two epochs.
**Source extraction:** We used our AIPS source extraction program HAPPY [@white1997] to identify and measure the properties of sources in the [[*FIRST*]{}]{} images. HAPPY evolved over the course of the project as a result of minor enhancements and bug fixes. To ensure a more uniform catalog, in 2007 we reprocessed all the data using the current version of HAPPY. That led to minor changes in the source lists.
![Distribution of rms noise values in individual [[*FIRST*]{}]{}survey images. The dark shaded histogram shows the noise distribution for JVLA fields observed on three days with particularly bad interference; their mean noise levels are $\sim2.5$ times higher. \[fig-gridrms\]](f4.eps){width="\linewidth"}
{width="0.5\linewidth"} {width="0.5\linewidth"}
**Flux calibration and CLEAN bias:** The flux density calibrators 3C286 and 3C48 were used for the North and South Galactic Caps, respectively. The co-adding procedure to create the final images from the individual grid images, as well as the source detection and parameterization procedures, remained unchanged throughout the survey period.
One of the systematic errors in VLA images that came to light from [[*FIRST*]{}]{} and NVSS was the so-called ‘CLEAN bias’. Sources extracted from CLEANed snapshot images invariably lost flux in the CLEANing process. Further analysis of the [[*FIRST*]{}]{} images for studies of sub-threshold sources through stacking [@white2007] showed that the CLEAN bias effect is really a snapshot bias that reduces the flux densities of even faint undetected sources. This was quantified by [@becker1995] and [@white2007] by adding artificial sources into the UV data and comparing the input and output flux densities for tens of thousands of point sources. We continue to add a CLEAN bias offset of 0.25 mJy to the peak flux densities of all sources [see @becker1995].
Since with the change of VLA hardware there was some concern that the CLEAN bias would be different, we repeated these CLEAN bias tests using artificial sources for each day of data taken in the final 2011 JVLA epoch. We found that the CLEAN bias was not stable and seemed to correlate with the amount of interference in the data. In particular, there were three days when the CLEAN bias was significantly higher than usual. The higher CLEAN bias on those days appeared to result from CLEANing too deeply, i.e., since there was more interference, the rms was higher and hence using the standard CLEAN limit was inappropriate. For those days the images were remade with a shallower CLEAN threshold, which brought the CLEAN bias back into line with the normal level.
The price of shallower CLEANing for those fields was an increase in the rms noise levels. Fig. \[fig-gridrms\] shows the distribution of the rms noise levels in all [[*FIRST*]{}]{} “grid” images (single pointings, before co-adding), with the rms distribution for images on the three noisy JVLA days highlighted. The mode of the noise distribution is 0.132 mJy, but the distribution has a tail toward higher rms values (most often the result of bright sources in the field). The extension to lower rms values is the result of multiply observed fields that have been combined to reduce the noise. The noise levels for JVLA fields observed on the three bad days are typically 2.5 times higher. Note that the [[*FIRST*]{}]{} catalog uses these noise measurements for all contributing grid images to compute the rms noise as a function of position in the co-added images [@white1997]; consequently, the final [[*FIRST*]{}]{} catalog has rms noise estimates that correctly reflect the elevated noise in these regions.
Epochs of Observations
----------------------
To enable the use of [[*FIRST*]{}]{} data for time-domain science, the 2014Dec17 release of the [[*FIRST*]{}]{} catalog includes observation epochs for the sources. Since the catalog is created from co-added images, each source may have contributions from many different pointings taken at different times. For some sources all the contributing observations come from a narrow time range of only a few minutes (when one or two consecutive grid images dominate the co-added source position), while others have significant contributions from pointings taken years apart (typically at the seams between observing seasons and near fields that were re-observed owing to problems with the original observations).
The catalog includes a mean weighted epoch ${\bar t}$, defined to be the average of the epochs of all the contributing pointings at the source position weighted by the same weights used to combine the overlapping maps. It also includes the weighted rms $\sigma(t)$ of the scatter of the pointing epochs about that mean, which is a measure of the effective spread in the observing epoch. Figure \[fig-rmsdist\](a) shows the cumulative distribution of $\sigma(t)$; the median rms is 1.8 days, and 90% of the $\sigma(t)$ values are less than four months, so most observations are well-localized in time. Plotting $\sigma(t)$ versus ${\bar t}$ (Fig. \[fig-rmsdist\]b) shows the tail of much higher epoch rms values for sources in the overlap regions between observing sessions.
Astrometry
==========
In our initial description of the [[*FIRST*]{}]{}survey [@becker1995] we evaluated the astrometric precision of the source catalog by comparing the positions of 46 Multi-Element Radio-Linked Interferometer Network (MERLIN) calibrators lying in the first survey strips and found the systematic errors to be $<0.05\arcsec$. An additional comparison with 4100 optical counterparts from the Automatic Plate Measuring (APM) machine scans of the Palomar Optical Sky Survey plates [@mcmahon1992] found the same limit for the optical-radio frame offset. For a survey in which the individual catalog entries have typical positional uncertainties of $0.3\arcsec$, any systematic position offsets of this magnitude were insignificant for all envisioned initial usages of the catalog. In the [[*FIRST*]{}]{} catalog paper [@white1997] we used a larger sample of radio calibrators and the increased size of the survey coverage area to conclude that any offset from the radio reference frame was $<0.03\arcsec$.
[cclcccccc]{} all & 100.0& NVSS & 15& 54.9 & 5.46 & 5.59 & $ {-}0.009\pm0.008$& $ 0.064\pm0.008 $\
isolated & 73.0 & NVSS & 15& 50.6 & 4.64 & 4.89 & $ {-}0.011\pm0.008$& $ 0.066\pm0.008 $\
all & 100.0& 2MASS & 2 & 8.0 & 0.42 & 0.43 & $ 0.021\pm0.002$ & $ 0.007\pm0.002 $\
pt & 37.1 & 2MASS & 2 & 8.3 & 0.34 & 0.34 & $ 0.020\pm0.002$ & $ 0.006\pm0.002 $\
pt, bright & 4.9 & 2MASS & 2 & 6.8 & 0.23 & 0.23 & $ 0.022\pm0.004$ & $ 0.009\pm0.004 $\
all & 93.0 & SDSS all & 2 & 32.9 & 0.36 & 0.37 & $ 0.021\pm0.001$ & $ 0.016\pm0.001 $\
all & 93.0 & SDSS galaxy & 2 & 26.5 & 0.38 & 0.39 & $ 0.021\pm0.001$ & $ 0.016\pm0.001 $\
all & 93.0 & SDSS star & 2 & 6.3 & 0.26 & 0.28 & $ 0.023\pm0.001$ & $ 0.014\pm0.001 $\
pt & 35.3 & SDSS all & 2 & 38.2 & 0.28 & 0.28 & $ 0.020\pm0.001$ & $ 0.014\pm0.001 $\
pt & 35.3 & SDSS galaxy & 2 & 29.6 & 0.30 & 0.30 & $ 0.020\pm0.001$ & $ 0.015\pm0.001 $\
pt & 35.3 & SDSS star & 2 & 8.6 & 0.21 & 0.22 & $ 0.022\pm0.001$ & $ 0.012\pm0.001 $\
pt, bright & 4.7 & SDSS all & 2 & 41.0 & 0.17 & 0.17 & $ 0.022\pm0.001$ & $ 0.012\pm0.001 $\
pt, bright & 4.7 & SDSS galaxy & 2 & 23.4 & 0.19 & 0.19 & $ 0.020\pm0.002$ & $ 0.013\pm0.002 $\
pt, bright & 4.7 & SDSS star & 2 & 17.7 & 0.15 & 0.16 & $ 0.025\pm0.002$ & $ 0.011\pm0.002 $\
JVLA, isolated & 4.5 & NVSS & 15& 52.0 & 4.62 & 4.99 & $ 0.005\pm0.030$ & $ 0.179\pm0.031 $\
JVLA & 6.4 & 2MASS & 2 & 6.7 & 0.47 & 0.49 & $ 0.000\pm0.007$ & ${-}0.029\pm0.008\phn $\
JVLA & 99.3 & SDSS all & 2 & 28.4 & 0.40 & 0.44 & $ 0.002\pm0.003$ & ${-}0.009\pm0.003\phn $\
With the survey now complete, large optical and infrared catalogs of the sky now available, and significant work over the intervening two decades invested in the creation of the International Celestial Reference Frame [ICRF2 – @ma2013], we have the opportunity to investigate small astrometric effects in some detail. In subsequent sections, we provide definitive matches of the [[*FIRST*]{}]{} survey’s final catalog to the NVSS, SDSS, and 2MASS catalogs; Table \[table-match\] summarizes the results of those matches, including the mean astrometric offsets between [[*FIRST*]{}]{} and other catalogs.
The positional offset for all [[*FIRST*]{}]{}-NVSS matches in Right Ascension is consistent with zero: $\Delta RA = -0.009 \pm
0.008\arcsec$ (avoiding source confusion by selecting isolated sources only, the value is $\Delta RA = -0.011 \pm 0.008\arcsec$). This agreement is unsurprising given that both surveys were conducted for the most part with the same instrument, but it eliminates the possibility that any errors have been introduced by the data reduction procedures.
Comparisons with the optical and near-IR catalogs, however, reveal an apparent discrepancy. In Table \[table-match\], the mean offset in RA between [[*FIRST*]{}]{} and either SDSS or 2MASS is consistent with $0.021\arcsec$, and the offset is determined with a very small uncertainty ranging from 1 to 4 mas for the various subsets. We note that an A-configuration survey of the SDSS Stripe 82 region [@hodge2011] also finds $\Delta RA = 0.020\arcsec$. Since the offset between the ICRF2 and optical reference frames has been established to be less than $0.003\arcsec$ in RA [@orosz2013; @assafin2013], it appears that VLA data suffer from a $+20$ mas offset.
However, using only [[*FIRST*]{}]{} data taken with the JVLA and matching to the SDSS catalog, we find $\Delta RA = 0.002\arcsec\pm 0.003$. The timing system, correlator, and data acquisition hardware and software are different between the VLA and the JVLA. Thus, we find strong evidence that positions derived from the old VLA system have a systematic offset in RA of $\sim +20$ mas with respect to the radio and optical reference frames.
Note that this systematic error is very much smaller than the positional uncertainties on sources in the [[*FIRST*]{}]{} catalog; for even the brightest sources, the random position errors are $\sim100$ mas or larger. This error is small enough to be irrelevant except in large-scale cross-matches that compare astrometric systems.
The declination offsets also show some systematic variations in Table \[table-match\], although different external catalogs have different offsets. The SDSS matches have a typical declination offset of 15 mas, while the 2MASS offset is smaller, $\sim7$ mas. The NVSS-[[*FIRST*]{}]{} declination offset stands in stark contrast with a difference of 60 mas. As for Right Ascension, in the JVLA region the remaining declination offsets compared with SDSS are consistent with zero. We discount the large discrepancies from the NVSS match, which we find in §\[NVSS\] to have highly non-Gaussian position difference distributions with long tails. With that caveat, we conclude from the SDSS and 2MASS matches that systematic errors in the [[*FIRST*]{}]{}declinations at all epochs are of order 0.02 or less.
![Distribution of sidelobe probability $P(S)$ in the [[*FIRST*]{}]{} catalog. 76% of the sources are in the lowest bin, $P(S) < 0.05$. \[fig-sidehist\]](f6.eps){width="\linewidth"}
{width="0.5\linewidth"} {width="0.5\linewidth"}
![Variation of mean sidelobe probability $P(S)$ with distance from bright sources. The sidelobe likelihood generally declines with the flux density of the bright source and with distance from the bright source, as expected. The dashed line shows the mean $P(S)$ for all the catalog sources that are not within 25 arcmin of a source brighter than 100 mJy. \[fig-sideradial\]](f8.eps){width="\linewidth"}
![Tests of the accuracy of the sidelobe probability $P(S)$. (a) [[*FIRST*]{}]{}-NVSS match rate as a function of $P(S)$. Only isolated [[*FIRST*]{}]{} sources with integrated flux densities brighter than 3.0 mJy are included. For low $P(S)$ values, 95% of the [[*FIRST*]{}]{} sources are detected in NVSS. (b) [[*FIRST*]{}]{}-SDSS match rate as a function of $P(S)$. The decline in the match rate with increasing sidelobe probability is roughly consistent with expectations assuming that the fraction of real sources in each bin is proportional to $1-P(S)$. The distribution is not expected to be exactly linear since the radio flux density distribution varies with $P(S)$, and the observed SDSS match rate depends on radio flux (Fig. \[fig-2MASS-SDSSfractions\]). (c) Sidelobe probability esimated using the SDSS match rates in (b). The effect of variation of match rate with radio flux was corrected using a fit to the distribution in Fig. \[fig-2MASS-SDSSfractions\]. The points would lie along the diagonal line if the catalog $P(S)$ values were perfect predictors of the sidelobe fraction. The sidelobe probability $P(S)$ quoted in the catalog ($x$-axis) appears to be slightly underestimated for moderate probabilities ($0.1 < P(S) < 0.6$). \[fig-sidelobes\]](f9.eps){width="\linewidth"}
The Sidelobe Flagging Algorithm {#sidelobes}
===============================
The original versions of the [[*FIRST*]{}]{} catalog (beginning with the 1995 October 16 release) included a sidelobe warning flag that indicated a likelihood that the source entry was actually a sidelobe of a bright nearby source rather than a real object [@white1997]. The algorithm for setting this flag used an oblique decision tree classifier[^5] [@murthy1994] that was trained using a set of sidelobes identified by visual examination of some [[*FIRST*]{}]{} images.
While this approach had some value in identifying potentially spurious sources, it was not very accurate. It was easy to find cases where sidelobes were not flagged or real sources were incorrectly flagged. In addition, the use of a binary yes/no flag for sidelobe flagging did not provide much guidance as to the actual likelihood that a source was spurious. Consequently, this catalog entry was considered somewhat unreliable and saw relatively little use.
[@white2005] developed a more sophisticated and useful variation on this algorithm. We used deep observations of a portion of the Galactic plane from a different survey to determine objectively which sources in a catalog were spurious (not seen in the much deeper data) and which were real (confirmed in the deeper data). That produced a reliable training set for the decision tree classifier. We then created multiple independent decision trees whose output was combined to obtain a sidelobe probability estimate for each source rather than a simple binary classification. This voting decision tree approach is described in more detail by [@white2000] and [@white2008].
For the 2008 and later releases of the [[*FIRST*]{}]{} catalog, we adopted a similar approach to computing sidelobe probabilities for the [[*FIRST*]{}]{}sources. We created a training set of 1905 sources (including 120 sidelobes). We matched [[*FIRST*]{}]{} with two deep radio surveys, the Spitzer First Look Survey [416 [[*FIRST*]{}]{} sources, 15 sidelobes; @condon2003] and the COSMOS deep survey pilot area [72 [[*FIRST*]{}]{} sources, 6 sidelobes; @schinnerer2004][^6]. We also included a less reliable but larger sample from our own survey of the Deeprange area [1356 [[*FIRST*]{}]{} sources, 99 sidelobes; @white2003]. Finally we augmented the set with a sample of 61 bright sources ($F>2$ Jy), none of which are sidelobes; this improved the performance of the classifier for bright objects.
The 15 parameters used for the classification include nine source properties (peak-to-integrated flux ratio, rms noise level, source major and minor axes compared with the synthesized beam, source position angle, ratio of source peak and integrated flux densities to the corresponding values in the “island” to which the source belongs[^7]), and six properties of the nearest bright source that could be creating sidelobes (positional offsets, flux ratios, and directions). Ten independent oblique decision trees were created using this training set. Their outputs are combined as described in [@white2000] to estimate a sidelobe probability, $P(S)$, for every [[*FIRST*]{}]{} source.
A histogram of the resulting sidelobe probabilities is shown in Figure \[fig-sidehist\]. The vast majority of the sources have low sidelobe probabilities; in fact, 70% of the objects have the minimum $P(S)$ value of 0.014. The mean sidelobe probability for the catalog is 0.097, which is an estimate of the fraction of sidelobes in the catalog.
Figure \[fig-sidedist\] shows the spatial distribution of objects in the vicinity of bright radio sources, which frequently give rise to nearby sidelobes. Applying a probability cut $P(S) < 0.1$ (Fig. \[fig-sidedist\]b) eliminates the vast majority of the obvious sidelobe pattern. The effect of the bright source can also be seen in the radial dependence of the mean $P(S)$ (Fig. \[fig-sideradial\]). The probability is lower in the closest bins because of the tendency of bright radio sources to have multiple components (doubles, triples, etc.); away from the center the probability declines with radius and with the bright source flux density.
The sidelobes are heavily concentrated in the vicinity of bright sources. The 10% of [[*FIRST*]{}]{} catalog sources that fall within 10 arcmin of a bright object ($F > 100$ mJy) have a mean sidelobe fraction $P(S) = 0.24$ and account for 25% of all the sidelobes in the catalog. The remaining 90% of [[*FIRST*]{}]{} sources (not near a bright object) have a much lower sidelobe fraction, $P(S) = 0.08$. We conclude that the sidelobe probability behaves qualitatively as expected, and that the $P(S)$ value does separate likely sidelobes from other sources in the vicinity of bright objects.
We rely on matches to external catalogs for quantitative assessments of the accuracy of the [[*FIRST*]{}]{} sidelobe probabilities. The NVSS-[[*FIRST*]{}]{} cross-match (§\[NVSS\]) provides one obvious test of $P(S)$. It is not definitive, however, because the 2.5 mJy NVSS detection limit is significantly shallower than [[*FIRST*]{}]{}. The great majority of sources with high $P(S)$ values are faint in the radio: the mean peak flux density for sources with $P(S)>0.1$ is only 1.35 mJy, and only $\sim4$% of those objects have flux densities above the NVSS detection threshold. That means that NVSS can only be used to confirm the accuracy of $P(S)$ for bright sources; it does not provide any information about the much more common faint sidelobes. Nonetheless, the comparison is useful. Figure \[fig-sidelobes\](a) shows the NVSS detection fraction for [[*FIRST*]{}]{} sources as a function of sidelobe probability. We include only [[*FIRST*]{}]{} sources that are bright enough that they ought to be detected by NVSS. The detection fraction declines as expected as $P(S)$ increases. The general trend toward fewer detections at higher $P(S)$ is clear and confirms that the sidelobe probabilities are reliable for brighter [[*FIRST*]{}]{} sources.
The cross-match between [[*FIRST*]{}]{} and SDSS (§\[SDSS\]) provides a more powerful test of the sidelobe probabilities. That may be surprising since the optical counterparts of most [[*FIRST*]{}]{} sources are too faint to be detected by SDSS; the absence of an optical counterpart does not reveal much about the reality of an individual radio source. However, the SDSS match fraction provides an accurate *statistical* measurement of the sidelobe fraction down to the detection limit of the [[*FIRST*]{}]{}survey. The complication in this case is that the fraction of sources with optical counterparts depends on the radio flux density, but that variation is relatively smooth and can be easily modeled.
Figure \[fig-sidelobes\](b) displays the fraction of [[*FIRST*]{}]{}sources with an SDSS counterpart as a function of the sidelobe probability. The distribution is well-behaved and clearly shows the expected decline with the increasing fraction of spurious sources (which naturally do not have SDSS counterparts). In this figure no corrections have been made for the dependence of the SDSS match fraction on radio flux density. For Figure \[fig-sidelobes\](c), we have both corrected for that flux dependence and also have inverted the distribution, using the corrected SDSS match fraction as a measurement of the sidelobe probability. This can then be compared directly with the catalog value for $P(S)$. The points would fall along the diagonal line if the catalog $P(S)$ values were perfect predictors of the sidelobe fraction. The actual sidelobe probabilities appear to be slightly higher than the catalog estimates, but the catalog $P(S)$ values do appear to be reasonably accurate.
In summary, the sidelobe probabilities $P(S)$ in the [[*FIRST*]{}]{}catalog give a useful measure of the likelihood that any source is spurious. The probabilities have been shown to be sufficiently accurate to be useful, and they do a good job of eliminating the spurious detections that tend to cluster around bright radio sources.
![The positional offsets in RA for all [[*FIRST*]{}]{}-NVSS radio source matches (in black) and for isolated [[*FIRST*]{}]{} sources defined as there being only one such source within 50(the NVSS synthesized beam FWHM) of an NVSS source. The distribution is highly symmetrical and centered very close to zero (see Table \[table-match\] for quantitative details). The distribution for all sources is significantly wider as single NVSS sources are often resolved into multiple components by [[*FIRST*]{}]{}.\[fig-NVSS\_astrometry\]](f10.eps){width="\linewidth"}
{width="0.5\linewidth"} {width="0.5\linewidth"}
![The ratio of the flux density offsets ([[*FIRST*]{}]{} divided by NVSS) between [[*FIRST*]{}]{} data taken with the VLA and the JVLA. All data are for sources with $\delta < 15^{\circ}$ to eliminate declination effects. The horizontal dashed line at $+3.3\%$ represents the expected offset for sources with a mean spectral index of $\alpha = -0.7$, given the difference in effective frequency for the two set of observations.\[fig-nonJVLA\_JVLA\_ratio\]](f12.eps){width="\linewidth"}
![Fraction of [[*FIRST*]{}]{} sources that are detected by NVSS as a function of the [[*FIRST*]{}]{} integrated flux density using a 15 matching radius. The vertical dashed line shows the NVSS 2.5 mJy detection limit. The black dotted line shows the distribution for all [[*FIRST*]{}]{} sources. The small fraction of bright unmatched sources are extended objects that are resolved into multiple components by [[*FIRST*]{}]{}. That is demonstrated clearly by the solid red line, which shows the distribution for isolated [[*FIRST*]{}]{} sources (having no neighbors within 50). \[fig-nvssdet\]](f13.eps){width="\linewidth"}
Comparisons with NVSS: Completeness and Reliability {#NVSS}
===================================================
The NRAO VLA Sky Survey [NVSS — @condon1998], also conducted at 20 cm, covered over $3\pi$ steradians of the sky north of $\delta =
- 40^{\circ}$ to a completeness limit outside of the Galactic plane of roughly 2.5 mJy. The synthesized beam size was $\sim 45\arcsec$ leading to rms positional uncertainties of 7for point sources at the catalog detection limit, with errors for brighter sources decreasing inversely with flux density to $\sim 1$ for the brightest sources[^8]. The large beam size allowed the detection of extended, low-surface-brightness objects that can be resolved out by the high-resolution (5) [[*FIRST*]{}]{} beam; in addition, for sources with angular sizes between $\sim 10\arcsec$ and 60, the [[*FIRST*]{}]{} flux densities underestimate the true integrated source intensity.
In Figure \[fig-NVSS\_astrometry\] we show the positional offsets in RA for all [[*FIRST*]{}]{}-NVSS radio source matches (in black) and for all isolated [[*FIRST*]{}]{} sources defined such that there is only one such source within 50 of the NVSS source position (roughly equal to the NVSS synthesized beam FWHM). The distribution is highly symmetrical and centered very close to zero (see Table \[table-match\] for quantitative details); the distribution of offsets in declination is indistinguishable from the RA distribution. The distribution for all sources is significantly wider as single NVSS sources are often resolved into multiple components by [[*FIRST*]{}]{}. Both distributions are distinctly non-Gaussian. Thus, rather than quote an rms in Table \[table-match\], we record the 68.3% percentile of the absolute value distribution, which is the equivalent of the rms for a Gaussian distribution.
In Figure \[fig-NVSS\_FIRST\_ratio\](a) we plot the ratio of [[*FIRST*]{}]{}source flux density to NVSS flux density for all matching sources at declinations $<15^{\circ}$ (imposed for comparison with Fig. \[fig-NVSS\_FIRST\_ratio\]b) as a function of NVSS flux density. The brighter of the peak or integrated [[*FIRST*]{}]{} flux density is used. The blue bar shows a ratio of unity, while the red dots show the median values for the bins indicated by the horizontal error bars (vertical error bars are smaller than the points). The 2.0 mJy cutoff on the left is the NVSS threshold; the curving cutoff in the lower left represents the [[*FIRST*]{}]{} threshold of 1.0 mJy. The $\sim 1-5\%$ deficit in [[*FIRST*]{}]{} flux density between 2.0 and 20 mJy is likely the result of diffuse flux from extended sources resolved out by the [[*FIRST*]{}]{} beam; the turn-up for the lowest two flux density points arises from the imposition of the 1 mJy [[*FIRST*]{}]{} threshold that biases the distribution upward. The rise above unity at NVSS flux densities $>50$ mJy could arise from a calibration offset of $\sim 2\%$ plus a (possibly dominant) contribution from the different bandwidths used in the two surveys convolved with the source spectral index distribution.
Figure \[fig-NVSS\_FIRST\_ratio\](b) displays the same plot as Figure \[fig-NVSS\_FIRST\_ratio\](a) for sources observed with the JVLA in 2011. For the JVLA data, the agreement with NVSS fluxes is actually somewhat better for flux densities greater than 4 mJy. In Figure \[fig-nonJVLA\_JVLA\_ratio\], we plot the ratio of the flux density offsets ([[*FIRST*]{}]{} divided by NVSS) between [[*FIRST*]{}]{}data taken with the VLA and with the JVLA. All data are for sources with $\delta < 15^{\circ}$ to eliminate any declination-dependent effects such as changes in the PSF. We believe there are several competing effects that produce the variations seen here. First, sources with typical spectral indices should be slightly brighter in the lower-frequency JVLA observations (1.335 vs. 1.400 GHz – see §\[hardware\]). The horizontal dashed line at $+3.3\%$ represents the expected offset for sources with a mean spectral index of $\alpha = -0.7$. That effect changes with flux, however, since the spectral indices become flatter for compact bright sources. A second effect is that the lower-frequency JVLA observations are also slightly lower resolution, which increases their sensitivity to extended emission. That is likely the reason why the dip seen from 3–10 mJy in Figure \[fig-NVSS\_FIRST\_ratio\](a) is not seen in Figure \[fig-NVSS\_FIRST\_ratio\](b).
This effect illustrates the sensitivity of large surveys to small changes in observing parameters. Subtle changes can lead to noticeable differences in large statistical samples of sources, so flux differences between the JVLA and VLA data at the level of 5–10% (Fig. \[fig-nonJVLA\_JVLA\_ratio\]) should be treated with caution. Correcting for these effects would require knowledge of the spectral indices and sizes of the sources, which are usually not available. Note, however, that these systematic effects are small compared with the noise for faint sources, so most studies can treat the JVLA and VLA data as having equivalent flux scales.
Figure \[fig-nvssdet\] shows the fraction of [[*FIRST*]{}]{} sources detected by NVSS as a function of the [[*FIRST*]{}]{} integrated flux density. The factor of 10 difference between the resolutions of the two surveys complicates the interpretation of this figure. Many NVSS sources are resolved by [[*FIRST*]{}]{} into multiple components. Those sources will appear as lower flux [[*FIRST*]{}]{} sources detected by NVSS, when in fact NVSS only detects the sum of the components; that accounts for most of the tail of detections at fainter fluxes. Moreover, such sources will often have large positional differences so that they do not match within the 15 matching radius we are using here; that accounts for the small fraction of bright [[*FIRST*]{}]{} sources that are undetected by NVSS. The detection fraction for isolated [[*FIRST*]{}]{}sources is also shown in Figure \[fig-nvssdet\]; essentially all bright isolated [[*FIRST*]{}]{} sources are detected by NVSS.
![The positional offsets in RA (black), and Dec (red) between [[*FIRST*]{}]{} radio sources and 2MASS objects. The distribution is highly symmetrical and centered very close to zero (see Table \[table-match\] for quantitative details).\[fig-2MASS\_astrometry\]](f14.eps){width="\linewidth"}
![ [[*FIRST*]{}]{}-2MASS matches as a function of [[*FIRST*]{}]{} flux density. The 2MASS sources have been divided into stellar objects (mainly AGN) and galaxies using the SDSS classification. (a) The raw counts of [[*FIRST*]{}]{} sources and 2MASS matches as a function of flux. Only sources with low sidelobe probabilities, $P(S) \le 0.02$, are included, which is why the number of sources drops near the detection limit.. (b) The fraction of 2MASS matches as a function of flux. The bulk of 2MASS sources detected by [[*FIRST*]{}]{} are galaxies. \[fig-2MASS\_star\_galaxy\_fractions\]](f15a.eps "fig:"){width="\linewidth"} ![ [[*FIRST*]{}]{}-2MASS matches as a function of [[*FIRST*]{}]{} flux density. The 2MASS sources have been divided into stellar objects (mainly AGN) and galaxies using the SDSS classification. (a) The raw counts of [[*FIRST*]{}]{} sources and 2MASS matches as a function of flux. Only sources with low sidelobe probabilities, $P(S) \le 0.02$, are included, which is why the number of sources drops near the detection limit.. (b) The fraction of 2MASS matches as a function of flux. The bulk of 2MASS sources detected by [[*FIRST*]{}]{} are galaxies. \[fig-2MASS\_star\_galaxy\_fractions\]](f15b.eps "fig:"){width="\linewidth"}
Match to the 2MASS Catalog {#2MASS}
==========================
The first deep image of the entire sky at 2$\mu$m was produced by the 2MASS Survey between 1997 and 2001 [@skrutskie2006]. The primary data products from the survey are an image atlas, and point and extended source catalogs containing over 470 million objects.
Astrometric data from the match of the [[*FIRST*]{}]{} catalog to the 2MASS point source catalog are reported in Table \[table-match\] and displayed in Figure \[fig-2MASS\_astrometry\]. We calculate the offsets between all [[*FIRST*]{}]{} source positions and any 2MASS object within $\pm 3\arcsec$ in each coordinate; this large box includes some chance coincidences but is highly complete even for extended objects. The astrometric offsets and rms widths are determined from Gaussian fits to the distribution in each coordinate. There is a statistically significant $+20$ mas offset in Right Ascension and a $+10$ mas offset in declination that persists even when only bright ($>10$ mJy) radio sources are used in the matching; as noted in §\[astrometry\], we believe that the RA offset, at least, arises from a systematic error in the VLA data acquisition system which the JVLA has corrected.
The rms uncertainties are the same in both coordinates. For the [[*FIRST*]{}]{} convolving beam size of 5.4, even bright point sources ($>50 \sigma$) will have an inherent rms positional uncertainty of $\sim
0.1\arcsec$ [@white1997]; the 2MASS 2.0 pixel size convolved with variable seeing led to an astrometric accuracy of $\lesssim 0.10\arcsec$ relative to the Hipparcos reference frame for objects with $K_s<14$ [@skrutskie2006]. The fact that the large majority of the 2MASS matches are extended galaxies (see Fig. \[fig-2MASS\_star\_galaxy\_fractions\]) plausibly makes up the remainder of the 0.23 value reported in Table \[table-match\].
In Figures \[fig-2MASS\_star\_galaxy\_fractions\] and \[fig-2MASS-SDSSfractions\], we show the fraction of [[*FIRST*]{}]{} radio sources with 2MASS counterparts as a function of radio flux density. The match fractions have been corrected for the effects of false matches using the density of 2MASS objects that fall between 7.5 and 8 from the [[*FIRST*]{}]{} source. The [[*FIRST*]{}]{}-2MASS match fraction falls from 100% at the brightest flux densities ($>5$ Jy) to less than $5\%$ at 40 mJy and then begins a steady rise to 10% near the survey threshold of 1 mJy. This is a consequence of the two population components that comprise the radio $\log N-\log S$ curve in the 1–1000 mJy range: “monsters” (radio-loud active galactic nuclei) and “normal” galaxies [@condon1992]; see the end of §\[SDSS\] for further discussion of the effects of the transition between populations. This plot includes only [[*FIRST*]{}]{} objects with low sidelobe probabilities, $P(S) \le 0.02$ (§\[sidelobes\]); that avoids contaminating the counts in the lowest flux bins with spurious sources. The fact that the 2MASS points all lie below the SDSS fraction of detections (Fig. \[fig-2MASS-SDSSfractions\]) is simply a consequence of the shallower depth of the 2MASS images.
![The fraction of [[*FIRST*]{}]{} radio sources with counterparts in the SDSS and 2MASS catalogs (using a 2 arcsec matching radius) as a function of peak radio flux density. Horizontal error bars represent the flux density ranges, while vertical error bars represent statistical uncertainties. Values are corrected for the false match rate. Only sources with low sidelobe probabilities, $P(S) \le 0.02$, are included. Both curves show a minimum where the radio source counts change from being dominated by AGN at high flux densities to being dominated by normal galaxies.\[fig-2MASS-SDSSfractions\]](f16.eps){width="\linewidth"}
Match to the SDSS Catalog {#SDSS}
=========================
In recognition of the high scientific value to a radio survey of having complementary optical data from which to derive radio source identifications, the original [[*FIRST*]{}]{} survey footprint was designed to largely overlap the (then-planned) Sloan Digital Sky Survey (SDSS). In the end, 93% of the [[*FIRST*]{}]{} sky coverage is also covered by SDSS. The extension to the original [[*FIRST*]{}]{}footprint, approved in 2008, was added in order to provide complementary data to a portion of the SDSS III survey.
The SDSS I and II projects collected imaging data between 2000 and 2008 over more than 10,000 deg$^2$ of the northern and southern Galactic caps in five colors, as well as obtaining spectra of over one million objects within the survey area. The images and catalogs resulting from this effort are summarized in Data Release 7 [@abazajian2009]. Data Release 10 in July 2013 increased the total sky coverage to over 14,500 deg$^2$ and brought the catalog to over one billion objects of which over 1.6 million have spectra [@ahn2014]. The [[*FIRST*]{}]{} survey covers 68% of the SDSS DR10 sky area.
![The positional offsets in RA (black), and Dec (red) between [[*FIRST*]{}]{} radio sources and SDSS objects. As for the 2MASS matches, the distribution is highly symmetrical and centered close to zero (see Table \[table-match\] for quantitative details).\[fig-SDSS\_astrometry\]](f17.eps){width="\linewidth"}
The first attempt at large-scale matching of SDSS and [[*FIRST*]{}]{}sources was published by [@ivezic2002]. The $\sim
0.1\arcsec$ astrometric offset in declination they found was subsequently corrected in the SDSS astrometry pipeline. Figure \[fig-SDSS\_astrometry\] and Table \[table-match\] show the results of the final match presented here for various optical and radio source properties. As with the 2MASS matches, we determine the astrometric offsets and rms widths by fitting a Gaussian to the distribution of SDSS sources found within $\pm 3\arcsec$ of a [[*FIRST*]{}]{} source in each coordinate. We report the match properties between all [[*FIRST*]{}]{} and SDSS objects as well as for subsets of the radio sources (point sources, bright $>10$ mJy sources) matched with optical objects separated into stellar and galaxy counterparts. As with 2MASS, we see a small but statistically significant $+20$ mas offset in Right Ascension and a $+10$ to $+15$ mas offset in declination (§\[astrometry\]).
The uncertainties in RA and declination are essentially identical. For the brighter ($S/N \gtrsim 50$) radio sources matched to point-like optical counterparts, the rms of $\sim 0.15\arcsec$ is consistent with the reported rms positional uncertainty for SDSS ($\sim 0.1\arcsec$), coupled with a similar uncertainty for [[*FIRST*]{}]{} sources.
[@ivezic2002] reported that roughly 30% of the initial batch of $10^5$ [[*FIRST*]{}]{} sources had optical counterparts in SDSS, the large majority of which (83%) were resolved optical objects (i.e., galaxies); galaxies accounted for $\sim 50\%$ of the counterparts at the brightest radio flux densities, rising to 90% at the survey threshold of $\sim 1$ mJy. The total fraction of SDSS matches in the completed catalogs, shown in Figure \[fig-SDSS\_star\_galaxy\_fractions\_linear\], is roughly consistent with this overall match rate.
![The fraction of SDSS matches to [[*FIRST*]{}]{} sources as a function of [[*FIRST*]{}]{} flux density using a matching radius of 2 arcsec. The SDSS sources have been divided into stellar objects (AGN) and galaxies. Only sources with low sidelobe probabilities, $P(S) \le 0.02$, are included. \[fig-SDSS\_star\_galaxy\_fractions\_linear\]](f18.eps){width="\linewidth"}
![The fraction of [[*FIRST*]{}]{} radio sources with counterparts in the SDSS divided by counterpart classification: stellar objects which, with the exception of a handful of radio stars, are quasars or other AGN, and galaxies. Only sources with low sidelobe probabilities, $P(S) \le 0.02$, are included. The fraction of sources with stellar counterparts declines monotonically from 100% to 3%, while those with galaxy counterparts transition from AGN-dominated systems at high flux densities to radio emission dominated by star formation near the survey threshold. \[fig-SDSS\_star\_galaxy\_fractions\]](f19.eps){width="\linewidth"}
Figures \[fig-SDSS\_star\_galaxy\_fractions\_linear\] and \[fig-SDSS\_star\_galaxy\_fractions\] show the SDSS match to the completed [[*FIRST*]{}]{} survey over a radio flux density range of a factor of $10^4$. At the bright end, the majority of radio sources have SDSS counterparts[^9]. Stellar counterparts (all but a handful of which are quasars) fall monotonically to $\sim 4\%$ of all [[*FIRST*]{}]{} sources at the survey threshold, while galaxy counterparts fall to a minimum of a 10% identification rate at 100 mJy and then rise again to $>30$% of all [[*FIRST*]{}]{} sources at 1 mJy; this reflects the long-established change in the radio source population mix as a function of flux density [@condon1992].
![Magnitudes and colors of SDSS stars (black) and galaxies (red) as a function of [[*FIRST*]{}]{} peak flux density. (a) Median $i$-band magnitudes. Stellar objects (mainly AGN) have radio and optical fluxes that decline in concert; the flattening around $i=21$ occurs as the SDSS detection limit is approached. Galaxies, on the other hand, optically brighten with decreasing radio flux density as the population shifts from high-redshift AGN in radio-bright galaxies to low-redshift star formation associated with mJy sources. (b) Median $g-r$ colors. Stellar objects (quasars) are typically more than 1 magnitude bluer than galaxies. Both quasars and galaxies get redder as the radio flux density decreases, presumably due to the increasing redshift of fainter radio sources. Galaxies begin to get bluer for $F<3$ mJy as the nearby star-forming galaxies come into view. \[fig-SDSS\_mags\_colors\]](f20.eps){width="\linewidth"}
The magnitudes and colors of the SDSS counterparts to [[*FIRST*]{}]{} sources also vary systematically with radio flux density (Figs. \[fig-SDSS\_mags\_colors\]a and \[fig-SDSS\_mags\_colors\]b). Changes in both of these quantities result from the transition from AGN-powered radio emission for the brightest radio sources to star-formation-powered radio emission for milliJansky radio sources. Around 1 Jy, the supermassive black hole “monsters” dominate the counterparts for both stellar objects and galaxies; objects appear as blue, point-like quasars when nuclear emission also dominates the optical, while radio galaxies have similar radio fluxes but are 10 times fainter in the optical. As the radio flux declines from 1 Jy to 1 mJy, quasars become both fainter and redder, mainly because the population is shifting to higher redshift. The magnitudes and colors of galaxies change little from 1 Jy to 100 mJy because the radio galaxy counterparts to bright radio sources are already close to the SDSS detection limit. However, below 100 mJy the median magnitudes of [[*FIRST*]{}]{}-selected galaxies actually get brighter as the radio flux decreases. This somewhat unexpected result is caused by the [[*FIRST*]{}]{} detection of nearby ($z<0.5$) star-forming galaxies. These objects are low luminosity radio sources but are bright optical sources. As star-forming galaxies take over the sample compared with low-luminosity radio galaxies, the median galaxy brightness increases. The galaxy colors initially get redder because of the declining contribution of blue nuclear emission, and then around 4 mJy the galaxy colors begin to become bluer as nearby galaxies with high star-formation rates are detected.
Conclusions: Lessons for future radio sky surveys {#conclusions}
=================================================
Rather than reiterate our results in a summary, we draw conclusions from the two-decade experience of the [[*FIRST*]{}]{} survey that may be of use for the next-generation projects to map the radio sky. We include remarks on scheduling, the continuing usefulness of uniform sky surveys, and the all-important matter of angular resolution in radio source identification, concluding with brief remarks on the value of a JVLA sky survey.
Scheduling
----------
A priority for any sky survey is uniformity. This is best achieved when the hardware, software, and researchers change little over the course of the project. Our original proposal to survey the radio sky with the VLA suggested arranging for a special, hybrid array to yield both high resolution (and thus, high astrometric accuracy) and high surface brightness sensitivity (to detect nearby galaxies), and devoting six months to the project, after which normal VLA operations would resume. In the event, two separate surveys were conducted, consuming approximately nine months of observing time; in the case of the [[*FIRST*]{}]{} survey, this was spread over eighteen years. As noted earlier, the allocated observing windows were not optimally matched to the sky area to be covered, meaning the observations often had to be carried out off the meridian (which would have been optimal). Queue scheduling, the antithesis of careful planned and optimized observing windows, should be avoided at all costs for survey observations.
Hardware and software changes accumulate the longer a survey takes. Examples from the [[*FIRST*]{}]{} experience are briefly summarized in §\[history\]; the changes to the researchers over two decades are best left to the reader’s imagination.
The legacy value of uniform sky surveys
---------------------------------------
The papers of record describing the two VLA surveys — [@becker1995] and [@condon1998] — have received over 4000 citations; both papers recorded their highest annual citation rates in 2014, nineteen and sixteen years after their publication, respectively. Over 6 million snapshot images have been downloaded from the [[*FIRST*]{}]{} survey web site alone. If those queries replaced three-minute snapshot observations, the observing time saved amounts to 75 times the total time invested in the survey by the VLA. And the number of images accessed continues to increase with time. Over the past two years (2012 June through 2014 June), more than 2.5 million image cutouts were extracted from our image server by 2500 different users around the world. Every 20 seconds our server delivers a cutout that is the the equivalent of a 1-minute snapshot observation with the current JVLA receivers; in 3 weeks our server distributes snapshots with a combined exposure time equivalent to the entire 4000 hours of VLA time that was allocated for the [[*FIRST*]{}]{} survey. The investment of observing time in the [[*FIRST*]{}]{} survey continues to pay dividends to the astronomical community.
Angular resolution and source identification {#resolution}
--------------------------------------------
Radio surveys require high angular resolution in order to have positions sufficiently accurate to obtain source identifications with objects at other wavelengths. Optical and infrared counterparts of even relatively bright radio sources are faint: e.g., only 33% of [[*FIRST*]{}]{} radio sources have an optical counterpart bright enough to be detected in SDSS. An important corollary to the fact that radio source identifications are faint is that potential counterparts are dense on the sky, and accurate radio positions are required to confidently associate the radio and optical objects.
### Signal-to-noise ratios and source positions
Can deeper radio observations at low resolution be a substitute for higher resolution observations? A common argument of advocates for lower resolution surveys is that as the signal-to-noise ratio (SNR) increases, the positions of catalog sources improve. Consequently one does not really need high resolution to do optical identifications. The prediction of this *SNR model* is that as the flux density increases, the positional error will decrease as $1/{SNR}$, allowing the optical counterpart to be confidently matched. Specifically, the NVSS description [@condon1998] gives this formula for the noise in RA or Dec for point sources: $${\sigma_{\hbox{1D}}}= \theta / ({SNR}\sqrt{2\ln 2}) \quad .
\label{eqn-sigmaoned}$$ Here $\theta$ is the resolution FWHM (45 for NVSS) and ${SNR}$ is the signal-to-noise ratio. The median NVSS rms noise for objects that match [[*FIRST*]{}]{} sources is 0.47 mJy. Note that this noise equation already has been increased by an empirical factor of $\sqrt{2}$ compared with the theoretical equation “to adjust the errors into agreement with the more accurate [[*FIRST*]{}]{} positions” (quoting the NVSS catalog description[^10]) This predicts ${\sigma_{\hbox{1D}}}\sim 7.6$ arcsec at the catalog detection limit (${SNR}=5$) and ${\sigma_{\hbox{1D}}}\sim 1$ arcsec at a flux density of 18 mJy.
The positional scatter in Eq. (\[eqn-sigmaoned\]) is a 1-dimensional uncertainty, giving the error in either RA or Dec. In a 2-dimensional distribution, many values will scatter outside the 1-sigma circle. The 90% confidence separation limit ${\sigma_{90}}$, which is typically more appropriate for catalog matching, is a constant factor $\sqrt{2\ln10}$ times larger than ${\sigma_{\hbox{1D}}}$: $${\sigma_{90}}= {\theta \over {SNR}} \sqrt{\ln 10 \over \ln 2} \quad .
\label{eqn-sigmaninety}$$ With this increase it is necessary for the NVSS flux density to exceed 40 mJy (${SNR}=85$) to reduce the predicted separation error to 1 arcsec.
{width="0.8\linewidth"}
{width="0.5\linewidth"} {width="0.5\linewidth"}
### Does the SNR model work for NVSS?
The above positional accuracy applies to perfect point sources (and perfect data). But how well does it work for real data? We can assess the accuracy of Eqn. (\[eqn-sigmaninety\]) using a comparison of the [[*FIRST*]{}]{} and NVSS data.
We selected a sample of all the [[*FIRST*]{}]{} sources that have an SDSS match within 0.7 arcsec and that have an NVSS match within 100 arcsec. We restricted the sample to sources with [[*FIRST*]{}]{} peak flux densities greater than the 2.5 mJy NVSS detection limit. For all these $\sim95,000$ sources, we computed the distance to the nearest NVSS source. The important thing about this sample is that the [[*FIRST*]{}]{} source matches the optical source position. That means that if NVSS is to identify the same counterpart, it needs to have a position close to the [[*FIRST*]{}]{} source position. There may be several [[*FIRST*]{}]{}source components associated with a single NVSS source, but only the [[*FIRST*]{}]{} sources that match optical counterparts are included.
How do the positional errors in Eqn. (\[eqn-sigmaninety\]) compare with reality? Fig. \[fig-nvsssep\] shows the position distance between the NVSS and [[*FIRST*]{}]{} positions as a function of the NVSS flux density. The positional differences decrease as expected as the flux densities increase. The blue line shows the 90% confidence separation limit ${\sigma_{90}}$ from Eq. (\[eqn-sigmaninety\]), simply assuming that all objects are point sources with the median NVSS rms value. Between 1 and 100 mJy, this line generally tracks the decline in positional differences as the flux density increases; however, there are still many points above the line.
The red histogram shows the empirical 50, 90, 95, and 99% confidence separation limits as a function of flux density, computed by determining the relevant percentile of the actual separations in each bin. The computed separation has been corrected for the effects of chance nearby NVSS associations. *The actual 90% confidence radius shows no improvement in the positions for flux densities greater than 4 mJy, and it is much larger than the predicted 90% curve.*
This empirical distribution does not look like the theoretical distribution. Remember that there is an optical counterpart near zero separation in these plots for every source. To find 90% of those counterparts using the NVSS positions, it is necessary to use a matching radius of approximately 7 arcsec ($\sim0.15\,\theta$) even for sources that are 100 times the rms noise level. The theoretical SNR model predicts that the positions for such bright sources ought to be much more accurate than that (${\sigma_{90}}= 0.8\arcsec$). To include 95% of the counterparts requires a matching radius of 15 ($\sim0.3\,\theta$), while finding 99% of the counterparts requires matching out to 39 ($\sim\theta$).
### Why are the low-resolution positions so inaccurate?
Why are the inaccuracies in the positions so much greater than the SNR model predictions? Real radio sources are not symmetrical objects. They have lobes, jets, and cores; star-forming galaxies have spiral arms. And there can be confusion when multiple radio sources get mixed together in a low-resolution beam. A low-resolution survey does indeed give a very accurate measurement of the mean flux-weighted position as the SNR increases. *However, the flux-weighted centroid is often not where the optical counterpart lies.* In many cases the counterpart is associated with some sharp structure within the radio source, and that structure may be far from the flux-weighted center.
![ FWHM resolution required to achieve reliable cross-matches at fainter magnitudes using $r$-band galaxy counts. The curves define the limits for 90% and 95% reliable identifications, and the resolutions of ASKAP-EMU, WODAN, and a VLA 3 resolution survey are shown. The depths of various $r$-band surveys are also shown by the vertical lines (Pan-STARRS, DESI, LSST, HSC). The SKA pathfinders are at best marginally sufficient for identifications at SDSS/Pan-STARRS depths, while a VLA S-band survey is usable with the much deeper surveys. \[fig-rcount\]](f23.eps){width="\linewidth"}
### Effect on optical identifications
This analysis demonstrates that matching at the 45 resolution of NVSS requires a matching radius of 15= 30% of the NVSS FWHM resolution to achieve 95% completeness. Our experience with the FIRST survey is similar: to get a reasonably complete list of optical identifications we had to use a matching radius of 2 $\sim$ 40% of the FIRST FWHM resolution. We argue this is a universal requirement for radio sources, at least for sources down to the sub-mJy regime: the matching radius that is required for realistic radio source morphologies is at least 30% of the FWHM resolution for 95% completeness.
The planned Square Kilometer Array (SKA) precursor surveys have relatively low resolution. The resolution for WODAN [@rottgering2011] is of order $15\times17$ arcsec, while the resolution for ASKAP-EMU [@norris2011] is 10 arcsec. For 95% completeness, WODAN will therefore require an optical matching radius of 4.8 arcsec and ASKAP-EMU will require 3 arcsec. Such large matching radii are a serious problem for optical matching. The cross-match between SDSS and [[*FIRST*]{}]{} shows that 19% of [[*FIRST*]{}]{} sources have a false (chance) SDSS counterpart within 4.8. For comparison, 33% of [[*FIRST*]{}]{}sources have a true match within 2. So one-third of the optical counterparts at SDSS depth will be false matches when using a 4.8 matching radius. Of course the false rate can be reduced somewhat by doing a careful analysis of the likelihood of association as a function of separation, but when the starting point is a sample that is contaminated by 33% false matches, the final list of identifications is going to be neither complete nor reliable.
The false matching problem will only get worse for deeper optical/IR data. For example, Pan-STARRS [PS1 – @kaiser2002] goes more than 1 magnitude deeper than SDSS in the red and also goes into the Galactic plane where the source density is much higher, demanding better resolution. We have used the sky density of objects in a preliminary PS1 catalog to compute the likelihood of false identifications in PS1 as a function of Galactic latitude. The left panel of Fig. \[fig-ps1false\] compares WODAN to a VLA S-band (2–4 GHz) B-configuration survey having 3 resolution. For WODAN, 10% of sources even in the extragalactic sky ($|b|>30^\circ$) will have a spurious counterpart in PS1. For most purposes that is an unacceptable level of contamination. In contrast, a VLA survey has only a $<1$% contamination rate in the extragalactic sky, and is usable even quite close to the Galactic plane.
The right panel of Fig. \[fig-ps1false\] turns this around and asks what FWHM resolution is required to achieve a 95% reliability ($\sim2\sigma$) in matches to the PS1 catalog. At $|b|>30^\circ$ a FWHM resolution of 11 is required. That is significantly higher resolution than WODAN and just at the resolution reached by ASKAP-EMU, but one easily satisfied by VLA surveys. In fact, a VLA survey with a resolution of 3 has 95% confident PS1 matches over 99% of the current PS1 catalog area, with only the most crowded areas of the Galactic plane requiring higher resolution.
The next generation of optical/IR surveys will be considerably deeper than Pan-STARRS. Fig. \[fig-rcount\] shows the resolution required as a function of magnitude using the $r$-band galaxy counts from the CFHTLS-D1 1 $\deg^2$ survey [@mccracken2003]. Since this does not include stars or redder galaxies, it is more optimistic (and less realistic) at the PS1 limit, but it shows the resolution required for deeper identifications. For 90% reliable identifications, a 3-resolution VLA survey can be used to $r=27.2$, ASKAP-EMU to $r=25.3$, and WODAN to $r=24.3$. For 95% reliable identifications, the magnitude limits are 24.9 (VLA), 22.1 (ASKAP-EMU), and 20.7 (WODAN). The SKA-precursor surveys are usable at the depth of SDSS and Pan-STARRS in the extragalactic sky, but fall well short of the required resolution at fainter magnitudes. The higher resolution VLA survey, by contrast, is useful at least to $r=26$.
The inescapable conclusion is that we need high resolution to get the accurate positions required for optical identifications. Deeper radio imaging is not a substitute for the requisite angular resolution.
A JVLA sky survey
-----------------
A new generation of radio sky surveys will soon be conducted by survey arrays such as LOFAR [@devos2009], ASKAP [@johnston2008], MEERKAT [@booth2009] and, ultimately, perhaps, the SKA [@schilizzi2010]. But the analysis above shows that none of the three SKA precursors provide the angular resolution necessary for unambiguous radio source identification or to resolve complex source regions. In addition, they operate at wavelengths of 20 cm or longer (initially, at least). It is thus worth considering whether the substantially enhanced JVLA might again be used for a sky survey.
The use of the JVLA for $\sim 7$ months could produce a $3\times10^4$ deg$^2$ survey at 2–4 GHz with an angular resolution two times that of [[*FIRST*]{}]{} and a sensitivity two times greater. This would produce over three times as many sources (taking into account the flux density falloff of most sources with frequency), one-quarter of which would have accurate spectral indices and all of which would have full Stokes parameters available. The positional accuracy for the three million sources would be better than 1.
The success of the [[*FIRST*]{}]{} survey is in large measure due to the generous support of a number of organizations. In particular, we acknowledge support from the NRAO, the NSF (grants AST 94-19906, AST 94-21178, AST-98-02791, AST-98-02732, AST 00-98259, and AST 00-98355), the Institute of Geophysics and Planetary Physics (operated under the auspices of the US Department of Energy by Lawrence Livermore National Laboratory under contract No. W-7405-Eng-48), the STScI, NATO, the National Geographic Society (grant NGS No. 5393-094), Columbia University, and Sun Microsystems.
This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation.
Funding for the SDSS and SDSS-II has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, the U.S. Department of Energy, the National Aeronautics and Space Administration, the Japanese Monbukagakusho, the Max Planck Society, and the Higher Education Funding Council for England. The SDSS Web Site is <http://www.sdss.org/>. Funding for SDSS-III has been provided by the Alfred P. Sloan Foundation, the Participating Institutions, the National Science Foundation, and the U.S. Department of Energy Office of Science. The SDSS-III web site is <http://www.sdss3.org/>.
Abazajian, K. N. et al. 2009, , 182, 543
Ahn, C. P., Alexandroff, R., Allende Prieto, C., et al. 2014, , 211, 17
Assafin, M., Vieira-Martins, R., Andrei, A. H., Camargo, J. I. B., & da Silva Neto, D. N. 2013, , 430, 2797
Becker, R. H., White, R. L., Helfand, D. J., 1995, , 450, 559
Booth, R. S., de Blok, W. J. G., Jonas, J. L., & Fanaroff, B. 2009, arXiv:0910.2935
Condon, J. J. 1992, , 30, 575
Condon, J. J., Cotton, W. D., Greisen, E. W., Yin, Q. F., Perley, R. A., Taylor, G. B., Broderick, J. J., 1998, , 115, 1693
Condon, J. J., Cotton, W. D., Yin, Q. F., et al. 2003, , 125, 2411
de Vos, M., Gunst, A. W., & Nijboer, R. 2009, IEEE Proceedings, 97, 1431
Gal-Yam, A., Ofek, E. O., Poznanski, D., et al. 2006, , 639, 331
Hodge, J. A., Becker, R. H., White, R. L., Richards, G. T., & Zeimann, G. R. 2011, , 142, 3
H[ö]{}gbom, J. A. 1974, , 15, 417
Ivezi[ć]{}, [Ž]{}., Menou, K., Knapp, G. R., et al. 2002, , 124, 2364
Johnston, S., Taylor, R., Bailes, M., et al. 2008, Experimental Astronomy, 22, 151
Kaiser, N., Aussel, H., Burke, B. E., et al. 2002, , 4836, 154
Kirkpatrick, S., Gelatt, C. D., & Vecchi, M. P. 1983, Science, 220, 671
Ma, C., Arias, F. E., Bianco, G., et al. 2013, VizieR Online Data Catalog, 1323, 0
McCracken, H. J., Radovich, M., Bertin, E., et al. 2003, , 410, 17
McMahon, R. G., & Irwin, M. J. 1992, Digitised Optical Sky Surveys, 174, 417
McMahon, R. G., White, R. L., Helfand, D. J., Becker, R. H., 2002, , 143, 1
Murthy, S. K., Kasif, S., & Salzberg, S. 1994, arXiv:cs/9408103
Norris, R. P., Hopkins, A. M., Afonso, J., et al. 2011, PASA, 28, 215
Ofek, E. O., Breslauer, B., Gal-Yam, A., et al. 2010, , 711, 517
Orosz, G., & Frey, S. 2013, , 553, A13
Reber, G., 1944, , 100, 279
R[ö]{}ttgering, H., Afonso, J., Barthel, P., et al. 2011, Journal of Astrophysics and Astronomy, 32, 557
Schilizzi, R. T., Dewdney, P. E. F., & Lazio, T. J. W. 2010, , 7733,
Schinnerer, E., Carilli, C. L., Scoville, N. Z., et al. 2004, , 128, 1974
Schinnerer, E., Smol[č]{}i[ć]{}, V., Carilli, C. L., et al. 2007, , 172, 46
Skrutskie, M. F. et al., 2006, , 131, 1163
Thyagarajan, N., Helfand, D. J., White, R. L., & Becker, R. H. 2011, , 742, 49
White, R. L., Becker, R. H., Helfand, D. J., Gregg, M. D., 1997, , 475, 479
White, R. L., Becker, R. H., Gregg, M. D., et al. 2000, , 126, 133
White, R. L., Helfand, D. J., Becker, R. H., et al. 2003, , 126, 706
White, R. L., Becker, R. H., & Helfand, D. J. 2005, , 130, 586
White, R. L., Helfand, D. J., Becker, R. H., Glikman, E., de Vries, W., 2007, , 654, 99
White, R. L. 2008, American Institute of Physics Conference Series, 1082, 37
York, D. G. at al. 2000, , 120, 1579
[^1]: <http://sundog.stsci.edu>
[^2]: <http://archive.stsci.edu/vlafirst/>
[^3]: The Karl G. Jansky Very Large Array is an instrument of the National Radio Astronomy Observatory, a facility of the National Science Foundation operated under cooperative agreement by Associated Universities, Inc.
[^4]: <http://sundog.stsci.edu/first/catalogs/history.html>
[^5]: The OC1 oblique decision tree software is available for download at [http://www.cbcb.umd.edu/\\textasciitilde{}salzberg/announce-oc1.html](http://www.cbcb.umd.edu/\textasciitilde{}salzberg/announce-oc1.html).
[^6]: The deeper imaging from [@schinnerer2007] was not available at the time.
[^7]: The islands are rectangular pixel regions surrounding sources as described in §3 of [@white1997].
[^8]: But see §\[resolution\] for a discussion of more realistic positional uncertainties for optical matching.
[^9]: In fact, just six of the 34 sources brighter than 3 Jy fail to match an SDSS object within 1. Three of these are components of M87 which, of course, does have a match; one falls outside the SDSS coverage; and the other two are 3C280 and PKS 2127${+}$04, both radio galaxies at $z=1$ that do fall below the SDSS threshold.
[^10]: <http://www.cv.nrao.edu/nvss/catalog.ps>
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study an uplink multiuser multiple-input multiple-output (MU-MIMO) system with one-bit analog-to-digital converters (ADCs). For such system, a supervised-learning (SL) detector has been recently proposed by modeling a non-linear end-to-end system function into a parameterized Bernoulli-like model. Despite its attractive performance, the SL detector requires a large amount of labeled data (i.e., pilot signals) to estimate the parameters of the underlying model accurately. This is because the amount of the parameters grows exponentially with the number of users. To overcome this drawback, we propose a semi-supervised learning (SSL) detector where both pilot signals (i.e., labeled data) and some part of data signals (i.e., unlabeled data) are used to estimate the parameters via expectation-maximization (EM) algorithm. Via simulation results, we demonstrate that the proposed SSL detector can achieve the performance of the existing SL detector with significantly lower pilot-overhead.'
author:
-
bibliography:
- 'ICC\_ref.bib'
title: 'Semi-Supervised Learning Detector for MU-MIMO Systems with One-bit ADCs'
---
Massive MIMO, one-bit ADC, MIMO detection, Machine Learning, Semi-Supervised Learning, EM Algorithm
Introduction
============
Massive multiple-input multiple-output (MIMO) is a promising technology for beyond 5G cellular systems where a large number of antennas at the BS is used to improve the capacity and energy-efficiency [@Lu]. In contrast, it can cause the hardware cost and the radio-frequency (RF) circuit power consumption to increase significantly [@Yang]. Especially, a high-resolution analog-to-digital converter (ADC) is the major problem as the power consumption of an ADC increases exponentially with the number of quantization bits and linearly with the baseband bandwidth[@Mezghani-2011]. To overcome this challenge, the use of low-resolution ADCs (e.g., 1$\sim$3 bits) for massive MIMO systems has received increasing attention over the past years. The one-bit ADC is particularly attractive as there is no need for an automatic gain controller, which reduces the hardware complexity significantly[@Hoyos]. In this case, simple zero-threshold comparators quantize the in-phase and quadrature components of the continuous-valued received signals separately. Although low-resolution ADCs provides the advantages, it gives rise to numerous technical challenges in channel estimation and MIMO detections.
For uplink MU-MIMO systems with one-bit ADCs, numerous channel estimation methods were developed as least-square (LS) based method [@Risi], maximum likelihood (ML) method[@Choi], zero-forcing (ZF) type method [@Choi] and Bussgang decomposition based method [@Li]. Also, regarding MIMO detections, the optimal ML detection was developed in [@Choi], and the low complexity methods were presented in [@Mollen; @Mollen2]. Inspired by coding theory, the MIMO detection problems have been reconstructed as an equivalent coding problem [@Hong]. Using the resulting model, a weighted minimum distance (wMD) decoding (i.e., an alternative expression of the ML detector) was presented. Very recently, supervised-learning (SL) detectors were proposed in [@Pohang_Sup; @Sup; @Reinfor] for the considered communication system with one-bit quantized signals. Especially, in our prior work [@Sup], we proposed the generative model, called Bernoulli-like model, by considering the traits of one-bit quantized signals. Despite its attractive performance, the SL detector in [@Sup] requires a large amount of pilot overhead to estimate the model parameters accurately. Thus, it is necessary to reduce a pilot overhead so that the SL detector will be used in practical systems.
In this paper, we study an uplink MU-MIMO system with one-bit ADCs where $K$ users with single-transmit antenna communicate with one BS with $N_{\rm r}$ receive antennas. Also, it is assumed that the BS is not aware of a channel state information (CSI) as in practical communication systems, and needs to estimate it using pilot signals during training phase (see Fig. \[phase\]). A block-fading channel is assumed in which the channel is static during the coherence time $T_c$ and changes independently in block-to-block. We assign the first $T_{t} < T_c$ time slots to the channel training phase and the remaining $ T_ {d} = T_ { c} - T_{t} $ time slots are dedicated to the data transmission phase as shown in Fig. \[phase\]. Inspired by semi-supervised learning[@zhu2009introduction], for such system, we propose a semi-supervised learning (SSL) detector which can significantly reduce the pilot-overhead of the existing SL detector in [@Sup]. The main idea of the proposed SSL detector is that it uses both pilot signals (i.e., labeled data) and some part of data signals (i.e., unlabeled) data to estimate the parameters of the underlying Bernoulli-like model via an efficient expectation-maximization (EM) algorithm. Via simulation results, we demonstrate that the proposed SSL detector can achieve the same performance of the SL detector with a significantly reduced pilot-overhead (e.g., $50\%$ overhead reduction).
This paper is organized as follows. In Section \[sec:Preliminaries\], we describe an uplink MU-MIMO system with one-bit ADCs and equivalent parallel binary discrete memoryless channels in a coding-theoretic viewpoint. In Section \[sec:Supervised\_Learning\], we briefly review a SL detector for the considered system. In Section \[sec:SSL\], we propose a novel SSL detector with parameter update rules which are built on EM algorithm. Section \[simulation\] provides the simulation results to verify the superiority of the proposed SSL detector. Finally, conclusion is provided in Section \[conclusion\].
[**Notation:**]{} Lower and upper boldface letters represent column vectors and matrices, respectively. Let $[a:b]\eqdef\{a,a+1,\ldots,b\}$ for any integers $a$ and $b>a$, and when $a=1$, it can be further shortened as $[b]$. For any $k \in [0:K-1]$, we let $g(k)=[b_0,b_1,\ldots,b_{K-1}]^{\transp}$ represent the $m$-ary expansion of $k$ where $k=b_0m^0+\cdots+b_{K-1}m^{K-1}$ for $b_i \in [0:m-1]$. We also let $g^{-1}(\cdot)$ denote its inverse function. For a vector, $g(\cdot)$ is applied element-wise. Likewise, if a scalar function is applied to a vector, it will be performed element-wise. $\Re(\av)$ and $\Im(\av)$ represent the real and complex part of a complex vector $\av$, respectively.
![Illustration of training, parameter-update, and data transmission phases within a coherence time.[]{data-label="phase"}](Phase.eps){width="9cm"}
Preliminaries {#sec:Preliminaries}
=============
In this section, we describe the system model and define an equivalent $N$ parallel binary discrete memoryless channels (DMCs).
System model {#system}
------------
We consider a single-cell uplink MU-MIMO system in which $K$ users with a single-antenna communicate with one BS with an array of $N_{\rm r} > K$ antennas. We denote $w_k \in \mathcal{W}=[0:m-1]$ as the user $k$’s message for $k \in [K]$, each of which contains $\log{m}$ information bits. Also Let $m$-ary constellation set by $\Sc=\{s_0,...,s_{m-1}\}$ with power constraint as $$\frac{1}{m}\sum_{i=0}^{m-1} |s_i|^2 = \SNR.$$ At time slot $t$, the user $k$ transmits the symbol ${\tilde x}_k[t]$ as $$\tilde{x}_k[t] = \Mc(w_k[t]) \in \Sc,$$ where $\Mc:\mathcal{W}\rightarrow \Sc$ denotes a modulation function. When all the $K$ users transmit the symbols ${\tilde \xv[t]}=[\tilde{x}_1[t],\ldots,\tilde{x}_K[t]]^{\transp}$, the BS receives the discrete-time complex-valued baseband signal vector ${\bf \tilde {r}[t]}\in\mathbb{C}^{N_{\rm r}}$, given by $${\bf \tilde{r}}[t] = {\bf \bar H}{\bf \tilde{x}}[t] +{\bf \tilde{z}}[t], \label{eq:system_complex}$$ where ${\bf \tilde H} \in \CC^{N_{\rm r} \times K}$ is the channel matrix between the BS and the $K$ users, for example, the $i$-th row of ${\bf \tilde H}$ is the channel vector between the $i$-th receiver antenna at the BS and the $K$ users. Also, ${\bf \tilde{z}}[t]=[{\tilde z}_1[t],\ldots,{\tilde z}_{N_{\rm r}}[t]]^{\transp}\in\mathbb{C}^{N_{\rm r}}$ denotes the noise vector whose elements are distributed as circularly symmetric complex Gaussian random variables with zero-mean and unit-variance, i.e., ${{\tilde z}_i}[t] \sim \Cc\Nc(0,1)$.
In the MIMO system with one-bit ADCs, each receiver antenna of the BS is equipped with RF chain followed by two one-bit ADCs that are applied to each real and imaginary part respectively. We define $\mbox{sign}(\cdot): \RR \rightarrow \{-1,1\}$ as the one-bit ADC quantizer function with $\hat{r}[t]=\mbox{sign}(\tilde{r}[t]) = 1$ if $\tilde{r}[t] \geq 0$, and $\hat{r}[t]=-1$, otherwise. Then, the BS receives the quantized output vector as $\hat{\rv}_{\rm R}[t] = \mbox{sign}({\rm Re}({\bf \tilde r}[t]))$ and $\hat{\rv}_{\rm I}[t] =\mbox{sign}({\rm Im}({\bf \tilde r}[t]))$. For the ease of representation, we rewrite the complex input-output relationship in into the equivalent real representation as $$\rv[t] = \mbox{sign}\left(\Hm\xv(\wv[t])+\zv[t]\right), \label{eq:obs1}$$ where $\rv[t]=[\hat{\rv}_{\rm R}^{\transp}[t],\hat{\rv}_{\rm I}^{\transp}[t]]^{\transp}$, $\xv(\wv[t])=[\Re(\tilde{\xv}[t])^{\transp},\Im(\tilde{\xv}[t])^{\transp}]^{\transp}$, $\zv[t]=[\Re(\tilde{\zv[t]})^{\transp},\Im(\tilde{\zv}[t])^{\transp}]^{\transp}\in\mathbb{R}^{N}$, and $$\Hm = \left[ {\begin{array}{cc}
\Re({\bf \tilde H}) & -\Im({\bf \tilde H}) \\
\Im({\bf \tilde H}) & \Re({\bf \tilde H})\\
\end{array} } \right] \in\mathbb{R}^{N\times 2K},$$ where $N=2N_{\rm r}$. This real system representation will be used in the sequel.
Equivalent N parallel B-DMCs {#system}
----------------------------
In [@Hong], it was shown that a real system representation can be transformed into an equivalent $N$ parallel B-DMCs via a coding-theoretic viewpoint. In the resulting $N$ parallel B-DMCs, the channel input/output and the channel transition probabilities are defined as follows.
[**Auto-encoding function:**]{} Given $\Hm$, we can create a spatial-domain code $\Cc=[\cv_0,\ldots,\cv_{m^K-1}]$, each of which is given by $$\cv_j= \left[\mbox{sign}\left(\hv_{1}^{\transp}\xv(g(j))\right),\ldots, \mbox{sign}\left(\hv_{N}^{\transp}\xv(g(j))\right)\right]^{\transp}\label{codecon}$$ where note that each codeword of $\Cc$ can be considered as a noiseless channel output in . In Fig. \[System\], the channel input $\qv$ of the equivalent channel is determined by the auto-encoding function $f(\cdot)$ such as $$\qv=f(\wv,\Hm)=\cv_j,$$ for $j=g^{-1}(\wv)\in [0 : m^K-1]$.
[**Effective channel:**]{} As shown in Fig. \[System\], the effective channel consists of the $N$ parallel BSCs with the channel input $\qv$ and the channel output $\rv$. This channel is specified by the following channel transition probabilities: For the $n$-th BSC, the transition probability, depending on user’s message $\wv=g(j)$ and the corresponding codeword $\cv_j$, are defined as $$\label{eq:epsilon}
\PP(r_n[t]|q_n=c_{j, n})=\begin{cases}
\epsilon_{j,n} & \text{if}~ r_n[t]\neq c_{j,n} \\
1-\epsilon_{j,n} & \text{if}~ r_nl[t]=c_{j,n}
\end{cases}$$ where the error-probability of the $n$-th BSC is computed as $$\epsilon_{j,n} \eqdef Q\left(|\hv_{n}^{\transp}\xv(g(j))|\right), \label{eq:epsilon_1}$$ where $Q(x) = \frac{1}{2\pi}\int_{x}^{\infty} \exp\left(-u^2/2\right) du$.
The purpose of this paper is to design a decoding function in Fig. \[System\] which decodes $\hat{\wv}[t]$ from an observation $\rv[t]$, by leveraging the equivalent effective channel (i.e., the channel transition probabilities in (\[eq:epsilon\])). We remark that the parameters of the transition probabilities are not known a priori and should be estimated with pilot signals during the training phase.
![Description of the equivalent $N$ parallel B-DMCs.[]{data-label="System"}](ComModel.eps){width="9cm"}
The Overview of SL Detector {#sec:Supervised_Learning}
===========================
In this section, we briefly review the supervised-learning (SL) detector proposed in [@Sup] with the assumption that a channel matrix $\Hm$ is not known. In the SL detector, thus, we need to estimate the parameters $\Cc$ and $\epsilon_{j,\ell}$ using pilot signals as in parameterized supervised learnings. From (\[eq:epsilon\]), we can define the generative model of $\rv[t]$, named [*Bernoulli-like*]{} model, which are fully described by the parameter vector $\thetav=[\thetav_0,\ldots,\thetav_{m^K-1}]$ where $\thetav_j = [\cv_j,\epsilonv_j]$, such as $$\begin{aligned}
p(\rv[t] |j,\thetav_{j})&\eqdef \PP(\rv[t]|g^{-1}(\wv[t])=j, \thetav_j))\nonumber\\
& =\prod_{n=1: r_n[t]\neq c_{j,n}}^{N}\epsilon_{j,n}\prod_{n=1: r_n[t] = c_{j,n}}^{N}(1-\epsilon_{j,n}) \label{eq:BLmodel}
$$ for $j \in [0:m^K-1]$. We remark that each class $j$ has its own probability distribution parameterized by $\thetav_j=[\cv_j, \epsilonv_j]$.
The SL detector in [@Sup] performs with the following two-phase during each coherence time $T_{c}$.
[**Parameter Estimation:**]{} In this phase, the parameter vector $\thetav$ is estimated using $T_t$ pilot signals. We first obtain the labeled data $\Lc$ such as $$\label{Pilots}
\Lc=\{(\rv[1], 0),\ldots,(\rv[T],0),\ldots,(\rv[T_t],m^K-1)\},$$ where $(\rv[t], j_t)$ represents the pilot signal corresponding to the label $j_t$. Since $T$ pilot signals are transmitted for each codeword, the overall pilot-overhead is equal to $T_{\rm t}=T\cdot {m^K}$. Also, for $t \in [T_t]$, the labels are determined as $$\label{eq:label}
j_t \eqdef \lfloor (t-1)/T \rfloor \in [0:m^K-1],$$ $\lfloor \cdot \rfloor$ denotes the floor function. In [@Sup], from the labeled data $\Lc$, the parameter vector $\thetav$ is determined via the optimal maximum-likelihood (ML) estimation as $$\begin{aligned}
\hat{c}_{j, n}&=\mbox{sign}\left(\sum_{t=j\cdot{T}+1}^{(j+1)\cdot{T}}r_{n}[t]\right) \label{eq:opt_mu}\\
\hat{\epsilon}_{j,n}&=\frac{1}{T}{\sum_{t=j\cdot{T}+1}^{(j+1)\cdot{T}} {\bf 1}_{\{\hat{\cv}_{j,n}\neq r_{n}[t]\}}}\label{eq:est_epsilon}\end{aligned}$$ for $n\in [N]$ and $j \in [0:m^K-1]$.
[**Data Detection:**]{} From the Bernoulli-like model parameterize by (\[eq:opt\_mu\]) and (\[eq:est\_epsilon\]), the ML detection performs as $$\hat{j}={\operatornamewithlimits{argmax}}_{j\in [0:m^K-1]}{p(\rv[t] |j,\thetav_{j})}.
\label{eq:detects}$$
The Proposed SSL Detector {#sec:SSL}
=========================
Despite its superior performance, the SL detector proposed in [@Sup] suffers from the heavy pilot-overhead because a larger number of pilot signals are required so that an empirical transition probability in (\[eq:est\_epsilon\]) is close to the true transition probability in . Moreover, this overhead becomes larger as the number of users $K$ increases, because the number of parameters to be estimated increases exponentially with the $K$ (see (\[eq:opt\_mu\]) and (\[eq:est\_epsilon\])). To address the above problem, we propose a semi-supervised learning (SSL) detector in which the parameter vector $\thetav$ is estimated by leveraging both data signals (i.e., unlabeled data $\Uc$) and pilot signals (i.e., labeled data $\Lc$). Here, the unlabeled data $\Uc$ is collected during $T_{u}$ time slots (see Fig. \[phase\]) such as $$\label{Pilots}
\Uc=\{\rv[T_t+1],\rv[T_t+2],\ldots,\rv[T_t+T_u]\}.$$ Also, we let $\Dc = \Lc \cup \Uc$ denote the observed data to be used for parameter-estimation in the proposed SSL detector.
{width="17cm" height="7cm"}
[**Parameter Estimation:**]{} In this phase, the parameter vector $\thetav=[\thetav_0,\ldots,\thetav_{m^K-1}]$ is updated from the given data $\Dc$ so that the conditional probabilities of the observations (i.e., the received binary signals) are maximized. This ML estimation is mathematically formulated as $$\begin{aligned}
\label{eq:ML}
\hat{\thetav} = {\operatornamewithlimits{argmax}}_{\thetav}{\log \PP(\Dc|\thetav)}.\end{aligned}$$ Note that from the Bernoulli-like model, we know the probability distribution $p(\rv[t]|j,\thetav_j)$ defined in (\[eq:BLmodel\]) for the given parameter $\thetav_j$, which will be used in the below. Also, the labels of the labeled data ate given as $\{j_t=\lfloor (t-1)/T\rfloor: t\in[T_t]\}$ in (\[eq:label\]).
For any fixed parameter $\thetav$, the objective function in (\[eq:ML\]) is represented as $$\begin{aligned}
&\log{\PP(\Dc|\thetav)}\nonumber\\
&=\log{\prod_{t=1}^{T_t}\PP(\rv[t],g^{-1}(\wv[t])= j_{t} |\thetav_{j_{t} }) \prod_{t=T_t+1}^{T_t+T_u} \PP(\rv[t]|\thetav)}\nonumber\\
&=\sum_{t=1}^{T_t} \log{\PP(j_{t} |\thetav_{j_{t} }) p(\rv[t]|j_{t} ,\thetav_{j_{t} })}\nonumber\\
&\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;+\sum_{t=T_t+1}^{T_t+T_u} {\log\left({\sum_{j=0}^{m^K-1} p(\rv[t], j|\thetav_{j})}\right)},\label{eq:likelihood}\end{aligned}$$ where recall that $p(\rv[t] |j,\thetav_{j})$ is defined in (\[eq:BLmodel\]), and $\PP(j_{t} |\thetav_{j_{t}}) = 1/m^K$ since the users’ messages are assumed to be generated uniformly and randomly. Definitely, the above objective function is non-convex especially due to the second-term caused by the unlabeled data and thus, the optimization problem in (\[eq:ML\]) is too complex to be solved. We thus solve it using [*Expectation-Maximization*]{} (EM) algorithm[@dempster1977maximum].
The EM algorithm consists of the following two steps, named expectation-step (E-step) and maximization-step (M-step), respectively: Given the up-to-date parameter vector $\thetav^i$, it finds the updated parameter vector $\thetav^{i+1}$.
[*E-step:*]{} In this step, we compute the following probability distribution using the latest parameter vector $\thetav^i$: $$\begin{aligned}
\label{eq:expect}
\gamma_{j}[t] &\eqdef \PP(g^{-1}(\wv[t])=j|\rv[t],\thetav_{j}^{i}).\end{aligned}$$ This is specified by considering the difference of the labeled and unlabeled data as follows:
- [*(Labeled Data)*]{} For $t \in [T_t]$ and $j \in [0:m^K-1]$, $$\gamma_{j}[t]={\bf 1}_{\{j=j_{t} \}}.$$
- [*(Unlabeled Data)*]{} For $t \in [T_t+1:T_t+T_u]$ and $j \in [0:m^K-1]$, $$\gamma_{j}[t]=\frac{p(\rv[t]| j,\thetav_j^{i})}{\sum_{j=0}^{m^K-1}{p(\rv[t]|j,\thetav_{j}^{i})}}.$$
[*M-step:*]{} In this step, we find an updated parameter vector $\thetav^{i+1}$ using the $\gamma_j [t]$ in the above as follows: $$\label{eq:obj_EM}
\thetav^{i+1}={\operatornamewithlimits{argmax}}_{\thetav}{\psi(\thetav|\thetav^i)},$$ where the objective function is defined as $$\begin{aligned}
&\psi(\thetav|\thetav^i)\nonumber\\
&\;\;\;\;\;\eqdef{\sum_{t=1}^{T_t+T_u}}\sum_{j=0}^{m^K-1}{\gamma_j[t]}\log{\PP(\rv[t],g^{-1}(\wv[t])=j|\thetav_j})\nonumber\\
&\;\;\;\;\;={\sum_{t=1}^{T_t+T_u}}\sum_{j=0}^{m^K-1}{\gamma_j[t]}(\log{p(\rv[t]|j,\thetav_j)}-K\log m),\label{eq:psi_def}\end{aligned}$$ where the second equality is from the Bayes rule and (\[eq:BLmodel\]). Note that $\gamma_j[t]$ in the above is constant with respect to $\thetav_j$. Also, from the Bernoulli-like model in (\[eq:BLmodel\]), the objective function in (\[eq:psi\_def\]) can be specified as $$\begin{aligned}
\psi(\thetav|\thetav^i)&={\sum_{t=1}^{T_t+T_u}} {\sum_{j=0}^{m^K-1}}-{\gamma_{j}[t]}{K\log{m}}\nonumber\\
&+{\sum_{j=0}^{m^K-1}}{\sum_{t=1}^{T_t+T_u}}{\sum_{n=1}^{N}}\Big(\gamma_{j}[t]{{\bf 1}_{\{r_n{[t]} \neq c_{j,n} \}}}\log{\epsilon_{ j,n}}\nonumber\\
&\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;+\gamma_{j}[t] {{\bf 1}_{\{r_n{[t]}=c_{j,n}\}}}{\log{(1-\epsilon_{j,n})}}\Big).\end{aligned}$$ Since the first-term in the above is constant with respect to $\thetav$, the parameter vector $\thetav$ can be optimized by only maximizing the second-term as follows: $$\begin{aligned}
\label{eq:lttrterm}
&(\hat{\epsilonv}^{i+1},\hat{\cv}^{i+1})\nonumber\\
&={\operatornamewithlimits{argmax}}_{(\epsilonv,\cv)}{\sum_{j=0}^{m^K-1}}{\sum_{n=1}^{N}}{\sum_{t=1}^{T_t+T_u}}\Big({\gamma_{j}[t]}{{\bf 1}_{\{r_n{[t]} \neq c_{j,n} \}}}{\log{\epsilon_{j, n}}}\nonumber\\
&\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;+{\gamma_{j}[t]}{{\bf 1}_{\{r_n{[t]}=c_{j, n}\}}}\log{(1-\epsilon_{j, n})}\Big).\end{aligned}$$ Obviously, we can see that maximizing is equivalent to maximizing the individual terms in : For each fixed $j$ and $n$, we have $$\begin{aligned}
\label{eq:indivterm}
&(\hat{\epsilon}_{j,n}^{i+1},\hat{c}_{j,n}^{i+1})={\operatornamewithlimits{argmax}}_{(\epsilon_{j,n}, c_{j,n})}{\sum_{t=1}^{T_t+T_u}}\Big({\gamma_{j}[t]}{{\bf 1}_{\{r_n{[t]} \neq c_{j, n} \}}}{\log{\epsilon_{j, n}}}\nonumber\\
&\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;+{\gamma_{j}[t]}{{\bf 1}_{\{r_e{[t]}=c_{j, n}\}}}\log{(1-\epsilon_{j, n})}\Big).\end{aligned}$$ To solve the above problem, we introduce the useful lemma in the below.
\[lem\_hj\] Suppose $a_{\ell}\geq 0$ for $1\leq{\ell}\leq{n}$, Then $\sum_{\ell=1}^{n}{a_{\ell}}\log{p_{\ell}}$ is maximized over all probability vectors $p=(p_1,\ldots,p_n)$ by $p_{\ell}=\frac{a_{\ell}}{\sum_{i=1}^{n}{a_{i}}}$. $\blacksquare$
First of all, we observe that the optimal $c_{j,n}$ should satisfy the following constraint for any $\epsilon_{j,n} < 0.5$: $$\label{eq:constraint}
{\sum_{t=1}^{T_t+T_u}}{\gamma_{j}[t]}{{\bf 1}_{\{r_n{[t]} \neq c_{j, n} \}}}<{\sum_{t=1}^{T_t+T_u}}{\gamma_{j}[t]}{{\bf 1}_{\{r_n{[t]}=c_{j, n}\}}}.$$ Also, we can see that this constraint is satisfied by assigning
- $\hat{c}_{j,n}^{i+1}=1$ if $\sum_{t=1}^{T_t+T_u}{r_{n}{[t]}}{{\gamma_{j}}[t]}>0$;
- $\hat{c}_{j,n}^{i+1}=-1$ if $\sum_{t=1}^{T_t+T_u}{r_{n}{[t]}}{{\gamma_{j}}[t]}<0$.
Equivalently, we obtain that $$\begin{aligned}
\label{eq:update1}
\hat{c}_{j, n}^{i+1}=\sign\left(\sum_{t=1}^{T_t+T_u}{\gamma_{j}[t]}r_n[t]\right) \mbox{ for } n \in [N].\end{aligned}$$ Next, applying Lemma \[lem\_hj\] in the below to , the error-probability $\epsilon_{j,n}^{i+1}$ is optimized as $$\begin{aligned}
\label{eq:update2}
&\hat{\epsilon}_{j, n}^{i+1}=\nonumber\\
&\frac{\sum_{t=1}^{T_t+T_u}{\gamma_{j}[t]}{{\bf 1}_{\{r_n [t] \neq \hat{c}_{j,n}^{i+1} \}}}}{\sum_{t=1}^{T_t+T_u}{\gamma_{j}[t]}{{\bf 1}_{\{r_n[t] \neq \hat{c}_{j, n}^{i+1} \}}}+\sum_{t=1}^{T_t+T_u}{\gamma_{j}[t]}{{\bf 1}_{\{r_n [t] \neq \hat{c}_{j,n}^{i+1} \}}}}.\end{aligned}$$ Finally, we can compute the log-likelihood using the updated parameter vector $\thetav^{i+1}$ as $$\begin{aligned}
\label{eq:llikelihood}
\log{\PP(\Dc|\thetav^{i+1})}&={\sum_{t=1}^{T_t}}{\log{\frac{1}{m^K}p(\rv[t]|j_{t} ,\thetav_{j_{t} }^{i+1})}}\nonumber\\
&+\sum_{t=T_t+1}^{T_t+T_u}\log{\frac{1}{m^K}}{\sum_{j=0}^{m^K-1}{p(\rv[t]|j,\thetav_{j}^{i+1})}},\end{aligned}$$ which is used to check the convergence of EM algorithm. The overall procedures are summarized in Fig. \[SSL concept\] and [**Algorithm 1**]{} where $\varepsilon \geq 0$ denotes the pre-determined threshold for the stopping criterion.
[**Data Detection:**]{} For $t\in[T_t+1:T_t+T_u]$, the SSL detector performs using the latest $\gamma_{j}[t]$ in as $$\hat{j}={\operatornamewithlimits{argmax}}_{j\in [0:m^K-1]}{\gamma_{j}[t]}.
\label{eq:detects}$$ Also, for $t\in[T_t+T_u+1:T_c]$, the detection process of the SSL detector is equivalent to that of the SL detector in Section \[sec:Supervised\_Learning\]. We remark that the performance-complexity tradeoff of the proposed SSL detector is controlled by the choice of $T_u$.
![$K=2$ and $N=8$. Performance comparisons of the proposed SSL detector, the SL detector, and MLD with CSIR.[]{data-label="comp1"}](K2N8.eps){width="9.5cm"}
\[mbdetector\]
**Input:**
- [*(Labeled data)*]{} $\Lc=\{(\rv[t],j_t): t\in[T_t]\}$
- [*(Unlabeled data)*]{} $\Uc=\{\rv[t]: t\in [T_t+1:T_t+T_u]\}$
**Output:** $\hat{\thetav}^{i+1}$
Simulation Results {#simulation}
==================
We evaluate the average bit-error rate (BER) performances of the proposed SSL detector and the conventional SL detector. For the simulations, a Rayleigh fading channel is considered where each element of a channel matrix [**H**]{} is drawn from an independent and identically distributed (i.i.d.) circularly symmetric complex gaussian random variable with zero mean and unit variance. a user is assumed to send binary data ($m=2$) and QPSK modulation is applied. A block fading duration (i.e., coherence time interval) is set to be $T_d=512, T_u=10\cdot T_t$ and $T_t=T\cdot{m^K}$.
Fig. \[comp1\] shows the BER performances of the SSL detector, SL detector, and maximum likelihood detection (MLD) with channel state information at a receiver (CSIR) in a condition of various training duration. It is notable that the performance of proposed SSL detector outperforms the conventional SL detector in the entire SNR regimes where the pilot-overhead is same. In particular, for $T=1$, the performance of the proposed SSL detector almost achieves that of the SL detector with $T=4$. This implies that the SSL detector reduces training span ($T_t$) considerably without degradation in performance, by making the best use of information from the generative model and data signals. Also, when compared with MLD in CSIR, this result shows that the proposed method allows the empirical conditional probability to converge into true conditional probability without increasing the number of pilots.
Conclusion
==========
In this paper, we presented a novel semi-supervised learning detector inspired by semi-supervised learning. Specifically, the proposed SSL detector updates parameters by using data signals through the maximum likelihood estimation under the Bernoulli-like model. Such parameter updates can significantly reduce pilot-overhead that is an issue in the existing SL detector. The simulation results demonstrated that the performance of the SSL detector almost achieves that of the SL detector, even with a quite lower pilot-overhead than that of the SL detector. We would like to emphasize that a SSL detector would be a strong practical framework in a field of machine learning based detector, in that compared with pilot signals, data signals are fairly cheap to obtain. On going work, we are investigating to develop more practical SSL detectors which require low complexity or are appropriate for time-varying channel system.
Acknowledgement {#acknowledgement .unnumbered}
===============
This work was supported by Samsung Research Funding & Incubation Center of Samsung Electronics under Project Number SRFC-IT1702-00.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Turkish Wikipedia Named-Entity Recognition and Text Categorization (TWNERTC) dataset is a collection of automatically categorized and annotated sentences obtained from Wikipedia. We constructed large-scale gazetteers by using a graph crawler algorithm to extract relevant entity and domain information from a semantic knowledge base, Freebase[^1]. The constructed gazetteers contains approximately 300K entities with thousands of fine-grained entity types under 77 different domains. Since automated processes are prone to ambiguity, we also introduce two new content specific noise reduction methodologies. Moreover, we map fine-grained entity types to the equivalent four coarse-grained types, *person, loc, org, misc*. Eventually, we construct six different dataset versions and evaluate the quality of annotations by comparing ground truths from human annotators. We make these datasets publicly available to support studies on Turkish named-entity recognition (NER) and text categorization (TC).'
author:
- |
H. Bahadir Sahin, [**Caglar Tirkaz**]{}, [**Eray Yildiz**]{},\
[**Mustafa Tolga Eren**]{}, [**Ozan Sonmez**]{}\
Huawei Turkey Research and Development Center, Umraniye, Istanbul, Turkey\
[[email protected]]{}\
[[hbahadirsahin, caglartirkaz, tolgaeren, osonmez]{}@gmail.com]{}
bibliography:
- './ref.bib'
title: 'Automatically Annotated Turkish Corpus for Named Entity Recognition and Text Categorization using Large-Scale Gazetteers'
---
Introduction
============
Named-entity recognition (NER) is an information extraction (IE) task that aims to detect and categorize entities to pre-defined types in a text. On the other hand, the goal of text categorization (TC) is to assign correct categories to texts based on their content. Most NER and TC studies focus on English, hence accessing available English datasets is not a issue. However, the annotated datasets for Turkish NER and TC are scarce. It is hard to manually construct datasets for these tasks due to excessive human effort, time and budget. In this paper, our motivation is to construct an automatically annotated dataset that would be very useful for NER and TC researches in Turkish.
The emergence of structured and linked semantic knowledge bases (KBs) provide an important opportunity to overcome these problems. Approaches that leverage such KBs can be found in literature [@heck2013leveraging; @gerber2013real; @hoffart2011robust; @mendes2011dbpedia]. However, using the structured data from KBs is a challenging task for linking named entities and domains to raw texts due to ambiguous texts and named entities [@cucerzan2007large].
In this work, we publish TWNERTC dataset in which named entities and categories of sentences have been automatically annotated. We use Turkish Wikipedia dumps as the text source[^2] and Freebase to construct a large-scale gazetteers to map fine-grained types to entities. To overcome noisy and ambiguous data, we leverage domain information which is given by Freebase and develop domain-independent and domain-dependent methodologies. All versions of datasets can be downloaded from our project web-page[^3]. Our main contributions are (1) the publication of Turkish corpus for coarse-grained and fine-grained NER, and TC research, (2) six different versions of corpus according to noise reduction methodology and entity types, (3) an analysis of the corpus and (4) benchmark comparisons for NER and TC tasks against human annotators. To the best of our knowledge, these datasets are the largest datasets available for Turkish NER ad TC tasks.
The rest of the paper is organized as follows: In Section 2, we briefly investigate the literature about NER, TC and datasets which are used in these research. In Section 3, we explain the construction of large-scale gazetteers by using Freebase. In Section 4, we explain how to use the gazetteers to automatically annotate and categorize Wikipedia texts to construct datasets along with dataset statistics, and noise reduction methodologies. Our evaluation about the quality of these constructed datasets are reported in Section 5.
Related Work
============
Named Entity Recognition (NER) and Text Classification (TC) are well-researched NLP tasks relevant to large amount of information retrieval and semantic applications. TC research predates to ’60s; however, it is accepted as a research field in ’90s with the advances in technology and learning algorithms [@sebastiani2002machine]. On the contrary, the classical NER task is defined in MUC [@chinchor1998overview] and CoNLL [@tjong2003introduction] conferences with coarse-grained entity types: person, location, organization and misc. In addition, few studies address the problem of fine-grained NER where the challenge is to capturing more than four entity types [@pasca2006organizing; @ekbal2010assessing; @yogatama2015embedding].
As research on NER has been pushing the limits of automated systems performing named-entity recognition, the need for annotated datasets and benchmarks is also increasing. Knowledge bases are important for NLP researches, since they provide a structured schema of topics that can be used to annotate entities with fine-grained types and/or categorize raw texts into related domains.
Steinmetz et al. [@steinmetz2013statistical] published benchmark evaluations that compare three datasets that use semantic information from KBs: DBpedia Spotlight [@mendes2011dbpedia], KORE50 [@hoffart2011robust; @yosef2011aida] and the Wikilinks Corpus [@singh2011large]. These datasets are in English and constructed with the aim of evaluating the performance of NER systems. The authors present the statistics of each dataset and baseline performances of various algorithms. There are other methodologies which leverages KBs to named entity extraction and linking; however, most of them are not available to public [@hoffart2011robust; @heck2013leveraging].
Constructing a comprehensive dataset for TC is tougher than NER since there is no limit for the number of categories that are represented in such sets. In general, there are many TC datasets available in English for many different problems such as sentiment analysis [@liu2015automated] and categorizing gender [@mukherjee2010improving]. The largest and the most popular dataset among them is Reuters Corpus Volume 1 (RCV1) which consists of manually categorized 800K news stories with over 100 sub-categories under three different main categories [@rose2002reuters]. This version the dataset has problems with document categories and suffers from lack of documentation about the dataset. Lewis et al. propose an improved version of this dataset with reduced categorization mistakes and provide a better documentation [@lewis2004rcv1].
Research on Turkish NER and TC are very limited compared to English and several other languages. The main reason is the lack of accessibility and usability of both Turkish NER and TC datasets. The most popular Turkish NER dataset is introduced by Gökhan et al. [@tur2003statistical]. This dataset contains articles from newspapers, approximately 500K words, and is manually annotated with coarse-grained entity types. Tatar and Çiçekli propose another coarse-grained NER dataset [@tatar2011automatic]; however, it contains only 55K words which makes this dataset to less preferable than previous dataset. More recent studies focus on Turkish NER in social media texts [@onal2015named; @kuccuk2014named; @demir2014improving; @celikkaya2013named]. Due to the research focus in the field, several Twitter-based coarse-grained NER datasets are published [@kuccuk2014named; @kuccuk2014experiments; @tantugrecognizing]. According to our knowledge, there is no literature available regarding to fine-grained NER in Turkish.
Turkish TC researchers tend to construct their own, case specific datasets in general [@amasyali2006automatic]. The newspapers are the main text source for such studies since they are easy to obtain and classify manually [@akkus2013categorization; @toraman2011developing; @kilincc2015ttc]. When the amount of annotated data is considered to train state-of-the-art learning algorithms, aforementioned Turkish datasets suffer from the lack of enough data. The main bottlenecks are requires human effort and time constraint, which limits the size and scope of the constructed datasets. In contrast, our aim is to provide larger, more comprehensive and insightful Turkish datasets for both NER and TC by using knowledge bases to create large-scale gazetteers and eliminating human factor in the annotation process.
In the next section, we will explain our dataset construction methodology starting with building gazetteers by using Freebase and Wikipedia. We will investigate how to build a graph crawler algorithm to crawl the knowledge in Freebase and to map its entity types and domain information into the raw texts automatically. We will also discuss noise reduction methods and propose three different versions of the dataset.
Dataset Construction Methodology
================================
Constructing Gazetteers
-----------------------
Gazetteers, or entity dictionaries, are important sources for information extraction since they store large numbers of entities and cover vast amount of different domains. KBs use a graph structure that is used to represent thousands of films and/or millions of songs in gazetteers. Hence, such graph structures can be efficient to construct large-scale gazetteers.
In our work, we use Freebase as the KB since it provides both quality and quantity in terms of entity types and domains for Turkish. Freebase has 77 domains that cover many different areas from music to meteorology with approximately 50M total entities. Among them, Turkish has 300K entities, and approximately 110K of them have a link to corresponding Turkish Wikipedia page.
By using KBs, one can eliminate the necessity of creating semantic schema design, collecting data and manually annotating raw texts. However, such large entity lists contain ambiguous and inaccurate information that can impact the quality. For instance, both film and boat domains contain “Titanic" in their entity lists, and such ambiguity could create false links. Considering the number of entities, the importance of disambiguating named entities becomes more important.
Our work is inspired from Heck et al.[@heck2013leveraging] who employ a method that takes advantage of user click logs of a web search engine in order to improve precision of gazetteers while maintaining recall. Since we do not have any sources to access such user click logs, we depend on attributes, such as entity domains, types and properties, that Freebase provides in order to improve the quality of our gazetteers. Note that a named entity’s domain, type and property are represented as */domain/type/property* in Freebase.
Freebase distinguishes entities that exist in more than one domain by unique *machine id* ([*mid* ]{}). For instance, Titanic is an entity in both film and boat domains with two *mids* depending on the domain. On the other hand, there are cases in which same *mids* can be associated with multiple domains. For instance the [*mid* ]{}of Titanic in the film domain is also used in the award domain. These cases occur when domains are closely related. Note that [*mid* ]{}is language independent. Hence, an entity has the same [*mid* ]{}regardless of its language; however, the information related to that entity can differ, e.g., missing equivalence of entities, translation differences.
Domains covered in Freebase have large amount of entities and related descriptive texts. However, for Turkish, we need to filter or merge some domains due to insufficient number of related raw texts. For instance, we merge “American football", “cricket", “ice hockey" domains under already existing *sports* domain.
{width="90.00000%"}
During annotation, due to ambiguous cases in Freebase, we first compute the domain and the entity type distribution of directly related (first order) relations of the selected named entity. Among the candidates, the entity type that contains the most first order relations in the knowledge graph is selected as the type of the entity. This ensures that the most informative entity type is selected for the annotated entity. For instance, while annotating the Wikipedia page of “Titanic (film)" the possible candidates are */film/film* and */award/award\_winning\_work*. Our method chooses */film/film* as the domain and type of the entity since it has more information compared to its competitor.
After determining the entity type of each entity, we form large-scale gazetteers for Turkish which contain 300K named-entities. Each entity in the gazetteer has its Wikipedia description, determined type and 1st-order relations. Note that some entities may not have Wikipedia description due to several reasons, e.g. song names or deleted Wikipedia pages; however, we use such entities in the annotating process.
Automatic Annotation Generation
===============================
Fine-Grained Annotations
------------------------
A knowledge graph consists of nodes and edges between nodes which are defined by the schema of a domain. Nodes are entities and edges represent relations between nodes, which are *properties* in Freebase. Examples of entities and relations are shown in [Figure \[fig:Freebase\_Graph\_Structure\]]{}, for “film" and “people" domains. The central node having the most number of relations in the film domain is the “film\_name". We call directly connected relations to the central node as first-order relations. Name of the director who directed the movie is “film\_director" relation. Since directors are people, they have also relations in “person" domain. Through such relationship we can get second-order relations about a movie. We benefit this graph structure along with the raw texts to create multi-purpose, annotated and categorized corpora. Inspired by the approach of [@heck2013leveraging], we create a graph crawling algorithm that is capable of categorizing sentences into Freebase domains and annotate named entities within the sentences with Freebase types and properties. This set of annotations are the *Fine-Grained Annotations* (FGA). Our method consists 5 steps and is applicable to both English and Turkish (with slight changes for sentence processing).
TWNERTC-DI TWNERTC-DD
--------------------------------------- ------------------ ------------------- -- --
\# Of Sentences 695949 696785
Largest Domains (\# sentences) People(137789) People(137907)
Smallest Domains (\# sentences) Exhibitions(132) Exhibitions (132)
\# of Tokens (with punctuation) 13445651 13458930
\# of Tokens (without punctuation) 10949015 10960244
\# of Tagged Tokens 2314333 2255320
\# of Unique Entity Types 3150 3447
Largest Domains (\# unique entities) People(1536) People (1011)
Smallest Domains (\# unique entities) Physics(58) Physics(12)
1. Select the central node from gazetteers. This node is “Central Pivot Node" (CPN). It is the main entity-type of the domain, e.g. film names in film domain.
2. Retrieve the descriptions provided by Wikipedia or Freebase. If Wikipedia has the corresponding page, fetch full texts from dump. Else if only Freebase description is available, use the description. Otherwise, return to the first step. Since several Turkish Wikipedia texts are direct copy of English version, language detection is applied on all texts and English texts are eliminated.
3. Extract sentences from raw texts and annotate sentences by longest-string pattern matching where the first-order relations of CPN are used resulting in IOB style annotation. We use Aktaş and Çebi’s sentence detection method to extract sentences from full texts [@aktacs2013rule].
4. Extend annotation process with second-order relations.
5. Categorize sentences by selecting the domain of entities that is used the most.
One should consider that higher order relations can create a long chain of relationships. In our work, we limit this chain with the second-order relations, since further relations provide less information while increasing ambiguity and inconsistency of automated annotations.
Statistics of the FGA
---------------------
TWNERTC contains 300K entities in total of which 110K have Turkish descriptions (from Freebase or Wikipedia dump). There are totally 700171 annotated sentences from 49 different domains with the largest and smallest domains being *people* (139K sentences) and *fashion* (493 sentences) subsequently. TWNERTC consists of 10997037 tokens without punctuation and among these tokens approximately 2M of them are annotated. 16K of the tags are unique which results in approximately 332 unique entity type per domain on the average. Note that, even if a sentence is categorized into the *film* domain, it can contain entities from other domains, such as *person* and *time*. *Location* is the domain having the highest number of unique tags (1051) whereas *physics* has the least amount of unique tags (15).
Disambiguate Noisy Entity Types
-------------------------------
We refine the gazetteers to minimize the effect of noise in the generated annotations and assigned domains. However, due to the nature of automated algorithm, we find that entity annotations have still inconsistent or missing types while categorization process is resulted with more accurate in general. In order to reduce remaining noise, we apply both *domain dependent* and *domain independent* noise reduction. Domain dependent approach finds the most common entity type of every entity according to the domain of the sentences. Then, entities are re-annotated with the common entity types. Domain independent approach follows the same process without using domain information while eliminating the noisy information. Statistics about these two versions of the dataset are presented in [Table \[tab:HEC\_HTC\_Statistics\_Noise\_Reduction\]]{}.
Transform FGA to Coarse-Grained Annotation
------------------------------------------
People Location Organization Misc
------------ -------- ---------- -------------- ---------
TWNERTC 353385 296318 75940 973477
TWNERTC-DI 345545 331943 90673 1172214
TWNERTC-DD 394291 361728 99075 1046818
FGA provides fine-grained annotations with many detailed entity types and properties. However, the amount of different entity types affects the learning algorithms negatively. Moreover, it is hard to evaluate the quality of the annotations when most works in literature performs coarse-grained entity recognition. Thus, we provide a coarse-grained version of the FGA datasets. In order to transform FGA to coarse-grained annotation (CGA), we map each fine-grained entity type in each domain to a coarse-grained entity type, i.e. *person*, *organization*, *location* and *misc*. We keep the IOB notations in types while converting the type. In this process, we eliminate several domains, such as meteorology, interests and chemistry, due to the lack of types that can be mapped to a coarse-grained version. This elimination process leaves 25 unique domains in CGA-based datasets.
In coarse version of the dataset, there are approximately 380K sentences. Similar to FGA statistics, the *people* domain has the highest number of sentences (104508) while *law* domain has the least number of sentences (454) among all 25 domains. The number of tokens is 7326286 with punctuation. Among these tokens, 851123 of them are annotated. Unlike FGA, CGA has only 4 types for each domain. 19 of the domains contains all 4 types whereas the remaining domains might contain 2 or 3 entity types such as geography and food.
We also transform the post-processed datasets we introduced in previous chapter to CGA. [Table \[tab:htc\_hec\_statistics\_cga\]]{} presents the details of created corpora with CGA. In total, we publish six datasets (one original and two post-processed with FGA, one original and two post-processed with CGA)[^4].
Evaluation
==========
To experimentally evaluate TWNERTC, five human annotators categorize and annotate test sets that are sampled from the datasets. We create six test sets (3 CGA, 3 FGA) with 10K-word for NER and one test with 2K sentences for TC. We compare annotations and domains of these sets with manually created ground truths.
For NER evaluation, we follow two different approaches for coarse-grained and fine-grained versions. While we evaluate the CGA versions against manually annotated ground-truths, it is an almost impossible to evaluate FGA versions. Hence, we train a fine-grained NER model to predict top-5 possible entity types for all entities and human annotators create ground-truths by using these predictions. For both cases, we extract 10K-word test sets for all three versions (original and post-processed). Note that, test sets are not identical but randomly selected sentences from the datasets. In addition, we exclude IOB tags since we prioritize evaluating entity type agreement in our evaluation results.
For evaluating CGA versions, given the sentence and the corresponding automatically created annotation, annotators are allowed to change any entity type with one of the five possible types, i.e. person, organization, location, misc or O (out). Finally, we merge the results of all annotators such that if at least 3 annotators agree on the same type for an entity, that type is the ground-truth. If there is no agreement, we keep the entity type as it is.
For evaluating FGA, we use a fine-grained entity recognizer, FIGER [@ling2012fine][^5], to create ground-truths. It is not an ideal evaluation approach since FIGER is designed for English and have not been tested for other languages according to our knowledge. However, more than thousands of different entity types are available in TWNERTC, and it is impracticable to ask human annotators to construct such fine-grained ground-truths manually from scratch. Therefore, we train a Turkish fine-grained model by using all remaining sentences and predicted possible types for entities in the test sets. Then, we ask human annotators to rank types of an entity from the most relevant to the least relevant. They are also allowed to suggest alternative types different than the given choices.
TWNERTC TWNERTC-DI TWNERTC-DD
---------------------------------- --------- ------------ ------------ --
\# of Annotated Entities 2376 1858 1965
\# of Ground-Truths for Entities 2891 2653 2 771
\# of Added Entity Type 872 958 926
\# of Removed Entity Type 537 163 120
\# of Changed Type 564 278 198
\# of Same Type 1275 1417 1647
We randomly sample 2K sentences from the unmodified corpus as TC test set. We train an internal classification algorithm with the rest of sentences and get top five predicted domains for the test sentences. We present predicted categories to human annotators and ask them to rank domains from the most relevant to the less relevant. They are also allowed to suggest different domains among the 49 domains which are represented in full corpus. Finally, we rank domains of each test sentence according to the annotator agreement and form the test set such that each sentence has 5 possible domains where first domain is the most relevant domain.
Evaluation Results for Coarse-Grained Annotations
-------------------------------------------------
In [Table \[tab:htc\_hec\_evaluation\_ner\]]{}, we present the number of entity types exist in automatically and manually annotated sets with the number of changes that annotators have made. We define changes from type *O* to any other type is an addition and opposite of this action is a removal. *Misc* is the most added, removed and changed type by annotators. It is an acceptable outcome since this specific type covers vast amount of entities except *person*, *location* and *organization*. We do not consider the number of added entity types as a major problem, since our gazetteers do not have infinite information and we have future plans to improve it. However, miss-annotated types are the real danger since if the amount of such mistakes increase, performance of the learning algorithms is affected negatively. In “change” type of , annotators change *misc* type to mainly *organization* and *person*.
As a conclusion, among the automatically annotated coarse-grained entity types, there are %76 matching ratio without *O* tags and manually added types. Additionally, we present precision, recall and F-score values in [Table \[tab:htc\_evaluation\_scores\_ner\]]{}. The dataset with domain-dependent post-process provides better NER performance compared to other two versions in general. Moreover, both post-processing methods improve the performance compared to the original dataset. Further, it can be observed that *misc* type is has the lowest F-Score among all types in all versions; however, this result is expected since *misc*’s coverage is larger than the other three types and makes it more vulnerable to mismatches.
Evaluation Results for Fine-Grained Annotations
-----------------------------------------------
As we discuss earlier, we train an external fine-grained algorithm, FIGER, to create Turkish NER models for all versions of the TWNERTC. By using the resulted models, we take five possible predicted type for each entity that are represented in test sets, and provide them as ground-truths to five human annotators. Obviously, FIGER designed to solve fine-grained NER in English; however, to evaluate the automated fine-grained types in a reasonable time, we leverage predictions of this algorithm.
In [Table \[tab:htc\_hec\_evaluation\_fine\_grained\_training\]]{}, we present the F1-scores of trained models on each test set. We provide these scores to give a better insight to researchers about ground-truths. Note that, strict represents the original F1-score formula, while loose macro and loose micro scores represent variations of the same formula [@ling2012fine]. It can be observed that the model trained with original TWNERTC performs poorly compared to post-processed versions. Since domain independent (DI) noise reduction method ensures that an entity can have only one type, it improves the performance more than the domain dependent (DD) method.
[Table \[tab:htc\_hec\_evaluation\_fine\_grained\_eval\]]{} presents the human annotators evaluation on automated fine-grained NER datasets, given FIGER predictions as possible ground-truths. Annotators rank the provided ground-truths and we check their ranking agreements. Eventually, top-1 agreement is hard to fulfill since our gazetteers contains thousands of entity types, and an entity may have more than ten different possible options. On the other hand, top-5 agreements provide promising results considering the amount of possible ground-truths.
Evaluation Results for TC
-------------------------
Original TWNERTC contains 49 different domains. The number of domains causes ambiguities in categorization among sentences depending on the context understanding. Thus, we evaluate three levels of accuracy. From [Table \[tab:htc\_hec\_evaluation\_tc\]]{}, we can see that automatically assigned domains have relatively low direct matches with annotators according to top-1 score. However, when we observe top-3, accuracy scores are doubled. Furthermore, error rates are less than %2 when all five ground truths are considered. Note that in top-3 and top-5, we only considered whether automatically assigned domain exists or not in the ground truth list.
Amount of the difference between from top-1 to top-3 scores is mainly caused by annotators’ different understanding of the sentence context. For instance, a sentence about Lionel Messi’s birth location is categorized as *people* in the test set. Whereas ground-truths start with *soccer* and followed by *people* and *location*. In addition, similar domains, such as *sports* and *soccer*, are also the reason of this difference.
In overall, results validates that automatically assigned domains are likely similar to what human annotators suggest to corresponding sentences. However, while automation process is limited to the context, humans can use their knowledge when suggesting domains. Hence, low accuracy in top-1 score is not a mistake of the methodology but a shortcoming in the process.
Conclusion and Future Research Directions
=========================================
We have described six, publicly available corpora for NER and TC tasks in Turkish. The data consists of Wikipedia texts which are annotated and categorized according to the entity and domain information extracted from Freebase. We explain the process to construct the datasets and introduce methodologies to eliminate noisy and incorrect data. We provide comprehensive statistics about dataset content. We analyzed subsets from these datasets and evaluate automatically created annotations and domains against manually created ground-truths. The final results show that automatic annotations and domains are quite similar to the ground-truths.
The obvious next step is to develop learning algorithms for NER and TC tasks to find baselines using traditional machine learning algorithms, and extending these baselines with approaches. Since TWNERTC provides a vast amount of structured data for researchers, deep learning methods can be exploited to solve fine-grained NER problem.
Acknowledgments
===============
This project is partially funded by 3140951 numbered TUBITAK-TEYDEB (The Scientific and Technological Research Council of Turkey – Technology and Innovation Funding Programs Directorate).
[^1]: https://www.freebase.com/
[^2]: https://dumps.wikimedia.org/
[^3]: [http://dx.doi.org/10.17632/cdcztymf4k.1]{}
[^4]: [http://dx.doi.org/10.17632/cdcztymf4k.1]{}
[^5]: https://github.com/xiaoling/figer
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this paper, we present our approaches for the FinSim 2020 shared task on “Learning Semantic Representations for the Financial Domain". The goal of this task is to classify financial terms into the most relevant hypernym (or top-level) concept in an external ontology. We leverage both context-dependent and context-independent word embeddings in our analysis. We submitted three systems which deploy Word2vec embeddings trained from scratch along with BERT embeddings pre-trained on a subset of the data (Financial Prospectus in English). We divide the test dataset into two subsets based on a domain rule. For one subset, we use cosine similarity to classify the term. This part is the same for the three systems. They differ for the second subset, where we use simple classifiers like Naive Bayes, on top of the embeddings, to arrive at a final prediction. Our best system ranks $1^{st}$ based on both the metrics, i.e., mean rank and accuracy.'
author:
- 'First Author$^1$[^1]'
- Second Author$^2$
- |
Third Author$^{2,3}$Fourth Author$^4$\
$^1$First Affiliation\
$^2$Second Affiliation\
$^3$Third Affiliation\
$^4$Fourth Affiliation\
{first, second}@example.com, [email protected], [email protected]
title: 'IJCAI–PRICAI–20 Example on typesetting multiple authors'
---
abstract, introduction, background/lit review, models, experimental setup, results,conclusion, references
Introduction
============
This short example shows a contrived example on how to format the authors’ information for [*IJCAI–PRICAI–20 Proceedings*]{}.
Background
==========
Methods
=======
A variety of methods, ranging from minimum distance to logistic regression, were employed. We broadly classify the set of techniques into supervised and unsupervised.
Supervised
----------
We had to use very simple supervised classifiers keeping in mind the scarce dataset. The features used for training were the average embedding for each term.
### Naïve Bayes
Naïve Bayes is a popular classical machine learning classifiers [@rish2001empirical]. The main assumption behind the model is that given the class labels. All features are conditionally independent of each other, hence the name Naïve Bayes. It is highly scalable, that is, takes less training time. It also works well on small datasets, making it a good baseline for our analysis. We used Bernoulli Naïve Bayes classifier from sklearn library.
### Logistic Regression
Logistic regression uses an equation as the representation, very much like linear regression, to model a binary dependent variable. The eight classes are treated as eight binary variables, which are assigned a probability between 0 and 1. Being a simple model, it works pretty well on small datasets. We used Logistic Regression classifier from sklearn library.
Unsupervised
------------
This section focused on word embeddings using the 150 pdfs provided to us. Terms were classified by calculating minimum distance, absolute distance and cosine similarities between term embeddings and label embeddings. These embeddings were also used as features in supervised learning. We use two main approaches:
### Word2vec
We use Word2vec embeddings [@mikolov2013efficient] for capturing semantic and syntactic properties of words. It is a dense low-dimensional representation of a word. We trained the embeddings on the given pdfs. Word2Vec represents each word as a vector. We tried different dimensions ranging from 50 to 500. A term is represented by an average of word embeddings of each of the words.
### BERT
Bidirectional Encoder Representations from Transformers (BERT) [@devlin2018bert] is the state-of-the-art language model, that has been found to be useful for numerous NLP tasks. It is deeply bidirectional (takes contextual information from both sides of the token) and learns a representation of text via self-supervised learning. BERT models pre-trained on large text corpora are available, and these can be fine-tuned for a specific NLP task. We fine-tuned BERT “write configuration”Base Uncased configuration[^2], which has 12 layers (transformer blocks), 12 attention heads and 110 million parameters. We then extracted embeddings from the last hidden layer for each word, which were used as word embeddings in unsupervised approaches.
Experimental Setup
==================
In this section, we quantitatively describe the dataset provided by the organizers and challenges accompanying it. We then mention the preprocessing steps briefly. Finally, we discuss the architecture and parameters of the unsupervised and supervised approaches in detail.
Data description
----------------
As a part of the task, we are provided with 100 terms with corresponding hypernym. As a common observation, majority of the terms contained the label within the term. For instance, consider the term “Convertible Bonds” and the corresponding label is bonds. Hence, such terms can be separately dealt using rule based approach. We were also provided with 150 pdfs, to equip us with large enough corpus to train word embeddings.
The dataset “table” comes with a lot of inherent challenges. Firstly, the dataset is too small for a supervised approach. Secondly, there were certain term in training data, which were not present in the provided corpus. Also, the corpus was provided in pdf format, and converting them to text added a lot of noise and sentence boundary detection proved to be a challenge.
Data preprocessing
------------------
Text preprocessing steps included removal of punctuation, stop words and special characters, followed by lower-casing, lemmatization and tokenization. We used nltk library[^3] [@loper2002nltk] for the same. The tokens were then converted to vectors using Word2vec embeddings. Finally, the average of all the word vectors is taken to create caption embeddings (as mentioned in section \[subsec:ann\]).
Data preprocessing
------------------
Text preprocessing steps included removal of punctuation, stop words and special characters, followed by lower-casing, lemmatization and tokenization. We used nltk library[^4] [@loper2002nltk] for the same. The tokens were then converted to vectors using Word2vec or BERT embeddings. Finally, the average of all the word vectors is taken to create final embedding for each term.
Model and parameters
--------------------
As mentioned in “mention section” some of the terms contained the label within them. On observing the training dataset, these terms can be directly classified into the corresponding label. But rule based approach, of directly classifying term into label creates problem in determining other 4 classes, and in case of unfortunate classification, we need to maximize our chances of getting better average rank. So we divided the dataset into two groups. In one group were the terms containing only one label. In the other group were terms in which the label came twice, or no label came.
[^1]: Contact Author
[^2]: https://github.com/google-research/bert
[^3]: https://pythonspot.com/category/nltk/
[^4]: https://pythonspot.com/category/nltk/
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
\
On behalf of the SuperB collaboration\
INFN Sezione di Pisa\
E-mail:
title: 'Status, open problems and prospects of the decay $B^+\rightarrow \ell^+ \nu_{\ell}$'
---
Introduction
============
SuperB [@SuperB] is a high luminosity $e^+e^-$ collider that will be able to indirectly probe NP at energy scales far beyond the reach of any accelerator planned or in existence. Just as detailed understanding of the SM was developed from stringent constraints imposed by flavour changing processes between quarks, the structure of any NP is severely constrained by flavour processes. The pattern of deviations from the SM can be used to test the NP. If NP is found at the LHC, then the many golden measurements from SuperB (of which $B^+ \rightarrow \ell^+\nu_{\ell}$ is an example) will help decode the subtle nature of the NP. However if no new particles are found at the LHC, SuperB will be able to search for NP at energy scales up to $100~{\rm TeV}$. In either scenario, flavour physics measurements that can be made at SuperB play an important role in understanding the nature of NP.
In the SM the purely leptonic $B$ meson decays $B^{+} \rightarrow \ell^{+}\nu_{\ell}$ proceed at the lowest order through an annihilation diagram with a $W^{+}$ exchange. The SM branching ratio (${\rm BR}$) can be calculated as [@Silverman] $$\label{eq:BF_Blnu}
{\rm BR}(B^{+}\rightarrow \ell^{+}\nu_{\ell})_{\rm SM} = \frac{G^2_F m_B m^2_{\ell}}{8\pi}\left(1 - \frac{m^2_{\ell}}
{m^2_B} \right)^2 f^2_B |V_{ub}|^2\tau_B~,$$ where $G_F$ is the Fermi constant, $m_{\ell}$ and $m_B$ are the lepton and $B^+$ masses, respectively, and $\tau_B$ is the $B^+$ lifetime. The ${\rm BR}$ is sensitive to the CKM matrix element $|V_{ub}|$ [@CKMmatrix] and the $B$ decay constant $f_B$. The SM estimate for ${\rm BR}(B^{+}\rightarrow \tau^{+}\nu_{\tau})$ is $(1.20 \pm 0.25)\times10^{-4}$, this assuming $\tau_B = 1.638\pm0.011~{\rm ps}$ [@TauB], $|V_{ub}| = (4.32 \pm 0.16 \pm 0.29)\times10^{-3}$ (errors are statistical and systematic, respectively) [@HFAG], and $f_B = 190 \pm 13 {\rm MeV}$ [@HPQCD]. The main uncertainties on the expected SM ${\rm BR}$ come from the $|V_{ub}|$ and $f_B$ parameters. To a very good approximation, helicity is conserved in $B^{+} \rightarrow \mu^+\nu_{\mu}$ and $B^{+} \rightarrow e^+\nu_{e}$ decays, leading to ${\rm BR}(B^{+}\rightarrow \mu^{+}\nu_{\mu}) = (5.4\pm1.1)\times10^{-7}$ and ${\rm BR}(B^{+}\rightarrow \mu^{+}\nu_{\mu}) = (1.3\pm0.4)\times10^{-11}$. However, reconstruction of $B^+\rightarrow \tau^+\nu_{\tau}$ decays is experimentally more challenging than $B^+\rightarrow \mu^+\nu_{\mu}$ or $B^+\rightarrow e^+\nu_{e}$ due to the large missing momentum from multiple neutrinos in the final state.
Purely leptonic $B$ decays are sensitive to NP, where additional heavy virtual particles replace the $W^{+}$ and contribute to the annihilation processes. Charged Higgs boson effects may greatly enhance or suppress the decay rate in some two-Higgs-doublet models [@TwoHiggsDoubletModel]. Similarly, there may be enhancements through mediation of leptoquarks in the Pati-Salam model of quark-lepton unification [@SUSY]. Direct test of Yukawa interactions in and beyond the SM are possible in the study of these decays, as annihilation processes proceed through the longitudinal component of the intermediate vector boson. In particular, in a SUSY scenario at large $\tan\beta$, non-SM effects in helicity-suppressed charged current interactions are potentially observable, being strongly $\tan\beta$-dependent and leading to [@TwoHiggsDoubletModel] $$\label{eq:BF_Blnu_NP}
\frac{{\rm BR}(B^+\rightarrow \ell^+\nu_{\ell})_{\rm NP}}{{\rm BR}(B^+\rightarrow \ell^+\nu_{\ell})_{\rm SM}} \simeq
\left( 1 - \tan^2\beta\frac{m^2_B}{M^2_H} \right)^2~,$$ where $M_H$ is the charged Higgs mass and ${\rm BR}(B^+\rightarrow \ell^+\nu_{\ell})_{\rm NP}$ is the NP expectation in the before mentionned NP models. As can be see from eq. \[eq:BF\_Blnu\_NP\], a measurement of the ${\rm BR}$ allows to set a constraint on the $\tan\beta - M_H$ plane.
Experimental Technique
======================
The recoil technique has been developed in order to search for rare $B$ decays with undetected particles, like neutrinos, in the final state. The technique consists of the reconstruction of one of the two B mesons ($B_{\rm tag}$), produced through the $e^+e^- \rightarrow \Upsilon(4S) \rightarrow B\bar{B}$ resonance, in a high purity hadronic or semi-leptonic final states, allowing to select a pure sample of $B\bar{B}$ events. Having identified the $B_{\rm tag}$, everything in the rest of the event (ROE) belongs by default to the signal B candidate ($B_{\rm sig}$), and so this technique provides a clean environment to search for rare decays. In this analysis, the $B_{\rm tag}$ is reconstructed in the hadronic modes (HD) $B\rightarrow D^{(*)}X$, where $X = n\pi + mK + pK^0_S + q\pi^0$ ($n+m+p+q < 6$), or semi-leptonic modes (SL) $B\rightarrow D^{(*)}\ell\nu$, ($\ell = e,~\mu$).
In the search for $B^+ \rightarrow \mu^+ \nu_{\mu}$ and $B^+ \rightarrow e^+ \nu_{e}$ decays, the signal is given by a single track identified as a muon and electron, respectively, in the ROE. In the search of $B^+ \rightarrow \tau^+ \nu_{\tau}$ decays, a single track as a muon, electron or pion is selected from the ROE, compatible with the $\tau^+ \rightarrow \mu^+ \nu_{\mu} \bar{\nu}_{\tau}$, $\tau^+ \rightarrow e^+ \nu_{e} \bar{\nu}_{\tau}$ and $\tau^+ \rightarrow \pi^+ \bar{\nu}_{\tau}$ decays, respectively. Furthermore, a single track and a neutral pion in the ROE is searched to reconstruct $\rho^+\rightarrow \pi^+\pi^0$ candidates compatible with the $\tau^+ \rightarrow \rho^+ \bar{\nu}_{\tau}$ decay.
One very important variable is the lepton momentum ($p'_{\ell}$) in the $B_{\rm sig}$ rest-frame, as the $B^+\rightarrow \ell^+\nu_{\ell}$ channels ($\ell = e,\mu$) produce monoenergetic leptons. This variable allows to separate $B^+\rightarrow \ell^+\nu_{\ell}$ from $B^+\rightarrow \tau^+(\rightarrow \ell^+ \nu_{\ell} \bar{\nu}_{\tau})\nu_{\tau}$ events, and provides additional discrimination against other sources of background. The closed kinematics of the hadronic recoil technique allow to easily calculate the $B_{\rm sig}$ rest frame from the reconstructed $B_{\rm tag}$ and beam energies. However, the semi-leptonic recoil technique poses a problem due to the presence of a neutrino in the $B_{\rm tag}$ reconstruction. As the only missing particle in the $B_{\rm tag}$ is a neutrino, it is possible to calculate CM angle between the $B_{\rm tag}$ and $D^{(*)}\ell$ momenta. Yet, as the $B_{\rm sig}$ and $B_{\rm tag}$ are back-to-back in the CM frame, this means that the $B_{\rm sig}$ momentum is contained in a cone around the $D^{(*)}\ell$ system. Using this information and the magnitude of the $B_{\rm sig}$ CM momentum ($p^*_{B} = \sqrt{(E^*_{\rm beam}/2)^2 - m^2_{B}}$, with $E^*_{\rm beam}$ the total beam energy in the CM-frame), it is possible to construct an estimator of $p'_{\ell}$ as the arithmetic average of the $p'_{\ell}$ calculated using all possible $B_{\rm sig}$ directions around the $D^{(*)}\ell$ system.
Finally, for these kind of decay modes with undetected particles in the final state, the most powerful variable for separating signal and background is the so-called extra energy, $E_{\rm extra}$, which is defined as the extra energy in the electromagnetic calorimeter not associated with the $B_{\rm tag}$ or $B_{\rm sig}$ candidates. For the signal this variable peaks strongly near zero.
Current Experimental Status
===========================
The latest state of the art results on $B^+ \rightarrow \ell^+\nu_{\ell}$ decay rates from both BaBar and Belle collaborations are summarized in table \[tab:Expe\_measurements\]. The current best knowledge on $B^+ \rightarrow \mu^+\nu_{\mu}$ and $B^+ \rightarrow e^+\nu_{e}$ channels are upper limits at $90\%$ C.L. In contrast, the $B^+ \rightarrow \tau^+\nu_{\tau}$ channel is a well established decay, with a value $(1.64 \pm 0.34)\times10^{-4}$ (combining all the experimental findings [@HFAG]), which is in agreement with the SM expectation. However, this last experimental result is a source of tension within the the CKM global fit. The indirect determination of $B^+ \rightarrow \tau^+\nu_{\tau}$ turns out to be at $~2.6\sigma$ ($~3.2\sigma$) from the experimental value, as estimated by the CKMfitter [@CKMFitter] (UTfit [@UTFIT]) collaboration. More precise experimental findings are needed to disentangle the current state of affairs.
Observable BaBar Belle
--------------------------------------------------------- --------------------------------------------------------------------------- -----------------------------------------------------------------------
${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})~~{\rm (SL)}$ $(1.7 \pm 0.8 \pm 0.2)\times 10^{-4}$ [@BaBar_BToenu_BTomunu_BTotaunu_SL] $(1.54^{+0.38+0.29}_{-0.37-0.31}\times 10^{-4}$ [@Belle_BTotaunu_SL]
${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})~~{\rm (HD)}$ $(1.8^{+0.57}_{-0.54} \pm 0.26)\times 10^{-4}$ [@BaBar_BTotaunu_HD] $(1.79^{+0.56+0.46}_{-0.49-0.51})\times 10^{-4}$ [@Belle_BTotaunu_HD]
${\rm BR}(B^+\rightarrow e^+\nu_{e})~~{\rm (SL)}$ $<0.8\times 10^{-5}$ [@BaBar_BToenu_BTomunu_BTotaunu_SL] —
${\rm BR}(B^+\rightarrow e^+\nu_{e})~~{\rm (HD)}$ $<1.9\times 10^{-6}$ [@BaBar_BToenu_BTomunu_HD] $< 0.98\times 10^{-6}$ [@Belle_BToenu_BTomunu_HD]
${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})~~{\rm (SL)}$ $<1.1\times 10^{-5}$ [@BaBar_BToenu_BTomunu_BTotaunu_SL] —
${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})~~{\rm (HD)}$ $<1.0\times 10^{-6}$ [@BaBar_BToenu_BTomunu_HD] $< 1.70\times 10^{-6}$ [@Belle_BToenu_BTomunu_HD]
: *Summary of the experimental findings on $B^+ \rightarrow \ell^+\nu_{\ell}$. The first and second errors are statistical and systematic. Upper limits are at $90\%$ C.L.*[]{data-label="tab:Expe_measurements"}
SuperB detector layout studies
==============================
Even though the expected SuperB increase in the instantaneous luminosity of a factor of $~100$ already promises significant improvements on the leptonic $B^+ \rightarrow \ell^+\nu_{\ell}$ decays, additional activities for detector optimization are currently ongoing. The SuperB baseline detector configuration is very similar to BaBar but the boost ($\beta\gamma$) is reduced from $0.56$ to $0.28$. This reduction increases the geometrical acceptance and so the reconstruction efficiency. A new layer is added to the vertex detector as close as possible to the beam pipe in order to not to degrade the time-dependent measurements. Additionally, the inclusion of two new devices that will increase further the geometrical acceptance of the detector is beging considered: a particle identification device (Fwd-PID) placed in the fordward region and an electromagnetic calorimeter (Bwd-EMC) located in the backward region, covering the polar angular regions of $(17,25)$ and $(152,167)$ degrees, respectively.
The Fwd-PID is a highly performant PID device for $K/\pi$ separation based on time-of-flight measurements, located in a region previously covered only by the tracking system. This new device will improve particle identification in a momentum region from $1.6$ to $5.0~{\rm GeV}$ where the tracking system alone is poorly performant. The Bwd-EMC will be used as a veto device, which means that no neutrals measured in it will be used to reconstruct the $B_{\rm tag}$ and $B_{\rm sig}$ candidates. Additional background suppression can be achieved by cutting on the total energy deposited in the Bwd-EMC, as the signal is expected to peak strongly at zero.
The SuperB fast simulation has been used to produce signal and the main background (generic $B\bar{B}$ decays) samples in the previously mentioned detector setups. This test showed that the reduced boost has the effect of increasing the signal efficiency by $\sim 7\%$ with an additional background suppression of $\sim 6\%$. The impact of the Fwd-PID device is to increase the signal and background reconstruction efficiencies by the same amount of $~2.5\%$, due to an increase of the tag-side kaons identification efficiency in the forward region. Finally, the impact of the Bwd-EMC is to reduce the backgrounds by $\sim 10\%$ with a negligible effect on the signal. The total effect is, at a fixed integrated luminosity, an increase in the total sample efficiency with a higher signal to background ratio $S/B$.
Expected sensitivities
======================
The $S/\sqrt{(S+B)}$ ratio, which would be the statistical significance of the ${\rm BR}$ measurement in a [*cut-and-count*]{} analysis, can be used as a measure of the expected sensitivities in SuperB. This ratio only takes into account the statistical uncertainties, and needs to be modified in order to consider the irreducible systematic uncertainties, $$\label{eq:Significane_syst}
{\rm Significance} = \frac{S}{\sqrt{(S+B+(\epsilon_{\rm syst}S)^2)}}~,$$ where $\epsilon_{\rm syst}$ is the total relative systematic error. No significant observation is expected at SuperB of the highly suppressed $B^+ \rightarrow e^+\nu_e$ decay, therefore it will be excluded from the subsequent discussion.
![\[fig:Ext\_BToTauNu\] [*left: Significance of the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ measurements as a function of integrated luminosity for the studied detector setups: BaBar (solid-black), SuperB baseline (dotted-black), Fwd-PID (red) and Bwd-EMC (blue). right: Excluded region on the $\tan\beta - M_H$ plane for the current (green) and expected sensitivities of SuperB at $75~{\rm ab}^{-1}$ (blue) from the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ (bottom) measurements.* ]{}](Extrapolation_BToTauNu_SuperB_Syst2.0Redu.eps "fig:"){width="7.0cm"} ![\[fig:Ext\_BToTauNu\] [*left: Significance of the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ measurements as a function of integrated luminosity for the studied detector setups: BaBar (solid-black), SuperB baseline (dotted-black), Fwd-PID (red) and Bwd-EMC (blue). right: Excluded region on the $\tan\beta - M_H$ plane for the current (green) and expected sensitivities of SuperB at $75~{\rm ab}^{-1}$ (blue) from the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ (bottom) measurements.* ]{}](BToTauNu_Pheno_plots_Extra_Syst2.0Redu.eps "fig:"){width="7.0cm"} ![\[fig:Ext\_BToTauNu\] [*left: Significance of the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ measurements as a function of integrated luminosity for the studied detector setups: BaBar (solid-black), SuperB baseline (dotted-black), Fwd-PID (red) and Bwd-EMC (blue). right: Excluded region on the $\tan\beta - M_H$ plane for the current (green) and expected sensitivities of SuperB at $75~{\rm ab}^{-1}$ (blue) from the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ (bottom) measurements.* ]{}](Extrapolation_BToMuNu_SuperB_Syst2.0Redu.eps "fig:"){width="7.0cm"} ![\[fig:Ext\_BToTauNu\] [*left: Significance of the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ measurements as a function of integrated luminosity for the studied detector setups: BaBar (solid-black), SuperB baseline (dotted-black), Fwd-PID (red) and Bwd-EMC (blue). right: Excluded region on the $\tan\beta - M_H$ plane for the current (green) and expected sensitivities of SuperB at $75~{\rm ab}^{-1}$ (blue) from the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (top) and ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ (bottom) measurements.* ]{}](BToMuNu_Pheno_plots_Extra_Syst2.0Redu.eps "fig:"){width="7.0cm"}
The irreducible systematic uncertainties on ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ (mainly due to $B_{\rm tag}$ and $B_{\rm sig}$ reconstruction efficiencies and $B\bar{B}$ counting) is $8.7\%$, which is currently a factor of $\sim 2$ smaller than the statistical error. It is evident that this measurement will be systematic dominated in the near future if no effort is made to reduce the systematic error. The uncertainty will saturate at $\sim 9\%$ already at $\sim50~{\rm ab}^{-1}$, which is only $2/3$ of the total expected dataset of SuperB. Experience has shown that systematics can be reduced with higher statistics, as it is possible to study larger control samples. It is then assumed that the systematic uncertainty can be reduced by a factor of two, which can be considered as a moderately conservative scenario. Under this hypothesis, we obtain the top-left plot of figure \[fig:Ext\_BToTauNu\], where we show the statistical significance as a function of the integrated luminosity, which gives an uncertainty of $~4.5\%$ at $75~{\rm ab}^{-1}$. In order to translate this into an excluded region in the $\tan\beta - m_H$ plane, it is needed as well to make some hypothesis on the systematic error of the SM branching ratio (see eq. \[eq:BF\_Blnu\_NP\]). The main uncertainties coming from $|V_{ub}|$ and $f_B$ (see eq. \[eq:BF\_Blnu\]), it will be assumed that the statistical error on $|V_{ub}|$ scales with luminosity and that the systematic component can be reduced by a factor of two. For the uncertainty on $f_B$ we use $1.5\%$, which is the expectated error for the SuperB era estimated by FLAG [@FLAG]. In the right-top plot of figure \[fig:Ext\_BToTauNu\] the excluded region (blue) for the expected sensitivities on the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$ at $75~{\rm ab}^{-1}$ is shown. For comparison we show (green) the excluded region with the current uncertainties. As can be seen, the excluded region can be significantly increased with the expected sensitivities at SuperB full dataset.
In the case of the $B^+\rightarrow \mu^+\nu_{\mu}$ channel, the irreducible systematic uncertainty is $\sim4.0\%$. As with $B^+\rightarrow \tau^+\nu_{\tau}$ decay, it is assumed that the systemtic errors can be reduced by a factor of two for the SuperB era, which gives the left-bottom plot of figure \[fig:Ext\_BToTauNu\]. As can be seen, the ${\rm BR}(B^+\rightarrow \mu^+\nu_{\mu})$ measurement will not be systematic dominated in contrast to the ${\rm BR}(B^+\rightarrow \tau^+\nu_{\tau})$. As shown in the bottom-right plot of figure \[fig:Ext\_BToTauNu\], the corresponding constraint on the $\tan\beta - M_H$ plane will be competitive with the one obtained from $B^+\rightarrow \tau^+\nu_{\tau}$ decays.
Summary
=======
In summary, we have investigated the reach of SuperB in the search of the $B^+\rightarrow \ell^+\nu_{\ell}$ decays with both the hadronic and semi-leptonic techniques. Preliminary results based on the SuperB fast simulation have shown a significant increase on the signal to background ratio due to the boost reduction and the impact of the Fwd-PID and Bwd-EMC devices. It has also been shown that under moderately conservative hypothesis on the evolution of the systematic uncertainties both $B^+ \rightarrow \tau^+\nu_{\tau}$ and $B^+ \rightarrow \mu^+\nu_{\mu}$ decays will give competitive an unprecedent reduction of the NP parameter space ($\tan\beta - M_H$ plane) for the expected SuperB sensitivities at $75~{\rm ab^{-1}}$ of data.
[99]{}
M. Bone [*et al.*]{} \[SuperB Collaboration\], arXiv:0709.0451 \[hep-ex\].
D. Silverman and H. Yao, Phys. Rev. D[**38**]{}, 214 (1988). N. Cabibbo, Phys. Rev. Lett. [**10**]{}, 531 (1963); M. Kobayashind [*et al.*]{}, Prog. Theor. Phys. [**49**]{}, 652 (1973). C. Amsler [*et al.*]{}, Physics Letters B [**667**]{}, 1 (2008). E. Barberio [*et al.*]{}, \[Heavy Flavor Averaging Group\], arXiv: hep-exp/0603003. E. Gamiz, [*et al.*]{} \[HPQCD Collaboration\], Phys. Rev. D[**80**]{}, 014503 (2009).
W. S. Hou, Phys. Rev. D [**48**]{}, 2342 (1993). A. G. Akeroyd [*et al.*]{}, J. Phys G [**29**]{} 2311, (2003).
J. Charles [*et al.*]{} \[CKMfitter Group\], Eur. Phys. J. C[**41**]{}, 1-131 (2005), (see http://ckmfitter.in2p3.fr). M. Bona [*et al.*]{} \[UTfit Collaboration\], JHEP 0507 (2005) 028, (see http://www.utfit.org).
B. Aubert, [*et al.*]{} \[BaBar Collaboration\], Phys. Rev. D[**81**]{}:051101, 2010. K. Hara, [*et al.*]{} \[Bella Collaboration\], arXiv: 1006.4201 \[hep-ex\].
P. del Amo Sanchez, [*et al.*]{} \[BaBar Collaboration\], arXiv:1008.0104 \[hep-ex\]. K. Ikado, [*te al.*]{} \[Bella Collaboration\], Phys. Rev. Lett. [**97**]{}:251802, 2006.
B. Aubert, [*et al.*]{} \[BaBar Collaboration\], Phys. Rev. D[**79**]{}:091101, 2009. Phys. Lett. B [**646**]{}, 67 (2007).
V. Lubicz, [*“The CKM analysis: inputs from theory”*]{}, talk given at [*The Xth Nicola Cabibbo International Conference on Heavy Quarks and Leptons, October 11-15, 2010, Frascati - Italy*]{}.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We consider the problem of electing a leader among nodes in a highly dynamic network where the adversary has unbounded capacity to insert and remove nodes (including the leader) from the network and change connectivity at will. We present a randomized algorithm that (re)elects a leader in $O(D\log n)$ rounds with high probability, where $D$ is a bound on the dynamic diameter of the network and $n$ is the maximum number of nodes in the network at any point in time. We assume a model of broadcast-based communication where a node can send only $1$ message of $O(\log n)$ bits per round and is not aware of the receivers in advance. Thus, our results also apply to mobile wireless ad-hoc networks, improving over the optimal (for deterministic algorithms) $O(Dn)$ solution presented at FOMC 2011. We show that our algorithm is optimal by proving that [*any*]{} randomized Las Vegas algorithm takes at least $\Omega(D\log n)$ rounds to elect a leader with high probability, which shows that our algorithm yields the best possible (up to constants) termination time.'
author:
- 'John Augustine[^1]'
- 'Tejas Kulkarni$^\ast$'
- 'Paresh Nakhe$^\ast$'
- 'Peter Robinson[^2]'
bibliography:
- 'papers.bib'
- 'papers1.bib'
- 'papers2.bib'
- 'leader.bib'
title: 'Robust Leader Election in a Fast-Changing World'
---
[^1]: Department of Computer Science and Engineering, Indian Institute of Technology Madras, Chennai, India. : [[email protected], [email protected], [email protected]]{}.
[^2]: Division of Mathematical Sciences, Nanyang Technological University, Singapore 637371. : [[email protected]]{}
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'E. Spitoni [^1]'
- 'G. Cescutti'
- 'I. Minchev'
- 'F. Matteucci'
- 'V. Silva Aguirre'
- |
\
M. Martig
- 'G. Bono'
- 'C. Chiappini'
date: 'Received xxxx / Accepted xxxx'
title: '2D chemical evolution model: the impact of galactic disc asymmetries on azimuthal chemical abundance variations'
---
Introduction
============
In recent years, integral field spectrographs (IFSs) have largely substituted long-slit spectrographs in studies designed to characterize the abundance distribution of chemical elements in external galaxies. IFSs have permitted for the first time to measure abundances throughout the entire two-dimensional extent of a galaxy (or a large part thereof) and, thus, to detect azimuthal and radial trends (Vogt et al. 2017).
In the last years, several observational works have been found evidence of significant azimuthal variations in the abundance gradients in external galaxies. S[á]{}nchez et al. (2015) and S[á]{}nchez-Menguiano et al. (2016) analyzed in detail the chemical inhomogeneities of the external galaxy NGC 6754 with the Multi Unit Spectroscopic Explorer (MUSE), concluding that the azimuthal variations of the oxygen abundances are more evident in the external part of the considered galaxy.
Vogt et al. (2017) studied the galaxy HCG 91c with MUSE and arrived to the conclusion that the enrichment of the interstellar medium has proceeded preferentially along spiral structures, and less efficiently across them.
Azimuthal variations have been detected in the oxygen abundance also in the external galaxy M101 by Li et al. (2013). Ho et al. (2017) presented the spatial distribution of oxygen in the nearby spiral galaxy NGC 1365. This galaxy is characterized by a negative abundance gradient for oxygen along the disc, but systematic azimuthal variations of $\sim$ 0.2 dex occur over a wide radial range of galactic radii and peak at the two spiral arms in NGC 1365. In the same work, the authors presented a simple chemical evolution model to reproduce the observations. Azimuthal variations can be explained by two physical processes: after a local self enrichment phase in the inter-arm region, a consequent mixing and dilution phase si dominant on larger scale (kpc scale) when the spiral density waves pass through.
Probing azimuthal inhomogeneities of chemical abundances has been attempted in the Milky Way system too. Balser et al. (2011), measuring H II region oxygen abundances, found that the slopes of the gradients differ by a factor of two in their three Galactic azimuth angle bins. Moreover, significant local iron abundance inhomogeneities have also been observed with Galactic Cepheids (Pedicelli et al. 2009; Genovali et al. 2014).
Balser et al. (2015) underlined the importance of azimuthal metallicity structure in the Milky Way disc making for the first time radio recombination line and continuum measurements of 21 HII regions located between Galactic azimuth $\phi$=90$^{\circ}$- 130$^{\circ}$. The radial gradient in \[O/H\] is -0.082 $\pm$ 0.014 dex kpc$^{-1}$ for $\phi$=90$^{\circ}$- 130$^{\circ}$, about a factor of 2 higher than the average value between $\phi$=0$^{\circ}$- 60$^{\circ}$. It was suggested that this may be due to radial mixing from the Galactic Bar.
Analyzing the Scutum Red-Supergiant (RSG) clusters at the end of the Galactic Bar, Davies et al. (2009) concluded that a simple one-dimensional parameterisation of the Galaxy abundance patterns is insufficient at low Galactocentric distances, as large azimuthal variations may be present. Combining these results with other data in the literature points towards large-scale ( $\sim$ kpc) azimuthal variations in abundances at Galactocentric distances of 3-5 kpc. It thus appears that the usual approximation of chemical evolution models assuming instantaneous mixing of metallicity in the azimuthal direction is unsubstantiated.
Azimuthal abundance gradients due to radial migration in the vicinity of spiral arms in a cosmological context have been studied in detail by Grand et al. (2012, 2014, 2016), and S[á]{}nchez-Menguiano et al. (2016).
Alternatively, Khoperskov et al. (2018) investigated the formation of azimuthal metallicity variations in the discs of spiral galaxies in the absence of initial radial metallicity gradients . Using high-resolution N -body simulations, they modeled composite stellar discs, made of kinematically cold and hot stellar populations, and study their response to spiral arm perturbations. They found that azimuthal variations in the mean metallicity of stars across a spiral galaxy are not necessarily a consequence of the reshaping, by radial migration, of an initial radial metallicity gradient. They indeed arise naturally also in stellar discs which have initially only a negative vertical metallicity gradient.
The aim of this paper is to develop a detailed 2D galactic disc chemical evolution model, able to follow the evolution of several chemical elements as in previous 1D models, but also taking into account azimuthal surface density variations. In this the paper when we refer to the thin and thick discs we mean the low- and high-\[$\alpha$/Fe\] sequences in the \[$\alpha$/Fe\]-\[Fe/H\] plane. Defining the thin and thick discs morphologically, rather than chemically, identifies a mixture of stars from both the low- and high-\[$\alpha$/Fe\] sequences, and vise versa (Minchev et al. 2015, Martig et al. 2016). It is, therefore, very important to make this distinction and avoid confusion. We follow the chemical evolution of the thin disk component, i.e. the low-$\alpha$ population. We assume that the oldest stars of that low-$\alpha$ component are associated with ages of $\sim$ 11 Gyr, in agreement with asteroseismic age estimates (Silva Aguirre et al. 2018).
Starting from the classical 1D Matteucci & Fran[ç]{}ois (1989) approach (the Galactic disc is assumed to be formed by an infall of primordial gas) we included 2D surface density fluctuation in the Milky Way disc chemo-dynamical model by Minchev et al. (2013) (hereafter MCM13), as well as using analytical spiral arm prescriptions.
Our paper is organized as follows. In Section 2, we describe the framework used for the new model. In Section 2.1 the adopted nucleosynthesis prescriptions are reported. In Section 2.2 the density fluctuation from the chemo-dynamical model by MCM13 are indicated. In Section 2.3 we present the analytical expressions for the density perturbations due to Galactic spiral arm. In Section 3 we presents our results with the density fluctuation from chemo-dynamical models and with an analytical spiral arm prescription are reported. Finally, our conclusions are drawn in Section 4.
A 2D galactic disc chemical evolution model
===========================================
The basis for the 2D chemical evolution model we develop in this section is the classical 1D Matteucci & Fran[ç]{}ois (1989) approach, in which the Galactic disc is assumed to be formed by an infall of primordial gas. The infall rate for the thin disc (the low-$\alpha$ sequence) of a certain element $i$ at the time $t$ and Galactocentric distance $R$ is defined as: $$\label{infall}
B(R,t,i)= X_{A_i} \, b(R) \, e^{-\frac{t}{\tau_D(R)}},$$ where $X_{A_i}$ is the abundance by mass of the element $i$ of the infall gas that here is assumed to be primordial, while the quantity $\tau_D(R)$ is the time-scale of gas accretion. The coefficient $b(R)$ is constrained by imposing a fit to the observed current total surface mass density $\Sigma_{D}$ in the thin disc as a function of the Galactocentric distance given by: $$\Sigma_D(R,t_G)=\Sigma_{D,0}e^{-R/R_{D}},
\label{mass}$$ where $t_G$ is the present time, $\Sigma_{D,0}$ is the central total surface mass density and $R_{D}$ is the disc scale length. The fit of the $\Sigma_D(R)$ quantity using the infall rate law of eq. (\[infall\]) is given by:
$$\sum_i \int_0^{t_G} X_{A_i} b(R) e^{-\frac{t}{\tau_D(R)}} dt = \Sigma_D (R,t_G),$$
The observed total disc surface mass density in the solar neighbourhood is $\Sigma_D (8 \mbox{ kpc}, t_G)=$ 54 M$_{\odot}$ pc$^{-2}$ (see Romano et al. 2000 for a discussion of the choice of this surface density). The infall rate of gas that follows an exponential law is a fundamental assumption adopted in most of the detailed numerical chemical evolution models in which the instantaneous recycling approximation (IRA) is relaxed.
An important ingredient to reproduce the observed radial abundance gradients along the Galactic disc is the inside-out formation on the disc (Spitoni & Matteucci 2011, Cescutti et al. 2007, Mott et al. 2013). The timescale $\tau_D(R)$ for the mass accretion is assumed to increase with the Galactic radius following a linear relation given by (see Chiappini et al. 2001):
$$\tau_{D}(R) = 1.033 R(\mbox{kpc}) - 1.27 \mbox{ Gyr}
\label{tau}$$
for Galactocentric distances $\geq$ 4 kpc. For the star formation rate (SFR) we adopt a Kennicutt (1998) law proportional to the gas surface density: $$\Psi(R,t) = \nu \Sigma_g^k(R,t),
\label{SFR}$$ where $\nu$ is the star formation efficiency (SFE) process and $\Sigma_g(R,t)$ is the gas surface density at a given position and time. The exponent $k$ is fixed to 1.5 (see Kennicutt 1998).
We divide the disc into concentric shells 1 kpc wide in the radial direction. Each shell is itself divided into 36 segments of width $\ang{10}$. Therefore at a fixed Galactocentric distance 36 zones have been created.
With this new configuration we can take into account variations of the SFR along the annular region, produced by density perturbations driven by spiral arms or bars. Therefore, an azimuthal dependence appears in eq. (\[SFR\]) and, which can be written as follows: $$\Psi(R,t,\phi) = \nu \Sigma_g^k(R,t,\phi).
\label{SFR2}$$
In this paper we will show results related to the effects of density fluctuations of the chemo-dynamical model of MCM13 and we will test the effects of an analytical formulation for the density perturbations created by spiral arm waves. The reference model without any density azimuthal perturbation is similar to the one by Cescutti et al. (2007), which as been shown to be quite successful in reproducing the most recent abundance gradients observed in Cepheids (Genovali et al. 2015).
Nucleosynthesis prescriptions
-----------------------------
In this work we present results for the azimuthal variations of abundance gradients for oxygen and iron. As done in a number of chemical evolution models in the past (e.g. Cescutti et al. 2006, Spitoni et al. 2015, 2019, Vincenzo et al. 2019), we adopt the nucleosynthesis prescriptions by Fran[ç]{}ois et al. (2004) who provided theoretical predictions of \[element/Fe\]-\[Fe/H\] trends in the solar neighbourhood for 12 chemical elements.
Fran[ç]{}ois et al. (2004) selected the best sets of yields required to best fit the data (details related to the observational data collection are in Fran[ç]{}ois et al. 2004). In particular, for the yields of Type II SNe they found that the Woosley & Weaver (1995) ones provide the best fit to the data: no modifications are required for the yields of iron, as computed for solar chemical composition, whereas for oxygen, the best results are given by yields computed as functions of the metallicity. The theoretical yields by Iwamoto et al. (1999) are adopted for the Type SNeIa, and the prescription for single low-intermediate mass stars is by van den Hoek & Groenewegen (1997).
Although Fran[ç]{}ois et al. (2004) prescriptions still provide reliable yields for several elements, we must be cautious about oxygen. Recent results have shown that rotation can influence the oxygen nucleosynthesis in massive stars (Meynet & Meader 2002) and therefore chemical evolution (Cescutti & Chiappini 2010), in particular at low metallicity. However, this does not affects our results being the data shown in this project relatively metal rich. Moreover, we are mostly interested in differential effects, rather than absolute values.
2D disc surface density fluctuations from the MCM13 model
---------------------------------------------------------
We consider the gas density fluctuations present in the Milky Way like simulation obtained by Martig et al. (2012) and chosen in MCM13 for their chemodynamical model. The simulated galaxy has a number of properties consistent with the Milky Way, including a central bar. MCM13 followed the disc evolution for a time period of about 11 Gyr, which is close to the age of the oldest low-$\alpha$ disc stars in the Milky Way. The classical 1D chemical evolution model is quite successful in reproducing abundance gradient along the Galactic disc (Cescutti et al. 2007).
The chemical evolution model used by MCM13 was very similar to the one adopted here; a comparison between its star formation history and that of the simulation was presented in Fig. A.1 by Minchev et al. (2014), showing good agreement. To extracted the gas density variations we binned the disk into 18 1-kpc-wide radial bins and 10$^{\circ}$-wide azimuthal bins at $|z|<$ 1 kpc. The time resolution is 37.5 Myr for 11 Gyr of evolution. All of the above is used for our new model described below.
With the aim of preserving the general trend of the 1D chemical evolution model, we introduce a density contrast function $f$ related to the perturbations originated by the MCM13 model. At a fixed Galactocentric distance $R$, time $t$ and azimuthal coordinate $\phi$, the new surface mass density is: $$\Sigma_D(R,t,\phi)=\Sigma_D(R,t) f(\phi,R,t).$$ We impose that the average value of the density contrast $f$ is 1, i.e.:
$$\langle f(\phi,R,t) \rangle_{\phi}=1.$$
This guarantees that, at a fixed Galactocentric distance $R$ and a time $t$, the average surface mass density is the one predicted by the 1 D chemical evolution model.
ISM density fluctuations from analytical spiral structure
---------------------------------------------------------
Here we investigate the effect of an analytical spiral arm formulation on the azimuthal variations of the abundance gradients.
In particular, we analyse steady wave spiral patterns. As suggested by Bertin et al (1989) and Lin & Shu (1966) when the number of important spiral modes of oscillation is small, the spiral structure is expected to have a highly regular grand design and to evolve in time in a quasi- stationary manner.
In this work, we consider the model presented by Cox & G[ó]{}mez (2002). The expression for the time evolution of the density perturbation created by spiral arms, referred to an inertial reference frame not corotating with the Galactic disk, in terms of the surface mass density is:
$$\Sigma_S(R,\phi,t)= \chi(R,t_G) M(\gamma),$$
where $\chi(R,t_G)$ is the present day amplitude of the spiral density: $$\chi(R,t_G)=\Sigma_{S,0} e^{-\frac{R-R_0}{R_{S}}},$$ while $M(\gamma)$ is the modulation function for the “concentrated arms” given by Cox & G[ó]{}mez (2002). The $M(\gamma)$ function can be expressed as follows:
$$M(\gamma)= \left(\frac {8}{3 \pi} \cos(\gamma)+\frac {1}{2} \cos(2\gamma) +\frac{8}{15 \pi} \cos(3\gamma) \right),
\label{MGAMMA}$$
$$\gamma(R,\phi,t)= m\left[\phi +\Omega_s t -\phi_p(R_0) -\frac{\ln(R/R_0)}{\tan(\alpha)} \right].
\label{gamma}$$
In eq. (\[gamma\]), $m$ is the number of spiral arms, $\alpha$ is the pitch angle, $R_S$ is the radial scale-length of the drop-off in density amplitude of the arms, $\Sigma_{0}$ is the surface arm density at fiducial radius $R_0$, $\Omega_s$ is the pattern angular velocity, with the azimuthal coordinate $\phi$ increasing counter-clockwise and a clockwise rotation, $\phi_p(R_0)$ is the coordinate $\phi$ computed at $t$=0 Gyr and $R_0$. An important feature of such a perturbation is that its average density at a fixed Galactocentric distance $R$ and time $t$ is zero,
$$\langle \Sigma_S \rangle_{\phi}= \Sigma_{S, 0} e^{-\frac{R-R_0}{R_{S}}} \langle M(\gamma) \rangle_{\phi}=0.$$
In Fig. \[MGAMMAF\] we show the modulation function $M(\gamma)$ of “concentrated arms” on the Galactic plane using the model parameters suggested by Cox & G[ó]{}mez (2002): $R_0=8$ kpc, $\alpha=\ang{15}$, $R_S=7$ kpc. The modulation function is computed at 5 Gyr assuming the angular velocity value of $\Omega_s$ = 20 km s$^{-1}$ kpc$^{-1}$ and $\phi_p(R_0)=0$. In this work we aim to investigate the effects of spiral arm density perturbations on the chemical enrichment by ejecta from stellar populations perfectly corotating with the Galactic disk. Our purpose here is the study of regular gas density perturbation linked to simple but reliable spiral arm descriptions.
To properly describe the temporal evolution of local density perturbations, the relative spiral arm speed pattern compared to the Galactic disk motion must be computed (further details will be provided in Section 3.2, in the Result discussion).
Cox & G[ó]{}mez (2002) provided a value for the spiral arm perturbation density at 8 kpc equal to $\rho_0= \frac{14}{11}$ m$_H$ cm$^{-3}$. Our implementation requires the surface density $\Sigma_{S,0}$, which can be recovered from the $z$ direction amplitude provided by Cox & G[ó]{}mez (2002) (their eq. 1), with the following relation:
$$\Sigma_{S,0}=2 \rho_0 \int_0^\infty \mbox{sech}^2\left(\frac{x}{H}\right)dx=2 H \rho_0,$$
where $H$ is the disc scale-height. Adopting $H$=180 pc (chosen to match the scale-height of the thin stellar disc proposed by Dehnen & Binney 1998, Model2; and in agreement with Spitoni et al. 2008) we obtain
$$\Sigma_{S,0}=21.16 \mbox{ M}_{\odot} \mbox{ pc}^{-2}.$$
It is important to underline that in our approach the time dependence of the density perturbation by the spiral arms is only in the modulation function $M(\gamma)$ through the term $\Omega_s t$ (see eqs. \[MGAMMA\] and \[gamma\]). Currently, there are no analytical prescriptions for the time evolution of both the amplitude of the spiral arm perturbation and its radial profile in the Galactic evolution context (spiral arm redshift evolution). Therefore, we make the reasonable assumption that during the Galactic evolution the ratio between the amplitude of the spiral density perturbation $\chi(R,t)$ and the total surface density $\Sigma_D(R,t)$ computed at the same Galactic distance $R$ remains constant in time, i.e. $ \frac{d}{dt} \, \left[
\chi(R,t)/\Sigma_D(R,t) \right]$=0, assuming a coeval evolution of these two structures in time. We define the dimensionless quantity $\delta_S(R,\phi,t)$ as the following ratio:
$$\delta_S(R,\phi,t)= \frac{ \Sigma_S(R,\phi,t)+ \Sigma_D(R,t)}{\Sigma_D(R,t)}=1 + \frac{ \Sigma_S(R,\phi,t)}{\Sigma_D(R,t)}.
\label{delta}$$
With the assumption that the ratio $\chi(R,t)/\Sigma_D(R,t)$ is constant in time, eq. (\[delta\]) becomes:
$$\delta_S(R,\phi,t) =1 + M(\gamma)\frac{
\chi(R,t_G)}{\Sigma_D(R,t_G)}.
\label{delta2}$$
If we include the contribution of the perturbation originated by spiral arm in the SFR driven by a linear Schmidt (1959) law (i.e. $\Psi= \nu \Sigma_g(R,t)$) we have that:
$$\Psi(R,t,\phi)_{d+s} = \nu \Sigma_g(R,t) \delta_S(R,\phi,t).
\label{SFR_D}$$
We are aware that this is a simplification to the more complex behavior seen in N-body simulations (Quillen et al. 2011, Minchev et al. 2012b, Sellwood and Carlberg 2014) and external galaxies (Elmegreen et al. 1992; Rix & Zaritsky 1995; Meidt et al. 2009), where multiple spiral patterns have been found. We will make use of this description in Section 3.2.2, where we will consider the simultaneous perturbation by a number of spiral patterns moving at different angular velocities.
![The modulation function $M(\gamma)$ of eq. (\[MGAMMA\]) for concentrated arms by Cox & G[ó]{}mez (2002) with $N=2$ spiral arms, fiducial radius $R_0=8$ kpc, pitch angle $\alpha=\ang{15}$, and $\phi_p(R_0)=0$.[]{data-label="MGAMMAF"}](Figure_1_5gyr.png)
As stated in the previous Section, the average modulation function over the azimuth $\phi$ at a fixed time $t$ and Galactocentric distance $R$ is null ($\langle M(\gamma) \rangle_{\phi}=0$). Therefore, in presence of a linear Schmidt (1959) law at a fixed Galactocentric distance the average value of $\Psi(R,t,\phi)_{d+s}$ over $\phi$ of the SFR defined in eq. (\[SFR\_D\]) is equal to the unperturbed SFR by the following expression:
$$\langle\Psi(R,t,\phi)_{d+s}\rangle_{\phi} =\Psi(R,t) \langle 1 + M(\gamma)\frac{ \chi(R)}{\Sigma_D(R,t_G)}\rangle_{\phi}=$$
$$=\Psi(R,t)\left(1+\langle M(\gamma) \rangle_{\phi}\frac{ \chi(R)}{\Sigma_D(R,t_G)}\right)=\Psi(R,t).$$
Here, we do not adopt a linear Schmidt (1959) law, and we use the SFR proposed by Kennicutt (1998) which exhibits the exponent $k$=1.5. Hence, the SFR in the Galactic disc in presence of spiral arm density perturbations becomes: $$\Psi_k(R,t,\phi)_{d+s} = \nu \Sigma_g(R,t)^k \delta_S(R,\phi,t)^{k}.
\label{SFR_k}$$
Roberts (1969) provided the exact shape of the steady gas distribution in spiral arms, finding an offset between the maximum of the stellar spiral arm and the maximum of the gas distribution driven by galactic shocks. In his Figure 7, it is shown that the regions of newly born luminous stars and the HII regions lie on the inner side of the observable gaseous spiral arm of HI. The presence of a small but noticeable offset between the gas and stellar spiral arms has been also found in the study of interactions between disc galaxies and perturbing companions in 3D N-body/smoothed hydrodynamical numerical simulations by Pettitt (2006).
Because of uncertainties related to the real magnitude of this offset (small offsets are predicted by Pettitt 2006), in our work we do not consider it, and the SFR is more enhanced in correspondence of the total density perturbation peak (see eq. \[delta2\] and the modulation function in Figure 1). We are aware that is true only near the corotation radius, however with our simpler approach we provide an upper limit estimate for the azimuthal abundance variations generated by steady spiral arms density perturbations. In presence of an off-set the density perturbation should be less “concentrated” and more smeared.
Our model in the presence of analytical spiral arms must be considered as a first attempt to include spiral structure in a classical chemical evolution model. As stated in Section 2.2, we will also present results for the azimuthal abundance variations originated by chemodynamical Milky Way like simulation in the presence of spiral arms and bar in a self consistent way. Our analytical spiral arms model is meant to break down the problem to understand the reason for the causes of azimuthal variations. Assuming that modes add linearly, we can approximate a realistic galactic disk by adding several spiral sets with different pattern speeds, as seen in observations (e.g., Meidt et al. 2009) and simulations (e.g., Masset & Tagger 1997, Quillen et al. 2011, Minchev et al. 2012a).
 
![Results for the chemical evolution model in which we consider the density fluctuation by the chemo-dynamical model by MCM13. Time evolution of the oxygen abundance gradient at $\phi$=0$^{\circ}$. []{data-label="minchev_grad_ev"}](ev_gradient_ivan.png)
![ [*Upper Panel*]{}: We present the average Fe abundances of Galactic Cepheids presented by Genovali et al. (2014) in bin of 15$^{\circ}$ for the azimuthal coordinate $\phi$ at different Galactocentric distances. [*Lower left Panel*]{}: Fe abundances as functions of the azimuthal coordinates computed at 6, 8, 10, 12 kpc predicted by the chemical evolution model in which we implemented the density fluctuation by the MCM13 model. [*Lower right Panel*]{}: residual of the Fe abundances predicted by our model computed after subtracting the average radial gradient. []{data-label="CEP1"}](cep1_large.png "fig:") ![ [*Upper Panel*]{}: We present the average Fe abundances of Galactic Cepheids presented by Genovali et al. (2014) in bin of 15$^{\circ}$ for the azimuthal coordinate $\phi$ at different Galactocentric distances. [*Lower left Panel*]{}: Fe abundances as functions of the azimuthal coordinates computed at 6, 8, 10, 12 kpc predicted by the chemical evolution model in which we implemented the density fluctuation by the MCM13 model. [*Lower right Panel*]{}: residual of the Fe abundances predicted by our model computed after subtracting the average radial gradient. []{data-label="CEP1"}](ivan_cep_new.png "fig:")
Results
=======
In this section we apply our 2D model by using surface density fluctuations from the MCM13 chemo-dynamical model and from an analytical prescription.
Density fluctuation from the MCM13 chemo-dynamical model
--------------------------------------------------------
In this section we present our results based on the new 2D chemical evolution model including the density mass fluctuation extracted from the chemo-dynamical model by MCM13.
Fig. \[mSFR\] shows the galactic disc SFR computed at 0.1, 0.7, 6, 11 Gyr after the start of disc formation, for the chemical evolution model in which we tested the effects of the density fluctuation by MCM13 in units of M$_{\odot}$ pc$^{-2}$ Gyr$^{-1}$. We notice that at early times (i.e the “1 Gyr” case reported in the upper left panel), the SFR is more concentrated in the inner Galactic regions, the SFR in the innermost regions decreases and the outer parts become more star forming active because of the “inside-out” prescription coupled with the inclusion of the density fluctuation. At the Galactic epoch of 1 Gyr after the start of disc formation, regions with the same Galactocentric distances have approximately the same SFR. Already after 0.7 Gyr of Galactic evolution, azimuthal star formation inhomogeneities are not negligible. Concerning the panel with the model results at 6 Gyr, azimuthal inhomogeneities are evident, in particular at 8 kpc the ratio between the maximum and the minimum values assumed by the SFR is SFR$_{max}$ /SFR$_{min}$=6.72.
In Fig. \[mSFR\] the bar and spiral arms features do not show up clearly, especially in early times. This is caused by the adopted inside out prescription (eq. \[tau\]) which leads to huge differences between the SFRs computed in inner and outer regions. In Fig. \[1SFR\], the galactic disc SFR($R$, $\phi$) is normalized to the maximum value SFR$_{R, \ max}$ of the annular region located at the Galactocentric distance $R$, i.e SFR($R$, $\phi$)/SFR$_{R, max}$, computed at 0.1, 0.7, 6, 11 Gyr after the start of disc formation, respectively. Here, different features related to density perturbations originated by spiral arms and bar can be noted.
In Fig. \[minchev\_av\] the main results related to the present day oxygen abundance azimuthal variation are presented. The top panel shows the azimuthal distribution of the residual of the oxygen abundances computed with our chemical evolution model at 4, 8, 12, 16, and 18 kpc after subtracting the average radial gradient (i.e. the one obtained with the reference model without any density perturbation). Throughout this paper we adopt the photospheric values of Asplund et al. (2009) as our solar reference. We see that the behavior is in excellent agreement with the observations by S[á]{}nchez et al. (2015); indeed, data show that outer regions display larger azimuthal variations, and the amplitude of the risidual variations are of the order of 0.1 dex (see Figure 7 by S[á]{}nchez et al. 2015) . In our model the maximum variations are $\sim$ 0.12 dex for the chemical evolution models computed at 18 kpc. Our results appears to have a bit less scatter.
In the lower panel of Fig. \[minchev\_av\] we present our “mock” observations. We draw oxygen abundances of different ISM regions at different Galactocentric distance at random azimuthal coordinates $\phi$. Hence, we add an error of $\sigma_{\phi}$=5$^{\circ}$ to alleviate the fact that our model presents a resolution of 10$^{\circ}$ in the azimuthal component $\phi$. Moreover, the average observational uncertainty associated to the oxygen abundances of $\sigma_{[O/H]}$ = 0.05 dex provided by S[á]{}nchez et al. (2015) has been considered. We define the “new” oxygen abundance including these uncertainties as follows: $$[\mbox{O/H}]_{new} = [\mbox{O/H}]+ U([-\sigma_{[O/H]}, \sigma_{[O/H]}]),
\label{Err1}$$ where $U$ is the random generator function. Similarly, we implement the uncertainty in the azimuthal component through the following relation: $$\phi_{new} = \phi + U([-\sigma_{\phi}, \sigma_{\phi}]).
\label{Err2}$$
Here, it is clearly visible the similarity between the S[á]{}nchez et al. (2015) observations and our results. To summarize, the inclusions of density perturbations taken from a self-consistent dynamical model at different Galactic times, leads to significant variations in chemical abundances in the outer Galactic regions.
In Fig. \[minchev\_grad\] we show results for the present day abundance gradient (after 11 Gyr of evolution) for iron computed for six azimuthal slices (as indicated) of width 10$^{\circ}$ at different azimuthal coordinates. In the same plot is indicated with a shaded grey area the maximum spread in the abundance ratio \[Fe/H\] obtained by the azimuthal coordinates we considered (0$^{\circ}$, 60$^{\circ}$, 120$^{\circ}$, 180$^{\circ}$, 240$^{\circ}$, and 300$^{\circ}$). As a consequence of the results presented above, the shaded area is larger towards external regions. We also overplot the data from Genovali et al. (2014) in order to compare to our model predictions. We notice that the predicted gradient is slightly steeper than the observed one in the external Galactic regions. However, we notice that the model lines pass within in the data standard deviation computed dividing the data by Genovali et al. (2014) in six radial bins.
In Fig. \[minchev\_grad\_ev\] we tested the effects of chemo-dynamical fluctuations on the time evolution of the oxygen abundance gradient at a fixed azimuth ($\phi$=0$^{\circ}$). In agreement with Minchev et al. (2018) the abundance gradient flattens in time, because of the chemical evolution model assumptions. As shown by Spitoni et al. (2015) and Grisoni et al. (2018), the inclusion of radial gas flows can in lead to even steeper gradients in time during the whole Galactic history.
![ Spiral pattern speed $\Omega_s$ and disk angular velocity $\Omega_d$ computed by Roca-F[à]{}brega et al. (2014) are indicated with light blue and violette lines, respectively. With the vertical long dashed red line we show the position of the corotation radius located at the Galactocentric distance $R=8.31$ kpc. Outer and Inner Lindblad resonances extracted by Roca-F[à]{}brega et al. (2014) simulation are also drawn with dotted magenta and dotted purple lines, respectively. []{data-label="omega"}](omega.png)
[c|cccc]{}\
Models &$m$ &$\alpha$& $\Omega_{s}$\
& & & \[km s$^{-1}$ kpc$^{-1}$\]\
\
S2A & 2 & 15$^{\circ}$ & 20\
S2B & 2 & 15$^{\circ}$ & 17.5\
S2C & 2 & 15$^{\circ}$ & 15\
S2D & 2 & 15$^{\circ}$ & 13.75\
S2E & 2 & 15$^{\circ}$ & 12.5\
S2F& 2 & 15$^{\circ}$ & 25\
S2G & 2 & 7$^{\circ}$ & 20\
S2H & 2 & 30$^{\circ}$ & 20\
S1A & 1 & 15$^{\circ}$ & 20\
S1B & 1 & 15$^{\circ}$ & 17.5\
S1C &1 & 15$^{\circ}$ & 15\
S1D & 1 & 15$^{\circ}$ & 13.75\
S1E & 1 & 15$^{\circ}$ & 12.5\
S1F& 1 & 15$^{\circ}$ & 25\
![ Results for the chemical evolution model in which we consider the density fluctuation associated with the analytical spiral arm formulation. [*Upper Panel*]{}: The azimuthal distribution of the residual of the oxygen abundances computed with our chemical evolution model at 4, 8, 12, 16, and 20 kpc (after subtracting the average radial gradient for a model with $R_S$=7, $R_D$=3.5, $\Sigma_0$=20, $\nu$=1.1, $\Omega_s$=20 km s$^{-1}$ kpc$^{-1}$, and $m$=2 spiral arm (model S2A in Table 1). [*Lower Panel*]{}: the time evolution of the \[O/H\] abundance as a function of the azimuthal coordinate computed at 8 kpc. []{data-label="SA"}](R3.png "fig:") ![ Results for the chemical evolution model in which we consider the density fluctuation associated with the analytical spiral arm formulation. [*Upper Panel*]{}: The azimuthal distribution of the residual of the oxygen abundances computed with our chemical evolution model at 4, 8, 12, 16, and 20 kpc (after subtracting the average radial gradient for a model with $R_S$=7, $R_D$=3.5, $\Sigma_0$=20, $\nu$=1.1, $\Omega_s$=20 km s$^{-1}$ kpc$^{-1}$, and $m$=2 spiral arm (model S2A in Table 1). [*Lower Panel*]{}: the time evolution of the \[O/H\] abundance as a function of the azimuthal coordinate computed at 8 kpc. []{data-label="SA"}](R3_t.png "fig:")
![ [*Upper Panel*]{}: Galactic disc SFR resulting from model S2A after 1 Gyr of evolution (see Table 1 and text for model details). The color code indicates the SFR in units of M$_{\odot}$ pc$^{-2}$ Gyr$^{-1}$. [*Lower Panel*]{}: same but computed at 11 Gyr.[]{data-label="SASFR"}](R3_SFR_1Gyr.png "fig:") ![ [*Upper Panel*]{}: Galactic disc SFR resulting from model S2A after 1 Gyr of evolution (see Table 1 and text for model details). The color code indicates the SFR in units of M$_{\odot}$ pc$^{-2}$ Gyr$^{-1}$. [*Lower Panel*]{}: same but computed at 11 Gyr.[]{data-label="SASFR"}](R3_SFR_11Gyr.png "fig:")
![ Disk angular velocity $\Omega_d$ computed by Roca-F[à]{}brega et al. (2014) is indicated with light blue and violette lines. With different horizontal solid lines are indicated the spiral pattern speed $\Omega_s$ adopted in our models (see text and Table 1 for model details). The vertical long dashed lines show the positions of the corotation radii assuming different $\Omega_s$ values.[]{data-label="DOS"}](DIF_OMEGA_S.png)
![ Present day residual azimuthal variations in oxygen abundance for the corotation regions (as indicated) of the different pattern speeds shown in Fig. \[S2\_grad\]. An increase in the effect is found as the corotation shifts to larger radius, i.e., for slower spiral patterns. Such a set of spirals with progressively slower patterns speeds as radius increases, can be a realistic representation of a galactic disk. []{data-label="CS2"}](cor_spir_2_lab.png)
![ As in Fig. \[SA\] but for model S1A, with with multiplicity $m$=1 of spiral arms.[]{data-label="S1A"}](R04.png "fig:") ![ As in Fig. \[SA\] but for model S1A, with with multiplicity $m$=1 of spiral arms.[]{data-label="S1A"}](R04_t.png "fig:")
In Fig. \[CEP1\] we compare the average iron abundance azimuthal variation in bins of $\phi$=15$^{\circ}$ presented by Genovali et al. (2014) computed at 6, 8, 10, and 12 kpc, respectively with our 2D chemical evolution model, resulting from the MCM13 density variations. We see that the observed azimuthal variations are for limited Galactocentric distances (6-12 kpc) and with a narrow range of azimuthal coordinates. Although it is evident that the observed amplitude of azimuthal variations are larger than the ones predicted by our models, more precise Galactic Cepheid data are required to make firm conclusions.
Moreover, other dynamical processes that we have not considered in this work had maybe played important roles in the evolution and in the building up of the Galactic gradients and their azimuthal variations - radial migration processes can already introduce some variations in about a Gyr (Quillen et al. 2018).
![ As in Fig. \[CS2\] but for models with $m=1$ multiplicity (see Table 1).[]{data-label="CS1"}](cor_spir_1_lab.png)
Density fluctuations from an analytical spiral arm formulation
--------------------------------------------------------------
In this Section we discuss the results of chemical evolution models with only analytical prescriptions for spiral arm density perturbations without including any density fluctuations from chemo-dynamical models. The primary purpose here is to test the effect of regular perturbations (i.e. spiral arms evolution described by an analytical formulation) on the chemical evolution of a Milky Way like galaxy. We underline that the results showed in the previous Section reflect more closely the complex behavior of the Milky Way. However, we are also interested to explore different spiral arm configurations which could characterize external galactic systems by varying the free parameters of the analytical expression of the spiral arms. In particular, we will show the effects on the azimuthal variations of abundance gradients for oxygen by varying:
i) the multiplicity $m$ of spiral arms;
ii) the spiral pattern speed, $\Omega_s$;
iii) the pitch angle $\alpha$.
For all model results that will be presented we assume the following Cox & Gomez (2002) prescriptions: the radial scale length of the drop-off in density amplitude of the arms fixed at the value of $R_S=7$ kpc, the pitch angle is assumed constant at $\alpha=\ang{15}$, and the surface arm density $\Sigma_0$ is 20 M$_{\odot}$ pc$^{-2}$ at the fiducial radius $R_0=8$ kpc; finally we assume $\phi_p(R_0)=0$.
The disk rotational velocity $\Omega_d(R)$ has been extracted from the simulation by Roca-F[à]{}brega et al. (2014) (see their left panel of Figure 1). The exponential fit of $\Omega_d(R)$ variations as a function of the Galactocentric distance $R$ (expressed in kpc) is: $$\Omega_d(R)=98.93 \, e^{-0.29 \, R}+ 11.11 \mbox{ [km } \mbox{s }
^{-1}\mbox{kpc}^{-1}].
\label{eqom}$$ We start by adopting the constant pattern angular velocity $\Omega_s$ = 20 km s$^{-1}$ kpc$^{-1}$ consistent with the Roca-F[à]{}brega et al. (2014) model. Similar value was first estimated from moving groups in the U-V plane by Quillen & Minchev (2005, 18.1 $\pm$ 0.8 km s$^{-1}$ kpc$^{-1}$) and a summary of derived values for the Milky Way can be found in Bland-Hawthorn & Gerhard (2016). In Fig. \[omega\] we show the $\Omega_s$ and $\Omega_d(R)$ quantities as well as the Outer and Inner Lindblad resonances as a function of the Galactocentric distance, the corotation radius is located at 8.31 kpc.
### Results with a single analytical spiral pattern
We begin our analysis discussing the results obtained with model S2A (see Table 1), which has a pattern speed of $\Omega_s$ = 20 km s$^{-1}$ kpc$^{-1}$, placing the corotation resonance at the solar radius.
The upper panel of Fig. \[SA\] shows the the oxygen abundance residual azimuthal variations after 11 Gyr of disc evolution for different Galactocentric distances. The average radial gradient is subtracted. As expected, larger abundance azimuthal variations are found near the corotation radius. In this region the chemical enrichment should be more efficient due to the lack of the relative gas-spiral motions. Higher SFR at the corotation radius caused by locally higher gas overdensity lasts for a longer time, therefore more massive stars can be created and more metals can be ejected into the local ISM under the spiral arm passage.
At 8 kpc we have $\Delta$\[O/H\] $\approx$ 0.05 dex. For other Galactocentric distances, away from the corotation, variations are much smaller. In the lower panel of Fig. \[SA\] we present the temporal evolution of the oxygen abundance azimuthal variations for the model S2A as a function of the azimuthal coordinate $\phi$ computed at 8 kpc. As expected, larger inhomogeneities are present at early times, decreasing in time.
As discussed in Section 2.2, we assume that during the Galactic evolution the ratio between the amplitude of the spiral density perturbation and the total surface density computed at the same radius $R$, remains constant in time. However, this analytical approach is not capable to put constraints on the temporal evolution of pattern speed.
Galactic chemical evolution is an integral process in time. The stronger spiral structure induced azimuthal variations at early times are, therefore, washed out by phase mixing.
Fig. \[SASFR\] depicts the SFR after 1 Gyr of evolution (upper panel) and at the present time (lower panel) on the galactic plane computed with the model S2A. Here, it is evident the way in which the spiral arm density perturbation affects and modulates SFR computed at the present time in unit of M$_{\odot}$ pc$^{-2}$ Gyr$^{-1}$. The shape of the two spiral arm over-densities is clearly visible in the SFR. This is in contrast to our results using the MCM13 density fluctuations (see Fig. \[mSFR\]), where multiple spiral density waves were present. Moreover, we can appreciate the inside-out disk formation: at later times the external regions become star formation active.
### The effect of different pattern speeds
In this Section we vary the spiral pattern speed, which has the effect of shifting the corotation resonance in radius. We argue that a combination of multiple spiral modes with different pattern speeds can be a realistic representation of a galactic disk. The horizontal and vertical lines in Fig. \[DOS\] show the different pattern speeds and corresponding corotation radii, respectively, used in this Section: it is clear that smaller $\Omega_{s}$ values lead to a more external corotation radius.
In Fig. \[S2\_grad\] we show the oxygen abundance gradients computed at different azimuths after 11 Gyr of disk evolution for models with spiral multiplicity $m$ = 2 and different spiral pattern speed $\Omega_{s}$ (see Table 1 for model details).
We notice that the more the corotation radius is shifted towards the external Galactic regions the more the oxygen azimuthal abundance variations are amplified near the corotation radius. This result is reasonable in the light of our previous findings presented above with our model assuming chemo-dynamical fluctuations by MCM13. We recall that larger variations in the chemical abundance of outer galactic regions have been found by observations in external galaxies (S[á]{}nchez et al. 2015).
In Fig. \[CS2\] we show the present day azimuthal residual of the oxygen abundances after subtracting the average radial gradient computed for the Galactic annular regions which include the relative corotation radius for the following models with $m=2$ multiplicity: S2A, S2B, S2C, S2D, S2F (see Table 1 for other parameter details). The model S2D computed at $R=$13 kpc has $\Delta$\[O/H\]$\approx$ 0.32 dex. Already in regions not so far from the solar neighbourd, the variations are important, i.e., model S2C whose corotation resides in the annular region centered at $R=$ 11 kpc, presents an oxygen abundance variation of $\Delta$\[O/H\] is $\approx$ 0.20 dex.
As discussed in Setion 2.3, it is well accepted that multiple patterns can be present in galactic disks (e.g., Meidt et al. 2009) including our own Milky Way (Minchev & Quillen 2006, Quillen et al. 2011), with slower patterns shifted to outer radii. This will have the effect of placing the corotation regions very similarly to what Fig. \[CS2\] presents and having corotating arms at all radii as found by Grand et al. (2012), Hunt et al. (2019). Therefore, the increasing scatter in abundance with galactic radius can be explained as the effect of multiple patterns propagating at the same time. Note that radial migration will introduce additional scatter, that can in principle be accounted for.
![ [*Upper Panel*]{}: The Galactic disc SFR related to the model S2G computed after 11 Gyr of Galactic evolution (see Table 1 and text for model details) with a pitch angle $\alpha=7^{\circ} $. The color code indicates the SFR in units of M$_{\odot}$ pc$^{-2}$ Gyr$^{-1}$. [*Lower Panel*]{}: as the upper panel but for the model S2H where the pitch angle $\alpha$ is 30$^{\circ}$.[]{data-label="SESFSFR"}](R1_SFR_11.png "fig:") ![ [*Upper Panel*]{}: The Galactic disc SFR related to the model S2G computed after 11 Gyr of Galactic evolution (see Table 1 and text for model details) with a pitch angle $\alpha=7^{\circ} $. The color code indicates the SFR in units of M$_{\odot}$ pc$^{-2}$ Gyr$^{-1}$. [*Lower Panel*]{}: as the upper panel but for the model S2H where the pitch angle $\alpha$ is 30$^{\circ}$.[]{data-label="SESFSFR"}](R2_SFR_11.png "fig:")
![ Effects of different pitch angles $\alpha$ on the azimuthal distribution of the residual of the oxygen abundances computed with our chemical evolution model at 4, 8, 12, 16, and 20 kpc In the upper panel the pitch able is set at the value of 7$^{\circ}$ (SE model in Table 1), while in the lower panel $\alpha=30 ^{\circ}$ (SF model in Table 1). []{data-label="SE_SF"}](R1.png "fig:") ![ Effects of different pitch angles $\alpha$ on the azimuthal distribution of the residual of the oxygen abundances computed with our chemical evolution model at 4, 8, 12, 16, and 20 kpc In the upper panel the pitch able is set at the value of 7$^{\circ}$ (SE model in Table 1), while in the lower panel $\alpha=30 ^{\circ}$ (SF model in Table 1). []{data-label="SE_SF"}](R2.png "fig:")
### Results with an $m$=1 spiral pattern
We want to test whether the intensity of the amplitude of the azimuthal chemical abundance variations is dependant on the number $m$ of spiral arms. In Table 1 we label as model S1A a model identical to the model S2A but with an $m$=1 spiral structure, i.e., having only one spiral arm. Such a mode arises naturally from the coupling of $m$=2 and $m$=3 modes as found by Quillen et al. (2011) and Minchev et al. (2012a) using pure N-body and SPH simulations, and is seen in external galaxies (Zaritsky & Rix 1997).
In the upper panel of Fig. \[S1A\] we notice that the abundance variations are larger than the ones obtained with the same model but $m$=2 (upper panel of Fig. \[SA\]): a fluctuation of about $\Delta$\[O/H\]=0.1 dex is seen at the corotation radius ($\sim$ 8 kpc).
In the same Figure is presented the time evolution of azimuthal abundance inhomogeneities for oxygen computed at 8 kpc with the model S1A at 2, 4, 6, 8, and 11 Gyr.
In Fig. \[S1\_grad\] we have the oxygen abundance gradients computed at different azimuths after 11 Gyr of disk evolution for models with spiral multiplicity $m$ = 1 and the same spiral pattern speeds $\Omega_{s}$ as in Fig. \[S2\_grad\] (see Table 1 for model details). We notice that around the corotation radii the azimuthal abundance variations are generally more evident for models with one spiral arm compared to ones with spiral multiplicity $m$ = 2.
In Fig. \[CS1\] we show the present day azimuthal residual of the oxygen abundances after subtracting the average radial gradient computed in annular regions which contain the corotation radii for models with $m=1$ multiplicity: S1A, S1B, S1C, S1D, S1F (see Table 1 for other parameter details). For the model S1D at the Galactic distance of 13 kpc we have $\Delta$\[O/H\]$\approx$ 0.40 dex, which is about $\approx 25$% larger than the S2D case. As found for the model with $m=2$, the oxygen abundance variations become important in regions not so far from the solar vicinity, i.e., model S1C whose corotation resides at $R=$ 11 kpc, $\Delta$\[O/H\] $\approx$ 0.23 dex.
### Results for different pitch angles
In this Section we consider different pitch angles $\alpha$ for the spiral arms in our Milky Way galaxy.
Recent work by Quillen et al. (2018) and Laporte et al. (2018) suggest that tightly wound spiral structure should be considered, based on modeling of phase-space structure found in the second Gaia data release (Gaia collaboration et al. 2018).
A smaller pitch angle gives rise to more tightly wound spiral structure. The upper panel of Fig. \[SESFSFR\] depicts the present time SFR computed with a pitch angle $\alpha=7^{\circ}$ (model S2G in Table 1), whereas the lower panel shows the case of $\alpha=30^{\circ}$ (model S2H in Table 1). For both panels the other model parameters as the same as model S2A. The spiral pattern is clearly visible in the SFR, and for the model S2G a tighter wound spiral structure is present.
In Fig \[SE\_SF\] we compare the azimuthal variations for models S2G and S2H. We see that the chemical variations are identical at the corotation radius and simply azimuthally shifted for other Galactocentric distances.
Conclusions
===========
In this paper we presented a new 2D chemical evolution model, able to trace azimuthal variations in the galactic disc density. We applied this model to (i) the density fluctuations arising in a disc formation simulation by Martig et al. (2012), used for the MCM13 Milky Way chemo-dynamical model, and (ii) the density perturbations originating from an analytical spiral arm formulation.
The main conclusions for density perturbation from Milky Way chemo-dynamical model by MCM13 can be summarized as follows:
- We found that the density fluctuations produce significant oxygen azimuthal variations in the abundance gradients of the order of 0.1 dex.
- The azimuthal variations are more evident in the external galactic regions, in agreement with the recent observations of the galaxy NGC 6754, using MUSE data (S[á]{}nchez et al. 2015).
In an effort to understand the above findings, we constructed simple analytical spiral arm models, for which we varied the pattern speed, multiplicity and pitch angle with the following main findings:
- The larger fluctuations in the azimuthal abundance gradients are found near the corotation radius, where the relative velocity with respect to the disk is close to zero.
- Larger azimuthal variations are found at corotation radii shifted to larger radii, i.e., slower pattern speeds.
- The variation is more enhanced for the model with only one spiral arm, which is expected to result from the combination of an $m$=2 and $m$=3 spiral structure.
- We found that the more significant azimuthal abundance variations seen at early times in presence of a regular, periodic perturbation tend to quench at later times. This is expected, as galactic chemical evolution is cumulative process and phase-mixing and radial migration tends to wipe structure with time.
Combining the effect of corotaton radii by assuming the simultaneous propagation of multiple spiral modes through galactic disks, we can obtain a realistic picture of azimuthal variations induced at stellar birth found in self-consistent models, such as the MCM13. Material spiral arms propagating near the corotation at all galactic radii have been described by a number of recent numerical work with different interpretations (see Grand et al. 2012, Comparetta & Quillen 2012, Hunt et al. 2019).
In future work we will improve the new 2D chemical evolution model introduced here by taking into account stellar radial migration of long-lived stars and the pollution to the ISM abundance introduced by them at radii and azimuths different than their birth places. We will also use this model to update the Galactic habitable zone results presented by Spitoni et al. (2014, 2017) and study the effect of spiral structure and the Galactic bar.
Acknowledgement {#acknowledgement .unnumbered}
===============
We thank the anonymous referee for various suggestions that improved the paper. E. Spitoni and V. Silva Aguirre acknowledge support from the Independent Research Fund Denmark (Research grant 7027-00096B). V. Silva Aguirre acknowledges support from VILLUM FONDEN (Research Grant 10118). G. Cescutti acknowledges financial support from the European Union Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No. 664931. This work has been partially supported by the EU COST Action CA16117 (ChETEC). I. Minchev acknowledges support by the Deutsche Forschungsgemeinschaft under the grant MI 2009/1-1. F. Matteucci acknowledges research funds from the University of Trieste (FRA2016).
[99]{} Asplund, M., Grevesse, N., Sauval, A. J., & Scott, P. 2009, ARA&A, 47, 481
Balser, D. S., Wenger, T. V., Anderson, L. D., Bania, T. M., 2015, ApJ, 806, 199
Bertin, G., Lin, C. C., Lowe, S. A., Thurstans, R. P., 1989, ApJ, 338, 78
Bland-Hawthorn, J., & Gerhard, O. 2016, ARAA, 54, 529 Cescutti, G., Chiappini, C., 2010, A&A, 515, A102
Cescutti, G., Fran[ç]{}ois, P., Matteucci, F., Cayrel, R., Spite, M., 2006, A&A, 448, 557
Cescutti G., Matteucci F., Fran[ç]{}ois, P., Chiappini C., 2007, A&A, 462, 943
Chiappini, C., Matteucci, F., Romano, D., 2001, ApJ, 554, 1044 Comparetta, J., & Quillen, A. C., 2012, arXiv:1207.5753 Cox, D. P., G[ó]{}mez, G. C. 2002, ApJS, 142, 261
Davies, B., Origlia, L., Kudritzki, R.-P., et al., 2009, ApJ, 696, 2014
Dehnen, W., Binney, J., 1998, MNRAS, 294, 429
Dib, S., Piau, L., Mohanty, S., & Braine, J. 2011, MNRAS, 415, 3439 Elmegreen, B. G., Elmegreen, D. M., & Montenegro, L. 1992, ApJs, 79, 37
Faure, C., Siebert, A., Famaey, B., 2014, MNRAS, 440, 2564 Fran[ç]{}ois P., Matteucci F., Cayrel R., et al. 2004, A&A, 421, 613
Gaia Collaboration, Katz, D., Antoja, T., et al. 2018, A&A, 616, A11
Genovali, K., Lemasle, B., Bono, G., et al. 2014, A&A, 566, A37 Genovali, K., Lemasle, B., da Silva, R., et al., 2015, A&A, 580, A17 Gerhard, O., 2011, Memorie della Societa Astronomica Italiana Supplementi, 18, 185
Grand, R. J. J., Kawata, D., Cropper, M. 2012, MNRAS, 426, 167 Grand, R. J. J., Kawata, D., Cropper, M. 2014, MNRAS, 439, 623 Grand, R. J. J., Springel, V., Kawata, D., et al., 2016, MNRAS, 460, L94 Grisoni, V., Spitoni, E., Matteucci, F., 2018, MNRAS, 481, 2570 Ho, I.-T., Seibert, M., Meidt, S. E., et al., 2017, ApJ, 846, 39
Hunt, J. A. S., Bub, M. W., Bovy, J., et al., 2019, arXiv:1904.10968 Iwamoto, K., Brachwitz, F., Nomoto, K., et al. 1999, ApJ Suppl. Ser., 125, 439
Laporte, C. F. P., Minchev, I., Johnston, K. V., & G[ó]{}mez, F. A. 2018, arXiv:1808.00451 Li, Y., Bresolin, F., Kennicutt, R. C., Jr., 2013, ApJ, 766, 17 Lin, C. C., & Shu, F. H., 1966, Proceedings of the National Academy of Science, 55, 229 Kennicutt, R. C., Jr. 1998, ApJ, 498, 541 Khoperskov, S., Di Matteo, P., Haywood, M., Combes, F., 2018, A&A, 611, L2
Martig, M., Bournaud, F., Croton, D. J., Dekel, A., Teyssier, R., 2012, ApJ, 756, 26
Martig, M., Minchev, I., Ness, M., Fouesneau, M., & Rix, H.-W. 2016, ApJ, 831, 139 Masset, F., & Tagger, M. 1997, A&A, 322, 442
Matteucci, F., Fran[ç]{}ois, P., 1989, MNRAS, 239, 885 Meidt, S. E., Rand, R. J., Merrifield, M. R. 2009, , 702, 277
Meynet, G., Maeder, A., 2002, A&A, 390, 561
Minchev, I., Anders, F., Recio-Blanco, A., et al., 2018, MNRAS, 481, 1645 Minchev, I., Chiappini, C., & Martig, M. 2013, A&A, 558, A9
Minchev, I., Chiappini, C., & Martig, M. 2014, A&A, 572, A92
Minchev, I., Famaey, B., Quillen, A. C., et al. 2012a, A&A, 548, A126
Minchev, I., Famaey, B., Quillen, A. C., et al. 2012b, A&A, 548, A127
Minchev, I., Martig, M., Streich, D., et al., 2015, ApJL, 804, L9 Mott, A., Spitoni E., Matteucci F., 2013, MNRAS, 435, 2918
Pedicelli, S., Bono, G., Lemasle, B., et al., 2009, A&A, 504, 81
Pettitt, A. R., Tasker, E. J., & Wadsley, J. W., 2016, MNRAS, 458, 3990 Quillen, A. C., Dougherty, J., Bagley, M. B., Minchev, I., & Comparetta, J. 2011, MNRAS, 417, 762
Quillen, A. C., Carrillo, I., Anders, F., et al. 2018, MNRAS, 480, 3132 Quillen, A. C., & Minchev, I. 2005, AJ, 130, 576 Ragan, S. E., Moore, T. J. T., Eden, D. J., et al., 2018, MNRAS, 479, 2361
Renaud, F., Bournaud, F., Emsellem, E., et al. 2015, MNRAS, 454, 3299 Rix, H.-W., & Zaritsky, D. 1995, ApJ, 447, 82
Roberts, W. W., 1969, ApJ, 158, 123 Roca-F[à]{}brega, S., Antoja, T., Figueras, F., et al., 2014, MNRAS, 440, 1950 Romano D., Karakas A. I., Tosi M., Matteucci F., 2010, A&A, 522, A32 Romano, D., Matteucci, F., Salucci, P., Chiappini, C., 2000, ApJ, 539, 235 S[á]{}nchez, S. F., Galbany, L., P[é]{}rez, E., et al. 2015, A&A, 573, A105
S[á]{}nchez-Menguiano, L., S[á]{}nchez, S. F., Kawata, D., et al., 2016, ApJL, 830, L40
Sellwood, J. A., & Carlberg, R. G. 2014, ApJ, 785, 137 Schmidt, M. 1959, ApJ, 129, 243
Siebert, A., Famaey, B., Binney, J., et al., 2012, MNRAS, 425, 2335 Silva Aguirre, V., Bojsen-Hansen, M., Slumstrup, D., et al. 2018, MNRAS, 475, 5487
Spitoni, E., Gioannini, L., & Matteucci, F. 2017, A&A, 605, A38
Spitoni E., Matteucci F., 2011, A&A, 531, A72 Spitoni, E., Matteucci, F., J[ö]{}nsson, H., Ryde, N., Romano, D. 2018, A&A, 612, A16 Spitoni, E., Matteucci, F., Sozzetti, A., 2014, MNRAS, 440, 2588
Spitoni, E., Recchi, S., Matteucci, F. 2008, A&A, 484, 743 Spitoni, E., Romano, D., Matteucci, F., Ciotti, L., 2015, ApJ, 802, 129 Spitoni, E., Silva Aguirre, V., Matteucci, F., Calura, F., & Grisoni, V., 2019, A&A, 623, A60
van den Hoek, L. B., Groenewegen, M. A. T. 1997, A&AS., 123, 305 Vincenzo, F., Spitoni, E., Calura, F., et al., 2019, arXiv:1903.03465 Vogt, F. P. A., P[é]{}rez, E., Dopita, M. A., Verdes-Montenegro, L., Borthakur, S., 2017, A&A, 601, A61
Woosley, S. E., & Weaver, T. A. 1995, ApJ, 101, 181
Zaritsky, D., & Rix, H.-W., 1997, ApJ, 477, 118
[^1]: email to: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We prove necessary and sufficient conditions for the validity of Whitney’s extension theorem in the ultradifferentiable Roumieu setting with controlled loss of regularity.'
address:
- 'A. Rainer: Fakultät für Mathematik, Universität Wien, Oskar-Morgenstern-Platz 1, A-1090 Wien, Austria'
- 'G. Schindl: Departamento de Álgebra, Análisis Matemático, Geometría y Topología, Facultad de Ciencias, Universidad de Valladolid, Paseo de Belén 7, 47011 Valladolid, Spain'
author:
- Armin Rainer
- Gerhard Schindl
date: 'September 4, 2017'
title: On the extension of Whitney ultrajets
---
[^1]
Introduction
============
Whitney’s extension theorem [@Whitney34a] provides conditions for the extension of jets defined in closed subsets of ${\mathbb{R}}^n$ to infinitely differentiable functions on ${\mathbb{R}}^n$. Its ultradifferentiable analogues ask for a precise determination how the growth rate of the jets is preserved by their extension. The growth rate of the jets, respectively of the derivatives of a smooth function, is measured by weight functions ${\omega}$. We denote by ${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n)$ the associated space of ultradifferentiable functions $f$ on ${\mathbb{R}}^n$; by definition, the growth rate of the sequence $(\|f^{({\alpha})}\|_{L^\infty({\mathbb{R}}^n)})_{{\alpha}\in {\mathbb{N}}^n}$ is regulated in terms of ${\omega}$. We use the letter ${\mathcal{B}}$ to emphasize that the bounds are global in ${\mathbb{R}}^n$. These classes of ultradifferentiable functions were introduced by Beurling [@Beurling61] and Björck [@Bjoerck66] and equivalently described by Braun, Meise, and Taylor [@BMT90]. Similarly, ${\mathcal{B}}^{\{{\omega}\}}(E)$ is the space of jets on the compact subset $E \subseteq {\mathbb{R}}^n$ with a growth rate regulated by ${\omega}$, so-called *ultrajets*. Precise definitions will be given in .
The weight functions ${\omega}$ which allow for an extension theorem preserving the class ${\mathcal{B}}^{\{{\omega}\}}$ have been fully characterized. We denote by $j_E^\infty$ the mapping which sends a smooth function to the infinite jet consisting of its partial derivatives of all orders restricted to $E$.
\[thm:preservingclass\] Let ${\omega}$ be a weight function. The following conditions are equivalent:
1. For every compact $E \subseteq {\mathbb{R}}^n$ the jet mapping $j^\infty_E : {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{{\omega}\}}(E)$ is surjective.
2. There is a compact $E \subseteq {\mathbb{R}}^n$ such that $j^\infty_E : {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{{\omega}\}}(E)$ is surjective.
3. ${\omega}$ is *strong*, i.e., $\int_{1}^\infty \frac{{\omega}(tu)}{u^2}\,du \le C{\omega}(t) + C$ for all $t>0$ and some $C>0$.
Note that a strong weight function is necessarily non-quasianalytic. is due to Bonet, Braun, Meise, and Taylor [@BBMT91] and Abanin [@Abanin01] (the latter showed the equivalence with (2)). Partial results have been contributed in earlier papers, e.g. Meise and Taylor [@MeiseTaylor88], Bonet, Meise, and Taylor [@BonetMeiseTaylor89a]. We want to mention that the statement remains true if the Roumieu type classes ${\mathcal{B}}^{\{{\omega}\}}$ are replaced by the Beurling type classes ${\mathcal{B}}^{({\omega})}$, but we shall only be concerned with the Roumieu case in this paper.
The purpose of this paper is to study the extension problem for weight functions ${\omega}$ which are not strong. In that case the extension involves a loss of regularity: the class is not preserved. So we are led to the following problem.
\[Q\] Let ${\omega}$ be a non-quasianalytic weight function. Let ${\sigma}$ be another weight function. Under which conditions is the jet mapping $j^\infty_E$ defined on ${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n)$ surjective onto ${\mathcal{B}}^{\{{\sigma}\}}(E)$ for all compact $E \subseteq {\mathbb{R}}^n$?
A complete answer has been given for the one-point set $E=\{0\}$, by Bonet, Meise, and Taylor [@BonetMeiseTaylor92], and for compact convex sets $E$, by Langenbruch [@Langenbruch94]. In these cases the mapping $j^\infty_{E} : {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{{\sigma}\}}(E)$ is surjective if and only if $$\label{intro:mixed}
{\;\exists}C>0 {\;\forall}t > 0 : \int_{1}^\infty \frac{{\omega}(tu)}{u^2}\,du \le C{\sigma}(t) + C.$$ So this condition is necessary for our problem.
We answer (for all compact $E \subseteq {\mathbb{R}}^n$) under three additional conditions. The first condition is that ${\omega}$ is concave. This has technical reasons, but it is not incongruous, since every strong weight function is equivalent to a concave one; cf. [@MeiseTaylor88 Proposition 1.3]. Secondly, we require that ${\sigma}(t) = o(t)$ as $t \to \infty$; again any strong weight function has this property.
To explain the third condition let us recall that any weight function ${\sigma}$ is associated with a family of weight sequences ${\mathfrak{S}}= \{S^x\}_{x>0}$ such that for the corresponding ultradifferentiable spaces we have $$\label{intro:rep}
{\mathcal{B}}^{\{{\sigma}\}}({\mathbb{R}}^n) = {\operatorname}{ind}_{x>0} {\mathcal{B}}^{\{S^x\}}({\mathbb{R}}^n)
\quad (\text{and } {\mathcal{B}}^{({\sigma})}({\mathbb{R}}^n) = {\operatorname}{proj}_{x>0} {\mathcal{B}}^{(S^x)}({\mathbb{R}}^n)).$$ The condition we require is that $$\label{intro:good}
{\;\forall}x >0 {\;\exists}y>0 {\;\exists}C \ge 1 {\;\forall}1\le j \le k : \frac{S^{x}_j}{jS^x_{j-1}} \le C\, \frac{S^{y}_k}{k S^y_{k-1}}.$$ The following is our main result.
\[main\] Let ${\omega}$ be a non-quasianalytic concave weight function. Let ${\sigma}$ be a weight function satisfying ${\sigma}(t) = o(t)$ as $t \to \infty$ and . Then the following conditions are equivalent:
1. For every compact $E \subseteq {\mathbb{R}}^n$ the jet mapping $j^\infty_E : {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{{\sigma}\}}(E)$ is surjective.
2. There is $C>0$ such that $\int_{1}^\infty \frac{{\omega}(tu)}{u^2}\,du \le C{\sigma}(t) + C$ for all $t>0$.
The implication (1) $\Rightarrow$ (2) follows from the aforementioned result of [@BonetMeiseTaylor92] and does not require the three additional conditions on ${\omega}$ and ${\sigma}$. We discuss the condition and its relation to other properties of the weight function in and . Let us emphasize that, while (1) and (2) in are invariant under equivalence of weight functions (two weight functions are equivalent if and only if they generate the same class), concavity and are not invariant. Thus, for the validity of is is enough that the assumptions on ${\omega}$ and ${\sigma}$ are satisfied up to equivalence of weight functions.
The problem put forward in has been solved for Denjoy–Carleman classes by Chaumat and Chollet [@ChaumatChollet94], where the growth rate of the derivatives is controlled by weight sequences $M$. Indeed, under suitable conditions on the weight sequences, [@ChaumatChollet94] proved that the jet mapping $j^\infty_E : {\mathcal{B}}^{\{N\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{M\}}(E)$ is surjective, for every compact $E \subseteq {\mathbb{R}}^n$, if and only if $${\;\exists}C>0 {\;\forall}k \in {\mathbb{N}}: \sum_{\ell \ge k} \frac{N_{\ell -1}}{N_\ell} \le C \, \frac{k M_{k-1}}{M_k}.$$ The case that the extension preserves the class (i.e., $M=N$) is due to Bruna [@Bruna80] (see also [@Lambert79]). We will see that our is a generalization of this result (under an additional assumption on $N$). In general, a class ${\mathcal{B}}^{\{{\omega}\}}$ cannot be represented as a class ${\mathcal{B}}^{\{M\}}$ for a weight sequence $M$, and vice versa, cf. Bonet, Meise, and Melikhov[@BMM07] and Rainer and Schindl [@RainerSchindl12].
The approach of [@ChaumatChollet94] was the starting point of our recent paper [@RainerSchindl16a] in which we obtained a generalization of their extension result for *admissible* unions of Denjoy–Carleman classes. By virtue of , we deduced a version of which however required an restrictive undesired condition on the involved weight functions.
In the present paper we surmount this problem by using the special cut-off functions which were constructed in [@BBMT91]. They are tailor-made for weight functions ${\omega}$; we actually need a modified version for two weight functions ${\omega}$ and ${\sigma}$ related by . Then we combine the resulting partition of unity $\{{\varphi}_i\}$ subordinate to a collection of Whitney cubes $Q_i$ with center $x_i$ with the technique of [@ChaumatChollet94] which is based on a extension method of Dynkin [@Dynkin80]. The extension of a ultrajet $F$ of class ${\mathcal{B}}^{\{{\sigma}\}}$ is defined as a linear combination $$\sum_i {\varphi}_i\, T_{\hat x_i}^{p(x_i)} F$$ of Taylor polynomials, where the degree $p(x_i)$ depends on the distance of $x_i$ to $E$ and $\hat x_i \in E$ realizes this distance. More precisely, the dependence of $p$ is through counting functions corresponding to the sequences in ${\mathfrak{S}}$, the family associated with ${\sigma}$. It is this part of the proof which necessitates the assumption .
The paper is structured as follows. We introduce weight functions, weight sequences, and the corresponding spaces of ultradifferentiable functions and jets in . A deeper analysis of the weights, their associated functions, and properties needed in the proof of the extension theorem follows in . We recall the construction of special cut-off functions due to [@BBMT91] in ; since we need a slight generalization for two weight functions ${\omega}$ and ${\sigma}$ satisfying , we indicate the required modifications in the proof. The main theorem \[main\] and its corollaries are proved in .
Spaces of ultradifferentiable functions and jets {#sec:spaces}
================================================
Weight functions {#weightfunction}
----------------
By a *weight function* we mean a continuous increasing function ${\omega}: [0,\infty) \to [0,\infty)$ with ${\omega}(0) =0$ and $\lim_{t \to \infty} {\omega}(t) = \infty$ that satisfies $$\begin{aligned}
& {\omega}(2t) = O({\omega}(t)) \quad\text{ as } t \to \infty, \label{om1}\\
& {\omega}(t) = O(t) \quad\text{ as } t \to \infty, \label{om2}\\
& \log t = o({\omega}(t)) \quad\text{ as } t \to \infty, \label{om3}\\
& {\varphi}(t) := {\omega}(e^t) \text{ is convex}. \label{om4}\end{aligned}$$ A weight function is called *non-quasianalytic* if $$\int_0^\infty \frac{{\omega}(t)}{1+t^2} \, dt <\infty.$$ Two weight functions ${\omega}$ and ${\sigma}$ are said to be *equivalent* if ${\omega}(t) = O({\sigma}(t))$ and ${\sigma}(t) = O({\omega}(t))$ as $t \to \infty$. For each weight function ${\omega}$ there is an equivalent weight function $\tilde {\omega}$ such that ${\omega}(t) = \tilde {\omega}(t)$ for large $t>0$ and $\tilde {\omega}|_{[0,1]} =0$. It is thus no restriction to assume that ${\omega}|_{[0,1]} =0$ when necessary.
The *Young conjugate* ${\varphi}^*$ of ${\varphi}$ is defined by $${\varphi}^*(t) := \sup_{s\ge 0} \big(st-{\varphi}(s)\big), \quad t \ge 0.$$ Assuming ${\omega}|_{[0,1]} =0$, we have that ${\varphi}^*$ is a convex increasing function satisfying ${\varphi}^*(0)=0$, $t/{\varphi}^*(t) \to 0$ as $t \to \infty$, and ${\varphi}^{**}={\varphi}$; cf. [@BMT90] and [@BBMT91 Remark 1.2].
The space of ultradifferentiable functions
------------------------------------------
Let ${\omega}$ be a weight function and ${\rho}>0$. We consider the Banach space ${\mathcal{B}}^{{\omega}}_{\rho}({\mathbb{R}}^n) := \{f \in C^\infty ({\mathbb{R}}^n) : \|f\|^{\omega}_{{\rho}}< \infty\}$, where $$\|f\|^{\omega}_{{\rho}} := \sup_{x \in {\mathbb{R}}^n,\,{\alpha}\in {\mathbb{N}}^n} |{\partial}^{\alpha}f(x)| \exp(-\tfrac{1}{{\rho}} {\varphi}^*({\rho}|{\alpha}|)),$$ and the inductive limit $${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) := {\operatorname}{ind}_{{\rho}\in {\mathbb{N}}} {\mathcal{B}}^{{\omega}}_{\rho}({\mathbb{R}}^n).$$ For weight functions ${\omega}$ and ${\sigma}$ we have ${\mathcal{B}}^{\{{\omega}\}} \subseteq {\mathcal{B}}^{\{{\sigma}\}}$ if and only if ${\sigma}(t) = O({\omega}(t))$ as $t \to \infty$, cf. [@RainerSchindl12 Corollary 5.17]; in particular, ${\omega}$ and ${\sigma}$ are equivalent if and only if ${\mathcal{B}}^{\{{\omega}\}} = {\mathcal{B}}^{\{{\sigma}\}}$. The space ${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n)$ contains non-trivial functions with compact support if and only if ${\omega}$ is non-quasianalytic (cf. [@BMT90] or [@RainerSchindl12]).
Weight sequences {#weights}
----------------
Let $\mu = (\mu_k)$ be a positive increasing sequence, $1 = \mu_0 \le \mu_1 \le \mu_2 \le \cdots$. We associate the sequences $M=(M_k)$ and $m = (m_k)$ defined by $$\label{def:M}
\mu_0 \mu_1 \mu_2 \cdots \mu_k = M_k = k!\, m_k, $$ for all $k \in {\mathbb{N}}$. We call $M$ a *weight sequence* if $M_k^{1/k} \to \infty$. A weight sequence $M$ is called *non-quasianalytic* if $$\sum_{k} \frac{1}{\mu_k} < \infty.$$ We say that $M$ has *moderate growth* if there exists $C>0$ such that $M_{j+k} \le C^{j+k} M_j M_k$ for all $j,k \in {\mathbb{N}}$, or equivalently, $$\mu_k \lesssim M_k^{1/k};$$ we refer to [@RainerSchindl16a Lemma 2.2] for a proof and more equivalent conditions. (For real valued functions $f$ and $g$ we write $f \lesssim g$ if $f \le C g$ for some positive constant $C$.)
Two weight sequences $M$ and $N$ are said to be *equivalent* if there is a constant $C>0$ such that $1/C \le M_k^{1/k}/N_k^{1/k} \le C$ for all $k$.
\(1) Some authors (e.g. [@ChaumatChollet94], [@RainerSchindl12]) prefer to work with *“sequences without factorials”*, that is $m_k$ instead of $M_k$.
\(2) Note that $\mu$ uniquely determines $M$ and $m$, and vice versa. In analogy we shall use $\nu \leftrightarrow N \leftrightarrow n$, ${\sigma}\leftrightarrow S \leftrightarrow s$, etc. That $\mu$ is increasing means precisely that $M$ is logarithmically convex (*log-convex* for short). Log-convexity of $m$ is a stronger condition: if $m$ is log-convex we shall say that $M$ is *strongly log-convex*.
\[lem:basicM\] Let $1 = \mu_0 \le \mu_1 \le \mu_2 \le \cdots$. Then:
1. $M_k^{1/k}$ is increasing, equivalently, $$\label{mucompare}
{\;\forall}k\in {\mathbb{N}}_{>0} : M_k^{1/k} \le \mu_k.$$
2. $M_jM_k\le M_{j+k}$ for all $k,j$.
3. If $M_k^{1/k} \to \infty$, then $\mu_k \to \infty$.
This is straightforward to check.
The space of ultradifferentiable functions
------------------------------------------
Let $M=(M_k)$ be a weight sequence and ${\rho}>0$. We consider the Banach space ${\mathcal{B}}^{M}_{\rho}({\mathbb{R}}^n) := \{f \in C^\infty ({\mathbb{R}}^n) : \|f\|^M_{{\rho}}< \infty\}$, where $$\|f\|^M_{{\rho}} := \sup_{x \in {\mathbb{R}}^n,\, {\alpha}\in {\mathbb{N}}^n} \frac{|{\partial}^{\alpha}f(x)|}{{\rho}^{|{\alpha}|} M_{|{\alpha}|}},$$ and the inductive limit $${\mathcal{B}}^{\{M\}}({\mathbb{R}}^n) := {\operatorname}{ind}_{{\rho}\in {\mathbb{N}}} {\mathcal{B}}^{M}_{\rho}({\mathbb{R}}^n).$$ Traditionally, ${\mathcal{B}}^{\{M\}}({\mathbb{R}}^n)$ is called a *Denjoy–Carleman class*. For weight sequences $M$ and $N$ we have ${\mathcal{B}}^{\{M\}} \subseteq {\mathcal{B}}^{\{N\}}$ if and only if $M_k^{1/k} \lesssim N_k^{1/k}$; one implication is obvious, the other follows from the existence of *characteristic* ${\mathcal{B}}^{\{M\}}$-functions, cf. [@RainerSchindl12 Lemma 2.9 and Proposition 2.12]. In particular, $M$ and $N$ are equivalent if and only if the corresponding classes coincide. By the Denjoy–Carleman theorem (e.g. [@Hoermander83I Theorem 1.3.8]), ${\mathcal{B}}^{\{M\}}({\mathbb{R}}^n)$ contains non-trivial elements with compact support if and only if $M$ is non-quasianalytic.
The connection between and
---------------------------
With any weight function ${\omega}$ we can associate a family of weight sequences $\{W^x\}_{x>0}$ such that ${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n)$ can be described us the union of the spaces ${\mathcal{B}}^{\{W^x\}}({\mathbb{R}}^n)$; see below.
With a weight function ${\omega}$ we associate a *weight matrix* ${\mathfrak{W}}= \{W^x\}_{x>0}$ by setting $$W^x_k := \exp(\tfrac{1}{x}{\varphi}^*(x k)), \quad k \in {\mathbb{N}};$$ cf. [@RainerSchindl12 5.5]. Moreover, we define $${\vartheta}^x_k := \frac{W^x_k}{W^x_{k-1}}. $$
\[lemma4\] We have:
1. Each $W^x$ is a weight sequence (in the sense of ).
2. ${\vartheta}^x \le {\vartheta}^y$ if $x \le y$, which entails $W^x \le W^y$. \[lemma4(2)\]
3. For all $x>0$ and all $j,k \in {\mathbb{N}}$, $W^x_{j+k} \le W^{2x}_{j} W^{2x}_{k}$ and $w^x_{j+k} \le w^{2x}_{j} w^{2x}_{k}$. \[eq:mW\]
4. For all $x>0$ and all $k \in {\mathbb{N}}_{\ge 2}$, ${\vartheta}^x_{2k} \le {\vartheta}^{4x}_{k}$.
5. ${\;\forall}{\rho}>0 {\;\exists}H\ge 1 {\;\forall}x >0 {\;\exists}C \ge 1 {\;\forall}k \in {\mathbb{N}}: {\rho}^k W^x_k \le C W^{Hx}_k$. \[5.10\]
6. If ${\omega}(t) = o(t)$ as $t \to \infty$ then $(w^x_k)^{1/k} \to \infty$ and ${\vartheta}^{x}_k/k \to \infty$ for all $x>0$.
(1)–(3) These are direct consequences of the properties of ${\varphi}^*$; cf. [@RainerSchindl12 5.5].
\(4) [@RainerSchindl16a Lemma 2.6].
\(5) [@RainerSchindl12 Lemma 5.9].
\(6) By [@RainerSchindl12 Corollary 5.15], we have $(w^x_k)^{1/k} \to \infty$. That also ${\vartheta}^x_k/k \to \infty$ follows from .
\[representation\] Let ${\omega}$ be a weight function and let ${\mathfrak{W}}= \{W^x\}_{x>0}$ be the associated weight matrix. Then, as locally convex spaces, $$\begin{aligned}
\label{eq:repR}
{\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) &= {\operatorname}{ind}_{x > 0} {\mathcal{B}}^{\{W^x\}}({\mathbb{R}}^n) = {\operatorname}{ind}_{x > 0} {\operatorname}{ind}_{{\rho}>0} {\mathcal{B}}^{W^x}_{\rho}({\mathbb{R}}^n).
\end{aligned}$$ We have ${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) = {\mathcal{B}}^{\{W^x\}}({\mathbb{R}}^n)$ for all $x>0$ if and only if $$\begin{aligned}
{\;\exists}H\ge 1 {\;\forall}t\ge 0 :
2{\omega}(t) \le {\omega}(Ht) + H. \label{om6}
\end{aligned}$$ Moreover, holds if and only if some (equivalently each) $W^x$ has moderate growth. It is no restriction to let the inductive limits in range only over $x, {\rho}\in {\mathbb{N}}$.
Let us emphasize that the fact that ${\mathcal{B}}^{\{{\omega}\}} = {\mathcal{B}}^{\{M\}}$ for some weight sequence $M$ if and only if ${\omega}$ satisfies is due to [@BMM07].
Whitney ultrajets
-----------------
Let $E$ be a compact subset of ${\mathbb{R}}^n$. We denote by ${\mathcal{J}}^\infty(E)$ the vector space of all jets $F= (F^{\alpha})_{{\alpha}\in {\mathbb{N}}^n} \in C^0(E,{\mathbb{R}})^{{\mathbb{N}}^n}$ on $E$. For $a \in E$ and $p \in {\mathbb{N}}$ we associate the Taylor polynomial $$\begin{gathered}
T^p_a : {\mathcal{J}}^\infty(E) \to C^\infty ({\mathbb{R}}^n,{\mathbb{R}}), ~ F \mapsto T^p_a F(x) := \sum_{|{\alpha}|\le p} \frac{(x-a)^{\alpha}}{{\alpha}!} F^{\alpha}(a), \end{gathered}$$ and the remainder $R^p_a F = ((R^p_a F)^{\alpha})_{|{\alpha}| \le p}$ with $$\begin{gathered}
(R^p_a F)^{\alpha}(x) := F^{\alpha}(x) - \sum_{|{\beta}| \le p-|{\alpha}|} \frac{(x-a)^{\beta}}{{\beta}!} F^{{\alpha}+{\beta}}(a), \quad a,x \in E. \end{gathered}$$ Let us denote by $j^\infty_E$ the mapping which assigns to a $C^\infty$-function $f$ on ${\mathbb{R}}^n$ the jet $j^\infty_E(f) := ({\partial}^{\alpha}f|_E)_{\alpha}$. By Taylor’s formula, $F =j^\infty_E(f)$ satisfies $$\label{Whitneyjets}
(R^p_a F)^{\alpha}(x) = o(|x-a|^{p-|{\alpha}|}) \quad \text{ for $a,x \in E$, $p\in {\mathbb{N}}$, $|{\alpha}| \le p$ as $|x-a|\to 0$.}$$ Conversely, if a jet $F \in {\mathcal{J}}^\infty(E)$ has this property, then it admits a $C^\infty$-extension to ${\mathbb{R}}^n$, by Whitney’s extension theorem [@Whitney34a] (for modern accounts see e.g. [@Malgrange67 Ch. 1], [@Tougeron72 IV.3], or [@Hoermander83I Theorem 2.3.6]).
\[def:ultrajets\] Let $E \subseteq {\mathbb{R}}^n$ be compact. Let $M=(M_k)$ be a weight sequence. For fixed ${\rho}>0$ we denote by ${\mathcal{B}}^{M}_{\rho}(E)$ the set of all jets $F$ such that there exists $C>0$ with $$\begin{gathered}
|F^{\alpha}(a)| \le C {\rho}^{|{\alpha}|} \, M_{|{\alpha}|}, \quad {\alpha}\in {\mathbb{N}}^n,~ a \in E,
\\
|(R^p_a F)^{\alpha}(b)| \le C {\rho}^{p+1} \, M_{p+1}\, \frac{|b-a|^{p+1-|{\alpha}|}}{(p+1-|{\alpha}|)!}, \quad p \in {\mathbb{N}},\, |{\alpha}| \le p,~ a,b \in E. \end{gathered}$$ The smallest constant $C$ defines a complete norm on ${\mathcal{B}}^{M}_{\rho}(E)$. We define $${\mathcal{B}}^{\{M\}}(E) := {\operatorname}{ind}_{{\rho}\in {\mathbb{N}}} {\mathcal{B}}^{M}_{\rho}(E).$$ An element of ${\mathcal{B}}^{\{M\}}(E)$ is called a *Whitney ultrajet of class ${\mathcal{B}}^{\{M\}}$ on $E$*.
Let ${\omega}$ be a weight function and ${\mathfrak{W}}= \{W^x\}_{x>0}$ the associated weight matrix. A jet $F$ is said to be a *Whitney ultrajet of class ${\mathcal{B}}^{\{{\omega}\}}$ on $E$* if $F \in {\mathcal{B}}^{\{W^x\}}(E)$ for some $x>0$; we set $${\mathcal{B}}^{\{{\omega}\}}(E)= {\mathcal{B}}^{\{{\mathfrak{W}}\}}(E)= {\operatorname}{ind}_{x >0} {\mathcal{B}}^{\{W^x\}}(E) = {\operatorname}{ind}_{x>0} {\operatorname}{ind}_{{\rho}>0} {\mathcal{B}}^{W^x}_{\rho}(E).$$
This definition of Whitney ultrajet of class ${\mathcal{B}}^{\{{\omega}\}}$ on $E$ coincides with the one given in [@BBMT91]. This follows from .
Notation for sequences
----------------------
The table summarizes our notation for sequences appearing in the paper. The three columns are mutually determined by the rule $$\mu_0 \mu_1 \mu_2 \cdots \mu_k = M_k = k!\, m_k$$ for $k \in {\mathbb{N}}$. (There will be no confusion by the fact that ${\sigma}$ usually denotes a weight function.)
---------- ---------- -----------------
$M$ $m$ $\mu$
$N$ $n$ $\nu$
$L$ $\ell$ ${\lambda}$
$W^x$ $w^x$ ${\vartheta}^x$
$S$ $s$ ${\sigma}$
$\dot S$ $\dot s$ $\dot {\sigma}$
---------- ---------- -----------------
More on weight functions and weight sequences {#sec:moreonweights}
=============================================
Functions associated with weight sequences {#hGaSi}
------------------------------------------
There are a few functions which one naturally associates with a weight sequence; cf. [@Mandelbrojt52], [@Komatsu73], [@ChaumatChollet94]. They will play an essential role in the proof of the extension theorem \[main\].
Let $m =(m_k)$ be a positive sequence satisfying $m_0 = 1$ and $m_k^{1/k} \to \infty$ (not necessarily log-convex). We associate the following functions $$\begin{aligned}
\label{h}
h_m(t) &:= \inf_{k \in {\mathbb{N}}} m_k t^k, \quad t > 0, \quad h_m(0):=0, \\
\label{counting2}
{\overline}{\Gamma}_m(t) &:= \min\{k : h_m(t) = m_k t^k\}, \quad t > 0, \\
\intertext{and, provided that $m_{k+1}/m_{k} \to \infty$,}
{\underline}{\Gamma}_m (t) &:= \min\Big\{k : \frac{m_{k+1}}{m_k} \ge \frac{1}{t} \Big\}, \quad t > 0.\end{aligned}$$
\[basic\] Let $m =(m_k)$ be a positive sequence satisfying $m_0 = 1$, $m_k^{1/k} \to \infty$, and $m_{k+1}/m_k \to \infty$. Then:
1. $h_m$ is increasing, continuous, and positive for $t>0$. For large $t$ we have $h_m(t) = 1$.
2. ${\underline}{\Gamma}_m$ is decreasing and ${\underline}{\Gamma}_m(t) \to \infty$ as $t\to 0$.
3. $k \mapsto m_k t^k$ is decreasing for $k \le {\underline}{\Gamma}_m(t)$. \[eq:ulGa3\]
4. $m_{k+1}/m_k \le n_{k+1}/n_k$ for all $k$ implies ${\underline}{\Gamma}_n \le {\underline}{\Gamma}_m$.
5. ${\underline}{\Gamma}_m
\le {\overline}{\Gamma}_m$. If $m$ is log-convex then ${\underline}{\Gamma}_m = {\overline}{\Gamma}_m$.
These facts are well-known and immediate from the definitions; we refer to [@Mandelbrojt52], [@Komatsu73], and [@ChaumatChollet94].
Let $M$ be a weight sequence satisfying $m_k^{1/k} \to \infty$. Then $m_{k}/m_{k-1} = \mu_k/k \to \infty$, in fact, we have $(k!\, m_k)^{1/k} = M_k^{1/k} \le \mu_k$, by .
So for such $M$ the functions $h_m$, ${\underline}{\Gamma}_m$, ${\overline}{\Gamma}_m$ are well-defined and enjoy the properties listed in . The sequence $m$ *will not be log-convex* in general, whence ${\underline}{\Gamma}_m$ and ${\overline}{\Gamma}_m$ fall apart. We need them both. It will crucial to be able to compare them, which is the content of the following lemma. Of course, we pay the price that we must switch from $m$ to another sequence $n$.
\[lem:m1\] Let $M$, $N$ be weight sequences satisfying $m_k^{1/k} \to \infty$ and $n_k^{1/k} \to \infty$. Assume that there exists $C\ge 1$ such that $\mu_j/j \le C \nu_k/k$ for all $j \le k$. Then, for all $t >0$, $$\label{eq:compare}
{\overline}{\Gamma}_n (Ct) \le {\underline}{\Gamma}_m(t).$$
Let $t >0$. If $k > {\underline}{\Gamma}_m(t)$, then $$\begin{aligned}
\frac{n_k}{n_{{\underline}{\Gamma}_m(t)}} = \frac{\nu_k}{k} \cdots \frac{\nu_{{\underline}{\Gamma}_m(t)+1}}{{\underline}{\Gamma}_m(t)+1}
\ge \Big(\frac{\mu_{{\underline}{\Gamma}_m(t)+1}}{C ({\underline}{\Gamma}_m(t)+1) }\Big)^{k - {\underline}{\Gamma}_m(t)} \ge (Ct)^{-k + {\underline}{\Gamma}_m(t)}
\end{aligned}$$ and thus $n_k (Ct)^k \ge n_{{\underline}{\Gamma}_m(t)} (Ct)^{{\underline}{\Gamma}_m(t)}$. It follows that ${\overline}{\Gamma}_n (Ct) \le {\underline}{\Gamma}_m(t)$.
We also need the following property.
\[lem:m2\] Let $M$, $N$, $L$ be weight sequences satisfying $m_k^{1/k} \to \infty$, $n_k^{1/k} \to \infty$, and $\ell_k^{1/k} \to \infty$. Assume that $$\label{eq:m1}
\mu_{2k} \lesssim \nu_k$$ and $$\label{eq:m2}
{\;\exists}C \ge 1 {\;\forall}1 \le j \le k : \frac{\nu_j}{j} \le C \frac{{\lambda}_k}{k}.$$ Then $$\label{eq:m3}
{\;\exists}D \ge 1 {\;\forall}t >0 : 2 {\underline}{\Gamma}_\ell(Dt) \le {\underline}{\Gamma}_m(t).$$
We first claim that and imply $$\label{eq:m4}
{\;\exists}C \ge 1 {\;\forall}1 \le h \le 2k : \frac{\mu_h}{h} \le C \frac{{\lambda}_k}{k}.$$ Note that is equivalent to $\frac{\mu_{2k}}{2k} \lesssim \frac{\nu_k}{k}$. Thus, if $h = 2j$ for $1 \le j \le k$, then $\frac{\mu_{h}}{h} = \frac{\mu_{2j}}{2j} \le C \frac{{\lambda}_k}{k}$. If $h$ is odd, then $\frac{\mu_h}{h} \le 2 \frac{\mu_{h+1}}{h+1} \le 2C \frac{{\lambda}_k}{k}$, since $\mu$ is increasing.
Now it is easy to see that implies .
Good weight functions
---------------------
Let us single out the weight functions whose associated weight matrix satisfies the conditions required in and .
\[def:good\] A weight function ${\omega}$ with associated weight matrix ${\mathfrak{W}}= \{W^x\}_{x >0}$ is called *good* if $$\label{eq:good}
{\;\forall}x >0 {\;\exists}y>0 {\;\exists}C \ge 1 {\;\forall}1\le j \le k : \frac{{\vartheta}^{x}_j}{j} \le C\, \frac{{\vartheta}^{y}_k}{k}.$$
\[rem:good\] (1) By , it is no restriction to assume $y \ge 2x$ in to the benefit that $w^x_{j+k} \le w^{y}_j w^{y}_k$ for all $j,k$, by .
\(2) We remark that is not invariant under equivalence of weight functions.
\(3) If $W^x$ is strongly log-convex, then is satisfied with $y = x$ and $C =1$.
\[cor:essence\] Let ${\omega}$ be a good weight function satisfying ${\omega}(t) = o(t)$ as $t \to \infty$. Let ${\mathfrak{W}}= \{W^x\}_{x>0}$ be the associated weight matrix. Then $$\begin{aligned}
\label{eq:essence1}
{\;\forall}x >0 {\;\exists}&y_3 \ge y_2 \ge y_1 \ge x {\;\exists}D\ge 1 {\;\forall}t>0 :
\notag \\
&
{\overline}{\Gamma}_{w^{y_3}}(D^3t)\le {\underline}{\Gamma}_{w^{y_2}}(D^2t)
\le {\overline}{\Gamma}_{w^{y_2}}(D^2t)
\le {\underline}{\Gamma}_{w^{y_1}}(Dt) \le \frac{{\underline}{\Gamma}_{{w^x}}(t)}{2}.
\end{aligned}$$ We may assume that $y_1 \ge 2x$ and $y_2 \ge 2y_1$ and hence $w^x_{j+k} \le w^{y_1}_j w^{y_1}_k$ and $w^{y_1}_{j+k} \le w^{y_2}_j w^{y_2}_k$ for all $j,k \in {\mathbb{N}}$.
The rightmost inequality in follows from , since ${\vartheta}^x_{2k} \le {\vartheta}^{4x}_{k}$ (for $k \ge 2$) by (4). The other inequalities are easy consequences of and (5). The supplement follows from (1).
The conjugate of a weight function
----------------------------------
The following conjugate will be important for the special partition of unity to be constructed in .
Let $\omega : [0,\infty) \to [0,\infty)$ satisfy ${\omega}(t) = o(t)$ as $t \to \infty$. We define $$\label{omegaconjugate}
\omega^{\star}(t):=\sup_{s\ge 0} \big(\omega(s)-st\big), \quad t>0.$$
Then $\omega^\star$ is decreasing, continuous, and convex with $\omega^{\star}(t) \to \infty$ as $t \to 0$, see [@PetzscheVogt84 Remark 1.5]. Since ${\omega}(t) = o(t)$ as $t \to \infty$, ${\omega}^\star(t)$ is finite for all $t$. If $\omega$ is concave and increasing, then, by [@PetzscheVogt84 Proposition 1.6], $$\label{omegaconjugate1}
\omega(t)=\inf_{s>0} \big(\omega^{\star}(s)+st\big), \quad t>0.$$
\[lem:omsistar\] Let ${\omega},{\sigma}: [0,\infty) \to [0,\infty)$ satisfy ${\omega}(t) = o(t)$ and ${\sigma}(t) = o(t)$ as $t \to \infty$. Suppose that ${\sigma}(t) = O({\omega}(t))$ as $t \to \infty$. Then $${\;\exists}C \ge 1 {\;\forall}t>0 : {\sigma}^\star (t) \le C {\omega}^\star(t/C) + C.$$
This is an easy computation.
The connection between and
---------------------------
With every positive sequence $M$ satisfying $M_0=1$ and $M_k^{1/k} \to \infty$ we associate a function ${\omega}_M$ by setting $${\omega}_M (t) = - \log h_M(1/t) = \sup_{k \in {\mathbb{N}}} \log \Big(\frac{t^k}{M_k}\Big), \quad t>0.$$ Then ${\omega}_M$ is increasing, convex in $\log t$, and zero for sufficiently small $t>0$. If $M$ is a weight sequence such that $\liminf_{k \to \infty} m_k^{1/k} >0$ and $\liminf_{k \to \infty} \mu_{Qk}/\mu_k >1$ for some $Q \in {\mathbb{N}}$, then ${\omega}_M$ is a weight function. See [@Komatsu73] and [@BMM07 Lemma 12]. The proof of the latter shows that ${\omega}_M(t) = o(t)$ as $t \to \infty$ provided that $m_k^{1/k} \to \infty$.
There is a connection between ${\omega}_M^\star$ and ${\omega}_m$. We found this in [@Debrouwere14 Lemma 5.7.8].
\[lem:star\] Let $M$ be a weight sequence such that $m_k^{1/k} \to \infty$. Then $$\label{star}
{\;\forall}t > 0 : {\omega}_M^\star(t) \le {\omega}_m\Big(\frac1t\Big) \le {\omega}_M^\star\Big(\frac {t}{e}\Big).$$
We have ${\omega}_M(t) = o(t)$ as $t \to \infty$ and so ${\omega}_M^\star$ is well-defined. For $s>0$, $$\begin{aligned}
\omega_M^{\star}(s):=\sup_{t\ge 0}\big(\omega_M(t)-st\big)
=\sup_{k\in{\mathbb{N}}}\sup_{t\ge 0}\Big(\log\Big(\frac{t^k}{M_k}\Big)-st\Big)=\sup_{k\in{\mathbb N}}\log\Big(\frac{k^k}{(es)^kM_k}\Big),\end{aligned}$$ by an easy calculation. Using $k! \le k^k \le e^k k!$ we find $$\omega_M^{\star}(s) \le \sup_{k\in{\mathbb{N}}}\log\Big(\frac{1}{s^km_k}\Big) = {\omega}_m\Big(\frac{1}{s}\Big)
\le \sup_{k\in{\mathbb N}}\log\Big(\frac{k^k}{s^kM_k}\Big) = {\omega}_M^\star\Big(\frac{s}{e}\Big)$$ as required.
Let ${\omega}$ be a weight function satisfying ${\omega}(t) = o(t)$ as $t \to \infty$. Let ${\mathfrak{W}}$ be the associated weight matrix. Then, for all $M \in {\mathfrak{W}}$ there exists $C\ge 1$ such that for all $t>0$ $$\label{eq:omMom1}
{\omega}^\star (t) \le C {\omega}_M^\star\Big(\frac{t}{C}\Big) + C
\quad \text{ and } \quad {\omega}_M^\star (t) \le C {\omega}^\star\Big(\frac{t}{C}\Big) + C$$ as well as $${\omega}^\star (t) \le C {\omega}_m\Big(\frac{C}{t}\Big) + C
\quad \text{ and } \quad
{\omega}_m(t) \le C {\omega}^* \Big(\frac{1}{eC t}\Big) + C.$$ In particular, $$\label{eq:key1}
\exp({\omega}^\star(t)) \le \Big(\frac{e}{h_m(t/C)}\Big)^C.$$
By [@RainerSchindl12 Lemma 5.7], for each $M \in {\mathfrak{W}}$, we have ${\omega}(t) = O({\omega}_M(t))$ and ${\omega}_M(t) = O({\omega}(t))$ as $t \to \infty$. So is a consequence of . The rest follows from .
In the proof of the following lemma log-convexity of the sequences was used.
\[lem:mg\] Let $M$ and $N$ be weight sequences such that $${\;\exists}C \ge 1 {\;\forall}k,j \in {\mathbb{N}}: M_{k+j} \le C^{k+j} N_j N_{k}. \label{mg0}$$ Then $h_M(t) \le h_{N}(Ct)^2$ for all $t>0$.
We need a corresponding version for the sequences $m,n$ which are not log-convex in general. This can be achieved by using the connection between ${\omega}_M^\star$ and ${\omega}_m$.
\[lem:hmodgrowth\] Let $M$ and $N$ be weight sequences satisfying and $m_k^{1/k} \to \infty$ and $n_k^{1/k} \to \infty$. Then there is a $D\ge 1$ such that $h_m(t) \le h_{n}(Dt)^2$ for all $t>0$.
By , $h_M(t) \le h_{N}(Ct)^2$ and hence $2\, {\omega}_{N}(t) \le {\omega}_M(Ct)$ for all $t>0$. Then $$\begin{aligned}
2\, {\omega}^\star_{N}(t) = \sup_{s \ge 0} \big( 2\,{\omega}_{N} (s) - 2ts\big)
\le \sup_{s \ge 0} \big( {\omega}_{M} (Cs) - 2ts\big) = {\omega}^\star_{M} \Big(\frac{2t}C\Big).
\end{aligned}$$ By , $$\begin{aligned}
2\, {\omega}_{n} \Big(\frac1t\Big) \le 2\,{\omega}_{N}^\star\Big(\frac te\Big)
\le {\omega}^\star_{M}\Big(\frac{2t}{eC}\Big) \le {\omega}_m\Big(\frac{eC}{2t}\Big).
\end{aligned}$$ This entails the statement.
The heirs of a weight function
------------------------------
We introduce notation for our convenience.
Let ${\omega}$ be a non-quasianalytic weight function. Then $$\label{eq:scion}
{\kappa}(t) = {\kappa}_{\omega}(t):=\int_1^{\infty}\frac{\omega(tu)}{u^2}du=t\int_t^{\infty}\frac{\omega(u)}{u^2}du, \quad t>0,$$ defines a weight function (possibly quasianalytic) satisfying ${\kappa}(t) = o(t)$ as $t \to \infty$; cf. [@BonetMeiseTaylor92 Remark 3.20]. Moreover, ${\kappa}$ is concave; see [@MeiseTaylor88 Proposition 1.3]. Since ${\omega}$ is increasing we have ${\kappa}\ge {\omega}$, which implies $K^x \le W^x$ for all $x>0$, where $\{K^x\}_{x>0}$ is the weight matrix associated with ${\kappa}$.
All weight functions ${\sigma}$ satisfying ${\sigma}(t) = o(t)$ and ${\kappa}(t) = O({\sigma}(t))$ as $t \to \infty$, i.e., $$\label{eq:heir}
{\;\exists}C >0 {\;\forall}t>0 : \int_1^{\infty}\frac{\omega(tu)}{u^2}du \le C {\sigma}(t) +C,$$ are called *heirs of the weight function ${\omega}$*. A *good heir of ${\omega}$* is a heir of ${\omega}$ which is a good weight function in the sense of . If ${\omega}$ itself is a heir of ${\omega}$, then ${\omega}$ is said to be a *strong* weight function.
In particular, ${\kappa}$ is a heir of ${\omega}$. By [@BonetMeiseTaylor92], the condition is necessary and sufficient for the surjectivity of $j^\infty_{\{0\}} : {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{{\sigma}\}}(\{0\})$. That a heir ${\sigma}$ satisfies ${\sigma}(t) = o(t)$ as $t \to \infty$ guarantees that we can work with the conjugate ${\sigma}^\star$.
\[lem:heir\] Let ${\omega}$ be a non-quasianalytic weight function and ${\sigma}$ a heir of ${\omega}$. Let ${\mathfrak{W}}= \{W^x\}_{x>0}$ and ${\mathfrak{S}}= \{S^x\}_{x>0}$ be the weight matrices associated with ${\omega}$ and ${\sigma}$, respectively. Then $${\;\exists}C\ge 1 {\;\forall}x >0 : S^{x} \le e^{1/x}\, W^{Cx}.$$
By , ${\omega}\le {\kappa}\le C {\sigma}+ C$ and hence ${\varphi}_{\omega}\le C {\varphi}_{\sigma}+C$. For the Young conjugates this means ${\varphi}_{\omega}^*(Ct) + C \ge C {\varphi}_{\sigma}^*(t)$ which entails the assertion.
Next we recall that can be equivalently stated with ${\omega}$ replaced by its harmonic extension. For a continuous function $u : {\mathbb{R}}\to {\mathbb{R}}$ with $\int_{\mathbb{R}}\frac{|u(t)|}{1+t^2}\, dt <\infty$, we define its *harmonic extension* $P_u : {\mathbb{C}}\to {\mathbb{R}}$ by $$P_u(x+iy) := \begin{cases}
\frac{|y|}{\pi} \int_{\mathbb{R}}\frac{u(t)}{(t-x)^2 + y^2}\, dt & \text{ if } y \ne 0,\\
u(x) & \text{ if } y =0.
\end{cases}$$ Then $P_u$ is continuous on ${\mathbb{C}}$ and harmonic in the open upper and lower half plane. If ${\omega}$ is a weight function, we extend ${\omega}$ to ${\mathbb{C}}$ by $z \mapsto {\omega}(|z|)$, and $P_{\omega}$ denotes the harmonic extension of $t \mapsto {\omega}(|t|)$. We have ${\omega}\le P_{\omega}$, cf. [@MeiseTaylor88 Remark 1.6].
\[lem:scionharmonicextension\] For a non-quasianalytic weight function ${\omega}$ and a weight function ${\sigma}$ the following conditions are equivalent:
1. ${\kappa}_{\omega}(t) = O({\sigma}(t))$ as $t \to \infty$.
2. $P_{\omega}(t) = O({\sigma}(t))$ as $t \to \infty$.
Concave and good weight functions {#sec:good}
---------------------------------
Let ${\omega}$ be a non-quasianalytic weight function. The weight function ${\kappa}= {\kappa}_{\omega}$ defined in is concave, and hence subadditive, since ${\kappa}(0)= 0$. Since ${\kappa}$ is the heir of ${\omega}$ which defines the largest function space among all heirs of ${\omega}$, it is of interest to find conditions which guarantee that ${\kappa}$ is a good heir of ${\omega}$.
Let us recall a result which relates concavity of a weight function with a condition on the associated weight matrix.
\[thm:subadditive\] Let ${\omega}$ be a weight function and let ${\mathfrak{W}}= \{W^x\}_{x >0}$ be the associated weight matrix. Then the following conditions are equivalent:
1. ${\omega}$ is equivalent to its least concave majorant.
2. $\exists C>0 {\;\exists}t_0 >0 {\;\forall}{\lambda}\ge 1 {\;\forall}t \ge t_0 : {\omega}({\lambda}t)\le C {\lambda}\, {\omega}(t)$.
3. $\forall x>0 {\;\exists}y>0 {\;\exists}D \ge 1 {\;\forall}1 \le j \le k : (w^x_j)^{1/j} \le D\, (w^y_k)^{1/k}$. \[eq:m6\]
The equivalence of the first two conditions can be found in [@PetzscheVogt84] and is based on [@Peetre70 Lemma 1]. The equivalence with the third condition was proved in [@RainerSchindl14] building on a result of [@FernandezGalbis06], by showing that the conditions are all equivalent to several stability properties of the corresponding spaces of ultradifferentiable functions.
\[thm:suffgood\] Let ${\omega}$ be a weight function. Assume that the associated weight matrix ${\mathfrak{W}}= \{W^x\}_{x >0}$ satisfies $$\label{eq:m5}
{\;\forall}x>0 {\;\exists}y>0 : {\vartheta}_k^x \lesssim (W^y_k)^{1/k}.$$ Then ${\omega}$ is a good weight function if and only if it is equivalent to its least concave majorant.
By and , for all $x>0$ there exists $y>0$ such that $(W^x_k)^{1/k} \le {\vartheta}_k^x \lesssim (W^y_k)^{1/k} \le {\vartheta}_k^y$ and consequently, $$(w^x_k)^{1/k} \lesssim \frac{{\vartheta}_k^x}{k} \lesssim (w^y_k)^{1/k} \lesssim \frac{{\vartheta}_k^y}{k}.$$ Then clearly the conditions and \[thm:subadditive\] are equivalent.
Note that is not invariant under equivalence of weight functions, in contrast to the three equivalent conditions in ; compare with (2).
Let ${\omega}$ be a non-quasianalytic weight function. Then ${\kappa}_{\omega}$, defined in , is a good heir of ${\omega}$ provided that its associated weight matrix satisfies .
This raises the following question.
\[q:concave\] Is every concave weight function equivalent to a good one?
A strong weight function ${\omega}$ is equivalent to the concave weight function ${\kappa}_{\omega}$. We will discuss the relation between strong and good weight functions in .
For the sake of completeness we remark that amounts to the following condition on the secants of ${\varphi}^*$: $${\;\forall}x>0 {\;\exists}y >0 {\;\exists}C>0 {\;\forall}k \in {\mathbb{N}}_{>0} :
\frac{{\varphi}^*(xk) - {\varphi}^*(xk-x)}{x} \le \frac{{\varphi}^*(yk)}{yk} + C.$$ A weight function ${\omega}$ is good if and only if $$\begin{aligned}
{\;\forall}x>0 {\;\exists}y >0 &{\;\exists}C>0 {\;\forall}1 \le j \le k :
\notag \\
\log k - \log j &\le \frac{{\varphi}^*(yk) - {\varphi}^*(yk-y)}{y} - \frac{{\varphi}^*(xj) - {\varphi}^*(xj-x)}{x} + C. \end{aligned}$$
A convenient partition of unity {#sec:partition}
===============================
In this section we construct a special partition of unity which will be a cornerstone for the extension theorem. The construction is based on a result of [@BBMT91].
Special bump functions
----------------------
The following proposition is due to [@BBMT91] in the case that ${\omega}$ is a strong concave weight function and ${\sigma}= {\omega}$. The proof of the general case (with ${\sigma}\ne {\omega}$) requires some slight modifications of the original proof of [@BBMT91]. We recall the main steps and detail the passages, where a transition from ${\omega}$ to ${\sigma}$ occurs.
In this section ${\mathfrak{W}}= \{W^x\}_{x>0}$ will always be the weight matrix associated with the weight function ${\omega}$.
\[proposition22BBMTWhitney\] Let ${\omega}$ be a non-quasianalytic concave weight function and let ${\sigma}$ be a heir of ${\omega}$. Then for each $n\in {\mathbb{N}}_{>0}$ there exist $m\in{\mathbb{N}}_{>0}$, $M>0$, and $0<r_0<1/2$ such that for all $0<r<r_0$ there are functions $f_{n,r}\in C^\infty({\mathbb R})$ satisfying the following properties: $$\begin{gathered}
\label{proposition22BBMTWhitneyequ}
0\le f_{n,r}\le 1,\quad
{{\operatorname}{supp}}f_{n,r} \subseteq \big[-\tfrac{9}{8}r,\tfrac{9}{8}r \big],\quad
f_{n,r}|_{[-r,r]} =1,
\\
\label{proposition22BBMTWhitneyequ1}
\sup_{x\in{\mathbb R}, \, j \in {\mathbb{N}}} \frac{|f^{(j)}_{n,r}(x)|}{W^m_j}\le M\exp\Big(\frac{1}{n}{\sigma}^{\star}(nr)\Big).\end{gathered}$$ The proof will show that $m=cn$ for some $c\in{\mathbb N}_{>0}$ independent of $n$.
Note that, in , ${\sigma}$ need not be a *good* heir of ${\omega}$.
The following two lemmas can be taken without modification from [@BBMT91]. They are based on Hörmander’s $L^2$-method to construct entire functions and on a Paley–Wiener theorem.
\[lemma23BBMTWhitney\] Let $\omega$ be a non-quasianalytic weight function. Then there exists $A>0$ such that for each $0<r\le 1$, each $k\in{\mathbb N}$, and each subharmonic function $u$ on ${\mathbb{C}}$ satisfying $$u(z)\le r|{\mathrm{Im}}(z)|-\frac{\omega(z)}{k} \quad \text{ for all } z \in {\mathbb{C}},$$ there exists an entire function $F$ on ${\mathbb{C}}$ with $F(0)=1$ and $$|F(z)|\le A\exp\Big(r|{\mathrm{Im}}(z)|-\frac{\omega(z)}{k}+3\log(1+|z|^2)\Big)\sup_{|w|\le 1}\exp(-u(w))$$ for all $z \in {\mathbb{C}}$.
\[lemma24BBMTWhitney\] Let $\omega$ be a non-quasianalytic weight function. There exists $L\in{\mathbb N}_{>0}$ such that for each $k\in{\mathbb N}_{>0}$ there exists $B>0$ such that for all $0<r<1/2$ the following holds. If there is an entire function $F$ with $F(0)=1$ such that $${\;\exists}M>0 {\;\forall}z\in{\mathbb{C}}: |F(z)|\le M\exp\Big(r|{\mathrm{Im}}(z)|-\frac{\omega(z)}{k}\Big),$$ then there exists $\psi \in {\mathcal{B}}^{\{\omega\}}({\mathbb R})$ with the following properties: $$\begin{gathered}
0\le\psi\le 1,\quad \psi(x)=0 \text{ for } x\le -r,\quad \psi(x)=1 \text{ for } x\ge r,
\\
\sup_{x\in{\mathbb R},\, j\in{\mathbb N}}\frac{|\psi^{(j)}(x)|}{W^{2Lk}_j}\le BM^2.\end{gathered}$$
Next we generalize [@BBMT91 Lemma 2.5]. For $T>1$ we define ${\omega}_T:{\mathbb R}\rightarrow[0,\infty)$ by $$\label{lemma25BBMTWhitneyequ}
{\omega}_T(t) := \begin{cases}
{\omega}(t) & \text{ if } |t| \ge T, \\
\frac{{\omega}'(T)}{2T}t^2-\frac{{\omega}'(T)}{2}T+{\omega}(T) & \text{ if } |t| \le T.
\end{cases}$$
\[lemma25BBMTWhitney\] Let $\omega$ be a non-quasianalytic concave weight function and let ${\sigma}$ be a heir of ${\omega}$. Then there is a $D>0$ such that for all $T > 1$, $$\label{lemma25BBMTWhitneyequ1}
\sup_{x\in{\mathbb{R}}}\frac{\partial}{\partial y}P_{{\omega}_T}(x+i)\le D\, \frac{{\sigma}(T)}{T}.$$
It suffices to follow the proof of [@BBMT91 Lemma 2.5] and replace the use of the estimate $\int_1^\infty \frac{{\omega}(tu)}{u^2}\, du \le C {\omega}(t) + C$ (that is [@BBMT91 1.7(1)]) by the estimate .
Let $\omega, {\omega}_T$ be as in and . We consider $h_T:{\mathbb{C}}\rightarrow{\mathbb R}$ given by $$h_T(z):= \begin{cases}
P_{{\omega}_T}(z+i) & \text{ if } {\mathrm{Im}}(z) \ge 0,\\
P_{{\omega}_T}(z-i) & \text{ if } {\mathrm{Im}}(z) < 0.
\end{cases}$$ If ${\omega}_T$ is replaced by ${\omega}$, then we will write $h$ for the corresponding function. By the symmetry of ${\omega}_T$ and of the Poisson kernel, $h_T$ is continuous on ${\mathbb{C}}$. We have $$\label{eq:hTh}
h_T(z) - {\omega}(T) \le h(z) \le h_T(z) \quad \text{ for all } z \in {\mathbb{C}}, ~T>1,$$ by [@BBMT91 2.7(2)]. The following generalizes [@BBMT91 Lemma 2.7].
\[lemma27BBMTWhitney\] Let $\omega$ be a non-quasianalytic concave weight function and let ${\sigma}$ be a heir of ${\omega}$. Then there exist $E,F,G>0$ such that for all $T>1$ and all $z \in {\mathbb{C}}$, $$\label{lemma27BBMTWhitneyequ}
E^{-1}h_T(z)-F{\omega}(T)\le{\sigma}(z),\quad {\omega}(z)\le h_T(z)+G.$$
The proof of [@BBMT91 Lemma 2.7] yields that there exists $G>0$ such that $$|P_{\omega}(z+w) - P_{\omega}(z)| \le G \quad \text{ for all } z,w \in {\mathbb{C}}, ~ |w|\le 1.$$ Together with this implies $${\omega}(z) \le P_{\omega}(z) \le P_{\omega}(z+i) + G = h(z) + G \le h_T(z) + G,$$ if ${\mathrm{Im}}(z) \ge 0$, and similarly for ${\mathrm{Im}}(z) <0$. This gives the second inequality in .
For the first inequality note that $P_{\omega}\le C {\sigma}+ C$, by . Then, by , $$h_T(z) - {\omega}(T) - G \le h(z) - G \le P_{\omega}(z) \le C {\sigma}(z) + C,$$ which easily implies the first inequality in .
Now we generalize [@BBMT91 Lemma 2.8].
\[lemma28BBMTWhitney\] Let $\omega$ be a non-quasianalytic concave weight function and let ${\sigma}$ be a heir of ${\omega}$. Then for each $n\in{\mathbb N}_{>0}$ there exist $m\in{\mathbb N}_{>0}$, $M>0$ and $0<r_0<1/2$ such that for all $0<r<r_0$ there are functions $g_{n,r}\in C^\infty({\mathbb{R}})$ satisfying the following properties: $$\begin{gathered}
\label{lemma28BBMTWhitneyequ}
0\le g_{n,r}\le 1,\quad g_{n,r}=0 \text{ for } x\le -r,\quad g_{n,r}(x)=1 \text{ for } x\ge r,
\\
\label{lemma28BBMTWhitneyequ1}
\sup_{x\in{\mathbb{R}},\, j\in{\mathbb N}}\frac{|g_{n,r}^{(j)}(x)|}{W^m_j}\le M\exp\Big(\frac{1}{n}{\sigma}^{\star}(n r)\Big).\end{gathered}$$
There is a constant $C$ such that ${\omega}\le C {\sigma}+C$. Let $A,L,D,E,F,G$ be the constants arising in , , , and . We can assume that $C,L,D,E,F$ are positive integers. For $n\in{\mathbb{N}}_{>0}$ let $$\label{eq:choiceofk}
k:=(2CEF+D)n \quad \text{ and } \quad m:=4Lk.$$ Choose $0<r_0<1/2$ such that the equation ${\sigma}(t)/t=r_0 k/D$ has a solution $t>1$. Fix $0<r<r_0$ and choose $T = T(k,r)>1$ such that $$\label{lemma28BBMTWhitneyequ2}
{\sigma}(T)=T\,\frac{rk}{D}.$$ Define $u_{n,r}:{\mathbb{C}}\rightarrow{\mathbb R}$ by $$u_{n,r}(z):=r |{\mathrm{Im}}(z)|-\frac{h_T(z)}{k}-\frac{G}{k}.$$ Then, by , for all $z \in {\mathbb{C}}$ $$\begin{aligned}
\label{lemma28BBMTWhitneyequ3}
u_{n,r}(z)&\le r|{\mathrm{Im}}(z)|-\frac{{\omega}(z)}{k}, \\
-u_{n,r}(z)&\le-r|{\mathrm{Im}}(z)|+\frac{E}{k}{\sigma}(z)+\frac{EF}{k}{\omega}(T)+\frac{G}{k}.\label{eq:BBMT2}\end{aligned}$$ By definition $u_{n,r}$ is subharmonic on the open upper and lower half plane. By and , we have $$-\frac{1}{k}\frac{\partial}{\partial y}h_T(x)
=-\frac{1}{k}\frac{\partial}{\partial y}P_{{\omega}_T}(x+i)\ge-\frac{D}{k}\frac{{\sigma}(T)}{T}
=-r,$$ for all $x\in{\mathbb R}$. Thus, for each non-negative $g\in C^\infty_c({\mathbb{C}})$, $$\int_{{\mathbb{C}}}u_{n,r}(z)\Delta g(z)\, d\lambda(z)
=2\int_{-\infty}^{\infty}\Big(r-\frac{1}{k}\frac{\partial}{\partial y}h_T(x)\Big)g(x)\, dx\ge 0,$$ whence $u_{n,r}$ is subharmonic on ${\mathbb{C}}$. By and , there is an entire function $F_{n,r}$ with $F_{n,r}(0)=1$ and $$\label{lemma28BBMTWhitneyequ4}
|F_{n,r}(z)|\le A\exp\Big(r|{\mathrm{Im}}(z)|-\frac{{\omega}(z)}{k}+3\log(1+|z|^2)\Big)\sup_{|w|\le 1}\exp(-u_{n,r}(w))$$ for all $z \in {\mathbb{C}}$. By and since ${\omega}\le C {\sigma}+C$, there is a constant $K(n)>0$ such that $$\begin{aligned}
\sup_{|w|\le 1}\exp(-u_{n,r}(w)) &\le K(n) \exp\Big(\frac{CEF}{k}{\sigma}(T)\Big).\end{aligned}$$ Using $\log (t) = o({\omega}(t))$ as $t \to \infty$ (i.e., ), we find that (for a possibly larger constant $K(n)$) $$\begin{aligned}
|F_{n,r}(z)| &\le K(n)\exp\Big(r|{\mathrm{Im}}(z)|-\frac{{\omega}(z)}{2k}\Big)\exp\Big(\frac{CEF}{k}{\sigma}(T)\Big)\end{aligned}$$ for all $z \in {\mathbb{C}}$. By , there is a constant $B(n)>0$ and functions $g_{n,r}\in C^\infty({\mathbb{R}})$ satisfying and $$\sup_{x \in {\mathbb{R}},\,j\in{\mathbb N}}\frac{|g_{n,r}^{(j)}(x)|}{W^{4Lk}_j}\le B(n)K(n)^2\exp\Big(\frac{2CEF}{k}{\sigma}(T)\Big).$$ By the definition of ${\sigma}^{\star}$, , and the choice of $k$ (see ), $$\begin{aligned}
\frac{1}{n} {\sigma}^{\star}(n r) &\ge \frac{1}{n} \big({\sigma}(T)-n rT\big)
={\sigma}(T)\Big(\frac{1}{n}-\frac{D}{k}\Big)
={\sigma}(T)\frac{2CEF}{k} \end{aligned}$$ which implies . The proof is complete.
Follow the proof of [@BBMT91 Proposition 2.2 (p.168)] and use .
A special partition of unity
----------------------------
Let $E \subseteq {\mathbb{R}}^n$ be a compact set. We denote by $d(Q,E)$ the Euclidean distance of a closed set $Q \subseteq {\mathbb{R}}^n$ to $E$, in particular, $d(x,E) = \inf\{|x-y| : y \in E\}$.
\[Whitneycubes\] Let $E \subseteq {\mathbb{R}}^n$ be a non-empty compact set. There exists a collection of closed cubes $\{Q_i\}_{i \in {\mathbb{N}}}$ with sides parallel to the axes satisfying the following properties:
1. ${\mathbb{R}}^n \setminus E = \bigcup_{i \in {\mathbb{N}}} Q_i$.
2. The interiors of the $Q_i$ are pairwise disjoint.
3. ${\operatorname}{diam} Q_i \le d(Q_i,E) \le 4 {\operatorname}{diam} Q_i$ for all $i \in {\mathbb{N}}$.
4. Let $Q_i^*$ be the closed cube which has the same center as $Q_i$ expanded by the factor $9/8$. For each $i \in {\mathbb{N}}$ the number of cubes $Q_j^*$ which intersect $Q_i^*$ is bounded by $12^{2n}$.
5. There exist $b_1, B_1>0$ (independent of $E$) such that for all $i,j \in {\mathbb{N}}$ with $Q_i^* \cap Q_j^* \ne \emptyset$ we have $b_1 {\operatorname}{diam} Q_i \le {\operatorname}{diam} Q_j \le B_1 {\operatorname}{diam} Q_i$.
For every $x \in {\mathbb{R}}^n$ we denote by $\hat x$ any point in $E$ with $|x-\hat x| = d(x,E)$.
\[cor:Whitneycubes\] In the setting of , let $x_i$ be the center of $Q_i$. Then for all $x \in Q^*_i$, $$\begin{gathered}
\frac{1}{2} d(x,E) \le d(x_i,E) \le 3 d(x,E),
\\
\frac{1}{3} {\operatorname}{diam} Q_i \le d(x,E) \le 9 {\operatorname}{diam} Q_i,
\end{gathered}$$ $$|\hat x_i-x|\le 2 d(x_i,E), \quad |\hat x_i-\hat x| \le 4 d(x_i,E).$$
All this follows easily from $$|x_i-x| \le \frac{9}{8} \frac{{\operatorname}{diam} Q_i}{2} \le \frac{9}{16} d(Q_i,E) \le \frac{9}{16} d(x_i,E). \qedhere$$
In analogy with [@BBMT91 Lemma 3.7] we may conclude the following.
\[Proposition6\] Let $E \subseteq {\mathbb{R}}^n$ be a non-empty compact set and let $\{Q_i\}_{i \in {\mathbb{N}}}$ be the family of cubes provided by . Let ${\omega}$ be a non-quasianalytic concave weight function and let ${\sigma}$ be a heir of ${\omega}$. Then for all $p\in{\mathbb N}_{>0}$ there exist $m\in{\mathbb{N}}_{>0}$, $M>0$, $0 < r_0 < 1/2$, and a family of smooth functions $\{\varphi_{i,p}\}_{i\in {\mathbb{N}}}$ satisfying
1. $0\le\varphi_{i,p}\le 1$ for all $i\in{\mathbb N}$,
2. ${{\operatorname}{supp}}\varphi_{i,p}\subseteq Q_i^*$ for all $i\in{\mathbb N}$,
3. $\sum_{i\in{\mathbb{N}}}\varphi_{i,p}(x)=1$ for all $x\in{\mathbb{R}}^n\setminus E$,
4. if $d(Q_i,E) \le r_0/B_1$, then for all ${\beta}\in{\mathbb{N}}^n$ and $x\in{\mathbb{R}}^n\setminus E$, $$|\varphi^{({\beta})}_{i,p}(x)|\le M W^m_{|{\beta}|}
\exp\Big(\frac{A_1(n)}{p}{\sigma}^{\star}\Big(\frac{b_1 p}{A_2(n)} {\operatorname}{diam} Q_i\Big)\Big),$$ for constants $A_1(n)\le A_2(n)$ only depending on $n$.
Let $p$ be a positive integer. Let $f_{p,r}$, for $0< r<r_0=r_0(p)$, be the functions provided by . The function $$g_{p,r}(x) := f_{p,r}(x_1) \cdots f_{p,r}(x_n), \quad x = (x_1,\ldots,x_n) \in {\mathbb{R}}^n,$$ satisfies $0 \le g_{p,r} \le 1$, has support in the cube centered at $0$ with sidelength $9r/4$ and equals $1$ in the cube centered at $0$ with sidelength $2r$. There exist $m$, $M$ such that for all $r<r_0$, ${\beta}\in {\mathbb{N}}^n$, and $x \in {\mathbb{R}}^n$ $$\begin{aligned}
\label{eq:part1}
|g^{({\beta})}_{p,r}(x)|\le M W^m_{|{\beta}|}\exp\Big(\frac{n}{p}{\sigma}^{\star}(pr)\Big),
\end{aligned}$$ thanks to (2) and (5).
Let $2r_i$ denote the sidelength of $Q_i$ and $x_i$ its center. If $r_i < r_0$, or equivalently, ${\operatorname}{diam} Q_i < 2 \sqrt n r_0$, then we define $$\psi_{i,p}(x):=g_{p,r_i}(x-x_i).$$ Then $$\label{eq:partition}
0 \le {\psi}_{i,p} \le 1,\quad {{\operatorname}{supp}}{\psi}_{i,p} \subseteq Q_i^*,
\quad {\psi}_{i,p}|_{Q_i} = 1.$$ Moreover, by , $$\label{eq:part11}
|\psi^{({\beta})}_{i,p}(x)|
\le M W^m_{|{\beta}|}\exp\Big(\frac{n}{p}{\sigma}^{\star}\Big(\frac{p}{2 \sqrt n} {\operatorname}{diam} Q_i\Big)\Big).$$ For those $i$ with $r_i \ge r_0$, we just choose arbitrary $C^\infty$-functions ${\psi}_{i,p}$ satisfying .
Then put $$\varphi_{1,p}:=\psi_{1,p},\quad \varphi_{i,p}:=\psi_{i,p}\prod_{k=1}^{i-1}(1-\psi_{k,p}),~i\ge 2.$$ It is easy to check that (1)–(3) are satisfied (cf. [@Bruna80 Lemma 3.3]).
Assume that $d(Q_i,E) \le r_0/B_1$. Then (3)&(5) guarantees that the diameters of the cubes which correspond to nontrival factors in the product which defines ${\varphi}_{i,p}$ satisfy ${\operatorname}{diam} Q_k \le r_0 < 2 \sqrt n r_0$. So for those factors we have the estimate . There are at most $12^{2n}$ such factors. Consequently, by (2) and (5), we get $$|\varphi^{({\beta})}_{i,p}(x)|\le M^{12^{2n}}
W^m_{|{\beta}|}\exp\Big(\frac{n 12^{2n}}{p}{\sigma}^{\star}\Big(\frac{b_1 p}{2 \sqrt n} {\operatorname}{diam} Q_i\Big)\Big).$$ This implies (4), since ${\sigma}^\star$ is decreasing.
The extension theorem {#sec:extension}
=====================
In this section we prove the implication $(2) \Rightarrow (1)$ in . We subdivide the proof into three parts for two reasons:
1. The proof is (by nature) quite technical. We hope that the subdivison improves the clarity of the presentation.
2. The organization into parts should make it easier to see, where in the line of arguments the particular assumptions are needed. The first two parts and prepare the stage with preliminary lemmas and estimates. This is the place, where we use that the heir ${\sigma}$ of ${\omega}$ is *good*. The actual proof of the extension theorem is given in the third part, i.e., .
In we deduce a consequence for Denjoy–Carleman classes and compare it with the result of [@ChaumatChollet94]. Finally, in we discuss the relation of strong and good weight functions.
Preliminaries, I {#sec:prelimI}
----------------
Let $E \subseteq {\mathbb{R}}^n$ be a compact set. Let $S=(S_k)$ be a weight sequence satisfying $s_k^{1/k} \to \infty$ and let $F= (F^{\alpha})_{{\alpha}}$ be a Whitney ultrajet of class ${\mathcal{B}}^{\{S\}}$ on $E$, i.e., there exist $C>0$ and ${\rho}\ge 1$ such that $$\begin{gathered}
|F^{\alpha}(a)| \le C {\rho}^{|{\alpha}|} \, S_{|{\alpha}|}, \quad {\alpha}\in {\mathbb{N}}^n,~ a \in E,
\label{jets1}
\\
|(R^k_a F)^{\alpha}(b)| \le C {\rho}^{k+1} \, |{\alpha}|!\, s_{k+1}\, |b-a|^{k+1-|{\alpha}|}, \quad k \in {\mathbb{N}},\, |{\alpha}| \le k,~ a,b \in E.
\label{jets2}\end{gathered}$$ The extension of $F$ will be of the form $$\label{extensionformula}
\sum_{i\in {\mathbb{N}}} {\varphi}_{i,p}(x) \, T_{\hat x_i}^{2 {\overline}{\Gamma}_{s'}(L d(x_i)) } F(x), \quad x \in {\mathbb{R}}^n \setminus E,$$ where
- $\{{\varphi}_{i,p}\}_{i \in {\mathbb{N}}}$ is a partition of unity provided by ($x_i$ is the center of the cube $Q_i$),
- $S'$ is a suitable weight sequence and $L$ is a constant, both depending on $S$.
For simplicity of notation we use the abbreviation $d(x) := d(x,E)$. Recall that $\hat x$ denotes any point in $E$ with $d(x) = |x-\hat x|$.
We begin with several estimates for the Taylor polynomials appearing in .
\[proposition10\] For $a_1,a_2 \in E$, $x \in {\mathbb{R}}^n$ and $|{\alpha}| \le q$, $$|(T^q_{a_1}F-T^q_{a_2}F)^{({\alpha})}(x)|\le C (2n^2 {\rho})^{q+1}|{\alpha}|! \, s_{q+1}(|a_1-x|+|a_1-a_2|)^{q+1-|{\alpha}|}.$$
This is straightforward; for details see [@ChaumatChollet94 Proposition 10].
\[proposition9\] Let $S$, $S'$ be weight sequences satisfying $s_k^{1/k} \to \infty$, $(s'_k)^{1/k} \to \infty$, and $$\label{eq:mass1}
{\;\exists}{\lambda}\le 1 {\;\forall}t>0 : 2 {\overline}{\Gamma}_{s'}(t) \le {\underline}{\Gamma}_s({\lambda}t).$$ Then there is a constant $D_1 = D_1(S,S') >1$ such that, for all Whitney ultrajets $F= (F^{\alpha})_{{\alpha}}$ of class ${\mathcal{B}}^{\{S\}}$ that satisfy and , all $L \ge D_1 {\rho}$, all $x \in {\mathbb{R}}^n$, and ${\alpha}\in {\mathbb{N}}^n$, $$\begin{aligned}
|(T_{\hat x}^{2 {\overline}{\Gamma}_{s'} (L d(x))} F)^{({\alpha})}(x)| &\le C (2L)^{|{\alpha}|+1} S_{|{\alpha}|}, \label{prop91}
\intertext{and, if $|{\alpha}| < 2 {\overline}{\Gamma}_{s'}(L d(x))$,}
|(T_{\hat x}^{2 {\overline}{\Gamma}_{s'}(L d(x))}F)^{({\alpha})}(x)-F^{\alpha}(\hat x)| &\le C (2L)^{|{\alpha}|+1} |{\alpha}|!\, s_{|{\alpha}|+1} d(x).
\label{prop92}\end{aligned}$$
For we may restrict to the case $|{\alpha}| \le 2 {\overline}{\Gamma}_{s'}(L d(x))$. By , $$\begin{aligned}
|(T_{\hat x}^{2 {\overline}{\Gamma}_{s'}(Ld(x))}F)^{({\alpha})}(x)|
&\le\sum_{\substack{{\alpha}\le {\beta}\\ |{\beta}|\le2 {\overline}{\Gamma}_{s'}(Ld(x))}}
\frac{|x-\hat{x}|^{|{\beta}|-|{\alpha}|}}{({\beta}-{\alpha})!} C {\rho}^{|{\beta}|} S_{|{\beta}|}
\notag \\
&\le C |{\alpha}|! \sum_{\substack{{\alpha}\le {\beta}\\ |{\beta}|\le2 {\overline}{\Gamma}_{s'}(Ld(x))}}
\frac{|{\beta}|!\,(n d(x))^{|{\beta}|-|{\alpha}|}}{|{\alpha}|!\, (|{\beta}|-|{\alpha}|)!} {\rho}^{|{\beta}|} s_{|{\beta}|}
\notag \\
&\le \frac{C |{\alpha}|! }{(n d(x))^{|{\alpha}|}} \sum_{\substack{{\alpha}\le {\beta}\\ |{\beta}|\le2 {\overline}{\Gamma}_{s'}(Ld(x))}}
(2n {\rho}d(x))^{|{\beta}|} s_{|{\beta}|}
\notag \\
&\le \frac{C|{\alpha}|! }{(n d(x))^{|{\alpha}|}} \sum_{j = |{\alpha}|}^{2 {\overline}{\Gamma}_{s'}(Ld(x))}
(2n^2 {\rho}d(x))^{j} s_{j}, \label{calculation} \end{aligned}$$ since the number of ${\beta}\in {\mathbb{N}}^n$ with $|{\beta}|=j$ is bounded by $n^j$. By , we may let $j$ run from $|{\alpha}|$ to ${\underline}{\Gamma}_s(L{\lambda}d(x))$ in the sum on the right-hand side of . For such $j$ we have $(L {\lambda}d(x))^{j} s_{j}\le (L {\lambda}d(x))^{|{\alpha}|} s_{|{\alpha}|}$, by , and hence $$\begin{aligned}
|(T_{\hat x}^{2 {\overline}{\Gamma}_{s'}(Ld(x))}F)^{({\alpha})}(x)|
&\le C S_{|{\alpha}|} \Big(\frac{L {\lambda}}{n}\Big)^{|{\alpha}|} \sum_{j = |{\alpha}|}^{{\underline}{\Gamma}_s(L{\lambda}d(x))}
\Big(\frac{2n^2 {\rho}}{L {\lambda}} \Big)^{j}. \end{aligned}$$ We obtain if $L$ is chosen such that $2n^2{\rho}/(L{\lambda})\le 1/2$; then $D_1= 4n^2/{\lambda}$.
For note that, if $|{\alpha}| < 2 {\overline}{\Gamma}_{s'}(L d(x))$, then $$(T_{\hat x}^{2 {\overline}{\Gamma}_{s'}(L d(x))}F)^{({\alpha})}(x)-F^{\alpha}(\hat x)
= \sum_{\substack{{\alpha}\le {\beta}\\ |{\alpha}| < |{\beta}|\le2 {\overline}{\Gamma}_{s'}(Ld(x))}}\frac{(x-\hat{x})^{{\beta}-{\alpha}}}{({\beta}-{\alpha})!} F^{\beta}(\hat x).$$ Thus the same arguments yield .
Preliminaries, II {#sec:prelimII}
-----------------
Let ${\sigma}$ be a good weight function and let ${\mathfrak{S}}$ be the associated weight matrix. Let $E$ be a compact subset of ${\mathbb{R}}^n$ and $F \in {\mathcal{B}}^{\{{\sigma}\}}(E)$. There exist $S \in {\mathfrak{S}}$, $C>0$ and ${\rho}\ge 1$ such that and hold. By , there are $\dot S, \ddot S, \dddot S \in {\mathfrak{S}}$ satisfying $S \le \dot S \le \ddot S \le \dddot S$ such that: $$\begin{aligned}
\label{eq:ext1}
{\;\exists}D = D(S)\ge 1 &{\;\forall}t >0 :
\notag \\
{\overline}{\Gamma}_{\dddot s}(D^3t) &\le {\underline}{\Gamma}_{\ddot s}(D^2t)
\le {\overline}{\Gamma}_{\ddot s}(D^2t)
\le {\underline}{\Gamma}_{\dot s}(Dt) \le \frac{{\underline}{\Gamma}_{s}(t)}{2},
\intertext{as well as}
\label{eq:mge1}
&{\;\forall}j,k \in {\mathbb{N}}: s_{j+k} \le \dot s_j \dot s_k,
\\ \label{eq:mge2}
&{\;\forall}j,k \in {\mathbb{N}}: \dot s_{j+k} \le \ddot s_j \ddot s_k.
\end{aligned}$$ Let $\{Q_i\}_{i\in {\mathbb{N}}}$ be the family of cubes provided by and let $b_1,B_1$ be the constants from . Let $x_i$ be the center of $Q_i$.
\[lem:H1\] There is a constant $C_1 = C_1(S)>0$ such that for all $L>C_1 {\rho}$, all ${\beta}\in {\mathbb{N}}^n$, and all $x \in Q_i^*$ with $d(x)<1$, $$\begin{aligned}
\label{H1}
|{\partial}^{\beta}(T_{\hat x_i}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F) (x)|
&\le C
L^{|{\beta}|+1} \ddot S_{|{\beta}|} \, h_{\ddot s}(L d(x_i)).\end{aligned}$$
It suffices to consider $|{\beta}| \le 2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) =: q$. Let $H_1$ denote the left-hand side of . By , $$\begin{aligned}
H_1
\le C (2n^2 {\rho})^{q+1} |{\beta}|! \,
s_{q+1} (|\hat x_i-x|+|\hat x_i-\hat x|)^{q+1-|{\beta}|}.\end{aligned}$$ By and , $$s_{q+1} \le \dot s_1 \dot s_{q} = \dot s_1 \dot s_{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i))} \le
\dot s_1 \ddot s_{{\overline}{\Gamma}_{\ddot s}(L d(x_i))}^2.$$ Together with , we conclude that $$\begin{aligned}
H_1
&\le C \dot s_1 (2n^2 {\rho})^{q+1}
|{\beta}|! \, \ddot s_{{\overline}{\Gamma}_{\ddot s}(L d(x_i))}^2 (6 d(x_i))^{q+1-|{\beta}|}.\end{aligned}$$ By the definition of ${\overline}{\Gamma}_{\ddot s}$, $$h_{\ddot s}(L d(x_i)) = \ddot s_{{\overline}{\Gamma}_{\ddot s}(L d(x_i))} (L d(x_i))^{{\overline}{\Gamma}_{\ddot s}(L d(x_i))} \le
\ddot s_{|{\beta}|} (L d(x_i))^{|{\beta}|} \quad \text{ for all } {\beta}.$$ Since $d(x_i) \le 3\, d(x)$, by , we find $$\begin{aligned}
H_1
&\le 36 \, C \dot s_1 n^2 {\rho}\Big(\frac{12n^2 {\rho}}{L}\Big)^{q}
d(x)\, L^{|{\beta}|}\, |{\beta}|!\,\ddot s_{|{\beta}|} \, h_{\ddot s}(L d(x_i)).\end{aligned}$$ If $L > 36 n^2 \dot s_1 \, {\rho}$ and $d(x)<1$, then follows.
\[lem:H2\] There is a constant $C_2 = C_2(S)>0$ such that for all $L>C_2 {\rho}$, all ${\beta}\in {\mathbb{N}}^n$, and all $x \in Q_i^*$ with $d(x)<1$, $$\begin{aligned}
\label{H2}
|{\partial}^{\beta}(T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x)|
&\le C \Big(\frac{3L D }{n}\Big)^{|{\beta}|+1} \dddot S_{|{\beta}|} h_{\dddot s}(3 LD d(x)). \end{aligned}$$
Let $H_2$ denote the left-hand side of . Using , and the fact that ${\underline}{\Gamma}_s$ is decreasing, it is easy to see that both $2 {\overline}{\Gamma}_{\ddot s}(L d(x_i))$ and $2 {\overline}{\Gamma}_{\ddot s}(L d(x))$ are majorized by ${\underline}{\Gamma}_s(L {\lambda}d(x))$ for some ${\lambda}<1$. So the degree of the polynomial $T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F$ is at most ${\underline}{\Gamma}_s(L {\lambda}d(x))$. Similarly the valuation of the polynomial is at least $2 {\overline}{\Gamma}_{\dddot s}(3 L D d( x)) =: 2 q$, indeed, using that ${\underline}{\Gamma}_{\ddot s}$ is decreasing $$2{\overline}{\Gamma}_{\ddot s}(Ld(x_i)) \ge 2{\underline}{\Gamma}_{\ddot s}(Ld(x_i)) \ge 2{\underline}{\Gamma}_{\ddot s}(3L d(x))
\ge 2{\overline}{\Gamma}_{\dddot s}(3LDd(x))$$ and analogously for $2{\overline}{\Gamma}_{\ddot s}(Ld(x))$. Thus, by the calculation in , $$\begin{aligned}
H_2
&\le \frac{C |{\beta}|!}{(n d(x))^{|{\beta}|}} \sum_{j =2 q }^{{\underline}{\Gamma}_s(L {\lambda}d(x))}
(2n^2 {\rho}d(x))^{j} s_{j}. \end{aligned}$$ By , $$s_j (L {\lambda}d(x))^j \le s_{2 q} (L {\lambda}d(x))^{2 q} \quad \text{ for } 2q \le j \le {\underline}{\Gamma}_s(L {\lambda}d(x)).$$ By , $s_{2q} \le
\dot s_{q}^2 \le \dddot s_{q}^2$. By the definition of $q$, $$h_{\dddot s}(3LD d(x)) = \dddot s_{q} (3LD d(x))^{q} \le \dddot s_{|{\beta}|} (3LD d(x))^{|{\beta}|} \quad \text{ for all }
|{\beta}|.$$ All this leads to $$\begin{aligned}
H_2
&\le \frac{C |{\beta}|!}{(n d(x))^{|{\beta}|}} \sum_{j =2 q }^{{\underline}{\Gamma}_s(L {\lambda}d(x))}
\Big(\frac{2n^2 {\rho}}{L{\lambda}} \Big)^{j} s_{2 q} (L {\lambda}d(x))^{2 q}
\\
&\le \frac{C |{\beta}|!}{(n d(x))^{|{\beta}|}} \sum_{j =2 q }^{{\underline}{\Gamma}_s(L {\lambda}d(x))}
\Big(\frac{2n^2 {\rho}}{L{\lambda}} \Big)^{j} \dddot s_{q}^2 ( L {\lambda}d(x))^{2 q}
\\
&\le C \Big(\frac{3LD}{n}\Big)^{|{\beta}|} |{\beta}|!\, \dddot s_{|{\beta}|} h_{\dddot s}(3LD d(x))
\Big(\frac{{\lambda}}{3D}\Big)^{2 q}
\sum_{j =2 q }^{{\underline}{\Gamma}_s(L {\lambda}d(x))}
\Big(\frac{2n^2 {\rho}}{L{\lambda}} \Big)^{j}. \end{aligned}$$ If we choose $L \ge 4n^2 {\rho}/{\lambda}$, then the sum is bounded by $2$, and follows, as ${\lambda}<1$ and $D \ge 1$.
The extension theorem {#sec:extproof}
---------------------
\[extensiontheorem\] Let ${\omega}$ be a non-quasianalytic concave weight function and let ${\sigma}$ be a good heir of ${\omega}$. Let $E$ be a compact subset of ${\mathbb{R}}^n$. Then the jet mapping $j^\infty_E : {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{{\sigma}\}}(E)$ is surjective.
We assume that the setup of holds. Assume $$\label{eq:L}
L > \max\{C_1,C_2\} \, {\rho}$$ so that and are valid.
Let $p\in {\mathbb{N}}$ be fixed (and to be specified later). Let $\{{\varphi}_{i,p}\}_{i\in {\mathbb{N}}}$ be the family of functions provided by , relative to the family of cubes $\{Q_i\}_{i \in {\mathbb{N}}}$ from , and let $r_0 = r_0(p)$ be the constant appearing in this proposition. Recall that $x_i$ denotes the center of $Q_i$.
We will show that an extension of class ${\mathcal{B}}^{\{{\omega}\}}$ of $F$ to ${\mathbb{R}}^n$ is provided by $$f(x) :=
\begin{cases}
\sum_{i\in {\mathbb{N}}} {\varphi}_{i,p}(x) \, T_{\hat x_i}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F(x), & \text{ if } x \in {\mathbb{R}}^n \setminus E, \\
F^0(x), & \text{ if } x \in E.
\end{cases}$$ Clearly, $f$ is $C^\infty$ in ${\mathbb{R}}^n \setminus E$.
In the following ${\mathfrak{W}}$ denotes the weight matrix associated with ${\omega}$.
There exist constants $K_j=K_j(S)$, $j = 1,2,3$, such that the following holds. If $p = K_1 L$ and $L>K_2 {\rho}$, then there exist weight sequences $W \in {\mathfrak{W}}$, $\tilde S \in {\mathfrak{S}}$ and a constant $M_1= M_1(S,L)>0$ such that for all $x \in {\mathbb{R}}^n \setminus E$ with $d(x) < (3B_1)^{-1} r_0$ and all ${\alpha}\in {\mathbb{N}}^n$, $$\label{eqclaim2}
|{\partial}^{\alpha}(f - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x)| \le C M_1^{|{\alpha}|+1} W_{|{\alpha}|} h_{\tilde s}(LK_3 d(x)),$$ where $C$ and ${\rho}$ are the constants from and (and $B_1$ stems from ).
#### *Proof of the claim*
By the Leibniz rule, $$\begin{aligned}
{\partial}^{\alpha}& (f - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x) \notag
\\
&=
\sum_{{\beta}\le {\alpha}} \binom{{\alpha}}{{\beta}} \sum_i {\varphi}_{i,p}^{({\alpha}-{\beta})}(x) \,
{\partial}^{\beta}(T_{\hat x_i}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x). \label{Leibniz}\end{aligned}$$ Using , $\ddot s \le \dddot s$ which entails $h_{\ddot s} \le h_{\dddot s}$, and the fact that $h_{\dddot s}$ is increasing, we conclude from and , that for $x \in Q_i^*$ with $d(x)<1$, $$\begin{aligned}
|{\partial}^{\beta}(T_{\hat x_i}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x_i)) } F - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x)|
\le C
(6D L)^{|{\beta}|+1} \dddot S_{|{\beta}|} \, h_{\dddot s }(3LD d(x)).\end{aligned}$$
By , there exist $W = W(p) \in {\mathfrak{W}}$ and $M = M(p)>0$ such that, provided that $d(Q_i,E) \le r_0/B_1$, we have, for all ${\beta}\in {\mathbb{N}}^n$ and $x\in{\mathbb{R}}^n\setminus E$, $$\begin{aligned}
|\varphi^{({\beta})}_{i,p}(x)| &\le M W_{|{\beta}|}
\exp\Big(\frac{A_1(n)}{p}{\sigma}^{\star}\Big(\frac{b_1 p}{A_2(n)} {\operatorname}{diam} Q_i\Big)\Big)
\\
&\le M W_{|{\beta}|}
\exp\Big(\frac{A_1(n)}{p}{\sigma}^{\star}\Big(\frac{b_1 p}{9 A_2(n)} d(x)\Big)\Big)
=: M W_{|{\beta}|} \Pi(p,x),\end{aligned}$$ by , since ${\sigma}^\star$ is decreasing (recall that ${{\operatorname}{supp}}({\varphi}_{i,p}) \subseteq Q_i^*$).
Let us assume that $x \in {\mathbb{R}}^n \setminus E$ satisfies $d(x) < (3B_1)^{-1} r_0$. Then, if $x \in Q_i^*$, $$d(Q_i,E) \le d(x_i) \le 3d(x) \le \frac{r_0}{B_1},$$ by . So, for all $i \in {\mathbb{N}}$, $x \in {\mathbb{R}}^n \setminus E$ with $d(x) < (3B_1)^{-1} r_0$, and ${\beta}\in {\mathbb{N}}^n$, $$|\varphi^{({\beta})}_{i,p}(x)|\le M W_{|{\beta}|}\, \Pi(p,x).$$ By and , we may assume that $\dddot S \le D_1\, W$ for some constant $D_1$. Then, by and , for $x \in {\mathbb{R}}^n \setminus E$ with $d(x) < (3B_1)^{-1} r_0$, $$\begin{aligned}
|{\partial}^{\alpha}& (f - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x)|
\\
&\le
\sum_{{\beta}\le {\alpha}} \frac{{\alpha}!}{{\beta}!({\alpha}-{\beta})!}
\cdot 12^{2n} \cdot M W_{|{\alpha}|-|{\beta}|}\Pi(p,x)
\cdot C (6DL)^{|{\beta}|+1} \dddot S_{|{\beta}|} \, h_{\dddot s}(3LD d(x))
\\
&\le 12^{2n}C M
\Big(\sum_{j=0}^{|{\alpha}|} \frac{|{\alpha}|!\, n^{|{\alpha}|+j}}{j!(|{\alpha}|-j)!}
(6DL)^{j+1} W_{|{\alpha}|-j}
\dddot S_{j}\Big) \, \Pi(p,x)
\, h_{\dddot s}(3LD d(x))
\\
&\le 6\, 12^{2n} DL CD_1 M n^{|{\alpha}|} W_{|{\alpha}|}
\Big(\sum_{j=0}^{|{\alpha}|} \frac{|{\alpha}|!\, }{j!(|{\alpha}|-j)!}
(6D L n)^{j}\Big) \,
\Pi(p,x)
\, h_{\dddot s}(3LD d(x))
\\
&= 6\, 12^{2n} DL CD_1 M (n (1 + 6DL n ))^{|{\alpha}|} W_{|{\alpha}|}
\Pi(p,x)
\, h_{\dddot s}(3LD d(x)), \end{aligned}$$ since $W_{|{\alpha}|-j} \dddot S_j \le D_1\, W_{|{\alpha}|-j} W_j \le D_1\, W_{|{\alpha}|}$, by (2). By , for each $\tilde S \in {\mathfrak{S}}$ there is a constant $H\ge 1$ such that $$\Pi(p,x)
\le \Big(\frac{e}{h_{\tilde s}(\frac{b_1 p d(x)}{9 A_2(n) H})} \Big)^{\frac{A_1(n) H}{p}}.$$ By , there is a constant $B\ge 1$ such that $h_{\dddot s}(t) \le h_{\tilde s}(Bt)^2$ provided that $\dddot S_{j+k} \le \tilde S_j \tilde S_k$ for all $j,k$. That such $\tilde S \in {\mathfrak{S}}$ exists follows from .
Let us choose $L$ according to and such that $p := 27 \,A_2(n) H B D L/b_1 \ge A_1(n) H$ is an integer. Then, since $h_{\tilde s} \le 1$, $$\begin{aligned}
\Pi(p,x) \, h_{\dddot s}(3LD d(x))
\le \frac{e h_{\dddot s}(3LD d(x))}{h_{\tilde s}(3 B L D d(x))}
\le e h_{\tilde s}(3 B L D d(x))\end{aligned}$$ and we obtain . (Note that $M$ depends on $p$, and hence on $L$, which results in the non-explicit dependence of $M_1$.) The claim is proved.
#### *End of proof*
By , we have with $S' := \ddot S$. We may additionally assume that $L \ge D_1 {\rho}$ for the corresponding constant $D_1$ in . So, by and , for $x \in {\mathbb{R}}^n \setminus E$ with $d(x) < (3B_1)^{-1} r_0$ and ${\alpha}\in {\mathbb{N}}^n$, $$\begin{aligned}
|f^{({\alpha})}(x)|
&\le
|(T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x))} F)^{({\alpha})}(x)| + |{\partial}^{\alpha}(f - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x)|
\notag \\ \label{final}
&\le
C M^{|{\alpha}|+1} W_{|{\alpha}|}
\end{aligned}$$ for a suitable constant $M=M(S,L)$; here we use that $h_{\tilde s} \le 1$.
Let us fix a point $a \in E$ and ${\alpha}\in {\mathbb{N}}^n$. Since ${\overline}{\Gamma}_{\ddot s}(t) \to \infty$ as $t \to 0$ (see ), we have $|{\alpha}| < 2 {\overline}{\Gamma}_{\ddot s}(L d(x))$ if $x \in {\mathbb{R}}^n \setminus E$ is sufficiently close to $a$. Thus, as $x \to a$, $$\begin{aligned}
&|f^{({\alpha})}(x) - F^{{\alpha}}(a)|
\\
&\le
|{\partial}^{\alpha}(f - T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x)) } F) (x)| +
|(T_{\hat x}^{2 {\overline}{\Gamma}_{\ddot s}(L d(x))}F)^{({\alpha})}(x)-F^{\alpha}(\hat x)| + |F^{\alpha}(\hat x) - F^{\alpha}(a)|
\\
& = O(h_{\tilde s}(L K_3 d(x))) + O(d(x)) + O(|\hat x - a|),
\end{aligned}$$ by , (where $S' =\ddot S$), and . Hence $f^{({\alpha})}(x) \to F^{{\alpha}}(a)$ as $x \to a$. We may conclude that $f \in C^\infty({\mathbb{R}}^n)$. After multiplication with a suitable cut-off function of class ${\mathcal{B}}^{\{{\omega}\}}$ with support in $\{x : d(x) < (3B_1)^{-1} r_0\}$, we find that $f \in {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n)$ thanks to , , and . The theorem is proved.
The proof of Theorem \[extensiontheorem\] shows that for each ${\rho}>0$ there exist $M({\rho})>0$ and a continuous linear extension operator ${\mathcal{B}}^S_{\rho}(E) \to {\mathcal{B}}^{W}_{M({\rho})}({\mathbb{R}}^n)$. This extension operator depends on ${\rho}$ and $S$ (through $L$ and $p$) and in general there is no continuous extension operator ${\mathcal{B}}^{\{{\sigma}\}}(E) \to {\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n)$, cf. [@Petzsche88] and [@schmetsValdivia00 p.223].
The extension theorem for Denjoy–Carleman classes {#sec:CC}
-------------------------------------------------
In this section we prove a consequence of the extension theorem \[extensiontheorem\] for Denjoy–Carleman classes and compare it with the result of [@ChaumatChollet94]. The *sine qua non* for the extension of jets of class ${\mathcal{B}}^{\{M\}}$ to a function of class ${\mathcal{B}}^{\{N\}}$ is the following condition: $$\label{eq:CC}
{\;\exists}C >0 {\;\forall}k \in {\mathbb{N}}: \sum_{\ell \ge k} \frac{1}{\nu_\ell} \le C\, \frac{k}{\mu_k}.$$ (This is true, if $M$ has moderate growth, which we shall have to assume in the main result of this section, . In general, the right condition seems to be ($*$) in [@SchmetsValdivia04] which is equivalent to provided that $M$ has moderate growth, see [@SchmetsValdivia04 2.(c)].) In the next lemma we show that implies for the associated weight functions ${\omega}_M$ and ${\omega}_N$. This is based on [@Komatsu73 Proposition 4.4] and was announced in [@ChaumatChollet94 p.39]. We include a full proof for the convenience of the reader.
\[lem:CCstrong\] Let $M$ and $N$ be weight sequences satisfying $\mu \le \nu$ and . Then $$\label{eq:strong}
{\;\exists}C >0 {\;\forall}t >0 : \int_1^\infty \frac{{\omega}_N(t u)}{u^2}\, du \le C {\omega}_M(t) + C.$$
Let ${\Sigma}_M(t) := \max\{ k : \mu_k \le t\}$. Then (cf. [@Mandelbrojt52 p.21]) $$\label{eq:Siom}
{\omega}_M(t) = \int_0^t \frac{{\Sigma}_M(u)}{u}\, du.$$ Similarly for ${\Sigma}_N$ and ${\omega}_N$. Since $\mu_k \le \nu_k$ for all $k$ we have ${\Sigma}_N \le {\Sigma}_M$. By , $N$ is non-quasianalytic and, by [@Komatsu73 Lemma 4.1], $$\frac{{\Sigma}_N(t)}{t} \to 0 \quad \text{ and } \quad \frac{{\omega}_N(t)}{t} \to 0 \quad \text{ as } \quad t \to \infty.$$ Fix $t \ge \nu_1$ and set $p := {\Sigma}_M(t)$ and $q := {\Sigma}_N(t)$; then $p \ge q \ge 1$, $\nu_{q+1}>t$, and $\mu_{p+1}>t$. Integration by parts yields $$\begin{aligned}
\int_t^\infty \frac{{\Sigma}_N(u)}{u^2}\, du
&= \frac{{\Sigma}_N(t)}{t} + \int_t^\infty \frac{d {\Sigma}_N(u) }{u}
\\
&= \frac{{\Sigma}_N(t)}{t} + \sum_{\ell = q+1}^\infty \frac{1}{\nu_{\ell}}
\\
&= \frac{{\Sigma}_N(t)}{t} + \sum_{\ell = q+1}^p \frac{1}{\nu_{\ell}} + \sum_{\ell = p+1}^\infty \frac{1}{\nu_{\ell}}
\\
&\le \frac{{\Sigma}_N(t)}{t} + \frac{p}{\nu_{q+1}} + C\, \frac{p+1}{\mu_{p+1}}
\\
&\le (2 + 2 C) \frac{{\Sigma}_M(t)}{t} \quad \text{ for } t \ge \nu_1.
\end{aligned}$$ Consequently, by integrating, $$\begin{aligned}
\int_{\nu_1}^s \int_t^\infty \frac{{\Sigma}_N(u)}{u^2}\, du\, dt
&=
s \int_s^\infty \frac{{\Sigma}_N(u)}{u^2}\, du + \int_{\nu_1}^s \frac{{\Sigma}_N(u)}{u}\, du
- \nu_1 \int_{\nu_1}^\infty \frac{{\Sigma}_N(u)}{u^2}\, du
\\
&\le
(2 + 2 C) \int_{\nu_1}^s \frac{{\Sigma}_M(t)}{t}\, dt
\\
&\le
(2 + 2 C) {\omega}_M(s) \quad \text{ for } s \ge \nu_1.
\end{aligned}$$ It follows that, for $s \ge \nu_1$, $$\begin{aligned}
s \int_s^\infty \frac{{\Sigma}_N(u)}{u^2}\, du \le (2 + 2 C) {\omega}_M(s) + \nu_1 \int_0^\infty \frac{{\Sigma}_N(u)}{u^2}\, du
\end{aligned}$$ Clearly, this also holds for all $0 <s<\nu_1$. By partial integration and , $$\int_s^\infty \frac{{\omega}_N(u)}{u^2} \, du = \frac{{\omega}_N(s)}{s} + \int_s^\infty \frac{{\Sigma}_N(u)}{u^2}\, du.$$ Thus, using ${\omega}_N \le {\omega}_M$, $$\begin{aligned}
s \int_s^\infty \frac{{\omega}_N(u)}{u^2}\, du \le (3 + 2 C) {\omega}_M(s) + \nu_1 \int_0^\infty \frac{{\Sigma}_N(u)}{u^2}\, du
\end{aligned}$$ which implies .
\[lem:MomM\] Let $M$ be a weight sequence of moderate growth such that $\liminf_{k \to \infty} m_k^{1/k}>0$ and $\liminf_{k \to \infty} \mu_{Qk}/\mu_k >1$ for some $Q \in {\mathbb{N}}_{\ge 2}$. Then ${\mathcal{B}}^{\{M\}}({\mathbb{R}}^n) = {\mathcal{B}}^{\{{\omega}_M\}}({\mathbb{R}}^n)$ and ${\mathcal{B}}^{\{M\}}(E) = {\mathcal{B}}^{\{{\omega}_M\}}(E)$ for each compact $E \subseteq {\mathbb{R}}^n$.
Since $M$ is log-convex, we have (cf. e.g. [@Mandelbrojt52]) $$\label{eq:MW1}
M_k = \sup_{t>0} \frac{t^k}{e^{{\omega}_M(t)}} = \sup_{s \in {\mathbb{R}}} \frac{e^{sk}}{e^{{\varphi}_{{\omega}_M}(s)}} = e^{{\varphi}_{{\omega}_M}^*(k)}.$$ So the first identity follows from , since ${\omega}_M$ is a weight function. The second identity is a consequence of [@RainerSchindl12 (5.11)] and .
\[thm:sequences\] Let $M$ and $N$ be weight sequences of moderate growth satisfying $\mu \lesssim \nu$. Assume that both the sequences $\mu_k/k$ and $\nu_k/k$ tend to infinity and are almost increasing in the sense that $$\label{eq:condai}
{\;\exists}C >0 {\;\forall}1 \le j \le k : \frac{\mu_j}{j} \le C \, \frac{\mu_k}{k}.$$ Then the following conditions are equivalent:
1. For every compact $E \subseteq {\mathbb{R}}^n$ the jet mapping $j^\infty_E : {\mathcal{B}}^{\{N\}}({\mathbb{R}}^n) \to {\mathcal{B}}^{\{M\}}(E)$ is surjective.
2. There is a $C>0$ such that $\int_{1}^\infty \frac{{\omega}_N(tu)}{u^2}\,du \le C\,{\omega}_M(t) + C$ for all $t>0$.
3. There is a $C>0$ such that $\sum_{\ell \ge k} \frac{1}{\nu_\ell} \le C\, \frac{k}{\mu_k}$ for all $k \in {\mathbb{N}}$.
\(3) $\Rightarrow$ (2) follows from since we may assume without loss of generality that $\mu \le \nu$ (otherwise we replace $(N_k)$ by an equivalent sequence $(C^k N_k)$).
\(2) $\Rightarrow$ (1) Since $M$ has moderate growth, $\mu_k/k \lesssim m_k^{1/k}$ tends to infinity and hence ${\omega}_M(t) = o(t)$ as $t \to \infty$. By (2), ${\omega}_M$ is a heir of ${\omega}_N$. The condition (and ) guarantees that ${\omega}_M$ is a good heir of ${\omega}_N$. Moreover, for $\nu$ implies that ${\omega}_N$ is equivalent to its least concave majorant; this follows from and since $\nu_k \lesssim N_k^{1/k}$ as $N$ has moderate growth (cf. [@RainerSchindl16a Lemma 2.2]). So and entail (1).
\(1) $\Rightarrow$ (3) This follows from [@ChaumatChollet94 Proposition 27]; an inspection of its proof shows that the general assumption of [@ChaumatChollet94] that all sequences are *strongly log-convex* (i.e. $m_k = M_k/k!$ is log-convex) is not needed. Alternatively, it is a consequence of [@SchmetsValdivia04 Theorem 1.1] thanks to [@SchmetsValdivia04 2.(c)].
By , the condition can be replaced by $$\label{eq:condai2}
{\;\exists}C >0 {\;\forall}1 \le j \le k : m_j^{1/j} \le C \, m_k^{1/k}.$$ should be compared with [@ChaumatChollet94 Theorem 30]. In the latter the sequences $M$ and $N$ are assumed to be strongly log-convex which entails the weaker condition (and ). On the other hand, in [@ChaumatChollet94 Theorem 30] moderate growth of $N$ is not required.
Strong and good weight functions {#sec:strong}
--------------------------------
In view of and the fact that every strong weight function is equivalent to a concave one it is natural to ask:
\[q:strong\] Is every strong weight function equivalent to a good one?
By , this holds true if the associated weight matrix satisfies . We do not know the general answer to this question. However, we can provide some more information on strong weight functions. We start with a corollary to .
\[cor:weakcond\] Let ${\omega}$ be a weight function and let ${\mathfrak{W}}= \{W^x\}_{x>0}$ be the associated weight matrix. Then ${\omega}$ is a strong weight function provided that $$\label{eq:weakcond}
{\;\exists}x,y>0 : \sum_{\ell \ge k} \frac{1}{{\vartheta}^y_\ell} \lesssim \frac{k}{{\vartheta}^x_k}.$$
This is immediate from since ${\omega}$ is equivalent to ${\omega}_{W^x}$ for all $x>0$, by [@RainerSchindl12 Lemma 5.7], and is invariant under equivalence.
\[rem:weakcond\] Together with and [@RainerSchindl16a Corollary 5.13], the corollary implies that actually is equivalent to ${\omega}$ being strong, provided that the associated weight matrix satisfies . We do not know if this is always true.
The next lemma is based on a construction from [@RainerSchindl16a] which stems from an idea in [@Petzsche88 Proposition 1.1].
\[lem:strongmatrix\] Let ${\mathfrak{M}}= \{M^x\}_{x>0}$ be a collection of non-quasianalytic weight sequences satisfying $\mu^x \lesssim \mu^y$ whenever $x \le y$. Assume that $${\;\forall}x >0 {\;\exists}y>0 {\;\exists}C>0 : \sum_{\ell \ge k} \frac{1}{\mu^y_\ell} \le C \, \frac{k}{\mu^x_k}.$$ Then there exists a collection of non-quasianalytic weight sequences ${\mathfrak{S}}= \{S^x\}_{x>0}$ with the following properties:
1. $1\le {\sigma}^x_k/k$ is increasing to $\infty$ for all $x>0$.
2. ${\sigma}^x \lesssim {\sigma}^y$ whenever $x \le y$.
3. ${\;\forall}x >0 {\;\exists}y>0 {\;\exists}C>0 : \sum_{\ell \ge k} \frac{1}{{\sigma}^y_\ell} \le C \, \frac{k}{{\sigma}^x_k}$.
4. ${\;\forall}x >0 {\;\exists}y>0 : {\sigma}^x \lesssim \mu^y$ and ${\;\forall}x >0 {\;\exists}y>0 : \mu^x \lesssim {\sigma}^y$.
With any positive increasing sequence $\mu=(\mu_k)$ satisfying $\mu_0 =1$ and $\sum_k 1/\mu_k < \infty$ we may associate a positive sequence ${\sigma}= {\sigma}(\mu)$ in the following way: we define $${\tau}_k := \frac{k}{\mu_k} + \sum_{j\ge k} \frac 1 {\mu_j}, \quad k \ge 1,$$ and set $${\sigma}_k := \frac{{\tau}_1 k}{{\tau}_k}, \quad k \ge 1, \quad {\sigma}_0 := 1.$$ Then, cf. [@RainerSchindl16a Lemma 4.2],
- ${\sigma}\lesssim \mu$.
- $\sum_{j\ge k} 1 /\mu_j \lesssim k/{\sigma}_k$.
- $1 \le {\sigma}_k/k$ is increasing to $\infty$ (in particular, $S$ is strongly log-convex).
- If $\mu'$ is an increasing positive sequence satisfying $\mu' \lesssim \mu$ and $\sum_{j\ge k} 1 /\mu_j \lesssim k/\mu'_k$, then $\mu' \lesssim {\sigma}$.
If we apply this construction to the sequences in ${\mathfrak{M}}$ we obtain a collection of weight sequences ${\mathfrak{S}}= \{S^x\}_{x>0}$ which satisfies the properties (1)–(4). By (4), there exists $x_0>0$ such that $S^{x}$ is non-quasianalytic for all $x\ge x_0$. If we set $S^x := S^{x_0}$ for all $x<x_0$, the collection ${\mathfrak{S}}$ is as desired.
Let ${\omega}$ be a weight function. Assume that the associated weight matrix ${\mathfrak{W}}= \{W^x\}_{x >0}$ satisfies $$\label{gmg}
{\;\forall}x>0 {\;\exists}y>0 : {\vartheta}_k^x \lesssim (W^y_k)^{1/k}.$$ Then the following conditions are equivalent:
1. ${\omega}$ is strong.
2. $\forall x >0 {\;\exists}y>0 {\;\exists}C>0 : \sum_{\ell \ge k} \frac{1}{{\vartheta}^y_\ell} \le C \, \frac{k}{{\vartheta}^x_k}$.
3. $\exists x >0 {\;\exists}y>0 {\;\exists}C>0 : \sum_{\ell \ge k} \frac{1}{{\vartheta}^y_\ell} \le C \, \frac{k}{{\vartheta}^x_k}$.
If ${\omega}$ is strong, then there exists a collection ${\mathfrak{S}}= \{S^x\}_{x>0}$ of strongly log-convex sequences such that $$\label{eq:sivt}
{\;\forall}x >0 {\;\exists}y>0 : {\sigma}^x \lesssim {\vartheta}^y \quad \text{ and } \quad {\;\forall}x >0 {\;\exists}y>0 : {\vartheta}^x \lesssim {\sigma}^y$$ and ${\omega}$ is good.
Note that is only needed for $(1) \Rightarrow (2)$.
For the equivalence of (1), (2), and (3), see , , and the references cited therein.
implies the statement about ${\mathfrak{S}}$. Goodness of ${\omega}$ follows either from , since a strong weight function is equivalent to a concave one, or from the strong log-convexity of the $S^x$ and : for each $x>0$ we find $y,z>0$ such that $$\frac{{\vartheta}^x_j}{j} \lesssim \frac{{\sigma}^y_j}{j} \le \frac{{\sigma}^y_k}{k} \lesssim \frac{{\vartheta}^z_k}{k},
\quad \text{ for } 1 \le j \le k. \qedhere$$
We remark that the condition entails ${\mathcal{B}}^{\{{\omega}\}}({\mathbb{R}}^n) = {\operatorname}{ind}_{x>0} {\mathcal{B}}^{\{S^x\}}({\mathbb{R}}^n)$, by .
\[2\][ [\#2](http://www.ams.org/mathscinet-getitem?mr=#1) ]{} \[2\][\#2]{}
[10]{}
A. V. Abanin, *On [W]{}hitney’s extension theorem for spaces of ultradifferentiable functions*, Math. Ann. **320** (2001), no. 1, 115–126.
A. Beurling, *Quasi-analyticity and general distributions*, Lecture notes, AMS Summer Institute, Stanford, 1961.
G. Bj[ö]{}rck, *Linear partial differential operators and generalized distributions*, Ark. Mat. **6** (1966), 351–407.
J. Bonet, R. W. Braun, R. Meise, and B. A. Taylor, *Whitney’s extension theorem for nonquasianalytic classes of ultradifferentiable functions*, Studia Math. **99** (1991), no. 2, 155–184.
J. Bonet, R. Meise, and S. N. Melikhov, *A comparison of two different ways to define classes of ultradifferentiable functions*, Bull. Belg. Math. Soc. Simon Stevin **14** (2007), 424–444.
J. Bonet, R. Meise, and B. A. Taylor, *Whitney’s extension theorem for ultradifferentiable functions of [R]{}oumieu type*, Proc. Roy. Irish Acad. Sect. A **89** (1989), no. 1, 53–66.
, *On the range of the [B]{}orel map for classes of nonquasianalytic functions*, Progress in functional analysis ([P]{}eñiscola, 1990), North-Holland Math. Stud., vol. 170, North-Holland, Amsterdam, 1992, pp. 97–111.
R. W. Braun, R. Meise, and B. A. Taylor, *Ultradifferentiable functions and [F]{}ourier analysis*, Results Math. **17** (1990), no. 3-4, 206–237.
J. Bruna, *An extension theorem of [W]{}hitney type for non-quasi-analytic classes of functions*, J. London Math. Soc. (2) **22** (1980), no. 3, 495–505.
J. Chaumat and A.-M. Chollet, *Surjectivité de l’application restriction à un compact dans des classes de fonctions ultradifférentiables*, Math. Ann. **298** (1994), no. 1, 7–40.
A. Debrouwere, *Analytic representations of distributions and ultradistributions*, Master’s thesis, Ghent University, 2014, <http://lib.ugent.be/fulltxt/RUG01/002/163/702/RUG01-002163702_2014_0001_AC.pdf>.
E. M. Dyn’kin, *[Pseudoanalytic extension of smooth functions. The uniform scale.]{}*, Transl., Ser. 2, Am. Math. Soc. **115** (1980), 33–58 (English).
C. Fern[á]{}ndez and A. Galbis, *Superposition in classes of ultradifferentiable functions*, Publ. Res. Inst. Math. Sci. **42** (2006), no. 2, 399–419.
L. H[ö]{}rmander, *The analysis of linear partial differential operators. [I]{}*, Grundlehren der Mathematischen Wissenschaften \[Fundamental Principles of Mathematical Sciences\], vol. 256, Springer-Verlag, Berlin, 1983, Distribution theory and Fourier analysis.
H. Komatsu, *Ultradistributions. [I]{}. [S]{}tructure theorems and a characterization*, J. Fac. Sci. Univ. Tokyo Sect. IA Math. **20** (1973), 25–105.
A. Lambert, *Quelques th[é]{}or[è]{}mes de d[é]{}composition des ultradistributions*, Ann. Inst. Fourier (Grenoble) **29** (1979), no. 3, x, 57–100.
M. Langenbruch, *Extension of ultradifferentiable functions*, Manuscripta Math. **83** (1994), no. 2, 123–143.
B. Malgrange, *Ideals of differentiable functions*, Tata Institute of Fundamental Research Studies in Mathematics, No. 3, Tata Institute of Fundamental Research, Bombay, 1967.
S. Mandelbrojt, *Séries adhérentes, régularisation des suites, applications*, Gauthier-Villars, Paris, 1952.
R. Meise and B. A. Taylor, *Whitney’s extension theorem for ultradifferentiable functions of [B]{}eurling type*, Ark. Mat. **26** (1988), no. 2, 265–287.
J. Peetre, *Concave majorants of positive functions*, Acta Math. Acad. Sci. Hungar. **21** (1970), 327–333.
H.-J. Petzsche, *On [E]{}. [B]{}orel’s theorem*, Math. Ann. **282** (1988), no. 2, 299–313.
H.-J. Petzsche and D. Vogt, *Almost analytic extension of ultradifferentiable functions and the boundary values of holomorphic functions*, Math. Ann. **267** (1984), no. 1, 17–35.
A. Rainer and G. Schindl, *Composition in ultradifferentiable classes*, Studia Math. **224** (2014), no. 2, 97–131.
, *Equivalence of stability properties for ultradifferentiable function classes*, Rev. R. Acad. Cienc. Exactas Fis. Nat. Ser. A Math. RACSAM. **110** (2016), no. 1, 17–32.
, *Extension of [W]{}hitney jets of controlled growth*, Math. Nachr. **290** (2017), no. 14-15, 2356–2374. J. Schmets and M. Valdivia, *Extension maps in ultradifferentiable and ultraholomorphic function spaces*, Studia Math. **143** (2000), no. 3, 221–250.
, *On certain extension theorems in the mixed [B]{}orel setting*, J. Math. Anal. Appl. **297** (2004), no. 2, 384–403, Special issue dedicated to John Horv[[á]{}]{}th.
E. M. Stein, *Singular integrals and differentiability properties of functions*, Princeton Mathematical Series, No. 30, Princeton University Press, Princeton, N.J., 1970.
J.-C. Tougeron, *Idéaux de fonctions différentiables*, Springer-Verlag, Berlin, 1972, Ergebnisse der Mathematik und ihrer Grenzgebiete, Band 71.
H. Whitney, *Analytic extensions of differentiable functions defined in closed sets*, Trans. Amer. Math. Soc. **36** (1934), no. 1, 63–89.
[^1]: Supported by the FWF-Projects P 26735-N25 and J 3948-N35
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Thallium monofluoride (TlF) is a prime candidate molecule for precision measurements aimed at discovering new physics. Optical cycling on the B $\leftarrow$ X transition around 271 nm enhances this potential. Hyperfine resolved ultraviolet spectra have been reported to determine the degree of rotational level mixing in the B-state and the efficiency of laser cooling on the B $\leftarrow$ X transition [@Norrgard2017]. Using these high-resolution spectra, the hyperfine structure in the B-state of TlF is re-analysed and the magnitude of the $\Lambda$-doubling in the B-state is discussed.'
author:
- Gerard Meijer
- 'Boris G. Sartakov'
title: 'On the $\Lambda$-doubling in the B$^3\Pi_1$ state of TlF'
---
\[sec:introduction\] Introduction and motivation
================================================
The ultraviolet spectrum of thallium fluoride was recorded and comprehensively analysed for the first time by Howell in 1937 [@Howell1937]. He identified the band system centered around 271 nm as a $^3\Pi_0$ – X$^1\Sigma^+$ system. Some 1700 cm$^{-1}$ below the $^3\Pi_0$ state he identified a $^3\Pi_1$ state. He rationalized why these states should appear in this energetic order (inverted) and that these are the only two low-lying states that can be reached from the electronic ground state [@Howell1937].
The first rotational analyses of the bands of the $^3\Pi_0$ – X$^1\Sigma^+$ and $^3\Pi_1$ – X$^1\Sigma^+$ systems were performed by Barrow and co-workers in 1958 [@Barrow1958]. In this work the assignment of the two electronically excited states was interchanged: the lowest one was referred to as the A$^3\Pi_0$ state and the upper one as the B$^3\Pi_1$ state. In their paper, this interchanged assignment is not commented upon. They do remark, however, that the A-X system consists of $R$ and $P$ branches “[*spreading in opposite directions from well marked origin gaps*]{}”, thereby implicitly stating that there is no $Q$-branch in the A-X system. They also make a one-sentence remark on the observation of a clear $Q$-branch in the B-X system [@Barrow1958]. Interestingly, already in 1950, Herzberg writes about the A- and B-state of TlF in a footnote to the 80-page Table of Molecular Constants in his classic book that “[*Howell exchanges $^3\Pi_1$ and $^3\Pi_0$ but the B $\leftarrow$ X system has strong $Q$-branches and must therefore have $\Delta \Omega$= $\pm$1*]{}” [@Herzberg1950].
Since then, Tiemann and coworkers have analysed the rotational structure in the B $\leftarrow$ X system up to high rotational quantum numbers (above $J=100$) and they have studied the predissociation of the B-state [@Wolf1987]. In their analysis, they extracted for the first time values for the $\Lambda$-doubling in the B-state and reported effective Dunham parameters for this [@Tiemann1988].
Our interest in the detailed energy level structure in the B-state of TlF was triggered by the observed anomolously large hyperfine splitting of the $J=1$ level [@Norrgard2017]. This splitting brings the lowest lying hyperfine levels in the B-state close in energy to where a fictitious $J=0$ level would be expected. We wondered, therefore, whether the B-state of TlF should be described as a $^3\Pi_0$ state after all. Our interest was also triggered by our recent spectroscopic study on the related AlF molecule. In that study, we measured and modelled the hyperfine splittings and $\Lambda$-doublings in all three $\Omega$-manifolds of the electronically excited a$^3\Pi$ state to kHz precision, using a standard Hamiltonian [@Truppe2019]. The model used in ref. [@Norrgard2017] to describe the TlF spectra is slightly different and although the match obtained between the observed and simulated low-$J$ Q-branch spectra of both TlF isotopes is good, it is not as perfect as one would expect it to be. We therefore set out to apply the same formalism as used for AlF to the published spectroscopic data on the TlF molecule [@Norrgard2017]. As described here, the hyperfine structure in the B-state unambiguously confirms that this state cannot be a $^3\Pi_0$ state. An excellent match between the observed and calculated hyperfine resolved spectra is found when the standard Hamiltonian for an isolated $^3\Pi_1$ state is used. Accurate information on the $\Lambda$-doubling as well as on the $g_F$-factors of the low-$J$ levels in the B-state of TlF is obtained.
\[sec:Hyperfine Structure\] Fit of the Hyperfine Structure
==========================================================
For our analysis, we used all the measured values for the hyperfine splittings of the $^{205}$TlF isotope as reported in Table I and Table II of ref. [@Norrgard2017], 48 for $e$-levels and 83 for $f$-levels. Different from the procedure followed in ref. [@Norrgard2017], we included the $e$- and $f$-levels in a single fit [@Truppe2019]. We did not include the reported hyperfine splittings of the less abundant $^{203}$TlF isotope, as fewer of these are tabulated and as there is very limited information on the $f$-levels for low values of $J$, which could bias the overall fit. Instead, after completing the fitting procedure, we used the isotopic scaling rules of the parameters found for $^{205}$TlF to determine the parameters for $^{203}$TlF, and to simulate the Q-branch spectra of both isotopes.
The hyperfine Hamiltonian, $H_{HFS}$, for each of the nuclei with nuclear spin ${\bf{I}}$ ($I_{Tl}$ = $I_F$ = $1/2$) of TlF can be written as $$H_{HFS}=a\,L_z\,I_z+b_F\,{\bf{I}}\cdot {\bf{S}}+\frac{1}{3}\,c\,(3\,S_z\,\,I_z-{\bf{S}}\cdot{\bf{I}}),$$ with the orbital, $a$, the Fermi contact, $b_F$, and the dipolar, $c$, parameters as originally defined by Frosch and Foley [@Frosch1952; @Carrington2003]. For a given electronic state, with a component $L_z=\Lambda$ of the total electron orbital angular momentum along the internuclear axis, with a total electron spin $S$ and with quantum number $\Omega$ – where $\Omega$ = $\Lambda$ + $\Sigma$, with $\Sigma$ the projection of $\bf{S}$ along the internuclear axis – the Hamiltonian can be expressed as $$H_{HFS}=I_z\left[a\,\Lambda+b_F(\Omega-\Lambda)+\frac{2}{3}\,c\,(\Omega-\Lambda)\right].$$ In the B-state of TlF, we can be certain that $\Omega=1$ ($\it{vide~infra}$) but the quantum numbers $\Lambda$ and $S$ are not well determined. It is commonly assumed that for the B-state $\Lambda=1$ and $S=1$ but any electronic state with $\Lambda=1, 2, \dots$ and with the appropriate value for the electron spin has an $\Omega=1$ component that can be admixed. The relatively short lifetime of the B-state of 99 $\pm$9 ns [@Hunter2012] suggests that $S=0$ electronic states contribute significantly to the B-state wavefunction. We therefore follow here the notation that is also used in ref. [@Norrgard2017], and write $$H_{HFS}=h_{\Omega}\,I_z,$$ implying that $h_1$ $\equiv$ $a$ when $\Lambda$=1 whereas $h_1$ will have contributions with different weights from the $a$, $b_F$ and $c$ terms when $\Lambda$ and $S$ have different values.
The lowest two, isolated $F=0$ and $F=1$ hyperfine levels in the B-state of TlF are located at an energy close to where the $J=0$ level would be expected if the B-state were a $^3\Pi_0$ state. We considered it important to use the analysis of the observed hyperfine structure to check whether the B-state might be a $\Omega=0$ state after all. For this, we reassigned the observed rotational transitions [@Norrgard2017] to those of a $^3\Pi_0$ $\leftarrow$ $^1\Sigma^+$ band and analysed the resulting hyperfine splittings with the Hamiltonian for an isolated $^3\Pi_0$ state. No satisfactory agreement could be obtained when the main two parameters, $h_0(Tl)$ and $h_0(F)$, were used. In an effort to better match the resulting hyperfine splittings, terms describing the interaction of each nucleus with the pure rotational angular momentum (parameters $C_I(Tl)$ and $C_I(F)$) as well as nuclear spin-spin interaction terms (a scalar one and a tensorial one, described with the parameters $D_0$ and $D_1$, respectively) were included [@Truppe2019]. Also in this case, no satisfactory agreement between theory and experiment could be obtained. We conclude, therefore, that based on the observed hyperfine structure, the B-state of TlF cannot be described as a $^3\Pi_0$ state.
We then analysed the hyperfine structure in the B-state of TlF using a standard Hamiltonian for an isolated $^3\Pi_1$ state [@Truppe2019]. This Hamiltonian is identical to the one for a $^1\Pi$ state. As both nuclei have a nuclear spin of $1/2$, one can formally not distinguish which of the two main $h_1$ parameters belongs to which nucleus; it is expected, however, that $h_1(Tl)$ is (much) larger than $h_1(F)$ and they are assigned accordingly. In our Hamiltonian, we also include the effective nuclear spin-rotation parameter $C_I$ as well as the corresponding $\Lambda$-doubling contribution $C'_I$ as defined by Brown and coworkers [@Brown1978]. Considering the relative magnitude of $h_1(Tl)$ and $h_1(F)$, these higher order $C_I$ and $C'_I$ terms are only included for the Tl nucleus. The $C'_I(Tl)$ term couples $e$- and $f$-levels and therefore, both sets of levels need to be treated in a single fit. It is this term that causes the so-called $"b"$-splitting [@Norrgard2017] to be different for $e$- and $f$-levels and as these splittings are tabulated up to high $J$-values, the value of $C'_I(Tl)$ can be accurately determined.
Parameter value (MHz) SD SD$\cdot$ sqrt(Q)
------------------- --------------------- ------ -------------------
$B $ 6687.879 – –
$q $ 2.423 – –
$D $ 0.010869 – –
$ H $ 8.1 $\cdot 10^{-8}$ – –
$h_1(Tl) $ 28789 34 42
$h_1(F) $ 861 17 20
$C_I(Tl) $ -7.83 0.43 1.57
$C'_I(Tl) $ 11.17 0.85 3.00
$\Delta \nu $ 2571 – –
: \[tab:HFS-Constants\] Rotational constant $B$, $\Lambda$-doubling parameter $q$, rotational centrifugal distortion parameters $D$ and $H$ and hyperfine parameters $h_1(Tl)$, $h_1(F)$, $C_I(Tl)$ and $C'_I(Tl)$ for $^{205}$TlF, as obtained from the best fit to the 131 hyperfine splittings of $^{205}$TlF listed in Table I and Table II of ref. [@Norrgard2017]. The four hyperfine parameters are given together with their standard deviation (SD) and the product of SD and $\sqrt{Q}$ (all values in MHz). The other parameters are kept fixed in the fit. The standard deviation of the fit is 8 MHz. The parameter $\Delta \nu$ is the shift of the vibrational band origin of $^{203}$TlF relative to $^{205}$TlF.
The parameters resulting from a best fit to the hyperfine splittings of $^{205}$TlF are presented in Table \[tab:HFS-Constants\]. For the four hyperfine parameters, the standard deviation (SD) as well as the product of the standard deviation with the square-root of the quality-factor Q is given; the latter is the better measure for the accuracy with which each parameter is determined in a fitting procedure in which various parameters are correlated [@Watson1977]. The standard deviation of the fit is 8 MHz.
The hyperfine parameters $h_1(Tl)$ and $h_1(F)$ are within the error bars the same as those found in the study by Norrgard $\it{et~al.}$ [@Norrgard2017]. In that study the hyperfine parameters were fitted separately for $e$- and $f$-levels, i.e. two more free parameters were used. A comparison of the higher order hyperfine parameters is not straightforward, as in ref. [@Norrgard2017] the $C'_I(Tl)$ term was not included, but two separate $C_I(Tl)$ terms for the $e$- and $f$-levels were used instead; the value we find for $C_I(Tl)$ is the average of the two separate values reported there [@Norrgard2017]. As mentioned before, the seperate fitting for the $e$- and $f$-levels is formally not valid and can result in a wrong correlation between hyperfine constants, leading to unrealistic error bars.
The rotational constants $B$, $D$ and $H$ as well as the $\Lambda$-doubling parameter $q$ for $^{205}$TlF are kept fixed in the fit of the hyperfine splittings at the tabulated values. The $\Lambda$-doubling is generally described by three terms in the Hamiltonian, parametrized by the $o$-, $p$- and $q$-parameters [@Brown1979]. In the case of an isolated $\Omega=1$ state only the $q$-term has non-vanishing matrix elements between sub-levels of the same $J$. As this term yields an energy difference between $e$- and $f$-levels given by $q J (J+1)$ its effect can be incorporated by using different rotational constants for the $e$- and $f$-levels. This is the approach used in ref. [@Norrgard2017] and the value for $q$ that is given in Table \[tab:HFS-Constants\] is the difference of their reported $B$-values for the $e$- and $f$-levels; the value for $B$ that is given in Table \[tab:HFS-Constants\] is the average of their $B$-values. The centrifugal distortion parameters $D$ and $H$ are those reported for the $f$-levels in ref. [@Norrgard2017].
We have used these $B$, $D$, $H$ and $q$ parameters to simulate the Doppler limited absorption spectra of the $v'=0$ $\leftarrow$ $v''=0$ band recorded by Tiemann and coworkers, tabulated in the PhD thesis of Wolf [@WolfPhD1987]. There, the frequencies of many lines in the $R$- and $P$-branch are listed, but in the $Q$-branch only the frequencies for isolated lines in a limited interval of high $J$-values are given. Our simulations show that their labeling of the $Q$-lines is one quantum number off. As a consequence, the value for the $\Lambda$-doubling parameter that they extracted for the $v''=0$ level of $q$ = 0.1 $\pm$ 0.3 MHz, is incorrect [@Tiemann1988].
![Reproduction of the measured $Q$-branch spectra of the B$^3\Pi_1$, $v$=0 $\leftarrow$ X$^1\Sigma^+$, $v''$=0 band of the $^{205}$TlF and $^{203}$TlF isotopes (middle) together with the simulation (lower), as presented in Fig. 2(b) of ref. [@Norrgard2017]. The upper spectrum shows our simulated $Q$-branch spectrum for $^{205}$TlF (blue) and $^{203}$TlF (red), using the parameters listed in Table \[tab:HFS-Constants\]. In our simulation, a rotational temperature of 7 K and a Lorentzian linewidth (FWHM) of 30 MHz is taken. The parameters for the $^{203}$TlF isotope are obtained from the parameters in Table I by using the isotopic scaling rules.[]{data-label="fig:Q-Branches"}](Q-Branches){width="1.0\linewidth"}
The hyperfine resolved fluorescence excitation spectrum of the $Q$-branch of the B$^3\Pi_1$, $v=0$ $\leftarrow$ X$^1\Sigma^+$, $v''=0$ band of both the $^{205}$TlF and $^{203}$TlF isotope is presented in Fig. 2(b) of ref. [@Norrgard2017], together with their simulated spectrum. Using the parameters listed in Table \[tab:HFS-Constants\], assuming a rotational temperature of 7 K and a Lorentzian lineshape with a full width at half maximum of 30 MHz, the simulated spectrum shown in Figure \[fig:Q-Branches\] (above part of their reproduced Fig. 2(b)) is obtained. For this simulation, the rotational constant of $^{205}$TlF in the X$^1\Sigma^+$, $v''=0$ state is taken from microwave spectroscopy as 6667.4 MHz [@Barrett1958; @Tiemann1988]. The hyperfine splittings in the electronic ground-state are below 1 MHz and the hyperfine levels belonging to a certain $J''$ are taken to be degenerate. The blue part in the simulated spectrum is the contribution from $^{205}$TlF and the red part is from $^{203}$TlF; both isotopes are assumed to be present in their natural abundance. The shift $\Delta \nu = 2571$ MHz of the vibrational band origin of $^{203}$TlF relative to $^{205}$TlF is determined from the isotope-splittings tabulated in ref. [@Norrgard2017]. The simulated spectrum reproduces even the finest details of the experimental spectrum, for both isotopes. This not only attests to the quality of the experimental spectrum, but it also unambiguously shows that the B-state of TlF is an $\Omega =1$ state.
\[sec:Lambda Doubling\] $\Lambda$–doubling for $J=1,2$
======================================================
In an isolated $\Omega=1$ state, the magnitude of the $\Lambda$-doubling for a given $J$-level is normally described by the term $q J(J+1)$, as stated before. In the B-state of TlF, however, the contribution of the $C'_I$ terms dominates the splitting between the opposite parity components of a given $F$-level for low $J$-values. For the $J=1$, $F=0$ level the analytical expression for the $\Lambda$-doubling is $$E_e - E_f = 2q + C'_I(Tl) + C'_I(F).$$ In fitting the hyperfine splittings, we did not include the $C'_I(F)$ term, i.e. we kept the value of $C'_I(F)$ fixed to zero. If we include $C'_I(F)$ in the fit, we find a value of 0.01 $\pm$ 0.30 MHz. Including this term, therefore, does not change the value of the $\Lambda$-doubling, but only adds a small uncertainty. The value that we find for the $\Lambda$-doubling of the $J=1$, $F=0$ level is 16 $\pm$ 1 MHz. Most importantly, the magnitude of the $\Lambda$-doubling does not depend on the (error bar of) the large $h_1(Tl)$ and $h_1(F)$ parameters.
$(J,F_1,F)$ $\Delta E_{\Lambda}$ (MHz) SD $g_F$ $(J,F_1,F)$ $\Delta E_{\Lambda}$ (MHz) SD $g_F$
------------- ---------------------------- ---- ------- ------------- ---------------------------- ---- -------
(1,1/2,0) 16 1 0 (2,5/2,2) -14 2 0.11
(1,1/2,1) 16 1 0.32 (2,5/2,3) -14 2 0.08
(1,3/2,1) -18 1 0.25 (2,3/2,1) 15 4 0.43
(1,3/2,2) -18 1 0.14 (2,3/2,2) 14 4 0.25
: \[tab:Lambda-Doubling\] Energy difference $\Delta E_{\Lambda}$ = $E_e$ – $E_f$ between the $e$- and $f$-components of the four hyperfine levels of the two lowest rotational levels in the B-state, labeled by ($J$, $F_1$, $F$), following the nomenclature in ref. [@Norrgard2017]. The $\Lambda$-doublet splittings are given together with their standard deviation (SD) (all values in MHz). The calculated $g_F$-values, assuming an isolated $^3\Pi_1$ state or a $^1\Pi$ state, are given in a separate column.
For the $F\neq 0$ levels, the analytical expressions for the $\Lambda$-doubling are more involved, and only the numerical values resulting from the fit are given in Table \[tab:Lambda-Doubling\] for the eight lowest energy $F$ levels. From this Table, it is seen that the values for the $\Lambda$-doubling are positive for the $F_1$ = $J - 1/2$ levels and negative for the $F_1$ = $J + 1/2$ levels. Interestingly, the magnitude of the $\Lambda$-doubling is seen to drop slightly in going from $J=1$ to $J=2$. The separation between the opposite parity components of these $F$-levels is only about ten times their homogeneous linewidth [@Hunter2012]. The $F\neq 0$ levels will experience a first-order Stark-shift in an external electric field. Assuming that the electric dipole moment in the B-state has a value that is comparable to the 4.2 Debye in the X$^1\Sigma^+$ state [@Boeckh1964], electric fields of a few V/cm will already lead to significant mixing of the opposite parity components of these $F$ levels.
For the shifting and splitting of the $F$-levels in a magnetic field, the magnetic $g_F$-factors need to be known. We know that in the B-state $\Omega = 1$ and we have calculated the $g_F$-factors using the formalism for an isolated $^3\Pi_1$ state as well as for a $^1\Pi$ state. We find the same values for these two cases, confirming that these two models are equivalent, and these $g_F$-values are given in Table \[tab:Lambda-Doubling\]. It should be noted that the presence of other electronic states close to the $^3\Pi_1$ (or $^1\Pi$) state can potentially influence these $g_F$-values.
\[sec:Conclusions\] Concluding remarks
======================================
The rigorous approach to incorporate the $\Lambda$-doubling and higher-order hyperfine structure terms in the Hamiltonian, in a combined fit of the $e$- and $f$-levels, is seen to describe the energy level structure in the B-state of TlF very well, unambiguously demonstrating that this is an $\Omega=1$ state. It is remarkable that this state can be so well described by the Hamiltonian of an isolated $^3\Pi_1$ state, even up to rotational energies of 0.25 eV above the lowest level, given the high density of electronically excited states nearby [@Balasubramanian1985; @Zou2008]. The large value for $h_1$ would be extraordinary for light diatomics, but might be common for molecules with heavy nuclei. Relativistic effects can lead to a decrease of the radius of the electron orbit and might cause this large value of $h_1(Tl)$. As a result, the values for $C_I(Tl)$ and $C'_I(Tl)$ are also larger than for light diatomics; the value for $C'_I(Tl)$ found here is about three orders of magnitude larger than the value for $C'_I(Al)$ found in the a$^3\Pi$ state of AlF [@Truppe2019]. The overall picture of the hyperfine structure in the B-state is the same as reported in ref. [@Norrgard2017] and, in particular, the rovibrational branching ratios reported there are not found to be different in our more rigorous analysis. We do find, however, that the experimental data also contain accurate information on the separation and ordering of the opposite parity components of the lowest $F$-levels, i.e. the levels that are most important for the laser cooling experiments [@Hunter2012].
\[sec:Acknowledgment\] Acknowledgment
=====================================
We acknowledge the constructive scientific discussion with Dave DeMille, Eric Norrgard and Larry Hunter on the details and differences of their and our analysis of the excellent experimental data that they presented in ref. [@Norrgard2017].
E. B. Norrgard, E. R. Edwards, D. J. McCarron, M. H. Steinecker, D. DeMille, S. S. Alam, S. K. Peck, N. S. Wadia, and L. R. Hunter, Physical Review A [**95**]{}, 062506 (2017).
H. G. Howell, Proceedings of the Royal Society of London A [**160**]{}, 242 (1937).
R. F. Barrow, H. F. K. Cheall, P. M. Thomas, and P. B. Zeeman, Proceedings of the Physical Society [**71**]{}, 128 (1958).
G. Herzberg, [*Molecular Spectra and Molecular Structure. I. Spectra of diatomic molecules.*]{}, 2nd ed. (Van Nostrand Reinhold Company Inc., 1950).
U. Wolf and E. Tiemann, Chemical Physics Letters [**133**]{}, 116 (1987).
E. Tiemann, Molecular Physics [**65**]{}, 359 (1988).
S. Truppe, S. Marx, S. Kray, M. Doppelbauer, S. Hofsäss, H. C. Schewe, N. Walter, J. Pérez-Ríos, B. G. Sartakov, and G. Meijer, Physical Review A [**xx**]{}, xx (2019): arXiv:1908:11774.
R. A. Frosch and H. M. Foley, Physical Review [**88**]{}, 1337 (1952).
A. Carrington and J. M. Brown, [*Rotational spectroscopy of diatomic molecules*]{} (Cambridge University Press, 2003).
L. R. Hunter, S. K. Peck, A. S. Greenspon, S. S. Alam, and D. DeMille, Physical Review A [**85**]{}, 012511 (2012).
J. M. Brown, M. Kaise, C. M. L. Kerr, and D. J. Milton, Molecular Physics [**36**]{}, 553 (1978).
J. K. G. Watson, Journal of Molecular Spectroscopy [**66**]{}, 500 (1977).
J. M. Brown and A. J. Merer, Journal of Molecular Spectroscopy [**74**]{}, 488 (1979).
U. Wolf, [*Laserspektroskopische Untersuchungen zur Prädissoziation der IIIa-VII-Verbindungen durch Tunneleffekt*]{}, Ph.D. thesis, Universität Hannover (1987).
A. H. Barrett and M. Mandel, Physical Review [**109**]{}, 1572 (1958).
R. von Boeckh, G. Gräff, and R. Ley, Zeitschrift für Physik [**179**]{}, 285 (1964).
K. Balasubramanian, The Journal of Chemical Physics [**82**]{}, 3741 (1985).
W. Zou and W. Liu, Journal of Computational Chemistry [**30**]{}, 524 (2008).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Recent progress in the development of superconducting nanowire single-photon detectors (SNSPDs) has delivered excellent performances, and has had a great impact on a range of research fields. The timing jitter, which denotes the temporal resolution of the detection, is a crucial parameter for many applications. Despite extensive work since their apparition, the lowest jitter achievable with SNSPDs is still not clear, and the origin of the intrinsic limits is not fully understood. Understanding its intrinsic behaviour and limits is a mandatory step toward improvements. Here, we report our experimental study on the intrinsically-limited timing jitter in molybdenum silicide (MoSi) SNSPDs. We show that to reach intrinsic jitter, several detector properties such as the latching current and the kinetic inductance of the devices have to be understood. The dependence on the nanowire cross-section and the energy dependence of the intrinsic jitter are exhibited, and the origin of the limits are explicited. System timing jitter of 6.0 ps at 532 nm and 10.6 ps at 1550 nm photon wavelength have been obtained.'
author:
- Misael Caloz
- Boris Korzh
- Edward Ramirez
- Christian Schönenberger
- 'Richard J. Warburton'
- Hugo Zbinden
- 'Matthew D. Shaw'
- Félix Bussières
title: 'Intrinsically-limited timing jitter in molybdenum silicide superconducting nanowire single-photon detectors'
---
Since their first demonstration [@Goltsman2001a], superconducting nanowire single-photon detectors (SNSPDs) have emerged as a key technology for optical quantum information processing [@Hadfield2009]. Their low dark count rate, fast response time, small jitter, and high efficiency favours their use in various demanding quantum optics applications such as quantum key distribution [@Boaron18], quantum networking [@Bussieres2014], device-independent quantum information processing [@Shalm2015a], deep-space optical communication [@Shaw14], IR-imaging [@Allman2015a; @Zhao2017], and integration in photonic circuits [@Sprengers2011; @Rath2015; @Ferrari2018].
One advance in the SNSPD field has been the introduction of amorphous superconductors such as tungsten silicide (WSi) [@Marsili2013] and molybdenum silicide (MoSi) [@korneeva2014; @Verma2015; @Caloz2017a; @Caloz2018]. SNSPDs based on these materials currently have the highest reported system detection efficiencies (SDE) (93% for WSi [@Marsili2013]). Their amorphous properties makes them materials of choice for applications where the film quality and yield are crucial, such as multi-mode coupled SNSPDs, or large arrays [@Allmaras2017].
The jitter is a crucial characteristic for time-resolved measurements such as light detection and ranging, high-speed quantum communication, and lifetime measurement of single-photon sources. It denotes the timing variation of the arrival time of the detection pulses. Assuming independent contributions [@You2013; @Calandri2016] the total measured jitter can be written as the following:\
$j_{system}^2 = j_{setup}^2 + j_{noise}^2 + j_{intrinsic}^2 + j_{geometric}^2$, where $j_{setup}$ includes the laser pulse width and measurement imprecisions, $j_{noise}$ is the contribution from the amplification and electronic parts, $ j_{intrinsic}$ includes the timing variation of the hotspot itself, and $ j_{geometric}$ is linked to the path the signal has to propagate depending on the photon absorption location in the nanowire [@Calandri2016].
A wide range of values have been reported for different geometries and materials, typically from few to hundreds of picoseconds [@Wu2017; @You2013; @Shcheslavskiy2016; @Verma2015]. A record value of 2.7 ps at 400 nm wavelength have been recently achieved with a NbN device [@Korzh2018], which demonstrates that there is still room for improvements. Recent study with MoSi meandered devices showed that low jitter value (26 ps at 1550 nm) is achievable with amorphous material [@Caloz2018]. Despite recent theoretical studies [@Sidorova2017; @Vodolazov2019], the lowest experimental jitter achievable with SNSPDs is still not clear, and the fundamental limits remain unknown. Understanding its intrinsic behaviour and limits is a mandatory step toward improvements. This letter addresses this question with MoSi-based devices.
We report on devices with a special design which reduce significantly the geometric jitter component. They have been fabricated and measured with cryogenic amplifiers and measurement setup that reduces the noise jitter component [@Korzh2018]. We show first that to reach intrinsically-limited jitter, different parameters have to be understood and optimized, namely the kinetic inductance ($L_k$) of the devices and its latching current ($I_{latch}$). Secondly, we tested devices with different cross-sections by varying their width and thickness to probe the nanowire cross-section dependence of the intrinsic jitter. Finally, we observed the photon-energy dependence of the jitter of MoSi SNSPDs.
The devices are fabricated out of 5, 7 and 9 nm-thick films of Mo$_{0.8}$Si$_{0.2}$ deposited by co-sputtering. The film is then patterned by a combination of e-beam lithography and reactive ion etching. A total of 80 different devices were measured for this study. The devices consist of a single 5 $\mu$m-long nanowire connected to a contact pad, through an meandered inductor, as illustrated in Fig. \[fig1\]. This nanowire design minimizes the geometric jitter component ($j_{geometric}$), while the series inductor is used to prevent the latching effect [@Korzh2018]. To probe the nanowire cross-section dependence of the intrinsic jitter ($j_{intrinsic}$), the nanowire width is varied from 60 nm to 200 nm, depending on the thickness. When the cross-section of the nanowire increases, the bias current needed to operate the detector increases as well, and eventually gets larger than $I_{latch}$, which prevents its operation. To cope with this problem, the devices are tested with different series inductances ranging from 100 nH to 3500 nH.
![(a) Scanning electron microscope image. The device is composed of a contact pad (in red), an inductor (in blue), and a nanowire connected to the ground. (b) Zoom of the 5 $\mu$m long MoSi nanowire.[]{data-label="fig1"}](fig1){width="45.00000%"}
The experiment was carried out using a pulse-tube cryocooler with a $\mathrm{^4He}$ sorption refrigerator reaching a base temperature just under 1 K.
The signal from the SNSPD was amplified with SiGe cryogenic amplifiers from Cosmic Microwave. For the slew rate versus the kinetic inductance characterisation and the energy-dependence measurements, the CITLF3 and the CITLF132 were used, respectively. The SNSPDs were biased with a low-noise current source through a 5 k$\Omega$ resistive bias-T at the input of the AC-coupled amplifier. In order to get higher latching currents, a shunt inductance of 1.2 $\mu$H was connected to the ground through a 50 $\Omega$ resistance.
To investigate the photon-energy dependence of the jitter, we used two second harmonic generation (SHG) crystals to frequency double the mode-locked lasers from 1064 nm and 1550 nm to 532 nm and 775 nm, respectively. After the crystal, the light was collimated and free-space coupled into the cryostat through a series of glass windows in the vacuum chamber and the heat shields at 40 K and 4 K, flood illuminating the device under test. The optical intensity was controlled with a circular metallic variable neutral-density filter. This configuration ensured that the converted and unconverted light co-propagated via the same path through the optical setup. After generation, filters were used to select 532, 775, 1064, and 1550 nm wavelength illumination.
The SNSPDs and laser synchronization signals were acquired simultaneously on a digital real-time oscilloscope with a sampling rate of 40 GS/s and a bandwidth of 12 GHz. The time delay between the two pulses was recorded for each acquisition. Histograms of 5000 detection delays were collected for each jitter measurement, which typically required a collection time of approximately 5 minutes. The histogram are fitted with an exponentially modified gaussian (EMG) function, and the jitter was obtained by taking the FWHM of this distribution.
The first part of this work consisted in understanding the latching current and noise jitter dependence on the electronic readout and device kinetic inductance. The desired mode of operation for SNSPDs is achieved only when the electric feedback is slower than the nanowire cooling time, which happens naturally if its kinetic inductance is large enough [@Kerman2009]. If this feedback is sped up by decreasing the kinetic inductance, the device will suffer from the latching effect where it is locked in a resistive state and can no longer detect photons. When the cross-section of the nanowire increases, *i.e.* when either its width and/or thickness increases, the switching current $I_{sw}$ increases as well, and can eventually get bigger than $I_{latch}$. The large kinetic inductance is necessary to slow down the electric feedback and prevents latching. However, if $L_k$ is too large, two problems arise: (i) the maximum count rate of the SNSPD is reduced, and (ii) the electrical signal coming out of the nanowire after a detection is slowed down, meaning a lower slew rate ($SR$) and consequently a larger noise jitter. The last point is crucial to reach intrinsically-limited jitter. The jitter induced by the gaussian noise of the cryogenic amplifier can be estimated by [@You2013; @Caloz2018]: $$\label{eq1}
j_{noise} = 2 \sqrt{2 \ln 2} \frac{\sigma_\mathrm{RMS}}{SR}$$ where $\sigma_\mathrm{RMS}$ is the amplifier RMS noise and $SR$ is the slew rate of the detection signal.
Fig. \[fig2\] shows the slew rate and the jitter induced by the noise for 120 nm wide, 7 nm thick nanowires, with different kinetic inductance ranging from 100 nH to 1000 nH. The slew rate is extracted from oscilloscope traces and is plotted in Fig. \[fig2\]a as a function of the bias current. Fig. \[fig2\]b shows the corresponding estimated jitter induced by the electrical noise as described in Eq. \[eq1\] as a function of the bias current.
![Dataset for 120 nm wide, 7 nm thick nanowires, with different kinetic inductance. (a) Slew rate of the signal rising edge for devices with different kinetic inductance (shown in the legend). (b) Estimated jitter induced by the electrical noise (described in Eq. \[eq1\]) as a function of the bias current. The stars indicate the latching current for the corresponding devices.[]{data-label="fig2"}](fig2){width="50.00000%"}
A compromise between $L_k$, $I_{latch}$, $SR$ and consequently $j_{system}$ has to be found. Regarding Fig. \[fig2\], it is clear that the best compromise is to reduce as much as possible $L_k$, while satisfying $I_{latch}>I_{sw}$. The lowest noise jitter we could achieve was estimated to be 5 ps. We performed this characterization for the three thicknesses, and obtained quantitatively the same results. The only way left to reduce the noise jitter is by decreasing the kinetic inductance of the device, which is incompatible with the latching effect, as explained above.
Once the latching current for a given $L_k$ is known, we selected devices with the lowest $L_k$ possible that still satisfied $I_{latch}>I_{sw}$ to ensure optimal performances. The best detectors for each thickness were selected for energy-dependence measurements as shown in Tab. \[tab2\]. We experimentally observed higher latching currents when using the CITLF1 amplifier, this allowed us to pick lower kinetic inductances resulting in lower system jitter. This improvement is probably due to electrical reflections going back and forth between the SNSPD and the amplifier but the exact explanations is left for future work.
Fig. \[fig3\]a shows the timing histogram for the 7 nm thick device measured for a bias current of $17.9\ \mu$A, as indicated by the circles in Fig. \[fig3\]b. The FWHM of the distributions are 6.0 $\pm$ 0.2 ps and 10.6 $\pm$ 0.2 ps at 532 nm and 1550 nm wavelength, respectively. A non-gaussian tail is clearly observed, it becomes more apparent for long wavelengths and low bias currents, this behaviour has also been reported in many studies [@Sidorova2017; @Caloz2018; @Korzh2018], but its origin remains unclear. The jitter energy dependence is plotted in Fig. \[fig3\]b. The 5 nm and 9 nm-thick devices exhibit qualitatively the same behaviour, the results are summarized in Tab. \[tab2\]. One notable difference between cross-sections is that we could obtain a jitter of 14.5 ps at 1550 nm for the 9 nm-thick device, while the 5 and 7 nm-thick devices showed values close to 10 ps.
{width="\textwidth"}
One clear signature of the intrinsic jitter can be seen in Fig.\[fig3\]b: for the same bias current, $j_{setup}$, $j_{noise}$, and $j_{geometric}$ are independent of the photon wavelength. This means that the difference between different wavelengths is only due to the $j_{intrinsic}$ and that we are probing the intrinsic component. The intrinsic behaviour becomes clearer for low bias currents and longer wavelengths.
The noise jitter of device \#2 was estimated to be 5 ps. By subtracting it to its system jitter of 6.0 ps, *i.e.* the noise contribution would be totally negligible, the remaining jitter goes down to roughly 3.5 ps. While this value seems very close to 2.7 ps of the NbN-based devices [@Korzh2018], the difference appears more clearly for longer wavelengths, where NbN achieved 4.6 ps at 1550 nm, while MoSi achieved 8.6 ps (after subtracting the noise). This shows that MoSi-based devices seem to yield larger jitter and the fact that this remaining jitter is purely intrinsic seems to point to a material difference for long wavelengths.
The energy-dependence data sets are currently being analysed following a recent theoretical framework including the intrinsic jitter behaviour and Fano fluctuations [@Allmaras2018]. While this model fits very well data with NbN devices, our study adds experimental inputs and could lead to a better understanding of intrinsic jitter mechanism and to an unified detection mechanism model.
----- ----------- ------- ------- ----- ----- ------ ---------
Thickness Width $L_k$
(nm) (nm) (nH) 532 775 1064 1550
\#1 5 120 200 6.2 6.5 8.8 10.7 ps
\#2 7 100 200 6.0 6.4 7.8 10.6 ps
\#3 9 80 250 7.0 7.3 9.5 14.4 ps
----- ----------- ------- ------- ----- ----- ------ ---------
: List of the selected devices for energy-dependence measurements, and their system jitter.[]{data-label="tab2"}
In conclusion, we reported intrinsically-limited timing jitter with MoSi SNSPDs. To reach fundamental limits, we showed that the kinetic inductance has to be minimized taking into account the latching current of the detector. We developed an experimental setup that minimize every component of the system jitter of our SNSPDs and allows us to probe and quantify the intrinsic jitter. The energy-dependence of the intrinsic jitter is shown and points to a material limitation of MoSi-based devices for wavelengths longer than 1064 nm. Finally, we observed that the intrinsic jitter is higher for thicker devices and longer wavelengths.
The authors would like to acknowledge the Swiss NCCR QSIT (National Center of Competence in Research - Quantum Science and Technology) for financial support. Part of this work was conducted at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
[26]{} ifxundefined \[1\][ ifx[\#1]{} ]{} ifnum \[1\][ \#1firstoftwo secondoftwo ]{} ifx \[1\][ \#1firstoftwo secondoftwo ]{} ““\#1”” @noop \[0\][secondoftwo]{} sanitize@url \[0\][‘\
12‘\$12 ‘&12‘\#12‘12‘\_12‘%12]{} @startlink\[1\] @endlink\[0\] @bib@innerbibempty [****, ()](http://scitation.aip.org/content/aip/journal/apl/79/6/10.1063/1.1388868) [****, ()](http://dx.doi.org/10.1038/nphoton.2009.230) @noop [****, ()]{} [****, ()](http://www.nature.com/nphoton/journal/vaop/ncurrent/full/nphoton.2014.215.html) [****, ()](http://link.aps.org/doi/10.1103/PhysRevLett.115.250402) in [**](http://www.osapublishing.org/abstract.cfm?URI=CLEO_SI-2014-SM4J.2) (, ) p. [****, ()](http://scitation.aip.org/content/aip/journal/apl/106/19/10.1063/1.4921318) @noop [****, ()]{} [****, ()](http://scitation.aip.org/content/aip/journal/apl/99/18/10.1063/1.3657518) @noop [****, ()]{} @noop [****, ()]{} [****, ()](http://dx.doi.org/10.1038/nphoton.2013.13) [****, ()](http://stacks.iop.org/0953-2048/27/i=9/a=095012) [****, ()](http://www.opticsexpress.org/abstract.cfm?URI=oe-23-26-33792) @noop [****, ()]{} @noop [****, ()]{} in [**](http://www.osapublishing.org/abstract.cfm?URI=CLEO_SI-2017-JTh3E.7) (, ) @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [****, ()]{} @noop [ ()]{}, [****, ()](\doibase 10.1103/PhysRevB.96.184504) [****, ()](https://link.aps.org/doi/10.1103/PhysRevApplied.11.014016) [****, ()](\doibase
10.1103/PhysRevB.79.100509) @noop [ ()]{},
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We discuss how to recognize the constellations seen in the Majorana representation of quantum states. Then we give explicit formulas for the metric and symplectic form on $SU(2)$ orbits containing general number states. Their metric and symplectic areas differ unless the states are coherent. Finally we discuss some patterns that arise from the Lieb-Solovej map, and for dimensions up to nine we find the location of those states that maximize the Wehrl-Lieb entropy.'
author:
- Anna Baecklund
- Ingemar Bengtsson
bibliography:
- 'sample.bib'
title: Four remarks on spin coherent states
---
[ address=[Teoretisk fysik, Kungliga Tekniska Högskolan, 106 91 Stockholm, Sweden]{} ]{}
[ address=[Fysikum, Stockholms Universitet, 106 91 Stockholm, Sweden]{} ]{}
Introduction
============
Much of quantum mechanics concerns the action of some group (perhaps under experimental control) on Hilbert space. The group $SU(2)$ provides a simple and instructive case. We will make four remarks that we believe are new, and worth making. They are detailed in the abstract.
To begin we fix a definite representation of $SU(2)$ by means of Schwinger’s oscillator representation [@Schwinger]. It starts with two commuting pairs of creation and annihilation operators
$$[a_+,\bar{a}_+ ] = [a_-,\bar{a}_- ] = 1 \ .$$
There are orthonormal basis states
$$|n_+,n_-\rangle = \frac{(\bar{a}_+)^{n_+}}{\sqrt{n_+!}}\frac{(\bar{a}_-)^{n_-}}{\sqrt{n_-!}}|0,0\rangle
\ . \label{factorials}$$
We refer to these states as number states. In terms of the oscillators we can write the $SU(2)$ Lie algebra generators as well as a number operator $\hat{n}$:
$$S_x = \frac{1}{2}(\bar{a}_+ a_- + \bar{a}_- a_+) \hspace{5mm}
S_y = \frac{1}{2i}(\bar{a}_+ a_- - \bar{a}_- a_+) \hspace{5mm}
S_z = \frac{1}{2}(\bar{a}_+ a_+ - \bar{a}_- a_-)$$
$$\hat{n} = \bar{a}_+ a_+ + \bar{a}_- a_- \ .$$
The Hilbert space is infinite dimensional but we restrict ourselves to eigenspaces of $\hat{n}$, that is we fix $n = n_++n_-$ and obtain an irreducible representation of dimension $N = n+1$. If the physics is that of a spin system we set $n_+ = j+m$ and $n_- = j-m$.
An alternative way of seeing how $SU(2)$ acts on ${\bf C}^{n+1}$ is to observe that the components of the vectors are in one-to-one correspondence with the coefficients of an $n$th order polynomial in an auxiliary complex variable $\zeta$ [@Majorana; @Penrose; @Bacry]. Up to an irrelevant complex factor such polynomials are determined by their $n$ complex roots, hence by $n$ possibly coinciding points in the complex plane taken in any order. Finally stereographic projection turns these $n$ points into $n$ unordered points on a sphere. If a root sits at $z_i$ the latitude and longitude of the corresponding point on the sphere are given by
$$z_i = \tan{\frac{\theta_i}{2}}e^{i\phi_i} \ .$$
The South Pole is at $z = \infty$. If the polynomial has $k$ roots there its degree is only $n-k$. A collection of $n$ unordered points on the sphere is called a constellation of stars, since—in Penrose’s original application—the sphere was literally to be identified with the celestial sphere [@Penrose].
The charming simplicity of the idea is compromised just a little by the care needed to ensure that an $SU(2)$ transformation in Hilbert space corresponds to a rotation of their celestial sphere. A state is described interchangeably as a vector $|\psi\rangle$ or as a polynomial $w(\zeta )$ in an auxiliary variable $\zeta$ through
$$|\psi \rangle = \sum_{k=0}^n \psi_k |n-k,k\rangle \hspace{5mm}
\leftrightarrow \hspace{5mm}
w(\zeta ) = \sum_{k=0}^n(-1)^k\sqrt{{n \choose k}}\psi_k\zeta^{n-k} \ .$$
Up to an irrelevant factor the polynomial admits the unique factorization
$$w(\zeta ) = (\zeta -z_1)(\zeta - z_2)\cdot \dots \cdot (\zeta -z_n) =
\sum_{k=0}^n(-1)^ks_k(z)\zeta^{n-k} \ .$$
Here $s_k(z)$ is the $k$th symmetric function of the $n$ roots $z_i$. These conventions answer all our needs. Given a constellation of $n$ stars on the sphere we can reconstruct the vector up to an irrelevant constant in terms of symmetric functions of the roots. To go the other way we must solve an $n$th order polynomial equation.
Spin coherent states are those for which all the stars coincide,
$$w(\zeta ) = (\zeta - z)^n \hspace{5mm} \leftrightarrow \hspace{5mm}
|z \rangle = \frac{1}{(1+|z |^2)^{\frac{n}{2}}} \sum_{k=0}^n
\sqrt{{n \choose k}}z^k|n-k,k\rangle \ .$$
Here we took care to normalize the states. An arbitrary normalized state is
$$|\psi\rangle = N \sum_{r=0}^\infty \frac{s_k(z_1,\dots ,z_n)}{\sqrt{{n \choose k}}}
|n-k,k\rangle \ , \label{Norm}$$
where $N = N(z_1,\dots , z_n)$ is a normalizing factor to be worked out. We then define the everywhere non-negative Husimi function
$$Q_\psi (z,\bar{z}) = \langle \psi|z\rangle \langle z|\psi \rangle
\hspace{5mm} \Rightarrow \hspace{5mm} \frac{n+1}{4\pi}\int d\Omega \ Q_\psi = 1 \ ,$$
where $d\Omega$ is the measure on the unit sphere and the state is normalized. The implication happens because the coherent states form a POVM. Hence, for all choices of the state the Husimi function is a probability distribution on a sphere—namely on the set of coherent states. Its zeroes occur antipodally to the roots of the polynomial $w(\zeta )$ that defines the state, at
$$\omega_i = - \frac{1}{\bar{z}_i} \ .$$
Finally the Wehrl entropy of a pure quantum state is defined as
$$S_W(\psi ) = - \frac{n+1}{4\pi}\int d\Omega \ Q_\psi
\ln{Q_\psi} \ .$$
Lieb conjectured [@Lieb] and Lieb and Solovej proved [@Solovej] (after an interval of 35 years during which many people thought about it) that this Wehrl entropy attains its minimum at the spin coherent states, thus singling out the latter as “classical”.
This was brief. Details can be found in books [@BZ], and elsewhere.
First remark: star gazing
=========================
Given a constellation of stars, can we recognize the corresponding quantum state without performing a calculation? Sometimes yes. We recognize the number states, we can sometimes see at a glance whether two states are orthogonal, and we can always recognize the time reversed state [@twistors].
To these cases we add constellations of type $k_{\rm N}$–$k_1$–$k_{\rm S}$, $k_{\rm N}$–$k_1$–$k_2$–$k_{\rm S}$, and so on, meaning that we place $k_{\rm N}$ stars at the North Pole, $k_i$ stars on regular polygons at some fixed latitudes, and $k_{\rm S}$ stars at the South Pole. (The vertices of the Platonic solids provide examples.) Let $q$ be a primitive $k_1$th root of unity. The configuration $k_{\rm N}$–$k_1$–$k_{\rm S}$ gives the polynomial
$$\zeta^{k_{\rm N}}(\zeta -\omega_1)(\zeta -\omega_1 q)\cdot \dots
(\zeta- \omega_1q^{k_1-1}) = \zeta^{k_{\rm N}}\left( \zeta^{k_1} + (-)^{k_1}\omega_1^{k_1}
\right) \ .$$
The equality holds because all but two of the symmetric functions in $q^r$ vanish. The resulting (unnormalized) vector is
$$|\psi \rangle = \sqrt{(k_{\rm N}+k_1)!k_{\rm S}!}
|k_{\rm N}+k_1,k_{\rm S}\rangle + \omega_1^{k_1}\sqrt{k_{\rm N}!(k_1 + k_{\rm S})!}
|k_{\rm N}, k_1 + k_{\rm S}\rangle \ .$$
In fact, by varying the latitude and rotating the polygon we sweep out the entire two dimensional subspace spanned by the two number states.
From the configuration $k_{\rm N}$–$k_1$–$k_2$–$k_{\rm S}$ we obtain a four parameter family of states in a subspace spanned by four number states. If $k_2 = k_1$ two of the number states coincide but there are still four free parameters, and we sweep out an entire subspace spanned by only three number states.
Second remark: orbits of number states
======================================
The Majorana representation is ideally suited to study orbits under $SU(2)$ [@Bacry]. To find the orbit to which a given constellation belongs, just perform an arbitrary rotation of the celestial sphere. The set of constellations that appear in this way is the orbit. Since the group is three dimensional, so is a typical orbit. Number states, where all the stars sit at an antipodal pair of points, are exceptional and form two dimensional orbits. Intrinsically they are spheres, with antipodal points identified if $k = n/2$.
Now recall that in projective Hilbert space (equipped with the Fubini-Study metric) a two dimensional subspace is intrinsically a Bloch sphere, of radius $1/2$, and also a Kähler manifold. The orbit of coherent states is a Kähler manifold too, but of a different radius. They form a rational curve in projective space [@Brody], while the subspaces form projective lines. What about the orbits containing general number states? Since they are isolated orbits under the isometry group it immediately follows from a theorem in differential geometry [@Hsiang] that they are minimal submanifolds of projective space. To work out their intrinsic geometry we place $n-k$ stars at the point $z_1$ and $k$ stars at the antipode, and calculate
$$\psi_1(z) = \langle \psi_1|z\rangle = \sqrt{{n \choose k}}|z_1|^{n-k}
\frac{(z-z_1)^k(z + 1/\bar{z}_1)^{n-k}}{(1+|z_1|^2)^{\frac{n}{2}}(1+|z|^2)^{\frac{n}{2}}}
\ .$$
Next we take three such states and evaluate the Bargmann invariant
$$\langle \psi_0|\psi_1\rangle \langle \psi_1|\psi_2\rangle \langle
\psi_2|\psi_0\rangle = \cos{D(0,1)}\cos{D(1,2)}\cos{D(2,0)}
e^{iA_{\Delta}}$$
to second order in the position of the stars. Here $D$ denotes the length of the geodetic edges of the triangle whose symplectic area is $A_\Delta$. Thus we obtain the intrinsic metric and the symplectic form on the orbit,
$$ds^2 = \frac{n+2k(n-k)}{4}\left[ \frac{4dz \otimes_{\rm sym}
d\bar{z}}{(1+|z |^2)^2}\right] \ ,
\hspace{8mm} \Omega = \frac{n-2k}{4}\frac{i4dz \wedge d\bar{z}}{(1+|z |^2)^2}
\ .$$
The metric on the unit sphere appears between square brackets. Metrically the area of the orbit grows as $k$ approaches $n/2$, but symplectically it shrinks. When $k = 0$ the states are coherent, the two areas agree and we have a Kähler metric on the coherent state orbit. For even $n$ and $k = n/2$ the symplectic form vanishes. In this sense—which makes more sense than it seems to at first sight—the coherent state orbit contains the “classical” states [@Klyachko; @Barnum; @Kus].
Third remark: the Lieb-Solovej map
==================================
In their proof of the Lieb conjecture [@Lieb] Lieb and Solovej [@Solovej] introduced a completely positive map that, in a sense, allows us to approach the classical limit in stages. It maps density matrices acting on ${\cal H}^{n+1}$ to density matrices acting on ${\cal H}^{m+1}$. If $m = n+1$ it is
$$\Phi^1(\rho ) = \frac{1}{n+2}\left( a_+^\dagger \rho a_+ + a_-^\dagger \rho a_-
\right) \ ,$$
and $\Phi^{m-n}$ is defined by iteration if $m > n+1$. It is easy to see that this map is trace preserving, and one also proves the key fact that
$$\Phi^{m-n}([S_i,\rho]) = [S_i,\Phi^{m-n}(\rho)] \ .$$
Therefore states on the same $SU(2)$ orbit in ${\cal H}^{n+1}$ will map to states on the same $SU(2)$ orbit in ${\cal H}^{m+1}$, and the resulting density matrices are isospectral. If $\rho$ is a pure coherent state the eigenvalues of $\Phi^{m-n} (|z\rangle \langle z|)$ can be computed. The proof then hinges on the beautiful theorem saying that the resulting spectrum majorizes all other occurring spectra. A modest illustration of this remarkable result seems called for.
Each time we apply the map $\Phi^1$ the rank of the density matrix goes up one step, so $\Phi^2$ applied to a pure state gives a spectrum described by a point in a two-dimensional simplex. The figures show results for 5 000 pure states chosen at random according to the Fubini-Study measure, for the initial dimensions 4 and 5. The straight lines that have been added are the spectra arising from linear combinations of two number states with $|n,k\rangle $ and $|n,k'\rangle$, $|k-k'| > 2$; for $n = 4$ the image can end up in the centre, but it is highly unlikely to do so. Images of the number states are marked by red dots, with the six coherent state images outermost. The latter do majorize all other spectra since all others lie in their convex hull, typically with a large margin. Interesting patterns arise in higher dimensions, but we have not done a systematic study.
Fourth remark: Maximizing the Wehrl-Lieb entropy
================================================
Once it is known that the Wehrl entropy attains its minimum at the coherent states it is irresistible to ask where it attains its maximum. In the Majorana representation the problem is to choose a constellation of points on the sphere that maximizes a particular function. Simpler, but still very difficult, relatives of this problem include the Thomson problem of minimizing the electrostatic potential of $n$ electrons on a sphere and the Tammes problem of maximizing the minimal chordal distance between the electrons. The possible physical motivation behind our problem is shared by the Queens of Quantum problem to maximize the distance between a pure state and the convex hull of the coherent states [@Giraud].
We have not attempted a full scale optimization of the Wehrl entropy. The reason should be evident to anyone who has seen this function written out explicitly as a function of the positions of the Majorana stars [@Lee; @Schupp]. Instead we have taken reasonable candidates for the maximum, including those known to solve the other problems, and then we have checked whether a local maximum of the Wehrl entropy results. In those cases where natural parameters appear in the constellations we have maximized over these parameters. For instance, when $n = 8$ the maximum of the Thomson and the Tammes problem is given by two squares on two distinct latitude circles. This is a configuration of type $0$–$4$–$4$–$0$, and by the first remark it lies in the subspace spanned by $|8,0\rangle$, $|4,4\rangle$, and $|0,8\rangle$. We can rotate the squares relative to each other and change the difference in latitude. The Thomson, Tammes and Wehrl problems are all solved by this configuration, but the latitudes differ. In this way we have convinced ourselves that the results in the accompanying table are correct. Full details can be found elsewhere [@Anna].
---------- --------------------- --------------------- --------------------- ---------------------
Number Maximum Queen of Thomson Tammes
of stars Wehrl Quantum
3 triangle triangle triangle triangle
4 tetrahedron tetrahedron tetrahedron tetrahedron
5 $1$–$4$–$0$ $1$–$4$–$0$ $1$–$4$–$0$ $1$–$4$–$0$
6 octahedron octahedron octahedron octahedron
7 $1$–$5$–$1$ $1$–$5$–$1$ $1$–$5$–$1$ $1$–$3$–$3$–$0$
8 $0$–$4$–$4$–$0$ looks odd $0$–$4$–$4$–$0$ $0$–$4$–$4$–$0$
9 $0$–$3$–$3$–$3$–$0$ $0$–$3$–$3$–$3$–$0$ $0$–$3$–$3$–$3$–$0$ $0$–$3$–$3$–$3$–$0$
---------- --------------------- --------------------- --------------------- ---------------------
We would like to compare the Wehrl and Queens of Quantum problem in Hilbert spaces of dimensions as high as those number of stars for which the Thomson problem has been studied [@Erber] However, in a note on spherical harmonics which is quite relevant here, Sylvester apologizes for treating some things sketchily because he was “very much pressed for time and within twenty-four hours of steaming back to Baltimore” [@Sylvester]. The twenty-first century is no less pressing than was the nineteenth, and we have not been able to go further.
Summary
=======
The Majorana representation of quantum states [@Majorana], in Hilbert spaces of dimension $N$, as consisting of constellations of $N-1$ unordered stars on a sphere [@Penrose], is very useful in all contexts where the group $SU(2)$ plays a dominating role. To set the stage we gave a brief discussion of how to see where we are in Hilbert space, given such a constellation. Next we gave explicit formulas for the metric and symplectic form on orbits through general angular momentum eigenstates. These orbits are interesting because they are symplectic but not Kähler, except for the coherent state orbit which goes through the highest weight states. A small observation on the Lieb-Solovej map followed. This map was introduced [@Solovej] in order to prove that coherent states minimize the Lieb-Wehrl entropy. Given that it is irresistible to ask for those states that maximize it. We presented results on this question for dimensions $N \leq 9$. It would be interesting to see results for higher dimensions.
We thank Johan Brännlund and Kate Blanchfield for asssistance. One of us thanks Andrei Khrennikov and Ekaterina Axelsson for yet another useful Växjö meeting.
[99]{}
J. Schwinger, [*On angular momentum*]{}, in L. C. Biedenharn and H. van Dam (eds.): Quantum Theory of Angular Momentum, Academic Press 1965.
E. Majorana, [*Atomi orientati in campo magnetico variabile*]{}, Nuovo Cim. [**9**]{} (1932) 43.
R. Penrose, [*A spinor approach to general relativity*]{}, Ann. Phys. (N.Y.) [**10**]{} (1960) 171.
H. Bacry, [*Orbits of the rotation group on spin states*]{}, J. Math. Phys. [**15**]{} (1974) 1686.
E. H. Lieb, [*Proof of an entropy conjecture of Wehrl*]{}, Commun. Math. Phys. [**62**]{} (1978) 35.
E. H. Lieb and J. P. Solovej, [*Proof of an entropy conjecture for Bloch coherent states and its generalizations*]{}, arXiv:1208.3632.
I. Bengtsson and K Życzkowski: [*Geometry of Quantum States*]{}, Cambridge UP 2006.
R. Penrose, [*Orthogonality of general spin states*]{}, Twistor Newsletter [**36**]{} (1993) 5.
D. C. Brody and L. P. Hughston, [*Geometric quantum mechanics*]{}, J. Geom. Phys. [**38**]{} (2001) 19.
W. Y. Hsiang, [*On compact homogeneous minimal submanifolds*]{}, Proc. Nat. Acad. Sci. U.S.A. [**56**]{} (1966) 5.
A. Klyachko, [*Dynamic symmetry approach to entanglement*]{}, in J.-P. Gazeau et al. (eds): Proc. NATO Advanced Study Inst. on Physics and Theoretical Computer Science, IOS, Amsterdam 2007.
H. Barnum, E. Knill, G. Ortiz, and L. Viola, [*Generalizations of entanglement based on coherent states and convex sets*]{}, Phys. Rev. [**A68**]{} (2003) 032308.
A. Sawicki, A. Huckleberry, and M. Kuś, [*Symplectic geometry of entanglement*]{}, Commun. Math. Phys. [**305**]{} (2011) 441.
O. Giraud, P. A. Braun, and D. Braun, [*Quantifying quantumness and the quest for queens of quantum*]{}, New. J. Phys. [**12**]{} (2010) 063005.
C. T. Lee, [*Wehrl’s entropy of spin states and Lieb’s conjecture*]{}, J. Phys. [**A**]{} (1988) 3749.
P. Schupp, [*On Lieb’s conjecture for the Wehrl entropy of Bloch coherent states*]{}, Commun. Math. Phys. [**207**]{} (1999) 481.
A. Baecklund: [*Maximization of the Wehrl Entropy in Finite Dimensions*]{}, MSc Thesis, KTH 2013.
T. Erber and G. M. Hockney, [*Complex systems: Equilibrium configurations of $N$ equal charges on a sphere ( $2 \leq N \leq 112$)*]{}, Adv. Chem. Phys. [**98**]{} (1997) 495.
J. J. Sylvester, [*Note on Spherical Harmonics*]{}, Phil. Mag. [**2**]{} (1876) 291.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Using the asymptotic conformal invariance of perturbative QCD we derive the expression of the coupling of external states to all conformal spin $p$ components of the forward elastic amplitude. Using the wave-function formalism for structure functions at small $x,$ we derive the perturbative coupling of the virtual photon for $p= 1$ , which is maximal for linear transverse polarization. The non-perturbative coupling to the proton is discussed in terms of “azimuthal matching” between the proton color dipoles and the $q\bar q$ configurations of the photon. As an application, the recent conjecture of a second QCD Pomeron related to the conformal spin-1 component is shown to rely upon a strong azimuthal matching of the $p= 1$ component in $\gamma^*$-proton scattering.'
author:
- |
N. Marchal[^1] and R. Peschanski\
Service de Physique Théorique CEA-Saclay\
F-91191 Gif-sur-Yvette Cedex, France\
title: 'Conformal couplings and “azimuthal matching” of QCD Pomerons'
---
-30pt -7pt 0.8cm -7pt
Conformal invariance of the BFKL equation
=========================================
As well-known the Balitskii, Fadin, Kuraev, Lipatov (BFKL) equation [@bfkl] expresses the elastic amplitude of two off-shell gluons in the high energy limit corresponding to the perturbative QCD resummation of leading logarithms. In terms of transverse coordinates (Fourier transforms of the four external gluon transverse momenta), the equation can be schematically written $\frac{\partial f}{\partial Y} {{\left({k, k',q;Y}\right)}} = K \otimes f,$ where $Y$ (in the case of structure functions $Y=\log 1/{x_{b_j}}$) is the whole rapidity range, $k,k',$ the two-dimensional initial gluon momenta and $q,$ the 2-momentum transfer. The BFKL integro-differential kernel $K$ is known to possess a global conformal $SL{{\left({2, \mathbb{C}}\right)}}$ invariance [@Li86]. The BFKL derivation is made in the framework of the leading log approximation but it is interesting to investigate the more general consequences of the asymptotic conformal invariance, which could be maintained at higher order. Indeed, for instance, next leading BFKL calculations could preserve an approximate conformal invariance [@brod]. Deviations from asymptotic conformal invariance could also be studied by comparison with the results obtained with this assumption.
The solution of the BFKL equation is for the 4-point gluon amplitude. For practical application to the proton structure functions, say, the conformal couplings of the BFKL solution with the $q\bar q$ states of the virtual photon and with the proton have to be explicited. This is the main purpose of our paper to formulate the most general coupling of external states and discuss the constraints imposed by the conformal symmetry of the BFKL equation.
The conformal symmetry of the BFKL equation [@Li86] is a powerful tool. Knowing that the kernel $K$ is invariant in the $SL{{\left({2, \mathbb{C}}\right)}}$ transformations, it is possible [@Li86; @exact] to solve exactly the BFKL equation by expanding over the $SL{{\left({2, \mathbb{C}}\right)}}$ unitary irreducible representations which are labelled by two quantum numbers, namely the “conformal dimension” $\gamma = \frac{1}{2} + i\nu$ and the “conformal spin”[^2] $p \in \Bbb{Z}.$ In the appropriate eigenbasis $K$ is diagonal with eigenvalues $$\epsilon (p ,\gamma) = \bar \alpha \ \chi_{p} (\gamma)
\label{omega}$$ where $\bar \alpha = \frac {\alpha N_c}{\pi}$ and \_p () & = & 2 (1) -(p+1-) - (p+)\
& = & \_[m=0]{}\^ { + - }. \[chi\] Using the expansion over the whole conformal basis leads to an expression for the structure function as $$F_2 (Y, Q^2) = \ \sum_{p} F_p (Y, Q^2)\equiv \ \sum_{p}\int_{\frac 12-i\infty}^{\frac 12+i\infty} d\gamma {{\left({ \frac{Q}{Q_0}}\right)}}^{2\gamma} e^{\ov{\alpha} \chi_p (\gamma)Y} f_p (\gamma)
\label{eq02}$$ where $f_p (\gamma)$ is obtained from the couplings of the different conformal spin components to the external sources. The aim of our paper is to discuss these functions $f_p (\gamma)$ taking into account the constraints due to conformal invariance.
In the expression (\[eq02\]), one usually sticks to the component $p=0$ which gives rise to the “hard” QCD Pomeron in the leading order BFKL formalism. In phenomenological applications, the perturbative coupling of the conformal component $p=0$ to the virtual photon has been known since a long time [@bjorken; @kT; @nikzak; @charme] and some models of the non-perturbative coupling to the proton have been discussed [@charme].
However, little has been done about higher conformal spins. They have been considered in two-jet production with large rapidity interval in hadron-hadron collisions [@del1] and in the forward jet production in Deep Inelastic Scattering [@del2] which correspond to two “hard” vertices with similar characteristic scales. We shall come back to the corresponding perturbative QCD calculations later on in the discussion. More recently, the general conformal coupling has been formally derived in the eikonal approximation [@vertex], leading to interesting selection rules. But higher spin components were expected to have no practical applications at high energy (rapidity interval) since they are at first sight power suppressed in energy. It is indeed the case for the processes considered in Refs. [@del1; @del2].
However, recently, it has been noticed [@pomerons] that the spin component $p= 1$ may have a non negligeable impact for processes corresponding to vertices with different characteristic scales and in particular for proton structure functions at moderate and large $Q^2.$ This is due to a “sliding” mechanism which shifts up its effective intercept and thus drastically changes the energy dependence. The $p= 1$ spin component may even be interpreted as the remnant of the well-known “soft” Pomeron in the high $Q^2$ region. This result is to be put in perspective with the two-Pomeron conjecture of Ref. [@dolan], where the “soft” Pomeron is considered to be higher-twist, while the “hard” Pomeron would represent some kind of leading-twist[^3]. Hence, it is worth studying in detail the constraints and properties of conformal couplings to QCD Pomerons, both from a perturbative (for the virtual photon) and non-perturbative (for the proton) points of view.
The next section [**2**]{} is devoted to the general formalism for the coupling to a generic conformal spin component of the BFKL solution. In section [**3**]{}, we derive the perturbative coupling to the virtual photon wave-function in terms of its $q\bar q$ configurations and introduce a class of models for the non-perturbative couplings to the proton satisfying appropriate constraints. Then, in section [**4**]{}, we make a phenomenological application to the two-pomeron conjecture based on conformal spin components of the proton structure functions, which leads to the necessity of a strong azimuthal “matching” condition which is discussed in detail. Summary and conclusions are presented in [**5**]{}.
Conformal impact factors
========================
Following Ref. [@Li86] the virtual photon-proton elastic BFKL scattering amplitude reads $${A}(s,-q^2)= i s \int\:\frac{d\omega}{2i\pi} \left( \frac{s}{Q^2} \right)^{\omega} f_{\omega}(q^2)\ ,
\label{eq3}$$ where $s/Q^2\approx 1/x,$ $q^2$ is the quadri momentum transfer squared and $$f_{\omega}(q^2) = \int \: d^2k\: d^2k' \: {\cal V}^{(1)}(k,q)\: \overline {\cal V}^{(2)}(k',q)\:
f_{\omega}(k,k',q)\ .$$ $f_{\omega} (k,k',q)$ is nothing else than the $Y \!\to \!\omega$ Mellin transformed of the two-gluon elastic amplitude verifying the BFKL evolution equation (see section [**1**]{}). ${\cal V}^{(1)}(k,q)$ and ${\cal V}^{(2)}(k',q)$ are the so-called [*impact factors*]{} describing the coupling of the initial states to the gluons.
After straightforward calculations using the conformal basis of eigenvectors [@Li86; @exact], one may write $$f_{\omega}(q^2)= \sum_p \int \: \frac {d\gamma}{2i\pi} \frac{c(p,\gamma)}{\omega \!- \!\epsilon (p ,\gamma)} \: V_1^{p,\gamma} (q)\ \overline{V}_2^{p,\gamma} (q)\ ,
\label{eq6}$$ with $$V_{1,2}^{p,\gamma} (q) =\frac{1}{(2\pi)^3} \int \: d^2 \rho \:d^2 \rho' \: d^2k \: {\cal V}^{(1,2)} (k,q)\ e^{ik\rho+i(q-k)\rho'} \: E^{p,\gamma}(\rho, \rho'),
\label{eq2}$$ where $\epsilon (p ,\gamma)$ is given in (\[omega\]) and $E^{p,\gamma}(\rho, \rho')$ are the $SL{{\left({2, \mathbb{C}}\right)}}$ eigenfunctions $$E^{p,\gamma}(\rho, \rho') =\left(\frac{\rho-\rho'}{\rho \rho'}\right)^{\gamma-p}\ \left(\frac{\bar \rho-\bar {\rho}'}{\bar {\rho} \bar {\rho}'}\right)^{\gamma+p}, \label{10}$$ and $$c(p,\gamma)= \frac { \nu^2 + p^2 }{ \left( \nu^2 + (p- \frac1{2})^2\right) \left( \nu^2 + (p+ \frac {1}{2})^2 \right)}\ ,$$ where $\gamma \equiv \frac 12 + i\nu.$ In the forward direction ($q= 0$) the formula (\[eq2\]) simplifies. After changing variables to $\rho+\rho'=2b$ and $\rho-\rho'=r$ and combining the relations (see [@Li86]) $$\int d^2b \: E^{p,\gamma} (b\!+\!\frac{r}{2}, b\!-\!\frac{r}{2})
= \frac{b_{p,\gamma}}{(2 \pi )^2} \ r ^{\gamma-p} \ \bar r ^{\gamma+p}$$ and (see [@gr]) $$\int d^2u \: u^{\gamma-p} \: \overline{u}^{\gamma+p} \: e^{\frac i2 (u+\bar u)} = 2\pi \int d| u| \: |u|^{1+2\gamma} J_{2p} (|u|) = 4^{1+\gamma}\pi \ \frac {\Gamma(\gamma+p+1)}{\Gamma(p-\gamma)}\ ,$$ one gets $$V_{1,2}^{p,\gamma} (q\!=\!0) = 2^{1+2\gamma} \ \frac {\Gamma(\gamma+p+1)}{\Gamma(p-\gamma)}
\ b_{p,\gamma}
\int \: d^2 k \: {\cal V}^{(1,2)} (k) \: k^{-(\gamma+p+1)} \overline{k}^{-(\gamma-p+1)}\ ,$$ where $b_{p,\gamma}$ is a $SL{{\left({2, \mathbb{C}}\right)}}$ constant given in Ref.[@Li86] and verifying $ \left| b_{p,\gamma} \right|^2 = \frac{ \pi^6 }{p^2 + \nu^2}.$
Using the relation $\Im m{A}(q^2=0) \equiv s \sigma_{tot} = \frac{s}{4\pi Q^2} F_2(Y,Q^2),$ one finally obtains $$\begin{gathered}
F_2 (x, Q^2) \sim \sum_{p} \int d\gamma \ x^{-\epsilon (p ,\gamma)} \left( \frac{Q}{Q_0} \right)^{2\gamma} \ V_1 \ \overline{V}_2\\
= \sum_{p} \int d\gamma \:\left| \frac{ \Gamma \left( p+\gamma \right)}{ \Gamma \left( p - \gamma +1 \right) } \right|^2 x^{-\epsilon (p,\gamma)} \left( \frac{Q}{Q_0} \right)^{2\gamma} \\
\times \int d^2 \kappa \: {\cal V}_1 (\kappa) \: \kappa^{-(\gamma+p+1)} \: \overline{\kappa}^{-(\gamma-p+1)}
\int d^2 \kappa_0 \: \overline{{\cal V}_2} (\kappa_0) \: \kappa_0^{-2+\gamma +p} \: \overline{\kappa_0}^{-2+\gamma-p}\ ,
\label{vertices}\end{gathered}$$ where one introduces the natural scaling variables $ k /Q = \kappa$ for the (photon) vertex $V_1$ and $ k_0/Q_0 = \kappa_0 $ for the (proton) vertex $V_2.$ Note that the Gamma function prefactors boil down to a factor $1$ on the integration line over the imaginary axis $\gamma = \frac 12 +i\nu.$
Let us consider for instance the first components ($p=0, 1$). By separation of modulus and azimuthal integration over $\kappa,$ they correspond to the two first coefficients of the Fourier expansion $${\cal V}_{1,2} (\kappa) = \alpha_{1,2} (\left| \kappa \right| ) + \beta_{1,2} (\left| \kappa \right| ) \cos (2\varphi ) + ...
\label{components}$$
In the case of proton structure functions and specializing to the two first components, one obtains $$\begin{gathered}
F_2 (x,Q^2) \sim \int d\gamma \ x^{-2 \bar \alpha \left( \Psi (1) -{\rm Re} \Psi (\gamma ) \right) } \left( \frac{Q}{Q_0} \right)^{2\gamma} f_0 (\gamma) \\
+\int d\gamma \ x^{-2 \bar \alpha \left( \Psi (1) - {\rm Re} \Psi (\gamma +1) \right) } \left( \frac{Q}{Q_0} \right)^{2\gamma} f_1 (\gamma) \\
+ \sum_{p \neq 0,1,-1}\int d\gamma \ldots
\label{eqf2}\end{gathered}$$ with $$f_0 (\gamma ) = \int_0^{\infty} d\left| \kappa \right| \left| \kappa \right|^{-1-2\gamma} \alpha_1(|\kappa| ) \int_0^{\infty} d\left| \kappa_0 \right| \left| \kappa_0 \right|^{-3+2\gamma}\alpha_2(|\kappa_0| )
\label{f0}$$ $$f_1 (\gamma ) = \int_0^{\infty} d\left| \kappa \right| \left| \kappa \right|^{-1-2\gamma} \beta_1(|\kappa| ) \int_0^{\infty} d\left| \kappa_0\right| \left| \kappa_0 \right|^{-3+2\gamma}\beta_2(|\kappa_0| )\ .
\label{f1}$$ Note a positivity constraint in the case of the eikonal coupling for which [@vertex] $${\cal V} (\kappa) \propto 4 \int d^2r \ \Phi (r)\ \sin^2 \left( \kappa r/2 \right)$$ where $\Phi (r)$ is the probability distribution of the $q\bar q$ configurations in coordinate space . Hence, a positivity condition ${\cal V} (\kappa)>0$ holds which leads to $\left| \beta \right|< \alpha.$ However $\beta$ can be negative as it is indeed the case in some processes like forward jet production in DIS [@del2]. Note also that the positivity constraint does not hold if there are not only $q\bar q$ configurations in the Fock space of the target (e.g. for the proton).
Conformal couplings to $q\bar q$ configurations
===============================================
Let us first derive the conformal couplings to the virtual photon. In the perturbative QCD framework and for the $p=0$ component, it is possible to derive the couplings from first order (virtual) gluon- (virtual) photon fusion graphs, thanks to the $k_T$-factorization property [@kT]. Our aim is to start from these results and derive the corresponding coupling to higher spin components. In fact, for the simple reason of the spin $1$ of the virtual photon, only the conformal spin $p= 1$ can be obtained from the transverse polarization components of the photon.
Interestingly, the factorization properties of QCD in the high-energy regime can be put into two equivalent forms [@charme]. As sketched in Fig.1, the perturbative[^4] coupling of the virtual photon to a dipole can be described by two different factorized formulae. One way is to use the $k_T$-factorization property [@kT] which relates the $\gamma^*$-dipole cross-section to the product of the impact factors $V$ by a $g^*$-dipole cross-section where $g^*$ is an off-mass-shell gluon. Another equivalent way is to use the photon wave-function formalism [@bjorken] which uses the $q\bar q$-dipole cross-section where the $q\bar q$ configurations are defined by the virtual photon wave function. The target dipole is considered to be small (or massive) in order to justify the (resummed) perturbative QCD calculations.
We shall thus make use of the relation (see Fig.1 and Ref. [@charme]) between the impact factors and the wave-functions [@bjorken] of the transverse photon in terms of $q\bar q$ configurations for both helicities. This relation reads [@charme] for $p=0$ $$\phi_{T}^{(p=0)}(\gamma)\equiv \frac 1{2\pi}\int rdrd\varphi {{\left({r^2Q^2}\right)}}^{1-\gamma} \int dz
{{\left({\vert\Psi^+_{T}{{\left({r,z}\right)}}\vert^2+\vert\Psi^-_{T}{{\left({r,z}\right)}}\vert^2}\right)}}
= C\ \frac {V_{T}^{(p=0)}}{\gamma}\ \frac {1} {v(1\!-\!\gamma)} \ ,
\label{relation}$$ where $$v(1\!-\!\gamma) \equiv
2^{2\gamma-3}\frac{\Gamma (1\!+\!\gamma)}
{\gamma (1\!-\!\gamma)\Gamma(2\!-\!\gamma)}.
\label{gluon}$$ $v(\gamma)$ is the factorized coupling of the off-mass-shell gluon to a dipole [@charme]. The light-cone wave functions of the transverse photon $\Psi^+_{T}$ for helicity $^+$ and $\Psi^-_{T}$ for helicity $^-$ are [@bjorken] $$\begin{aligned}
\Psi^+_T {{\left({z,r,Q^2}\right)}} = \sqrt C\ z\ e^{i\varphi} \epsilon K_1{{\left({\epsilon r}\right)}} \\
\Psi^-_T {{\left({z,r,Q^2}\right)}} = \sqrt C\ {{\left({1- z}\right)}}\ e^{-i\varphi} \epsilon K_1{{\left({\epsilon r}\right)}} ,
\label{waves}\end{aligned}$$ where $K_1$ is the Bessel function. By definition $\epsilon \equiv Q\sqrt{z(1-z)}$ and the normalization is $C= \frac{\alpha_{em}N_c e^2}{4\pi\alpha_s}.$
Now, for an arbitrary combination of both helicities, one finds contributions to two Fourier components in azimuthal angle (see (\[components\])), namely $$\begin{gathered}
\left| \eta_+ \Psi_T^+ + \eta_- \Psi_T^- \right|^2 \sim \left| \eta_+ z e^{i\varphi} + \eta_- (1-z) e^{-i\varphi} \right|^2 \epsilon^2 K_1^2{{\left({\epsilon r}\right)}} \\= \left\{\underbrace{\eta_+^2 z^2 +\eta_-^2 (1-z)^2}_{(p=0)} + \underbrace{2\eta_+ \eta_- z (1-z)}_{(p= 1)}\ \cos 2\varphi\right\}\ \epsilon ^2 K_1^2{{\left({\epsilon r}\right)}}\ .
\label{combine}\end{gathered}$$ Normalizing to $\eta_+^2 + \eta_-^2 =1,$ It is easy to realize that the two linearly polarized components $ \eta_+ = \pm \eta_- = \frac{1}{\sqrt{2}}$ give opposite contributions[^5] to the component $p= 1.$ The coupling to the linearly polarized photon are obtained by inserting the appropriate $z$-dependent factor in the expression of the wave function contribution to the $p= 1$ component. Projecting on the $p=1$ azimuthal Fourier component, one writes $$\begin{gathered}
\phi_T^{(p=1)}(\gamma)\equiv\frac{1}{2\pi}\int \ 2\cos\varphi\ rdrd\varphi{{\left({r^2Q^2}\right)}}^{1-\gamma} \int dz\: 2\Re e{{\left({\Psi^+_{T}\ {\Psi^-_{T}}^*}\right)}}{{\left({r,z}\right)}}
\nonumber\\
= \frac{\alpha_{em}N_c e^2}{4\pi\alpha_s} \int rdr {{\left({r^2Q^2}\right)}}^{1-\gamma} \int dz \: 2z(1-z)\epsilon^2 K_1^2 (\epsilon r)
\nonumber\\
\sim \int du \: u^{3-2\gamma} \: K_1^2 (u)\ \times \ \int dz \: 2z^{\gamma} (1-z)^{\gamma},\end{gathered}$$ or, noting that the only difference between the two components come from the $z$-dependent factors, $$\frac {\phi_T^{{(p=1)}}}{\phi_T^{(p=0)}} = \frac {V_T^{{(p=1)}}}{V_T^{(p=0)}}\equiv \ \frac{\gamma}{\gamma + 1}.
\label{p1}$$ One finally gets: [[([ \_T\^[(p=0)]{}\
\_T\^[[(p=1)]{}]{} ]{})]{}]{} &\~& [[([r\^2Q\^2]{})]{}]{}\^[1-]{} dz [[([ z\^2+(1-z)\^2\
2z(1-z)]{})]{}]{} \^2 K\_1\^2 (r)\
& = & 2\^[1-2]{} [[([\
1]{})]{}]{} . \[phis\] Using (\[p1\]) and the relation (\[relation\]), it is easy to write down the similar relation for the impact factors $V_T^{(p=0,1)}.$
As we just saw, the perturbative photon couplings to the non zero conformal spins requires a non zero linear polarization of the $q\bar q$ wave-function of the transverse photon to be dynamically active in the reaction. In other terms, the $p= 1$ BFKL component requires a maximal azimuthal correlation while the $p=0$ one is completely decorrelated azimuthally. Partial azimuthal (de)correlation can be obtained by a mixture of different BFKL components. This will in general depend on the dynamical features of the overall reaction. For instance [@del2], forward jet production in DIS can lead to some azimuthal correlation at small rapidity interval where the higher spin component $p= 1$ are still present. However the general prediction is a significant azimuthal decorrelation due to the strong dominance of the $p=0$ component in this case.
In the case of proton structure functions, however, the “sliding mechanism” is able [@pomerons] to promote the higher spin components, especially for $p= 1,$ to be still important at high energy (and relatively low $Q^2$) and thus to keep rather strong azimuthal correlations present in that region. This implies a discussion of the non-perturbative couplings. An important remark is that the “sliding mechanism” is also expected for perturbative couplings when a large ratio exists between the characteristic scales of both vertices. It would thus also be interesting to study such processes where we would predict an increase of the azimuthal correlations accompanying the expected “sliding mechanism”.
The non-perturbative couplings, e.g. to the proton, are in general beyond our present theoretical knowledge. It is already true for the leading $p=0$ conformal components, where there are some ambiguities [@charme] in the way one is able to factorize the perturbative from the non-perturbative couplings. This is all the more true for the non-leading $p= 1$ component which, to our knowledge, are for the first time studied for proton structure functions in the present paper. For the sake of definiteness, we will follow some reasonable theoretical and phenomenological requirements which we now indicate:
i\) The interaction of the proton are governed (at small $x$) by its $q\bar q$ configurations. This can be interpreted as color dipole configurations [@dipoles; @charme]. Compared with those of the virtual photon, their quantum fluctuations around the proton size $Q_0$ are expected to be smaller.
ii\) The coupling of the proton will be asked to obey the “sliding mechanism”, that is to verify the convergence and analyticity properties found in Ref.[@pomerons]. In particular, no singularity with $\gamma > -1$ should appear in the $p=1$ coupling.
iii\) Within conditions i) and ii), the $p=0$ ($\alpha_2$ in (\[components\])) and $p=1$ ($\beta_2$ in (\[components\])) couplings will be assumed to be equal up to a normalization which will be determined phenomenologically and represent the necessary degree of [*azimuthal correlation*]{} for practical relevance.
iv\) Concerning the abovementionned sign ambiguity of the $p=1$ vertices, it is removed for the contribution to structure functions which ought to be positive. Thus the product of the photon and proton vertices is considered to be positive.
We shall now propose a convenient class of parametrizations of the proton couplings $\alpha_2, \beta_2$ in equations (\[vertices\],\[components\]) satisfying the requirements i)-iv). Noting [@gr] the relation $$\int_0^{\infty} d\kappa_0\: \frac{\kappa_0 ^{q-1+2\gamma}}{1+ \kappa_0 ^{2q}} \equiv \frac1 {2q}\ B\left(\frac 12 \!+\!\frac {\gamma}q, \frac 12 \!-\!\frac {\gamma}q\right)
=\frac {\pi}{2q \cos\left(\frac {\gamma}q\right)}\ ,
\label{grad}$$ we are led to choose $$\alpha_2, \beta_2 \propto \frac{ \kappa_0 ^{{\left({q+2}\right)}}}{1+\kappa_0 ^{2q}} ;\ f_0(\gamma), f_1 (\gamma) \propto \frac 1{ \cos\left(\frac {\gamma}q\right)}\ .
\label{models}$$ Eventually, one may vary the peaking of the distribution around $\kappa_0 =1$ by changing the values of $q.$ It is interesting to note that for $q \ge 2$ the gauge invariance constraint [@Li86] $\alpha_2(0)= \beta_2(0)=0 $ is automatically verified. One can also multipoly by a polynomial expression in $\gamma.$ This can be used to satisfy the constraints, in particular the analyticity ones by cancelling poles at $\gamma>-1.$
The two-Pomeron conjecture and azimuthal matching
=================================================
As already mentionned, the non-zero conformal spin components are generally neglected in the phenomenology related to the BFKL equation. Indeed, at ultra-high energy $Y\to\infty,$ the structure function components in formula (\[eq02\]) are driven by the saddle-points at $\gamma = \frac 12.$ It is easy to realize that the corresponding intercepts $\chi_p(\frac 12)$ are all negative for $p \ne 0.$ In the same time their effective anomalous dimension $\frac 12$ means that they all contribute to a leading-twist behaviour. However, it has been remarked in Ref. [@pomerons] that at large but finite values of $Y$ or $Q^2$ the corresponding saddle-points [*slide*]{} away from $\gamma = \frac 12$ and generate contributions with very different $Y$ and $Q^2$ behaviour from the ultra-asymptotic ones. In particular the $p =1$ component is still increasing with energy (positive intercept) and their $Q^2$ behaviour mimic an higher-twist behaviour, i.e. they decrease like a negative power of $Q^2.$ Both features allowed the authors of [@pomerons] to look for the possibility that the $p = 1$ component could be interpreted as the high $Q^2$ remnant of the “soft” pomeron considered as an higher-twist contribution from the point of view of the operator product expansion of QCD. This would provide a QCD framework for the two-Pomeron hypothesis proposed in [@dolan] to describe the phenomenological features of structure functions in a different, Regge approach.
Let us now investigate how the phenomenological discussion can be influenced by the determination of the conformal couplings derived in the previous sections. In order to analyze the phenomenology of structure functions in a manner similar to Refs. [@pomerons; @dolan], we have to introduce our determination (\[phis\]) of the perturbative coupling to the photon and discuss the proton coupling using, for instance, the family of parametrizations (\[models\]). In the discussion, however, it is important to take into account the ambiguity of the separation between perturbative and non-perturbative couplings discussed in [@charme] for the $p=0$ component. Let us recall the problem and extend its lessons to the $p= 1$ component.
We will consider the following parametrization[^6] of the functions $f_p$ to be inserted in (\[eq02\]): $$f_0 (\gamma) = \phi^{(p=0)}_T (\gamma) \otimes \frac{\gamma (\gamma+1)}{ \cos \frac{\pi \gamma}{q} }
\label{I}$$ $$f_1 (\gamma) = \phi_T^{(p=1)}(\gamma) \otimes {\cal N}_I\ \frac {\gamma (\gamma+1)}{ \cos \frac{\pi \gamma}{q}},
\label{I'}$$ where the non perturbative coupling has been chosen in order to satisfy the analyticity and convergence constraints in a minimal way. Assuming the same analytic form for the non-perturbative $p=1$ proton coupling than $p=0,$ the arbitrary normalization ${\cal N}_{I}$ quantify the relative weight which we want to evaluate. The value $q=4$ has been choosen for convenience. $q>2$ at least is needed to verify the constraints ii). We checked that the results are rather independent of these choices, provided the constraints are satisfied. Note that $f_1$ is “softer” at $\gamma=0$ than $f_0$ due to the relative factor $\gamma/(\gamma+1).$
On a physical point of view, the non-perturbative vertices in formulae (\[I\],\[I’\]) can be interpreted [@npr; @charme] as related to the wave functions of the [*primordial dipole*]{} configurations in the proton. In the QCD dipole model [@dipoles] the BFKL dynamics can be expressed in terms of the dipole-dipole cross-section. Translating this model in the case of $\gamma^*$-proton scattering, it amounts to consider this cross-section averaged both over the $q\bar q$ configurations of the photon and the [*primordial dipole*]{} configurations of the proton.
In order to check[^7] the sliding mechanism advocated in [@pomerons], we display in Fig.2 the normalization independent plot $\frac{\partial \ln F_p}{\ov{\alpha }\partial Y} $ as a function of $\frac{\partial \ln F_p}{\partial \ln Q^2}$ for large $Y$ and different values of $\ln Q^2/Q_0^2.$ On the same plot and for the same values is also shown the corresponding results for the Regge parametrization of [@dolan]. As discussed in [@pomerons], the results (the black circles in Fig.1) gives the location in a two-dimensional representation where the effective intercept is plotted as a function of the effective saddle-point $\gamma_c.$ They can be shown [@pomerons] to be situated near the curves defined by the functions $\epsilon_p(\gamma),$ independantly of the peculiar form of the factors $f_{0,1}.$ The sizeable sliding of the $p= 1$ component is proven by the shift of the corresponding points with respect to the ultra-asymptotic value at $\gamma_c=\frac 12.$ Moreover the evolution at large $ Q^2$ meets the phenomenological determination of the two Pomeron components of [@dolan] for $\log \left(Q^2/Q_0^2\right) \sim 8,10$ and reasonable values of the parameter $\bar \alpha \sim .4.$
In order to determine the relative strength of the $p=1$ and $p=0$ components, and thus the rôle of the conformal prefactors $f_{0,1}$ (see formulae (\[I\] , \[I’\])) we have considered the 2-Pomeron fit (“hard” and “soft”) of [@dolan] in the large $Q^2$ region where it meets[^8] the behaviour of the two ($p=0$ and $p=1$) conformal spin components. For instance we show in Fig.3 the $Y$ dependence at fixed large $Q^2.$
The results indicate large normalizations, namely ${\cal N}_I \sim 50.$ Some other tests show that the normalization is always large of the same order. Thus, the 2-Pomeron conjecture (as seen from a QCD point of view) is obtained only if a strong dynamical enhancement favours the non-perturbative coupling of the $p=1$ component. Since the perturbative coupling is maximal for linearly polarized transverse photon and limited in size because of positivity constraints (the photon has only $q \bar q$ configurations) the relevance of the $p=1$ coupling relies on the existence of a non-perturbative mechanism enhancing considerably the matching between the proton primordial dipole configurations and the azimuthal polarization of the virtual photon.
We shall now speculate on such a non-perturbative mechanism based on azimuthal matching in $\gamma^*$-proton scattering.
The mechanism is the following (see Fig.4, for a schematic representation). It is known from a long time [@bjorken] that deep-inelastic lepton proton scattering is not necessarily dominated by a “hard” process if the energy is large with respect to the photon virtuality, e.g. if $x \sim Q/W$ is small. Indeed the [*effective*]{} virtuality is $\hat Q = Q \sqrt {z(1-z)},$ where $z$ (resp. $1-z$) is the momentum fraction of the quark (resp. antiquark) in the virtual $q\bar q$ state configurations of the virtual photon. This is explicit, for instance, in the wave-functions (\[waves\]). Thus, if the favoured $q\bar q$ configurations are particularly assymetric (aligned jet[@bjorken] configurations) one even may reach the situation where the quark or the antiquark in the pair have so small momentum that $\hat Q$ is of order unity and the reaction dominated by a “soft” process.
However, the experimental results seem not to favour the aligned jet mechanism since a “hard” component shows up which is precisely the one which could be described by the $p=0$ component. Yet, for the $p=1$ component, it is not excluded however that a partial jet alignement can take place, at least at moderate $Q^2.$ At high $Q^2$ one could then expect that assymetric configurations are substantially favoured in a kind of “hard/soft” compromise: the effective virtualities $\hat Q$ are smaller than $Q$ while remaining in the “semi-hard” regime. As a consequence, one expects a substantial [*azimuthal matching*]{} between the $q\bar q$ configurations of the virtual photon and the $q\bar q$ (or primary dipole [@npr]) configurations in the proton, see Fig.3. This [*azimuthal matching*]{} may give a strong dynamical enhancement for the coupling of the linearly polarized components of the photon to the proton. By this azimuthal enhancement one could find the qualitative justification for the two-pomeron description to be based on the two conformal spin components. On the other hand, in the absence of such a mechanism the $p=1$ components, even if increasing with energy due to the sliding phenomenon, would not be coupled enough to the proton to give rise to a sizable component. We shall in conclusion discuss possible tests of azimuthal alignment which is certainly deserving further study.
Summary and Conclusions
=======================
Let us briefly summarize our results:
i\) Taking into account that the non-zero (indeed the $p= 1$) conformal spin components of the BFKL QCD Pomeron could be phenomenologically relevant in deep-inelastic lepton proton scattering, we have given the formal expression of the proton structure functions’ conformal spin components in terms of the appropriate [*impact factors*]{}.
ii\) We have computed the perturbative [*impact factor*]{} for the $p= 1$ component at the virtual photon vertex using a general relation between impact factors and $q\bar q$ wave functions. the key result is that the coupling is maximal for linear azimuthal polarization and zero for circular (or no) polarization.
iii\) In order to be phenomenologically relevant as a “second” Pomeron contribution in $\gamma^*$-proton scattering, a strong [*azimuthal matching*]{} with the primordial dipole $q\bar q$ configurations of the proton is required. This non-perturbative mechanism could be associated in part with jet alignment [*a la Bjorken*]{} for the $p= 1$ component, while it is expected to be weak or absent for the $p=0$ one.
Some comments are in order. The large enhancement (a factor $\sim 50$) of the non-perturbative coupling to the $p=1$ component that we found necessary to match with the two-Pomeron parametrization of Ref.[@dolan] is consistent with the key point of this kind of phenomenological analysis: the mismatch between the “hard” and “soft” Pomerons is more important than in models with only one effective Pomeron singularity. In particular, at intermediate values of the virtuality $Q^2,$ both contributions are important. This is the reason why the “hard” component has a large intercept $\epsilon (p\!=\!0,\frac 12) \simeq 0.4$ in agreement with the theoretical range of values and larger than the effective intercept $\epsilon (p\!=\!0,\frac 12) \leq .3,$ see for instance [@npr; @charme]. The validity of a non-negligeable “soft” Pomeron coupling at high $Q^2$ is thus to be checked in further study.
A specific feature of the $p=1$ component is its special azimuthal properties. the question arises whether it is possible to isolate it using azimuthal correlation properties. For the total inclusive process, leading to the determination of the structure function itself, this seems uneasy. The relevant azimuthal axis in the photon-dipole center-of-mass frame, see Fig.4, can be very different from the photon-proton one, and thus, in particular, the s-channel helicity conservation which seems to be an approximate property of the “soft” Pomeron coupling is not in contradiction with the conformal spin properties of the $p=1$ component.
A possible test of the [*azimuthal matching*]{} could be performed in forward jet production in Deep Inelastic Scattering. Indeed, while the commonly considered configuration with similar scales for the photon probe and the jet is expected to lead to small azimuthal correlation at high rapidity interval [@del2], the case with a larger scale ratio is expected to lead to stronger azimuthal correlation due to the enhancement with energy of the higher conformal spin component responsible for the azimuthal matching in the considered formalism. Indeed, a practical way of checking the azimuthal correlations could be to fix a certain range of high[^9] $Q^2$ for the photon virtuality and vary the tranverse momentum of the forward jet down to the lower admissible value to select a jet. By this way, one increases the ratio $Q^2/k_T^2$ of scales and thus enhance the energy behaviour of the $p=1$ component. Moreover, since the model implies a strong mismatch between the “soft” and “hard” Pomerons at intermediate scale, one expects the development of a stronger (and perhaps different in sign!) azimuthal correlation than the perturbatively predicted azimuthal correlations (with negative sign) studied in ref. [@del2]. A similar method can be proposed at Tevatron analyzing azimuthal correlations between two jets (1) and (2) in different hemispheres, as analyzed in ref. [@del1], with the prediction that it will increase together with the ratio $k_T^{(1)}/k_T^{(2)}.$
[**ACKNOWLEDGEMENTS**]{}
We are grateful to Stephane Munier and Henri Navelet for stimulating discussions and suggestions. One of us (N.M.) wishes to thank the “Service de Physique Théorique de Saclay” and its staff for the kind hospitality during the period of “stage”.
[99]{} L.N.Lipatov, [*Sov.J. Nucl.Phys.*]{} [**23**]{} (1976) 642; V.S.Fadin, E.A.Kuraev and L.N.Lipatov, [*Phys. lett.*]{} [**B60**]{} (1975) 50; E.A.Kuraev, L.N.Lipatov and V.S.Fadin, [*Sov.Phys. JETP*]{} [**44**]{} (1976) 45, [**45**]{} (1977) 199; I.I.Balitsky and L.N.Lipatov, [*Sov.J. Nucl.Phys.*]{} [**28**]{} (1978) 822. L.N. Lipatov [*Zh.Eksp.Teor.Fiz.*]{} [**90**]{} (1986) 1536 (Eng. trans. [*Sov.Phys. JETP*]{} [**63**]{} (1986) 904). S.J. Brodsky, V.S. Fadin, V.T. Kim, L.N. Lipatov and G.B. Pivovarov, hep-ph/9901229. H.Navelet and R.Peschanski, [*Nucl.Phys.*]{} [**B507**]{} (1997) 353; [**B515**]{} (1998) 269. H.Navelet, S.Wallon, [*Nucl. Phys.* ]{}[**B522**]{} (1998) 237. J.D.Bjorken, J.Kogut and Soper, [*Phys.Rev.*]{} [**D3**]{} (1971) 1382. S. Catani, M. Ciafaloni, F. Hautmann, [*Nucl. Phys*]{}. [**B366**]{} (1991) 135. J. C. Collins, R. K. Ellis, [*Nucl. Phys.*]{} [**B360**]{} (1991) 3. E. M. Levin, M. G. Ryskin, Yu. M. Shabelskii, A. G. Shuvaev, [*Sov. J. Nucl. Phys.*]{} [**53**]{} (1991) 657. N.N.Nikolaev, B.G.Zakharov, [*Zeit. für. Phys.*]{} [**C49**]{} (1991) 607; [*Phys. Lett.*]{} [**B332**]{} (1994) 184. S. Munier and R. Peschanki, [*Nucl. Phys.*]{} [**B524** ]{}(1998) 377. V.Del Duca, C.R.Schmidt, [*Nucl. Phys.Proc.Suppl.*]{} [**39BC**]{} (1995) 137. J.Bartels, V.Del Duca, M.Wüsthoff, [*Zeit. für. Phys.*]{} [**C76**]{} (1997) 75. H.Navelet, R.Peschanski, [*Nucl. Phys.*]{} [**B515**]{} (1998) 269. S. Munier and R. Peschanki, [*QCD, conformal invariance and the two Pomerons*]{}, hep-ph/9811400, to be published in [*Eur. Jour. Phys.*]{} [**C**]{}. A.Donnachie and P.V.Landshoff, [*Phys. Lett.*]{} [**B457**]{} (1998) 408. I.S.Gradstein and I.M.Ryzhik, [*Table of integrals and products*]{}, ed. A.Jeffrey (Acad. Press, N.Y., 1994). A.H.Mueller, [*Nucl. Phys.*]{} [**B415**]{} (1994) 373; A.H.Mueller and B.Patel, [*Nucl. Phys.*]{} [**B425**]{} (1994) 471; A.H.Mueller, [*Nucl. Phys.*]{} [**B437**]{} (1995) 107. H. Navelet, R. Peschanski, Ch. Royon, [*Phys. Lett.*]{} [**B366**]{} (1995) 329. H. Navelet, R. Peschanski, Ch. Royon, S. Wallon, [*Phys. Lett.*]{} [**B385**]{} (1996) 357.
[**FIGURE CAPTIONS**]{}
[**Figure 1**]{} [*The two factorization schemes of the $\gamma ^*$-dipole cross-section*]{}
The photon ($\gamma ^*$) -dipole ($d$) cross-section, as given by the QCD dipole model corresponding to the perturbative QCD resummation at small $x$, admits two equivalent factorization schemes (see text). First Scheme: $k_T$-factorization of the $g^*$-dipole cross-section with transverse impact factors $V_T^{(p=0,1)}$; Second Scheme: wave-function factorization of the $(q\bar q)-d$ cross-section, where the virtual photon transverse wave functions $\Psi _T^{(+,-)}$ are described on the basis of its $(q\bar q)$ configurations. The two conformal spin components $(p=0,1)$ of the transverse impact factors can be expressed in terms of the wave functions for left (+) and right (-) helicities, see Eqns. (\[p1\],\[phis\]).
[**Figure 2**]{} [*Plot of effective intercept vs. effective dimension at fixed large $Y$*]{}
The effective intercept $\partial \ln F_{0,1} / \bar \alpha \partial Y$ plotted vs. the effective anomalous dimension $\partial \ln F_{0,1} / \partial \ln Q^{2}$ is compared to the functions $\epsilon_{0,1}\equiv \bar \alpha \chi _{0,1}\left( \gamma \right)$ (see Eqns. (\[omega\],\[chi\])). They are computed at $\bar \alpha = .15$ for fixed $Y=10$ and 4 values of $\ln {\displaystyle{Q^{2} / Q_{0}^{2}}}=\left\{ 4,6,8,10\right\} .$ The weight in the integrals (\[eq02\]) corresponds to Eqns. (\[I\],\[I’\]).
Black circles: numerical results; White circles: ultra asymptotic saddle points at $\gamma =\frac 12;$ Full lines: the functions $\epsilon _{p}\left( \gamma \right)$ for $\left(p= 0,1\right);$ Dotted lines; results from the Regge fit of Ref. [@dolan] corresponding to the same value of $Y$ and $\ln Q^{2}/Q_{0}^{2},$ with $Q_0 \sim 200 {\rm MeV}.$ Arrows indicate the direction of increasing $Q.$
[**Figure 3**]{} [*The structure function spin components $F_{2(p=0,1)}$ at fixed large $Q^{2} .$*]{}
The structure function components $F_{2(p=0,1)}$ are displayed as a function of $Y=\log 1/x$ and compared with the parametrization of the two-Pomeron model of Ref. [@dolan] at $Q^2=1000 {\rm GeV}^2.$ This value is choosen to correspond to $\ln {\displaystyle{Q^{2} / Q_{0}^{2}}}\sim 10 .$
Continuous line: “hard Pomeron” component of Ref.[@dolan]; Long-Dashed line: Spin $0$ component.
Short-Dashed line: “soft Pomeron” component of Ref.[@dolan]; Dashed line: Spin $ 1$ component.
[**Figure 4**]{} [*Azimuthal matching of photon and dipole $q\bar q$ configurations*]{}
The photon ($\gamma ^*$) -dipole ($d$) reaction is represented in the center of mass frame. The azimuthal angle between $q\bar q$ configurations of both colliding systems is the angle $\phi$ between the two planes. The quark (resp. antiquark) momentum fraction in the virtual photon is $z$ (resp. $1-z$) (the similar variable for the dipole configurations has been already averaged).
[**Figure 1**]{}
[**Figure 3**]{}
[**Figure 4**]{}
[^1]: Stagiaire de Diplôme d’Etudes Approfondies de Physique Théorique (ENS,Paris)
[^2]: The conformal spin can be half-integer but only integer values contribute to the structure functions [@nawa]. It can also in principle take negative half-integer and integer values [@Li86], but the decomposition over positive eigenvalues is complete and thus sufficient to describe the conformal expansion.
[^3]: Note however that the paper [@dolan] is written in the conventional Regge formalism while the study of [@pomerons] is made in the framework of the BFKL equation and its conformal invariant setting.
[^4]: In the non-perturbative regime, some modifications of the discussion have to be introduced [@charme] due to the fact that the intermediate gluon $g^*$ may be soft enough to be included in the non-perturbative input. In that case the two pictures lead to two different parametrizations.
[^5]: Note an overall sign ambiguity, which has to be fixed by the calculation of both vertices in the process. For instance the overall sign is negative in the forward jet case [@del2].
[^6]: In [@charme], two different models were introduced, depending whether the factorization between perturbative and non-perturbative couplings is assumed at the intermediate gluon level (model I in [@charme]) or at the quark level (model II in [@charme]). This ambiguity relies on the possibility of the gluon coupling to the $q\bar q$ configurations of the photon (with its typical singularity in $1/\gamma$) to be present (model I) or absorbed (model II) in the non-perturbative coupling to the proton. We checked that the results we obtained in the framework of model I are very similar for model II up to a renormalization of the $p=1$ component.
[^7]: We used the model I parametrization, but the results are the same for model II or by changing $q>2.$
[^8]: This comparison is to be taken only with a grain of salt since it is made in a region where the “soft” component is weak and thus not directly determined by data. A determination at small $Q^2$ would be more precise but then, the non-perturbative corrections are expected to be important and may invalidate a correct evaluation of the normalization in a BFKL framework.
[^9]: However, in practice $Q^2$ is limited by the necessity of a large rapidity interval with the forward jet.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Let $\mathbb{A}=
\begin{pmatrix}
A & 0 \\
0 & A \\
\end{pmatrix}
$ be a $2\times2$ diagonal operator matrix whose each diagonal entry is a bounded positive (semidefinite) linear operator $A$ acting on a complex Hilbert space $\mathcal{H}$. In this paper, we derive several $\mathbb{A}$-numerical radius inequalities for $2\times 2$ operator matrices whose entries are bounded with respect to the seminorm induced by the positive operator $A$ on $\mathcal{H}$. Some applications of our inequalities are also given.
address:
- '$^{[1]}$ University of Sfax, Sfax, Tunisia.'
- '$^{[2]}$ P.G. Department of Mathematics, Utkal University, Vanivihar, Bhubaneswar 751004, India.'
author:
- '$^{1}$ and $^{2}$'
title: 'Further inequalities for the $\mathbb{A}$-numerical radius of certain $2 \times 2$ operator matrices'
---
Introduction and Preliminaries {#s1}
==============================
Throughout this article, $(\mathcal{H},\langle\cdot\mid\cdot\rangle)$ stands for a complex Hilbert space with associated norm $\|\cdot\|$. If $\mathcal{M}$ is a given linear subspace of $\mathcal{H}$, then $\overline{\mathcal{M}}$ denotes its closure in the norm topology of $\mathcal{H}$. Further, the orthogonal projection onto a closed subspace $\mathcal{S}$ of $\mathcal{H}$ will be denoted by $P_{\mathcal{S}}$. Let $\mathcal{B}(\mathcal{H})$ denote the $C^{\ast}$-algebra of all bounded linear operators acting on $\mathcal{H}$ with the identity operator $I_{\mathcal{H}}$ (or simply $I$ if no confusion arises). If $T\in\mathcal{B}(\mathcal{H})$, then $\mathcal{N}(T), \mathcal{R}(T)$ and $T^*$ are denoted by the kernel, the range and the adjoint of $T$, respectively. An operator $T\in \mathcal{B}(\mathcal{H})$ is called positive (semi-definite) $\langle Ax\mid x\rangle\geq 0$, for every $x\in \mathcal{H}$. For the rest of this paper, by an operator we mean a bounded linear operator. Further, we suppose that $A\in\mathcal{B}(\mathcal{H})$ is a nonzero positive operator which induces the following semi-inner product $$\langle\cdot\mid\cdot\rangle_{A}:\mathcal{H}\times \mathcal{H}\longrightarrow\mathbb{C},\;(x,y)\longmapsto \langle x\mid y\rangle_{A}:=\langle Ax\mid y\rangle=\langle A^{1/2}x\mid A^{1/2}y\rangle.$$ Here $A^{1/2}$ denotes the square root of $A$. Notice that the seminorm induced by ${\langle \cdot\mid \cdot\rangle}_A$ is given by ${\|x\|}_A=\|A^{1/2}x\|$, for all $x\in\mathcal{H}$. It can checked that ${\|\cdot\|}_A$ is a norm on $\mathcal{H}$ if and only if $A$ is injective, and that the seminormed space $(\mathcal{H}, {\|\cdot\|}_A)$ is complete if and only if $\mathcal{R}(A)$ is a closed subspace of $\mathcal{H}$.
Let $T \in \mathcal{B}(\mathcal{H})$. An operator $S\in\mathcal{B}(\mathcal{H})$ is called an $A$-adjoint of $T$ if $\langle Tx\mid y\rangle_A=\langle x\mid Sy\rangle_A$ for all $x,y\in \mathcal{H}$ (see [@acg1]). Thus, the existence of an $A$-adjoint of $T$ is equivalent to the existence of a solution of the equation $AX = T^*A$. Notice that this kind of equations can be investigated by using a theorem due to Douglas [@doug] which establishes the equivalence between the following statements:
- The operator equation $TX=S$ has a bounded linear solution $X$.
- $\mathcal{R}(S) \subseteq \mathcal{R}(T)$.
- There exists a positive number $\lambda$ such that $\|S^*x\|\leq \lambda \|T^*x\|$ for all $x\in \mathcal{H}$.
Moreover, among many solutions of $AX=S$, it has only one, say $Q$, which satisfies $\mathcal{R}(Q) \subseteq \overline{\mathcal{R}(T^{*})}$. Such $Q$ is said the reduced solution of the equation $TX=S$. If we denote by $\mathcal{B}_{A}(\mathcal{H})$, the subspace of all operators admitting $A$-adjoints, then by Douglas theorem, we have $$\mathcal{B}_{A}(\mathcal{H})=\left\{T\in \mathcal{B}(\mathcal{H})\,;\;\mathcal{R}(T^{*}A)\subset \mathcal{R}(A)\right\}.$$ If $T\in \mathcal{B}_A(\mathcal{H})$, the reduced solution of the equation $AX=T^*A$ will be denoted by $T^{\sharp_A}$. We mention here that, $T^{\sharp_A}=A^\dag T^*A$ in which $A^\dag$ is the Moore-Penrose inverse of $A$ (see [@acg2]). In addition, if $T \in \mathcal{B}_A({\mathcal{H}})$, then $T^{\sharp_A} \in \mathcal{B}_A({\mathcal{H}})$, $(T^{\sharp_A})^{\sharp_A}=P_{\overline{\mathcal{R}(A)}}TP_{\overline{\mathcal{R}(A)}}$ and $((T^{\sharp_A})^{\sharp_A})^{\sharp_A}=T$. Moreover, If $S\in \mathcal{B}_A(\mathcal{H})$, then $TS \in\mathcal{B}_A({\mathcal{H}})$ and $(TS)^{\sharp_A}=S^{\sharp_A}T^{\sharp_A}.$ Furthermore, for every $T \in \mathcal{B}_A({\mathcal{H}})$, we have $$\label{diez}
\|T^{\sharp_A}T\|_A = \| TT^{\sharp_A}\|_A=\|T\|_A^2 =\|T^{\sharp_A}\|_A^2.$$ An operator $U\in \mathcal{B}_A(\mathcal{H})$ is called $A$-unitary if $\|Ux\|_A=\|U^{\sharp_A}x\|_A=\|x\|_A$, for all $x\in \mathcal{H}$. It is worth mentioning that, an operator $U\in \mathcal{B}_A(\mathcal{H})$ is $A$-unitary if and only if $U^{\sharp_A} U=(U^{\sharp_A})^{\sharp_A} U^{\sharp_A}=P_{\overline{\mathcal{R}(A)}}$ (see [@acg1]). For an account of the results, we invite the reader to consult [@acg1; @acg2].
An operator $T$ is called $A$-bounded if there exists $\lambda>0$ such that $ \|Tx\|_{A} \leq \lambda \|x\|_{A},\;\forall\,x\in \mathcal{H}.$ By applying Douglas theorem, one can see that the subspace of all operators admitting $A^{1/2}$-adjoints, denoted by $\mathcal{B}_{A^{1/2}}(\mathcal{H})$, is equal the collection of all $A$-bounded operators, i.e., $$\mathcal{B}_{A^{1/2}}(\mathcal{H})=\left\{T \in \mathcal{B}(\mathcal{H})\,;\;\exists \,\lambda > 0\,;\;\|Tx\|_{A} \leq \lambda \|x\|_{A},\;\forall\,x\in \mathcal{H} \right\}.$$ Notice that $\mathcal{B}_{A}(\mathcal{H})$ and $\mathcal{B}_{A^{1/2}}(\mathcal{H})$ are two subalgebras of $\mathcal{B}(\mathcal{H})$ which are, in general, neither closed nor dense in $\mathcal{B}(\mathcal{H})$. Moreover, we have $\mathcal{B}_{A}(\mathcal{H})\subset \mathcal{B}_{A^{1/2}}(\mathcal{H})$ (see [@acg1; @acg3]). Clearly, $\langle\cdot\mid\cdot\rangle_{A}$ induces a seminorm on $\mathcal{B}_{A^{1/2}}(\mathcal{H})$. Indeed, if $T\in\mathcal{B}_{A^{1/2}}(\mathcal{H})$, then it holds that $$\label{semii}
\|T\|_A:=\sup_{\substack{x\in \overline{\mathcal{R}(A)},\\ x\not=0}}\frac{\|Tx\|_A}{\|x\|_A}=\sup\big\{{\|Tx\|}_A\,; \,\,x\in \mathcal{H},\, {\|x\|}_A =1\big\}<\infty.$$
Saddi [@saddi] in 2012 defined the $A$-numerical radius of an operator $T\in\mathcal{B}(\mathcal{H})$ by $$\begin{aligned}
\omega_A(T)
&:= \sup \left\{|\langle Tx\mid x\rangle_A|\,;\;x\in\mathcal{H},\|x\|_A = 1\right\}.\end{aligned}$$
Faghih-Ahmadi and Gorjizadeh [@fg] in 2016 showed that for $T\in\mathcal{B}_{A^{1/2}}(\mathcal{H})$, we have $$\label{newsemi}
\|T\|_A=\sup\left\{|\langle Tx\mid y\rangle_A|\,;\;x,y\in \mathcal{H},\,\|x\|_{A}=\|y\|_{A}= 1\right\}.$$
We notice here that it may happen that ${\|T\|}_A$ and $\omega_A(T)$ are equal to $+ \infty$ for some $T\in\mathcal{B}(\mathcal{H})$ (see [@feki01]). However, ${\|\cdot\|}_A$ and $\omega_A(\cdot)$ are equivalent seminorms on $\mathcal{B}_{A^{1/2}}(\mathcal{H})$. More precisely, In 2018, Baklouti et al. [@bakfeki01] showed that for every $T\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$, we have $$\label{refine1}
\tfrac{1}{2} \|T\|_A\leq\omega_A(T) \leq \|T\|_A.$$
For the sequel, for any arbitrary operator $T\in {\mathcal B}_A({\mathcal H})$, we write $$\Re_A(T):=\frac{T+T^{\sharp_A}}{2}\;\;\text{ and }\;\;\Im_A(T):=\frac{T-T^{\sharp_A}}{2i}.$$ Recently, in 2019 Zamani [@zamani1 Theorem 2.5] showed that if $T\in\mathcal{B}_{A}(\mathcal{H})$, then $$\begin{aligned}
\label{zm}
\omega_A(T) = \displaystyle{\sup_{\theta \in \mathbb{R}}}{\left\|\Re_A(e^{i\theta}T)\right\|}_A=\displaystyle{\sup_{\theta \in \mathbb{R}}}{\left\|\Im_A(e^{i\theta}T)\right\|}_A.\end{aligned}$$
In 2020, the concept of the $A$-spectral radius of $A$-bounded operators was introduced by the first author in [@feki01] as follows: $$\label{newrad}
r_A(T):=\displaystyle\inf_{n\geq 1}\|T^n\|_A^{\frac{1}{n}}=\displaystyle\lim_{n\to\infty}\|T^n\|_A^{\frac{1}{n}}.$$ Here we want to mention that the proof of the second equality in can also be found in [@feki01 Theorem 1]. Like the classical spectral radius of Hilbert space operators, it was shown in [@feki01] that $r_A(\cdot)$ satisfies the commutativity property, i.e. $$\label{commut}
r_A(TS)=r_A(ST),$$ for all $T,S\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$. For the sequel, if $A=I$, then $\|T\|$, $r(T)$ and $\omega(T)$ denote respectively the classical operator norm, the spectral radius and the numerical radius of an operator $T$.
An operator $T\in\mathcal{B}(\mathcal{H})$ is called $A$-selfadjoint if $AT$ is selfadjoint, i.e., $AT = T^*A$ and it is called $A$-positive if $AT\geq0$. If $T$ is $A$-positive, we will write $T\geq_{A}0$. In recent years, several results covering some classes of operators on a complex Hilbert space $\big(\mathcal{H}, \langle \cdot\mid \cdot\rangle\big)$ were extended to $\big(\mathcal{H}, {\langle \cdot\mid \cdot\rangle}_A\big)$. Of course, the extension is not trivial since many difficulties arise. For instance, as mentioned above, it may happen that ${\|T\|}_A = \infty$ for some $T\in \mathcal{B}(\mathcal{H})$. Moreover, no operator admits an adjoint operator for the semi-inner product ${\langle \cdot\mid \cdot\rangle}_A$. In addition, for $T \in \mathcal{B}_A({\mathcal{H}})$, we have $(T^{\sharp_A})^{\sharp_A}=P_{\overline{\mathcal{R}(A)}}TP_{\overline{\mathcal{R}(A)}}\neq T$. For further details about $A$-numerical radius, interested readers can follow [@bakfeki01; @bakfeki04; @bfeki; @feki03; @zamani1; @NSD] and the references therein.
In this paper, we consider the ${2\times 2}$ operator diagonal matrix $\mathbb{A}=\begin{pmatrix}
A &0\\
0 &A
\end{pmatrix}$. Clearly, $\mathbb{A}\in \mathcal{B}(\mathcal{H}\oplus \mathcal{H})^+$. So, $\mathbb{A}$ induces the following semi-inner product $$\langle x, y\rangle_{\mathbb{A}}= \langle \mathbb{A}x, y\rangle=\langle x_1\mid y_1\rangle_A+\langle x_2\mid y_2\rangle_A,$$ for all $x=(x_1,x_2)\in \mathcal{H}\oplus \mathcal{H}$ and $y=(y_1,y_2)\in \mathcal{H}\oplus \mathcal{H}$.
Recently, several inequalities for the $\mathbb{A}$-numerical radius of $2 \times 2$ operator matrices have been established by P. Bhunia et al. and Rout et al. when $A$ is a positive injective operators (see [@BPN; @rout]). Moreover, different upper and lower bounds of $\mathbb{A}$-numerical radius when $A$ is a positive semidefinite operator has been recently investigated by the first author in [@feki04]. In this article, we will continue working in this direction and we will prove several new $\mathbb{A}$-numerical radius inequalities of certain $2 \times 2$ operator matrices. The inspiration for our investigation comes from [@dolat; @HirKit; @HirKit2]. Further, some of the obtained results cover and extend the recent work of Al-Dolat et al. [@dolat].
Results {#s2}
=======
In this section, we present our results. Throughout this section $\mathbb{A}$ is denoted to be the $2\times 2$ operator diagonal matrix whose each diagonal entry is the positive operator $A$. To prove our first result, the following lemmas are required.
([@feki01])\[ll2020\] Let $T\in\mathcal{B}(\mathcal{H})$ is an $A$-self-adjoint operator. Then, $$\|T\|_{A}=\omega_A(T)=r_A(T).$$
Now, we are in a position to prove our first result in this paper.
\[lm5\]([@feki02]) Let $\mathbb{T}=\begin{pmatrix}
T_{11}&T_{12} \\
T_{21}&T_{22}
\end{pmatrix}$ be such that $T_{ij}\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$ for all $i,j\in\{1,2\}$. Then, $\mathbb{T}\in \mathcal{B}_{\mathbb{A}^{1/2}}(\mathcal{H}\oplus \mathcal{H})$ and $$r_\mathbb{A}\left(\mathbb{T}\right)\leq r\left[\begin{pmatrix}
\|T_{11}\|_A & \|T_{12}\|_A \\
\|T_{21}\|_A & \|T_{22}\|_A
\end{pmatrix}\right].$$
([@bfeki; @feki04])\[lemma1\] Let $P, Q, R, S\in\mathcal{B}_{A^{1/2}}(\mathcal{H})$. Then, the following assertions hold
- $\omega_{\mathbb{A}}\left[\begin{pmatrix}
P & 0 \\
0 & S
\end{pmatrix}\right] = \max\big\{\omega_{A}(P), \omega_{A}(S)\big\}$.
- ${\left\|\begin{pmatrix}
P & 0 \\
0 & S
\end{pmatrix}\right\|}_{\mathbb{A}} = {\left\|\begin{pmatrix}
0 & P \\
S & 0
\end{pmatrix}\right\|}_{\mathbb{A}} = \max\big\{{\|P\|}_{A}, {\|S\|}_{A}\big\}$.
- If $P, Q, R, S\in\mathcal{B}_{A}(\mathcal{H})$, then ${\begin{pmatrix}
P & Q \\
R & S
\end{pmatrix}}^{\sharp_{\mathbb{A}}} = \begin{pmatrix}
P^{\sharp_A} & R^{\sharp_A} \\
Q^{\sharp_A} & S^{\sharp_A}
\end{pmatrix}$.
Now, we are ready to prove the following theorem.
\[thm102\] Let $P, Q, R, S\in \mathcal{B}_{A}(\mathcal{H})$. Then, for $\lambda\in [0, 1]$, we have $$\begin{aligned}
\label{ineq106}
&\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]\\
&\leq \frac{1}{2}\bigg[\omega_A(P)+2\omega_A(S)+\sqrt{\lambda^2 \omega^2_A(P)+\|Q\|_A^2}
+\sqrt{(1-\lambda)^2 \omega^2_A(P)+\|R\|_A^2}\bigg].\end{aligned}$$
Let $\mathbb{T}=\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}$. It is not difficult to see that $\Re_\mathbb{A}(e^{i\theta}\mathbb{T})$ is an $\mathbb{A}$-selfadjoint operator. So, by Lemma \[ll2020\] we have $${\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}=\omega_\mathbb{A}\Big(\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\Big).$$ Then, we see that $$\begin{aligned}
&{\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}\\
&=\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
e^{i\theta}P+e^{-i\theta}P^{\sharp_A}&e^{i\theta}Q+e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R+e^{-i\theta}Q^{\sharp_A}&e^{i\theta}S+e^{-i\theta}S^{\sharp_A}
\end{pmatrix}\right]\\
&\leq\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}\right]\\
&+\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R&0
\end{pmatrix}\right]+\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
0&0 \\
0&e^{i\theta}S+e^{-i\theta}S^{\sharp_A}
\end{pmatrix}\right]\\
&\leq\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R&0
\end{pmatrix}\right]+\omega_A(S),\end{aligned}$$ where the last inequality follows by Lemma \[lemma1\] (i) together with the triangle inequality. Moreover, it can be observed that $$\begin{aligned}
\mathbb{A}\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}&=\begin{pmatrix}
\lambda (e^{i\theta}AP+e^{-i\theta}AP^{\sharp_A})&e^{i\theta}AQ \\
e^{-i\theta}AQ^{\sharp_A}&0
\end{pmatrix}\\
&=\begin{pmatrix}
\lambda \left(e^{i\theta}(P^{\sharp_A})^*A+e^{-i\theta}P^*A\right)&e^{i\theta}(Q^{\sharp_A})^*A \\
e^{-i\theta}Q^*A&0
\end{pmatrix}\\
&=\begin{pmatrix}
\lambda (e^{-i\theta}P^{\sharp_A}+e^{i\theta}P)&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}^*\mathbb{A}.\end{aligned}$$ Hence $\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}$ is $\mathbb{A}$-selfadjoint operator. Similarly one can show that $\begin{pmatrix}
(1-\lambda) (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R&0
\end{pmatrix}$ is $\mathbb{A}$-selfadjoint operator. So by applying Lemma \[ll2020\] we see that $$\begin{aligned}
{\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}
&\leq\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R&0
\end{pmatrix}\right]+\omega_A(S)\\
&=\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
\lambda P^{\sharp_A}&e^{i\theta}I \\
Q^{\sharp_A}&0
\end{pmatrix}\begin{pmatrix}
e^{-i\theta}I&0 \\
\lambda P&Q
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) P^{\sharp_A}&e^{i\theta}I \\
e^{2i\theta}R&0
\end{pmatrix}\begin{pmatrix}
e^{-i\theta}I&0 \\
(1-\lambda) P&e^{-2i\theta}R^{\sharp_A}
\end{pmatrix}\right]+\omega_A(S).\end{aligned}$$ So, by using Lemma \[lm5\] we get $$\begin{aligned}
&{\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}\\
&\leq \frac{1}{2}r\left[\begin{pmatrix}
\lambda \|e^{i\theta}P+e^{-i\theta}P^{\sharp_A}\|_A & \|Q\|_A \\
\|Q\|_A& 0
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r\left[\begin{pmatrix}
(1-\lambda) \|e^{i\theta}P+e^{-i\theta}P^{\sharp_A}\|_A & \|R\|_A \\
\|R\|_A&0
\end{pmatrix}\right]+\omega_A(S)\\
&\leq\frac{1}{2}\bigg[\omega_A(P)+2\omega_A(S)+\sqrt{\lambda^2 \omega_A^2(P)+\|Q\|_A^2}+\sqrt{(1-\lambda)^2 \omega_A^2(P)+\|R\|_A^2}\bigg].\end{aligned}$$ Thus, by taking the supremum over all $\theta\in\mathbb{R}$ in the last inequality and then using \[zm\] we get desired result.
Our second result reads as follows.
\[thm101\] Let $P, Q, R, S\in \mathcal{B}_{A}(\mathcal{H})$. Then, for $\lambda\in [0, 1]$, we have $$\begin{aligned}
&\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]\\
&\leq \frac{1}{2}\left(\|P\|_A+2\omega_A(S)+\sqrt{\|\lambda^2 PP^{\sharp_A}+QQ^{\sharp_A}\|_A}+\sqrt{\|(1-\lambda)^2 PP^{\sharp_A}+R^{\sharp_A}R\|_A}\right).\end{aligned}$$
Let $\mathbb{T}=\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}$. Then by using an argument similar to that used in the proof of the previous theorem we get $$\begin{aligned}
{\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}
&\leq\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R&0
\end{pmatrix}\right]+\omega_A(S)\\
&=\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
\lambda (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{i\theta}Q \\
e^{-i\theta}Q^{\sharp_A}&0
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) (e^{i\theta}P+e^{-i\theta}P^{\sharp_A})&e^{-i\theta}R^{\sharp_A} \\
e^{i\theta}R&0
\end{pmatrix}\right]+\omega_A(S).\end{aligned}$$ So, by using we infer that $$\begin{aligned}
{\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}
&\leq \frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
e^{-i\theta}I&0 \\
\lambda P&Q
\end{pmatrix}\begin{pmatrix}
\lambda P^{\sharp_A}&e^{i\theta}I \\
Q^{\sharp_A}&0
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
e^{-i\theta}I&0 \\
(1-\lambda) P&e^{-2i\theta}R^{\sharp_A}
\end{pmatrix}\begin{pmatrix}
(1-\lambda) P^{\sharp_A}&e^{i\theta}I \\
e^{2i\theta}R&0
\end{pmatrix}\right]+\omega_A(S)
\\
&\leq \frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
\lambda e^{-i\theta}P^{\sharp_A} & I \\
\lambda^2 PP^{\sharp_A}+QQ^{\sharp_A}& \lambda e^{i\theta}P
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r_{\mathbb{A}}\left[\begin{pmatrix}
(1-\lambda) e^{-i\theta}P^{\sharp_A} & I \\
(1-\lambda)^2 PP^{\sharp_A}+R^{\sharp_A}R&(1-\lambda) P
\end{pmatrix}\right]+\omega_A(S).\end{aligned}$$ So, by using Lemma \[lm5\] we get $$\begin{aligned}
&{\left\|\Re_\mathbb{A}(e^{i\theta}\mathbb{T})\right\|}_\mathbb{A}\\
&\leq \frac{1}{2}r\left[\begin{pmatrix}
\lambda \|P\|_A & 1 \\
\|\lambda^2 PP^{\sharp_A}+QQ^{\sharp_A}\|_A& \lambda \|P\|_A
\end{pmatrix}\right]\\
&\;\;+\frac{1}{2}r\left[\begin{pmatrix}
(1-\lambda) \|P\|_A & 1 \\
\|(1-\lambda)^2 PP^{\sharp_A}+R^{\sharp_A}R\|_A&(1-\lambda) \|P\|_A
\end{pmatrix}\right]+\omega_A(S)
\\
&=\frac{1}{2}\bigg[\|P\|_A+2\omega_A(S)+\sqrt{\|\lambda^2 PP^{\sharp_A}+QQ^{\sharp_A}\|_A}+\sqrt{\|(1-\lambda)^2 PP^{\sharp_A}+R^{\sharp_A}R\|_A}\bigg].\end{aligned}$$ So, by taking the supremum over all $\theta\in\mathbb{R}$ in the last inequality and then using \[zm\] we get desired result.
The following Lemma is useful in the sequel. Notice that its assertions has been recently done by Bhunia et al. in [@BPN] and by Rout et al. in [@rout] when $A$ is a positive injective operator.
\[lem100\] Let $T,S\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$. Then,
- $ \omega_\mathbb{A}\left[\begin{pmatrix}
T&S\\
S&T
\end{pmatrix}\right]= \max\{\omega_A(T+S),\omega_A(T-S)\}.$
- $ \omega_\mathbb{A}\left[\begin{pmatrix}
T&-S\\
S&T
\end{pmatrix}\right]= \max\{\omega_A(T+iS),\omega_A(T-iS)\}.$
In order to prove Lemma \[lem100\] we need the following result.
\[weak\]([@bfeki]) Let $T\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$. Then, $$\omega_A(U^{\sharp_A}TU)=\omega_A(T),$$ for any $A$-unitary operator $U\in\mathcal{B}_A(\mathcal{H})$.
Now, we state the proof of Lemma \[lem100\].
(i)Let $\mathbb{U}=\frac{1}{\sqrt{2}}\begin{pmatrix}
I & I\\
-I & I
\end{pmatrix}$. By using Lemma \[lemma1\] (iii), we see that $$\mathbb{U}^{\sharp_\mathbb{A}}=\frac{1}{\sqrt{2}}\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &-P_{\overline{\mathcal{R}(A)}}\\
P_{\overline{\mathcal{R}(A)}} &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}.$$ So, by using the fact that $AP_{\overline{\mathcal{R}(A)}}=A$, we can verify that $\|\mathbb{U}x\|_{\mathbb{A}}=\|\mathbb{U}^{\sharp_{\mathbb{A}}}x\|_{\mathbb{A}}=\|x\|_{\mathbb{A}}$ for all $x\in \mathcal{H}\oplus \mathcal{H}$. Hence $\mathbb{U}$ is $\mathbb{A}$-unitary. So, by Lemma \[weak\] we see that $$\begin{aligned}
\omega_\mathbb{A}\left[\begin{pmatrix}
T&S\\
S&T
\end{pmatrix}\right]
&= \omega_\mathbb{A}\left[\mathbb{U}^{\sharp_\mathbb{A}}\begin{pmatrix}
T&S\\
S&T
\end{pmatrix}\mathbb{U}\right]\\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &0\\
0 &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}
\begin{pmatrix}
T-S &0\\
0&T+S
\end{pmatrix}\right]\\
&=\omega_{\mathbb{A}}\left[
\begin{pmatrix}
T-S &0\\
0&T+S
\end{pmatrix}\right]\\
&=\max\{\omega_A(T+S),\omega_A(T-S)\},
\end{aligned}$$ where the last equality follows from Lemma \[lemma1\] (i).
0.1 cm (ii)By considering the $\mathbb{A}$-unitary operator $\mathbb{U}=\frac{1}{\sqrt{2}}\begin{pmatrix}
I & iI\\
iI & I
\end{pmatrix}$ and proceeding as above we get the desired result.
As an application of of Lemma \[lem100\] (i) together with Theorems \[thm102\] and \[thm101\] we state the following corollary.
Let $P,Q\in \mathcal{B}_{A}(\mathcal{H})$. Then for all $\lambda\in [0,1]$, we have $$\omega_A\left(P\pm Q\right)\leq \min\{\mu,\nu \},$$ where $$\begin{aligned}
\mu
& =\omega_A(P)+\frac{1}{2}\left(\|P\|_A+\sqrt{\|\lambda^2 PP^{\sharp_A}+QQ^{\sharp_A}\|_A}+\sqrt{\|(1-\lambda)^2 PP^{\sharp_A}+Q^{\sharp_A}Q\|_A}\right),\end{aligned}$$ and $$\begin{aligned}
\nu
& =\frac{3}{2}\omega_A(P)+\frac{1}{2}\bigg[\sqrt{\lambda^2 \omega^2_A(P)+\|Q\|_A^2}
+\sqrt{(1-\lambda)^2 \omega^2_A(P)+\|Q\|_A^2}\bigg].\end{aligned}$$
Now we state the following theorem.
\[them10\] Let $\mathbb{T}=\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}$ be such that $P, Q, R, S\in \mathcal{B}_{A}(\mathcal{H})$. Then, $$\begin{aligned}
\label{upper1}
\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]
&\leq \max\{\omega_A(P), \omega_A(S)\}+\frac{\omega_A(Q+R)+\omega_A(Q-R)}{2}.\end{aligned}$$
Using triangle inequality we have $$\begin{aligned}
\label{sho}
\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]&\leq \omega_{\mathbb{A}}\left[\begin{pmatrix}
P&0 \\
0&S
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\right]\nonumber\\
&=\max\{\omega_A(P), \omega_A(S)\}+\omega_{\mathbb{A}}\left[\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\right],\end{aligned}$$ where the last equality follows by Lemma \[lemma1\] (i). Let $\mathbb{U}=\frac{1}{\sqrt{2}}\begin{pmatrix}
I & -I\\
I & I
\end{pmatrix}$. By proceeding as above we prove that $\mathbb{U}$ is $\mathbb{A}$-unitary. So, by Lemma \[weak\] we see that $$\begin{aligned}
\omega_\mathbb{A}\left[\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\right]
&= \omega_\mathbb{A}\left[\mathbb{U}^{\sharp_\mathbb{A}}\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\mathbb{U}\right]\nonumber\\
&=\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &0\\
0 &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}
\begin{pmatrix}
R+Q &-R+Q\\
R-Q &-R-Q
\end{pmatrix}\right]\nonumber\\
&=\frac{1}{2}\omega_{\mathbb{A}}\left[
\begin{pmatrix}
R+Q &-R+Q\\
R-Q &-R-Q
\end{pmatrix}\right]\nonumber\\
&\leq\frac{1}{2}\omega_{\mathbb{A}}\left[
\begin{pmatrix}
R+Q &0\\
0 &-R-Q
\end{pmatrix}\right]
+\frac{1}{2}\omega_{\mathbb{A}}\left[
\begin{pmatrix}
0 &-R+Q\\
R-Q &0
\end{pmatrix}\right]
\nonumber\\
&=\frac{1}{2}\Big(\omega_A(Q+R)+\omega_A(Q-R)\Big),
\end{aligned}$$ where the last equality follows from Lemma \[lem100\]. So, $$\label{qq2}
\omega_\mathbb{A}\left[\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\right]\leq \frac{1}{2}\Big(\omega_A(Q+R)+\omega_A(Q-R)\Big).$$ Combining together with yields to the desired result.
Very recently Theorem \[them10\] has been proved by Rout et al. in [@rout] when $A$ is a positive injective operator. Moreover, our approach here is different from theirs.
Our next objective is to present an improvement of the inequality . To do this, we need the following lemma.
\[p4200\]([@HORN]) Let $T=[t_{ij}]\in M_n(\mathbb{C})$ be such that $t_{ij}\geq 0$ for all $i, j=1, 2,\ldots, n$. Then $$\omega(T)= \frac{1}{2}r([t_{ij}+t_{ji}]).$$
\[them100\] Let $\mathbb{T}=\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}$ be such that $P, Q, R, S\in \mathcal{B}_{A}(\mathcal{H})$. Then, $$\begin{aligned}
&\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]\\
&\leq \frac{1}{2}\bigg(\omega_A(P)+\omega_A(S)+\sqrt{(\omega_A(P)-\omega_A(S))^2+(\omega_A(Q+R)+\omega_A(Q-R))^2}\bigg).\end{aligned}$$ Moreover, the inequality is sharper than the inequality .
It follows from [@feki04 Theorem 2.3] that $$\begin{aligned}
\omega_\mathbb{A}\left(\mathbb{T}\right)
&\leq\omega\left[\begin{pmatrix}
\omega_\mathbb{A}\left({P}\right) & \omega_{\mathbb{A}}\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\\
\omega_{\mathbb{A}}\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix} & \omega_\mathbb{A}\left({S}\right)
\end{pmatrix} \right] \\
&=r\left[\begin{pmatrix}
\omega_\mathbb{A}\left({P}\right) & \omega_{\mathbb{A}}\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}\\
\omega_{\mathbb{A}}\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix} & \omega_\mathbb{A}\left({S}\right)
\end{pmatrix}\right]\\
&=\frac{1}{2}\bigg[\omega_A(P)+\omega_A(S)+\sqrt{(\omega_A(P)-\omega_A(S))^2+4\omega^2_{\mathbb{A}}\begin{pmatrix}
0&Q \\
R&0
\end{pmatrix}}\bigg],
\end{aligned}$$ where the last equality follows from Lemma \[p4200\]. So, by applying we get $$\begin{aligned}
&\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]\\
&\leq \frac{1}{2}\bigg[\omega_A(P)+\omega_A(S)+\sqrt{(\omega_A(P)-\omega_A(S))^2+(\omega_A(Q+R)+\omega_A(Q-R))^2}\bigg].\end{aligned}$$ This proves the desired inequality. Moreover, it can be observed that $$\begin{aligned}
& \max\{\omega_A(P), \omega_A(S)\}+\frac{\omega_A(Q+R)+\omega_A(Q-R)}{2}\\
&=\frac{\omega_A(P)+\omega_A(S)+|\omega_A(P)-\omega_A(S)|}{2} + \frac{\omega_A(Q+R)+ \omega_A(Q-R)}{2}\\
&\geq \frac{1}{2}\bigg(\omega_A(P)+\omega_A(S)+\sqrt{(\omega_A(P)-\omega_A(S))^2+(\omega_A(Q+R)+\omega_A(Q-R))^2}\bigg).\end{aligned}$$ Hence, the proof is complete.
As a consequence of Theorem \[them100\] we state the following corollary.
\[cor100\] Let $P, Q, R, S\in \mathcal{B}_{A}(\mathcal{H})$. Then, $$\begin{aligned}
\label{ineq103}
r_A(PQ+RS)
&\leq \frac{1}{2}\big[\omega_A(PQ)+\omega_A(RS)\big]\\
&+\frac{1}{2}\sqrt{\big[\omega_A(PQ)-\omega_A(RS)\big]^2+\big[\omega_A(QR+PS)+\omega_A(QR-PS)\big]^2}.\end{aligned}$$
In order to prove Corollary \[cor100\], we need the following lemma.
\[kkkk2020\] If $T\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$, then $$r_A(T)\leq \omega_A(T).$$
Now we are in a position to establish Corollary \[cor100\].
It can be observed that $$\begin{aligned}
r_A(PQ+RS)
&=r_{\mathbb{A}}\left[\begin{pmatrix}
PQ+RS&0 \\
0&0
\end{pmatrix}\right]\\
&=r_{\mathbb{A}}\left[\begin{pmatrix}
P&R \\
0&0
\end{pmatrix}\begin{pmatrix}
Q&0 \\
S&0
\end{pmatrix}\right]\\
&=r_{\mathbb{A}}\left[\begin{pmatrix}
Q&0 \\
S&0
\end{pmatrix}\begin{pmatrix}
P&R \\
0&0
\end{pmatrix}\right]\quad (\text{by }\,\eqref{commut})\\
&=r_{\mathbb{A}}\left[\begin{pmatrix}
QP&QR \\
SP&SR
\end{pmatrix}\right]\\
&\leq \omega_{\mathbb{A}}\left[\begin{pmatrix}
QP&QR \\
SP&SR
\end{pmatrix}\right]\quad(\text{by Lemma } \ref{kkkk2020}).\end{aligned}$$ So, by applying Theorem \[them100\] we reach the required result.
Notice that by letting $Q=S=I$ in Corollary \[cor100\] we get $$\begin{aligned}
r_A(P+R)
&\leq \frac{1}{2}\big[\omega_A(P)+\omega_A(R)\big]\\
&+\frac{1}{2}\sqrt{(\omega_A(P)-\omega_A(R))^2+(\omega_A(P+R)+\omega_A(P-R))^2}.\end{aligned}$$
To establish further upper bounds for the $\mathbb{A}$-numerical radius of the operator matrix $\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}$, we need the following lemmas.
\[lmm05\]([@feki02]) Let $\mathbb{T}=\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}$ be such that $P, Q, R, S\in \mathcal{B}_{A^{1/2}}(\mathcal{H})$. Then, $\mathbb{T}\in \mathcal{B}_{\mathbb{A}^{1/2}}(\mathcal{H}\oplus \mathcal{H})$ and $$\|\mathbb{T}\|_\mathbb{A}\leq \left\|\begin{pmatrix}
\|P\|_A & \|Q\|_A \\
\|R\|_A & \|S\|_A
\end{pmatrix}\right\|.$$
\[a5so\] Let $T,S\in \mathcal{B}_{A}(\mathcal{H})$. Then, $$\|T^{\sharp_A}S\|_A=\|S^{\sharp_A}T\|_A.$$
By using the fact that $P_{\overline{\mathcal{R}(A)}}A=AP_{\overline{\mathcal{R}(A)}}=A$ together with , we see that $$\begin{aligned}
\|T^{\sharp_A}S\|_A
&=\|S^{\sharp_A}P_{\overline{\mathcal{R}(A)}}TP_{\overline{\mathcal{R}(A)}}\|_A\\
&=\sup\left\{|\langle AP_{\overline{\mathcal{R}(A)}}x\mid (S^{\sharp_A}P_{\overline{\mathcal{R}(A)}}T)^{\sharp_A}y\rangle|\,;\;x,y\in \mathcal{H},\,\|x\|_{A}=\|y\|_{A}= 1\right\}\\
&=\sup\left\{|\langle S^{\sharp_A}P_{\overline{\mathcal{R}(A)}}Tx\mid y\rangle_A|\,;\;x,y\in \mathcal{H},\,\|x\|_{A}=\|y\|_{A}= 1\right\}\\
&=\sup\left\{|\langle AP_{\overline{\mathcal{R}(A)}}Tx\mid Sy\rangle|\,;\;x,y\in \mathcal{H},\,\|x\|_{A}=\|y\|_{A}= 1\right\}\\
&=\sup\left\{|\langle S^{\sharp_A}Tx\mid y\rangle_A|\,;\;x,y\in \mathcal{H},\,\|x\|_{A}=\|y\|_{A}= 1\right\}\\
&=\|S^{\sharp_A}T\|_A.\end{aligned}$$ This proves the desired result.
Now, we are in a position to state the following theorem.
\[theorem:upper 3\] Let $P,Q,R,S\in \mathcal{B}_{A}(\mathcal{H})$. Then, $$\omega_\mathbb{A}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]\leq \min\{\mu,\nu \},$$ where $$\begin{aligned}
\mu
& =\frac{\sqrt{2}}{2}\sqrt{\|P\|_A^2+\|Q\|_A^2+\sqrt{(\|P\|_A^2-\|Q\|_A^2)^2 +4\|P^{\sharp_A}Q\|_A^2} } \\
&\quad\quad\quad+\frac{\sqrt{2}}{2}\sqrt{\|R\|_A^2+\|S\|_A^2+\sqrt{(\|R\|_A^2-\|S\|_A^2)^2 +4\|S^{\sharp_A}R\|_A^2} },\end{aligned}$$ and $$\begin{aligned}
\nu
& =\frac{\sqrt{2}}{2}\sqrt{\|P\|_A^2+\|R\|_A^2+\sqrt{(\|P\|_A^2-\|R\|_A^2)^2+4\|PR^{\sharp_A}\|_A^2 }} \\
&\quad\quad\quad+\frac{\sqrt{2}}{2}\sqrt{\|Q\|_A^2+\|S\|_A^2+\sqrt{(\|Q\|_A^2-\|S\|_A^2)^2 +4\|SQ^{\sharp_A}\|_A^2} }.\end{aligned}$$
We first prove that $$\label{ffirst}
\omega_\mathbb{A}\left[\begin{pmatrix}
P&Q \\
0&0
\end{pmatrix}\right]\leq \frac{\sqrt{2}}{2}\sqrt{\|P\|_A^2+\|Q\|_A^2+\sqrt{(\|P\|_A^2-\|Q\|_A^2)^2+4\|P^{\sharp_A}Q\|_A^2 }}\,.$$ By using together with we see that $$\begin{aligned}
\omega_\mathbb{A}\left[\begin{pmatrix}
P&Q\\
0& 0
\end{pmatrix}\right]
&\leq \left\|\begin{pmatrix}
P&Q\\
0 & 0
\end{pmatrix}\right\|_\mathbb{A}\\
&=\left\|\begin{pmatrix}
P&Q\\
0& 0
\end{pmatrix}^{\sharp_\mathbb{A}}
\begin{pmatrix}
P&Q\\
0& 0
\end{pmatrix}\right\|_\mathbb{A}^{\frac{1}{2}}\\
&=\left\|\begin{pmatrix}
P^{\sharp_A}& 0\\
Q^{\sharp_A}& 0
\end{pmatrix}\begin{pmatrix}
P&Q\\
0& 0
\end{pmatrix}\right\|_\mathbb{A}^{\frac{1}{2}}\\
&=\left\|\begin{pmatrix}
P^{\sharp_A}P & P^{\sharp_A}Q\\
Q^{\sharp_A}P& Q^{\sharp_A}Q
\end{pmatrix}\right\|_\mathbb{A}^{\frac{1}{2}}.\end{aligned}$$ So, by using Lemma \[lmm05\] together with Lemma \[a5so\] we see that $$\begin{aligned}
\omega_\mathbb{A}^2\left[\begin{pmatrix}
P&Q\\
0& 0
\end{pmatrix}\right]
&\leq \left\|\begin{pmatrix}
\|P^{\sharp_A}P\|_A & \|P^{\sharp_A}Q\|_A\\
\|Q^{\sharp_A}P\|_A& \|Q^{\sharp_A}Q\|_A
\end{pmatrix}\right\|\\
&=\left\|\begin{pmatrix}
\|P\|_A^2& \|P^{\sharp_A}Q\|_A\\
\|P^{\sharp_A}Q\|_A& \|Q\|_A^2
\end{pmatrix}\right\|\\
&=r\left[\begin{pmatrix}
\|P\|_A^2& \|P^{\sharp_A}Q\|_A\\
\|P^{\sharp_A}Q\|_A& \|Q\|_A^2
\end{pmatrix}\right]\\
&= \frac{1}{2}\left(\|P\|_A^2+\|Q\|_A^2+\sqrt{(\|P\|_A^2-\|Q\|_A^2)^2+4\|P^{\sharp_A}Q\|_A^2}\right).\end{aligned}$$ This proves . Let $\mathbb{U}=\begin{pmatrix}
0&I \\
I&0
\end{pmatrix}.$ In view of Lemma \[lemma1\] (iii) we have $\mathbb{U}\in \mathcal{B}_{\mathbb{A}}(\mathcal{H}\oplus \mathcal{H})$ and $\mathbb{U}^{\sharp_{\mathbb{A}}}=\begin{pmatrix}
0&P_{\overline{\mathcal{R}(A)}}\\
P_{\overline{\mathcal{R}(A)}}&0
\end{pmatrix}.$ Further, it can be seen that $\mathbb{U}$ is $\mathbb{A}$-unitary operator. So, by using Lemma \[weak\] together with we get $$\begin{aligned}
\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]
& \leq\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\begin{pmatrix}
0 &0\\
R&S
\end{pmatrix}\right] \\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\mathbb{U}^{\sharp_{\mathbb{A}}}\begin{pmatrix}
0 &0\\
R&S
\end{pmatrix}\mathbb{U}\right] \\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &0\\
0 &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}
\begin{pmatrix}
S &R\\
0 &0
\end{pmatrix}\right]\\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[
\begin{pmatrix}
S &R\\
0 &0
\end{pmatrix}\right]\\
& =\frac{\sqrt{2}}{2}\sqrt{\|P\|_A^2+\|Q\|_A^2+\sqrt{(\|P\|_A^2-\|Q\|_A^2)^2 +4\|P^{\sharp_A}Q\|_A^2} } \\
&+\frac{\sqrt{2}}{2}\sqrt{\|R\|_A^2+\|S\|_A^2+\sqrt{(\|R\|_A^2-\|S\|_A^2)^2 +4\|S^{\sharp_A}R\|_A^2} }.\end{aligned}$$ Now, since $\omega_\mathbb{A}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]=\omega_\mathbb{A}\left[\begin{pmatrix}
P^{\sharp_A} & R^{\sharp_A} \\
Q^{\sharp_A} & S^{\sharp_A}
\end{pmatrix}\right]$, then by using similar arguments as above we obtain $$\begin{aligned}
&\omega_\mathbb{A}\left[\begin{pmatrix}
P&Q \\
R&S
\end{pmatrix}\right]\\
& \leq\frac{\sqrt{2}}{2}\sqrt{\|P^{\sharp_A}\|_A^2+\|R^{\sharp_A}\|_A^2+\sqrt{(\|P^{\sharp_A}\|_A^2-\|R^{\sharp_A}\|_A^2)^2+4\|(P^{\sharp_A})^{\sharp_A}R^{\sharp_A}\|_A^2 }} \\
&+\frac{\sqrt{2}}{2}\sqrt{\|Q^{\sharp_A}\|_A^2+\|S^{\sharp_A}\|_A^2+\sqrt{(\|Q^{\sharp_A}\|_A^2-\|S^{\sharp_A}\|_A^2)^2 +4\|(S^{\sharp_A})^{\sharp_A}Q^{\sharp_A}\|_A^2} }\\
& =\frac{\sqrt{2}}{2}\sqrt{\|P\|_A^2+\|R\|_A^2+\sqrt{(\|P\|_A^2-\|R\|_A^2)^2+4\|PR^{\sharp_A}\|_A^2 }} \\
&+\frac{\sqrt{2}}{2}\sqrt{\|Q\|_A^2+\|S\|_A^2+\sqrt{(\|Q\|_A^2-\|S\|_A^2)^2 +4\|SQ^{\sharp_A}\|_A^2} }\,.\end{aligned}$$ Hence, the proof is complete.
Notice that Theorem \[theorem:upper 3\] is stated without any details by Rout et al. in [@rout] when $A$ is a positive injective operator.
Our next result reads as follows.
Let $P, Q\in\mathcal{B}_A(\mathcal{H}).$ Then $$\omega_{\mathbb{A}}\left[\begin{pmatrix}
P & Q\\
O & O
\end{pmatrix}\right]\geq \frac{1}{2}\max\{\alpha,\beta\},$$
where $\alpha=\omega_A(P+ Q)+\omega_A(P- Q)$ and $\beta=\omega_A(P+ iQ)+\omega_A(P- iQ)$.
Let $\mathbb{U}=\begin{pmatrix}
O & I\\I & O
\end{pmatrix}$. By using Lemma \[lemma1\] (iii), we see that $$\mathbb{U}^{\sharp_\mathbb{A}}=\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &0\\
0 &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}\mathbb{U}=\begin{pmatrix}
0 & P_{\overline{\mathcal{R}(A)}}\\
P_{\overline{\mathcal{R}(A)}} & 0
\end{pmatrix}.$$ So, by using the fact that $AP_{\overline{\mathcal{R}(A)}}=A$, we can verify that $\|\mathbb{U}x\|_{\mathbb{A}}=\|\mathbb{U}^{\sharp_{\mathbb{A}}}x\|_{\mathbb{A}}=\|x\|_{\mathbb{A}}$ for all $x\in \mathcal{H}\oplus \mathcal{H}$. Hence $\mathbb{U}$ is $\mathbb{A}$-unitary. So, using Lemma \[lem100\] (i) we observes that $$\begin{aligned}
\max\{\omega_A(P+Q),\omega_A(P-Q)\}=&\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
Q &P
\end{pmatrix}\right]\\
& \leq\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\begin{pmatrix}
0 &0\\
Q&P
\end{pmatrix}\right] \\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\mathbb{U}^{\sharp_{\mathbb{A}}}\begin{pmatrix}
0 &0\\
Q&P
\end{pmatrix}\mathbb{U}\right] \\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &0\\
0 &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}
\begin{pmatrix}
P &Q\\
0 &0
\end{pmatrix}\right]\\
&=\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&Q\\
0 &0
\end{pmatrix}\right]+\omega_{\mathbb{A}}\left[
\begin{pmatrix}
P &Q\\
0 &0
\end{pmatrix}\right]\\
=&2\omega_{\mathbb{A}}\left[
\begin{pmatrix}
P &Q\\
0 &0
\end{pmatrix}\right].
\end{aligned}$$
On the other hand, by considering the $\mathbb{A}$-unitary operator $\mathbb{V}=\begin{pmatrix}
I & 0\\
0 & -I
\end{pmatrix}$ and proceeding as above we see that $$\begin{aligned}
\max\{\omega_A(P+iQ),\omega_A(P-iQ)\}=&\omega_{\mathbb{A}}\left[\begin{pmatrix}
P&-Q\\
Q &P
\end{pmatrix}\right]\quad(\text{by Lemma } \ref{lem100}\;(ii))\\
&\leq 2\omega_{\mathbb{A}}\left[
\begin{pmatrix}
P &Q\\
0 &0
\end{pmatrix}\right].
\end{aligned}$$ Hence we get our desired result.
In order to prove the general situation, we need the following lemma.
([@feki004])\[n1\] Let $T, S\in\mathcal{B}_{A}(\mathcal{H})$. Then, $$\begin{aligned}
\label{SK1}
\omega_A(TS\pm ST^{\sharp_A}) \leq2\|T\|_A\,\omega_A(S).\end{aligned}$$
Based on Lemma \[lem100\], the forthcoming theorem is an application of the inequality .
Let $P, Q,R,S\in\mathcal{B}_A(\mathcal{H})$ be such that $\mathcal{N}(A)^\perp$ is an invariant subspace for $P,Q,R$ and $S$. Then $$\omega_\mathbb{A}\left[\begin{pmatrix}
P & Q\\
R & S
\end{pmatrix}\right]\geq \frac{1}{2}\max\{\mu,\nu\}.$$ where $$\mu=\max\{\omega_A(Q+R+S+P) ,\omega_A(Q+R-S-P) \},$$ and $$\nu=\max\big\{\omega_A\big(Q-R+i(S+P)\big) ,\omega_A\big(Q-R-i(S+P)\big)\big\}.$$
Let $\mathbb{U}=\begin{pmatrix}
O & I\\I & O
\end{pmatrix}$. By using Lemma \[lemma1\] (iii), we see that $$\mathbb{U}^{\sharp_\mathbb{A}}=\begin{pmatrix}
P_{\overline{\mathcal{R}(A)}} &0\\
0 &P_{\overline{\mathcal{R}(A)}}
\end{pmatrix}\mathbb{U}=\begin{pmatrix}
0 &P_{\overline{\mathcal{R}(A)}}\\
P_{\overline{\mathcal{R}(A)}} &0
\end{pmatrix}.$$ So, by using the fact that $AP_{\overline{\mathcal{R}(A)}}=A$, we can verify that $\|\mathbb{U}x\|_{\mathbb{A}}=\|\mathbb{U}^{\sharp_{\mathbb{A}}}x\|_{\mathbb{A}}=\|x\|_{\mathbb{A}}$ for all $x\in \mathcal{H}\oplus \mathcal{H}$. Hence $\mathbb{U}$ is $\mathbb{A}$-unitary. Moreover, clearly $(\mathbb{U}^{\sharp_{\mathbb{A}}})^{\sharp_{\mathbb{A}}}=\mathbb{U}^{\sharp_{\mathbb{A}}}$. Now, let $\mathbb{T}=\begin{pmatrix}
P & Q\\
R & S
\end{pmatrix}$. Since, $\mathcal{N}(A)^\perp$ is an invariant subspace for $P,Q,R$ and $S$, then $PP_{\overline{\mathcal{R}(A)}}=P_{\overline{\mathcal{R}(A)}}P$, $QP_{\overline{\mathcal{R}(A)}}=P_{\overline{\mathcal{R}(A)}}Q$, $RP_{\overline{\mathcal{R}(A)}}=P_{\overline{\mathcal{R}(A)}}R$ and $SP_{\overline{\mathcal{R}(A)}}=P_{\overline{\mathcal{R}(A)}}S$. So, a short calculation shows that $$\mathbb{U}^{\sharp_{\mathbb{A}}}\mathbb{T}^{\sharp_{\mathbb{A}}}+\mathbb{T}^{\sharp_{\mathbb{A}}}(\mathbb{U}^{\sharp_{\mathbb{A}}})^{\sharp_{\mathbb{A}}}=\begin{pmatrix}
Q^{\sharp_A}+R^{\sharp_A} & S^{\sharp_A}+P^{\sharp_A}\\
S^{\sharp_A}+P^{\sharp_A} & Q^{\sharp_A}+R^{\sharp_A}
\end{pmatrix} =\begin{pmatrix}
Q+R & S+P\\
S+P & Q+R
\end{pmatrix}^{\sharp_{\mathbb{A}}}.$$ By using Lemma \[n1\] we see that $$\omega_{\mathbb{A}}(\mathbb{U}^{\sharp_{\mathbb{A}}}\mathbb{T}^{\sharp_{\mathbb{A}}}+\mathbb{T}^{\sharp_{\mathbb{A}}}(\mathbb{U}^{\sharp_{\mathbb{A}}})^{\sharp_{\mathbb{A}}})
\leq 2\omega_{\mathbb{A}}(\mathbb{T}^{\sharp_{\mathbb{A}}}),$$ which, in turn, implies that $$\begin{aligned}
\omega_{\mathbb{A}}\left[\begin{pmatrix}
P & Q\\
R & S
\end{pmatrix} \right]
&\geq \frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
Q+R & S+P\\
S+P & Q+R
\end{pmatrix} \right] \\
&=\frac{1}{2}\max\{\omega_A(Q+R+S+P) ,\omega_A(Q+R-S-P) \}:=\frac{1}{2}\mu,\end{aligned}$$ where the last equality follows by applying Lemma \[lem100\] (i). On the other hand, by choosing $\mathbb{U}=\begin{pmatrix}
O & I\\-I & O
\end{pmatrix}$ and proceeding as above we obtain $$\begin{aligned}
\omega_{\mathbb{A}}\left[\begin{pmatrix}
P & Q\\
R & S
\end{pmatrix} \right]
&\geq \frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
Q-R & -(S+P)\\
S+P & Q-R
\end{pmatrix} \right] \\
&=\frac{1}{2}\max\big\{\omega_A\big(Q-R+i(S+P)\big) ,\omega_A\big(Q-R-i(S+P)\big)\big\}:=\frac{1}{2}\nu,\end{aligned}$$ where the last equality follows by applying Lemma \[lem100\] (ii). This completes the proof of the theorem.
In the next result, Lemma \[lem100\] enables us to present another application of the inequality .
\[app1\] Let $P, Q\in\mathcal{B}_A(\mathcal{H})$ be such that $\mathcal{N}(A)^\perp$ is an invariant subspace for $P$ and $Q$. Then $$\omega_\mathbb{A}\left[\begin{pmatrix}
P & Q\\
O & O
\end{pmatrix}\right]\geq \frac{1}{2}\max\left\{\omega_A(P+ iQ),\omega_A(P- iQ)\right\}.$$
Let $\mathbb{T}=\begin{pmatrix}
P & Q\\
0 & 0
\end{pmatrix}$ and $\mathbb{U}=\begin{pmatrix}
O & -I\\I & O
\end{pmatrix}$. It is not difficult to verify that $\mathbb{U}$ is $\mathbb{A}$-unitary. Moreover, by using the fact that $\mathcal{N}(A)^\perp$ is an invariant subspace for $P$ and $Q$, it can be seen that $$\mathbb{U}^{\sharp_{\mathbb{A}}}\mathbb{T}^{\sharp_{\mathbb{A}}}+\mathbb{T}^{\sharp_{\mathbb{A}}}(\mathbb{U}^{\sharp_{\mathbb{A}}})^{\sharp_{\mathbb{A}}}=\begin{pmatrix}
Q^{\sharp_A} & P^{\sharp_A}\\
-P^{\sharp_A} & Q^{\sharp_A}
\end{pmatrix} =\begin{pmatrix}
Q & -P\\
P & Q
\end{pmatrix}^{\sharp_{\mathbb{A}}}.$$ By using Lemma \[n1\] we see that $$\omega_{\mathbb{A}}(\mathbb{U}^{\sharp_{\mathbb{A}}}\mathbb{T}^{\sharp_{\mathbb{A}}}+\mathbb{T}^{\sharp_{\mathbb{A}}}(\mathbb{U}^{\sharp_{\mathbb{A}}})^{\sharp_{\mathbb{A}}})
\leq 2\omega_{\mathbb{A}}(\mathbb{T}^{\sharp_{\mathbb{A}}}),$$ which, in turn, implies that $$\begin{aligned}
\omega_{\mathbb{A}}\left[\begin{pmatrix}
P & Q\\
0 & 0
\end{pmatrix} \right]
&\geq \frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
Q & -P\\
P & Q
\end{pmatrix}^{\sharp_{\mathbb{A}}} \right] \\
&=\frac{1}{2}\omega_{\mathbb{A}}\left[\begin{pmatrix}
Q & -P\\
P & Q
\end{pmatrix}\right] \\
&=\frac{1}{2}\max\left\{\omega_A(P+ iQ),\omega_A(P- iQ)\right\},\end{aligned}$$ where the last equality follows by applying Lemma \[lem100\](ii).
As an application of the above theorem, we can derive the following $A$-numerical radius inequality.
\[f12\] Let $T\in \mathcal{B}_A(\mathcal{H})$ be such $\mathcal{N}(A)^\perp$ is an invariant subspace for $T$. Then $$\omega_A(T)\leq 2\min\left\{\omega_\mathbb{A}\left[\begin{pmatrix}
\Re_A(T) & 0\\
\Im_A(T) & O
\end{pmatrix}\right], \omega_\mathbb{A}\left[\begin{pmatrix}
0 & -i\Im_A(T)\\
\Re_A(T) & 0
\end{pmatrix}\right] \right\}.$$
To prove Theorem \[f12\], we need the following lemma.
\[lemf\] Let $T,S\in \mathcal{B}_A(\mathcal{H})$ and $\mathbb{A}=\begin{pmatrix}
A &0\\
0 &A
\end{pmatrix}$. Then $$\begin{aligned}
\omega_A(T\pm iS)\leq 2\,\omega_{\mathbb{A}}\left[\begin{pmatrix}
0&T\\
iS &0
\end{pmatrix}\right].\end{aligned}$$
Let $\mathbb{X}=\begin{pmatrix}
I &I\\
0 &0
\end{pmatrix}$ and $\mathbb{Y}=\begin{pmatrix}
0 &T\\
S &0
\end{pmatrix}$. It can be observed that $$\mathbb{X}\mathbb{Y}\mathbb{X}^{\sharp_\mathbb{A}}=\begin{pmatrix}
TP_{\overline{\mathcal{R}(A)}}+ SP_{\overline{\mathcal{R}(A)}} &0\\
0 &0
\end{pmatrix},$$ and $$\begin{aligned}
\|\mathbb{X}\|_{\mathbb{A}}^2=\|\mathbb{X}\mathbb{X}^{\sharp_\mathbb{A}}\|_{\mathbb{A}}
&=\left\|\begin{pmatrix}
2P_{\overline{\mathcal{R}(A)}}&0\\
0 &0
\end{pmatrix}\right\|_{\mathbb{A}}\\
&=2\|P_{\overline{\mathcal{R}(A)}}\|_A=2.\end{aligned}$$ So, by using the fact that $AP_{\overline{\mathcal{R}(A)}}=A$ we see that $$\begin{aligned}
\label{inff}
\omega_A(T+S)
&=\omega_A(TP_{\overline{\mathcal{R}(A)}}+ SP_{\overline{\mathcal{R}(A)}})\\
&=\omega_\mathbb{A}\left[\begin{pmatrix}
TP_{\overline{\mathcal{R}(A)}}+ SP_{\overline{\mathcal{R}(A)}} &0\\
0 &0
\end{pmatrix}\right]\\
& = \omega_\mathbb{A}(\mathbb{X}\mathbb{Y}\mathbb{X}^{\sharp_\mathbb{A}})\quad(\text{by Lemma } \ref{lemma1})\nonumber\\
&\leq \|\mathbb{X}\|_{\mathbb{A}}^2\omega_\mathbb{A}(\mathbb{Y})\quad(\text{by \cite[Lemma 4.4.]{zamani1}})\nonumber\\
&= 2\omega_\mathbb{A}(\mathbb{Y}).\end{aligned}$$ This gives $$\label{c2}
\omega_A(T+S)\leq2\omega_\mathbb{A}\left[\begin{pmatrix}
0 &T\\
S &0
\end{pmatrix}\right].$$ By replacing $S$ by $-S$ in and then using the fact that $\omega_\mathbb{A}\left[\begin{pmatrix}
0 &T\\
S &0
\end{pmatrix}\right]=\omega_\mathbb{A}\left[\begin{pmatrix}
0 &T\\
-S &0
\end{pmatrix}\right]$ we get $$\label{5arita}
\omega_A(T\pm S)\leq 2\,\omega_{\mathbb{A}}\left[\begin{pmatrix}
0&T\\
S &0
\end{pmatrix}\right].$$ Finally, by replacing $S$ by $iS$ in , we reach the desired results.
Now we are ready to prove Theorem \[f12\]. The following theorem
Clearly $T$ can written as $T=\Re_A(T)+i\Im_A(T)$ where $$\Re_A(T):=\frac{T+T^{\sharp_A}}{2}\;\;\text{ and }\;\;\Im_A(T):=\frac{T-T^{\sharp_A}}{2i}.$$ So, $T^{\sharp_A}=[\Re_A(T)]^{\sharp_A}-i[\Im_A(T)]^{\sharp_A}$. Moreover, a short calculation reveals that $(T^{\sharp_A})^{\sharp_A}=[\Re_A(T)]^{\sharp_A}+i[\Im_A(T)]^{\sharp_A}$. Further, since $\mathcal{N}(A)^\perp$ is an invariant subspace for $T$, then clearly $\mathcal{N}(A)^\perp$ is also an invariant subspace for $\Re_A(T)$ and $\Im_A(T)$. So, by applying Theorem \[app1\] we see that $$\begin{aligned}
&\omega_\mathbb{A}\left[\begin{pmatrix}
\Re_A(T) & 0\\
\Im_A(T) & O
\end{pmatrix}\right]\\
&=\omega_\mathbb{A}\left[\begin{pmatrix}
[\Re_A(T)]^{\sharp_A} & [\Im_A(T)]^{\sharp_A}\\
O & O
\end{pmatrix}\right]\\
&\geq \frac{1}{2}\max\left\{\omega_A\Big([\Re_A(T)]^{\sharp_A}+ i[\Im_A(T)]^{\sharp_A}\Big),\omega_A\Big([\Re_A(T)]^{\sharp_A}- i[\Im_A(T)]^{\sharp_A}\Big)\right\}\\
&=\frac{1}{2}\max\left\{\omega_A(T^{\sharp_A}),\omega_A\left((T^{\sharp_A})^{\sharp_A}\right)\right\}\\
&=\frac{1}{2}\omega_A(T).\end{aligned}$$ On the other hand, by applying Lemma \[lemf\] we get $$\begin{aligned}
&\omega_\mathbb{A}\left[\begin{pmatrix}
0 & -i\Im_A(T)\\
\Re_A(T) & 0
\end{pmatrix}\right]\\
&=\omega_\mathbb{A}\left[\begin{pmatrix}
0 & [\Re_A(T)]^{\sharp_A}\\
i[\Im_A(T)]^{\sharp_A} & O
\end{pmatrix}\right]\\
&\geq \frac{1}{2}\max\left\{\omega_A\Big([\Re_A(T)]^{\sharp_A}+ i[\Im_A(T)]^{\sharp_A}\Big),\omega_A\Big([\Re_A(T)]^{\sharp_A}- i[\Im_A(T)]^{\sharp_A}\Big)\right\}\\
&=\frac{1}{2}\omega_A(T).\end{aligned}$$
[10]{} M. Al-Dolat, I. Jaradat, B. Al-Husban, A novel numerical radius upper bounds for $2\times 2$ operator matrices, Linear and Multilinear Algebra (2020) [ https://doi.org/10.1080/03081087.2020.1756199]( https://doi.org/10.1080/03081087.2020.1756199)
Linear Algebra Appl. 428 (7) (2008) 1460-1475.
Integral Equations and Operator Theory, 62 (2008), pp.11-28.
Acta Sci. Math. (Szeged) 75:3-4(2009), 635-653.
Linear Algebra Appl. 555 (2018) 266-284.
Linear Multilinear Algebra 68(4) 845-866 (2020).
Bull. Iran. Math. Soc. (2020). <https://doi.org/10.1007/s41980-020-00392-8>
P. Bhunia, K. Paul, R.K. Nayak, On inequalities for $A$-numerical radius of operators, Electron. J. Linear Algebra, Volume 36, pp. 143-157, 2020.
Proc. Amer. Math. Soc. 17 (1966) 413-416.
, [Spectral radius of semi-Hilbertian space operators and its applications]{}, Ann. Funct. Anal. (2020) <https://doi.org/10.1007/s43034-020-00064-y>.
arXiv:2002.02905v1 \[math.FA\] 7 Feb 2020.
, [A note on the $A$-numerical radius of operators in semi-Hilbert spaces]{}, Arch. Math. (2020). <https://doi.org/10.1007/s00013-020-01482-z>
, Some $\mathbb{A}$-numerical radius inequalities for $d\times d$ operator matrices, arXiv:2003.14378 \[math.FA\] 31 Mar 2020.
, Some numerical radius inequalities for semi-Hilbertian space operators, arxiv:2001.00398v2 \[math.FA\].
Italian journal of pure and applied mathematics n. 36-2016 (73-78).
O. Hirzallah, F.Kittaneh, K. Shebrawi, *Numerical radius inequalities for $2\times 2$ operator matrices*, Studia Mathematica **210** (2012), 99–115.
O. Hirzallah, F.Kittaneh, K. Shebrawi, *Numerical Radius Inequalities for Commutators of Hilbert Space Operators*, Numerical Functional Analysis and Optimization 32(7):739-749
N. C. Rout, S. Sahoo, D. Mishra, *Some $A$-numerical radius inequalities for semi-Hilbertian space operators*, Linear Multilinear Algebra (2020) <https://doi.org/10.1080/03081087.2020.1774487>
, On $A$-numerical radius inequalities for $2\times 2$ operator matrices, arXiv:2004.07494 \[math.FA\] 16 Apr 2020.
The Australian Journal of Mathematical Analysis and Applications. Volume 9, Issue 1, Article 5, (2012), pp. 1–12.
Linear Algebra Appl. 578(2019) 159-183.
, *Topics in Matrix Analysis*, Cambridge University Press, Cambridge, 1991.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '*Solving inverse problems continues to be a central challenge in computer vision. Existing techniques either explicitly construct an inverse mapping using prior knowledge about the corruption, or learn the inverse directly using a large collection of examples. However, in practice, the nature of corruption may be unknown, and thus it is challenging to regularize the problem of inferring a plausible solution. On the other hand, collecting task-specific training data is tedious for known corruptions and impossible for unknown ones. We present MimicGAN, an unsupervised technique to solve general inverse problems based on image priors in the form of generative adversarial networks (GANs). Using a GAN prior, we show that one can reliably recover solutions to underdetermined inverse problems through a surrogate network that learns to mimic the corruption at test time. Our system successively estimates the corruption and the clean image without the need for supervisory training, while outperforming existing baselines in blind image recovery. We also demonstrate that MimicGAN improves upon recent GAN-based defenses against adversarial attacks and represents one of the strongest test-time defenses available today.*'
author:
- |
Rushil Anirudh, Jayaraman J. Thiagarajan, Bhavya Kailkhura, Timo Bremer\
{anirudh1, jjayaram, kailkhura1, bremer5}@llnl.gov
bibliography:
- 'refs.bib'
title: 'MimicGAN: Corruption-Mimicking for Blind Image Recovery & Adversarial Defense'
---
oldmaketitlemaketitle
Introduction
============
Related Work
============
Proposed Approach
=================
Experiments
===========
Discussion
==========
In this paper, we presented [MimicGAN]{} , an entirely unsupervised system that can recover images from *unknown* corruptions by introducing a corruption-mimicking surrogate in addition to a GAN prior. Such a system moves us a step closer towards a *universal* solution to inverse problems. This can lead to the design of image recovery systems that can be deployed as a pre-processing step for achieving robust classification in the *wild*. [MimicGAN]{} can recover realistic images over a wide array of extremely ill-posed inversion tasks, while being competitive with supervised approaches. Furthermore, we showed that it improves over recent GAN-based defenses for adversarial attacks, thereby producing state-of-the-art defense against the strongest attacks.
Disclaimer {#disclaimer .unnumbered}
==========
This document was prepared as an account of work sponsored by an agency of the United States government. Neither the United States government nor Lawrence Livermore National Security, LLC, nor any of their employees makes any warranty, expressed or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recommendation, or favoring by the United States government or Lawrence Livermore National Security, LLC. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States government or Lawrence Livermore National Security, LLC, and shall not be used for advertising or product endorsement purposes.
****
CelebA Dataset {#celeba-dataset .unnumbered}
==============
Fashion-MNIST Dataset {#fashion-mnist-dataset .unnumbered}
=====================
| {
"pile_set_name": "ArXiv"
} |
---
abstract: '$\phi$-meson–nucleus bound state energies and absorption widths are calculated for seven selected nuclei by solving the Klein-Gordon equation with complex optical potentials. Essential input for the calculations, namely the medium-modified $K$ and $\Kbar$ meson masses, as well as the density distributions in nuclei, are obtained from the quark-meson coupling model. The attractive potential for the $\phi$-meson in the nuclear medium originates from the in-medium enhanced $K\Kbar$ loop in the $\phi$-meson self-energy. The results suggest that the $\phi$-meson should form bound states with all the nuclei considered. However, the identification of the signal for these predicted bound states will need careful investigation because of their sizable absorption widths.'
author:
- 'J. J. Cobos-Martínez'
- 'K. Tsushima'
- 'G. Krein'
- 'A. W. Thomas'
date:
-
-
title: |
LFTC-17-1/1, ADP-17-23/T1029
$\Phi$-meson–nucleus bound states
---
Introduction
============
The properties of light vector mesons at finite baryon density, such as their masses and decay widths, have attracted considerable experimental and theoretical interest over the last few decades; see Refs. [@Hatsuda:1994pi; @Leupold:2009kz; @Hayano:2008vn] for recent reviews. In part this has been related to their imputed potential to carry information on the partial restoration of chiral symmetry. In 2007 the KEK-E325 collaboration reported a 3.4% mass reduction of the $\phi$-meson [@Muto:2005za] and an in-medium decay width of $\approx 14.5$ MeV at normal nuclear matter density. These conclusions were based on the measurement of the invariant mass spectra of $e^{+}e^{-}$ pairs in 12 GeV p+A reactions, with copper and carbon being used as targets [@Muto:2005za].
Even though this result may indicate a signal for partial restoration of chiral symmetry in nuclear matter, it is not possible to draw a definite conclusion solely from this. In fact, recently, a large in-medium $\phi$-meson decay width ($>$30 MeV) has been extracted at various experimental facilities [@Ishikawa:2004id; @Mibe:2007aa; @Qian:2009ab; @Wood:2010ei; @Polyanskiy:2010tj], without observing any mass shift. It is therefore evident that the search for evidence of a light vector meson mass shift in nuclear matter is indeed a complicated issue and further experimental efforts are required in order to understand the phenomenon better. Indeed, the J-PARC E16 collaboration [@JPARCE16Proposal; @Csorgo:2014sat] intends to perform a more systematic study for the mass shift of vector mesons with higher statistics than the above-mentioned experiment at KEK-E325.
However, either complementary or alternative experimental methods are desired. The study of the $\phi$-meson–nucleus bound states is complementary to the invariant mass measurements, where only a small fraction of the produced $\phi$-mesons decay inside the nucleus and may be expected to provide extra information on the $\phi$-meson properties at finite baryon density. Along these lines and motivated by the 3.4% mass reduction reported by the KEK-E325 experiment, the E29 collaboration at J-PARC has recently put forward a proposal [@JPARCE29Proposal; @JPARCE29ProposalAdd] to study the in-medium mass modification of the $\phi$-meson via the possible formation of $\phi$-nucleus bound states [@Buhler:2010zz; @Ohnishi:2014xla] using the primary reaction $\overline{p}p\rightarrow \phi\phi$. Furthermore, there is also a proposal at the Thomas Jefferson National Accelerator Facility (JLab), following the 12 GeV upgrade, to study the binding of $\phi$ (and $\eta$) to $^{4}$He [@JLabphi]. This new experimental approach [@Buhler:2010zz; @Ohnishi:2014xla; @Csorgo:2014sat; @JLabphi] for the measurement of the $\phi$ meson mass shift in nuclei, will produce a slowly moving $\phi$-meson [@Buhler:2010zz; @Ohnishi:2014xla; @Csorgo:2014sat; @JLabphi], where the maximum nuclear matter effect can be probed. In this way, one may indeed anticipate the formation of a $\phi$-nucleus bound state, where the $\phi$-meson is trapped inside the nucleus.
Meson-nucleus systems bound by attractive strong interactions are very interesting objects; see Refs. [@Krein:2017usp; @Metag:2017yuh] for recent reviews on the subject . First, they are strongly interacting exotic many-body systems and to study them serves, for example, to understand better the multiple-gluon exchange interactions, including QCD “van der Waals” forces [@Appelquist:1978rt], which are believed to play a role in the binding of the $J/\Psi$ and other exotic heavy-quarkonia to matter (a nucleus) [@Brodsky:1989jd; @Luke:1992tm; @Sibirtsev:1999jr; @Gao:2000az; @Beane:2014sda; @Brambilla:2015rqa; @Gao:2017hya; @Kawama:2014pja; @Iijima:2014cza; @Proceedings:2014rfa; @Ohnishi:2014xla; @Csorgo:2014sat]. Second, they provide unique laboratories for the study of hadron properties at finite density, which may not only lead to a deeper understanding of the strong interaction [@Hatsuda:1994pi; @Leupold:2009kz; @Hayano:2008vn; @Krein:2017usp; @Metag:2017yuh] but of the structure of finite nuclei as well [@Stone:2016qmi; @Guichon:2006er].
A downward mass shift of the $\phi$-meson in a nucleus is directly connected with the possible existence of an attractive potential between the $\phi$-meson and the nucleus, the strength of which is expected to be of the same order as that of the mass shift. Along these lines, various authors predict a downward shift of the in-medium $\phi$ meson mass and a broadening of the decay width, many of them focusing on the self-energy of the $\phi$ meson due to the kaon-antikaon loop. Ko [*et al.*]{} [@Ko:1992tp] used a density-dependent kaon mass determined from chiral perturbation theory and found that at normal nuclear matter density, $\rho_0$, the $\phi$ mass decreases very little, by at most $2\%$, and the width $\Gamma_\phi \approx 25$ MeV and broadens drastically for large densities. Hatsuda and Lee calculated the in-medium $\phi$ mass based on the QCD sum rule approach [@Hatsuda:1991ez; @Hatsuda:1996xt], and predicted a decrease of 1.5%-3% at $\rho_0$. Other investigations also predict a small downward mass shift and a large broadening of the $\phi$ width at $\rho_0$: Ref. [@Klingl:1997tm] reports a negative mass shift of $ < 1\%$ and a decay width of 45 MeV; Ref. [@Oset:2000eg] predicts a decay width of 22 MeV but does not report a result on the mass shift; and Ref. [@Cabrera:2002hc] gives a rather small negative mass shift of $\approx 0.81\%$ and a decay width of 30 MeV. More recently, Ref. [@Gubler:2015yna] reported a downward mass shift of $< 2\%$ and a large broadening width of 45 MeV at $\rho_{0}$; and finally, in Ref. [@Cabrera:2016rnc], extending the work of Refs. [@Oset:2000eg; @Cabrera:2002hc], the authors reported a negative mass shift of $3.4\%$ and a large decay width of 70 MeV at $\rho_0$. The reason for these differences may lie in the different approaches used to estimate the kaon-antikaon loop contributions to the $\phi$-meson self-energy and this might have consequences for the formation of $\phi$-meson–nucleus bound states.
From a practical point of view, the important question is whether this attraction, if it exists, is sufficient to bind the $\phi$ to a nucleus. A simple argument can be given as follows. One knows that for an attractive spherical well of radius $R$ and depth $V_{0}$, the condition for the existence of a nonrelativistic $s$-wave bound state of a particle of mass $m$ is $V_{0}>\frac{\pi^2\hbar^2}{8mR^2}$. Using $m = m_{\phi}^{*}$, where $m_{\phi}^{*}$ is the $\phi$-meson mass at normal nuclear matter density found in Ref. [@Muto:2005za] and $R= 5$ fm (the radius of a heavy nucleus), one obtains $V_{0}>2$ MeV. Therefore, the prospects of capturing a $\phi$-meson seem quite favorable, provided that the $\phi$-meson can be produced almost at rest in the nucleus.
An initial calculation of possible $\phi$-nucleus bound states was carried out in Ref. [@YamagataSekihara:2010rb] for a few nuclei. However, the theoretical potential on which this study was based [@Cabrera:2002hc] was too weak, with only two bound states being found. In order to remedy this, the real part of the potential was scaled, without any theoretical basis, so as to simulate a 3$\%$ mass reduction of the $\phi$-meson, that is, approximately equal to that reported in Ref. [@Muto:2005za]. This (scaled) potential was mainly used to study the sensitivity of the formation spectra to the potential strength [@Cabrera:2002hc]. Here, it was found that, as expected, whether or not the formation of the $\phi$-meson bound state is possible depends on the strength of the attractive potential between the $\phi$-meson and the nucleus.
In previous work [@Cobos-Martinez:2017vtr] we studied the $\phi$-meson mass shift and decay width in nuclear matter, based on an effective Lagrangian approach, by evaluating the $K\Kbar$ loop contribution in the $\phi$ self-energy, with the in-medium $K$ and $\Kbar$ masses explicitly calculated by the quark-meson coupling (QMC) model [@Saito:2005rv]. Here we extend our previous initial study [@Cobos-Martinez:2017vtr] to seven selected nuclei, showing details of the calculated nuclear potential and computing the $\phi$-nucleus bound state energies and absorption widths by solving the Klein-Gordon equation. The nuclear density distributions for heavy nuclei studied (except for $^4$He), as well as the medium modification of the $K$ and $\Kbar$ masses, are explicitly calculated using the QMC model [@Saito:1996sf].
This paper is organized as follows. In Sec. \[sec:nuclmatt\] we briefly discuss the computation and present results for the mass shift and decay width of the $\phi$-meson in infinite (symmetric) nuclear matter. Using the results of Sec. \[sec:nuclmatt\], together with the density profiles of the nuclei to be studied, in Sec. \[sec:finitenuclei\] we present results for the real and imaginary parts of the scalar $\phi$-nucleus potentials, as well as the corresponding bound state energies and absorption widths. Finally, Sec. \[sec:summary\] is devoted to a summary and discussion.
\[sec:nuclmatt\] $\phi$-meson self-energy in infinite nuclear matter
====================================================================
The $\phi$-meson property modifications in nuclear matter, such as its mass and decay width, are strongly correlated to its coupling to the $K\Kbar$ channel, which is the dominant decay channel in vacuum. Therefore, one expects that a significant fraction of the density dependence of the $\phi$-meson self-energy in nuclear matter might arise from the in-medium modification of the $K\Kbar$-loop in the $\phi$-self-energy intermediate state.
Here we use the effective Lagrangian approach of Ref. [@Klingl:1996by] and briefly review the computation of the $\phi$-meson self-energy in vacuum and in nuclear matter made in Ref. [@Cobos-Martinez:2017vtr]. The interaction Lagrangian $\mathcal{L}_{\text{Int}}$ involves $\phi K\Kbar$ and $\phi\phi K\Kbar$ couplings dictated by a local gauge symmetry principle: $$\label{eqn:Lint}
\mathcal{L}_{\text{Int}} = \mathcal{L}_{\phi K\Kbar} + \mathcal{L}_{\phi\phi K\Kbar},$$ where $$\label{eqn:phikk}
\mathcal{L}_{\phi K\Kbar} = {\mathrm{i}}g_{\phi}\phi^{\mu}
\left[\Kbar(\partial_{\mu}K)-(\partial_{\mu}\Kbar)K\right],$$ and $$\label{eqn:phi2kk}
\mathcal{L}_{\phi\phi K\Kbar} = g^2_{\phi} \phi^\mu\phi_\mu \Kbar K.$$ We use the convention for the isospin doublets: $$\label{eqn:isospin}
K=\left(\begin{array}{c} K^{+} \\ K^{0} \end{array} \right),\;
\overline{K}=\left(K^{-}\;\overline{K}^{0}\;\right).$$
![\[fig:phise\] $K\Kbar$-loop contribution to the $\phi$ meson self-energy.](phi_selfenergy)
We note that the use of the effective interaction Lagrangian of [Eq. (\[eqn:Lint\])]{} without the term given in [Eq. (\[eqn:phi2kk\])]{} may be considered as being motivated by the hidden gauge approach in which there are no four-point vertices, such as [Eq. (\[eqn:phi2kk\])]{}, that involve two pseudoscalar mesons and two vector mesons [@Lin:1999ve; @Lee:1994wx]. This is in contrast to the approach of using the minimal substitution to introduce vector mesons as gauge particles where such four-point vertices do appear. However, these two methods have been shown to be consistent if both the vector and axial vector mesons are included [@Yamawaki:1986zz; @Meissner:1986tc; @Meissner:1987ge; @Saito:1987ba]. Therefore, we present results with and without such an interaction. We consider first the contribution from the $\phi K\Kbar$ coupling given by Eq. (\[eqn:phikk\]) to the scalar part of the $\phi$ self-energy, $\Pi_{\phi}(p)$, by evaluating the diagram of [FIG. \[fig:phise\]]{}.
For a $\phi$-meson at rest the scalar self-energy is given by $$\label{eqn:phise}
{\mathrm{i}}\Pi_{\phi}(p)=-\frac{8}{3}g_{\phi}^{2}\int{\hspace{-0.4em}\ensuremath{\frac{\mathrm{d}^{4}q}{(2\pi)^{4}}}\,}\vec{q}^{\,2}
D_{K}(q)D_{K}(q-p),$$ where $D_{K}(q)=\left(q^{2}-m_{K}^{2}+{\mathrm{i}}\epsilon\right)^{-1}$ is the kaon propagator; $p=(p^{0}=m_{\phi},\vec{0})$ is the $\phi$-meson four-momentum vector at rest, with $m_{\phi}$ the $\phi$-meson mass; $m_{K} (=m_{\Kbar})$ is the kaon mass; and $g_{\phi}$ is the coupling constant.
The integral in [Eq. (\[eqn:phise\])]{} is divergent but it will be regulated using a phenomenological form factor, with cutoff parameter $\Lambda_{K}$, as in Refs. [@Krein:2010vp; @Cobos-Martinez:2017vtr]. The sensitivity of the results to the cutoff value is analyzed below.
The coupling constant $g_{\phi}$ is determined in Ref. [@Cobos-Martinez:2017vtr] from the experimental value for the $\phi \to K\overline{K}$ decay width in vacuum, corresponding to the branching ratio of $83.1\%$ of the total decay width (4.266 MeV) [@PDG:2015]. For the $\phi$ mass in vacuum, $m_{\phi}$, we use its experimental value $m_{\phi}^{\text{expt}}=1019.461$ MeV [@PDG:2015]; while for the kaon mass $m_{K}$ we use the average of the experimental values [@PDG:2015] of the positive-charged and neutral kaons, $m_{K^{+}}^{\text{expt}}=493.677$ MeV and $m_{K^{0}}^{\text{expt}}=497.611$, respectively. We note that the effect of this tiny mass ambiguity on the in-medium kaon (antikaon) properties is negligible. Then, we obtain the coupling $g_{\phi}=4.539$. The $\Lambda_{K}$-dependent $\phi$-meson bare mass $m_{\phi}^{0}$ is fixed by fitting the physical $\phi$-meson mass $m_{\phi}^{\text{expt}}$ [@Cobos-Martinez:2017vtr], and the values obtained are given in the caption to [TABLE \[tab:phippties\]]{}.
The mass and decay width of the $\phi$-meson in vacuum ($m_{\phi}$ and $\Gamma_{\phi}$), as well as in nuclear matter ($m_{\phi}^{*}$ and $\Gamma_{\phi}^{*}$), are determined self-consistently in Ref. [@Cobos-Martinez:2017vtr] from $$\begin{aligned}
\label{eqn:phimass}
m_{\phi}^{2}&=&\left(m_{\phi}^{0}\right)^{2}+\Re\Pi_{\phi}(m_{\phi}^{2}), \\
\label{eqn:phidecay}
\Gamma_{\phi}&=&-\frac{1}{m_{\phi}}\Im\Pi_{\phi}(m_{\phi}^{2}).\end{aligned}$$ The nuclear density dependence of the $\phi$-meson mass and decay width is driven by the intermediate-state kaon and antikaon interactions with the nuclear medium. This effect enters through $m_{K}^{*}$ in the kaon propagators in [Eq. (\[eqn:phise\])]{}. The in-medium mass, $m_{K}^{*}$, is calculated within the QMC model [@Cobos-Martinez:2017vtr], which has proven to be very successful in studying the properties of hadrons in nuclear matter and finite nuclei. For a more complete discussion of the model see Refs. [@Tsushima:1997df; @Guichon:1989tx; @Saito:2005rv]. Here we just make a few necessary comments. In order to calculate the in-medium properties of $K$ and $\Kbar$, we consider infinitely large, uniformly symmetric, spin-isospin-saturated nuclear matter in its rest frame, where all the scalar and vector mean field potentials, which are responsible for the nuclear many-body interactions, become constant in the Hartree approximation [@Cobos-Martinez:2017vtr]. In [FIG. \[fig:mk\]]{} we present the resulting in-medium kaon Lorentz scalar mass (=antikaon Lorentz scalar mass), calculated using the QMC model, as a function of the baryon density. The kaon effective mass at normal nuclear matter density $\rho_{0}=0.15$ fm$^{-3}$ has decreased by about 13%. We also recall, in connection with the calculation of the in-medium $K\Kbar$-loop contributions to the $\phi$-meson self-energy, that the isoscalar-vector $\omega$ mean field potentials arise both for the kaon and antikaon. However, they have opposite signs and cancel each other. Equivalently, they can be eliminated by a variable shift in the loop calculation [@Tsushima:1997df; @Guichon:1989tx; @Saito:2005rv] of the $\phi$ self-energy, and therefore we do not show them here.
[c]{}\
In [FIG. \[fig:Xphi\]]{}, we present the $\phi$-meson mass shift (upper panel) and decay width (lower panel) as a function of the nuclear matter density, $\rho_{B}$, for three values of the cutoff parameter $\Lambda_{K}$. As can be seen, the effect of the in-medium kaon and antikaon mass change yields a negative mass shift for the $\phi$-meson. This is because the reduction in the kaon and antikaon masses enhances the $K\Kbar$-loop contribution in nuclear matter relative to that in vacuum. For the largest value of the nuclear matter density, the downward mass shift turns out to be a few percent at most for all values of $\Lambda_{K}$. On the other hand, we see that $\Gamma_{\phi}^{*}$ is very sensitive to the change in the kaon and antikaon masses: it increases rapidly with increasing nuclear matter density, up to a factor of $\sim 20$ enhancement for the largest value of $\rho_{B}$. At normal nuclear matter density, $\rho_{0}$, we see that the negative kaon and antikaon mass shift of 13% [@Cobos-Martinez:2017vtr] induces a downward mass shift of the $\phi$-meson of just $\approx$ 2%, while the broadening of the $\phi$-meson decay width is an order of magnitude larger than its vacuum value.
For completeness, and in connection with the paragraph just after [Eq. (\[eqn:isospin\])]{}, we show in [FIG. \[fig:xiOn\]]{} the impact of adding the $\phi\phi K \Kbar$ interaction of Eq. (\[eqn:phi2kk\]) on the in-medium $\phi$ mass and decay width. We have used the notation that $\xi = 1 (0)$ means that this interaction is (not) included in the calculation of the $\phi$ self-energy. As can be seen, one still gets a downward shift of the in-medium $\phi$ mass when $\xi = 1$ as well as a significant broadening of the decay width. In both cases, though, the absolute values are slightly different from the $\xi = 0$ case. In [TABLE \[tab:phippties\]]{} we present the values for $m_{\phi}^{*}$ and $\Gamma_{\phi}^{*}$ at normal nuclear matter density $\rho_{0}$ with and without the gauged Lagrangian of [Eq. (\[eqn:phi2kk\])]{}. In both cases, the effect of adding Eq. (\[eqn:phi2kk\]) can be compensated by the use of a larger cutoff $\Lambda_{K}$ (compare the first and the last columns of [TABLE \[tab:phippties\]]{}).
$\Lambda_{K}= 2000$ $\Lambda_{K}= 3000$ $\Lambda_{K}= 4000$
--------------------- --------------------- --------------------- ---------------------
$m_{\phi}^{*}$ 1000.9 (1009.5) 994.9 (1004.3) 990.4 (1000.6)
$\Gamma_{\phi}^{*}$ 34.8 (37.8) 32.8 (36.0) 31.3 (34.7)
: \[tab:phippties\] $\phi$ mass and width at normal nuclear matter density $\rho_{0}$ with and without the gauged Lagrangian of [Eq. (\[eqn:phi2kk\])]{}. The values in parentheses are computed by adding the gauged Lagrangian of [Eq. (\[eqn:phi2kk\])]{}. All quantities are given in MeV. For the cutoff values of $\Lambda_{K}=2000,3000$, and 4000 MeV we find the $\phi$-meson bare mass $m^0_\phi$ values of 1074.0 (1023.4) MeV, 1132.9 (1024.6) MeV, and 1213.6 (1025.6) MeV, respectively.
[c]{}\
The results described above support a small downward mass shift and a large broadening of the decay width of the $\phi$-meson in a nuclear medium. Furthermore, they open experimental possibilities for studying the binding and absorption of $\phi$-mesons in nuclei. Although the mass shift found in this study may be large enough to bind the $\phi$-meson to a nucleus, the broadening of its decay width will make it difficult to observe a signal for the $\phi$-nucleus bound state formation experimentally. We explore this further in the following section.
\[sec:finitenuclei\] $\phi$-nuclear bound states
================================================
---------- -- --
\[37mm\]
\[37mm\]
\[37mm\]
\[37mm\]
---------- -- --
In this section we discuss the situation where the $\phi$-meson is placed in a nucleus. The nuclear density distributions for $^{12}$C, $^{16}$O, $^{40}$Ca, $^{48}$Ca, $^{90}$Zr, and $^{208}$Pb are obtained using the QMC model [@Saito:1996sf]. For $^{4}$He, we use the parametrization for the density distribution obtained in Ref. [@Saito:1997ae]. Then, using a local density approximation we calculate the $\phi$-meson complex potentials for a nucleus $A$, which can be written as $$\label{eqn:Vcomplex}
V_{\phi A}(r)= U_{\phi}(r)-\frac{{\mathrm{i}}}{2}W_{\phi}(r),$$ where $r$ is the distance from the center of the nucleus and $U_{\phi}(r)=\Delta m_{\phi}(\rho_{B}(r)) \equiv m^{*}_{\phi}(\rho_{B}(r))-m_\phi$ and $W_{\phi}(r)=\Gamma_{\phi}(\rho_{B}(r))$ are, respectively, the $\phi$-meson mass shift and decay width in a nucleus $A$. As usual, $\rho_{B}(r)$ is the baryon density distribution for the particular nucleus.
In Figure \[fig:VphiA\] we present the $\phi$-meson potentials calculated for the seven nuclei selected, for three values of the cutoff parameter $\Lambda_{K}$, $2000, 3000$ and $4000$ MeV. One can see that the depth of the real part of the potential, $U_\phi(r)$, is sensitive to the cutoff parameter, varying from -20 MeV to -35 MeV for $^{4}$He and from -20 MeV to -30 MeV for $^{208}$Pb. In addition, one can see that the imaginary part does not vary much with different values of $\Lambda_{K}$. These observations may well have consequences for the feasibility of experimental observation of the expected bound states.
Using the $\phi$-meson potentials obtained in this manner, we next calculate the $\phi$-meson–nuclear bound state energies and absorption widths for the seven nuclei selected. Before proceeding, a few comments on the use of [Eq. (\[eqn:kg\])]{} are in order. In this study we consider the situation where the $\phi$-meson is produced nearly at rest. Then, it should be a very good approximation to neglect the possible energy difference between the longitudinal and transverse components of the $\phi$-meson wave function $\psi_{\phi}^{\mu}$. After imposing the Lorentz condition, $\partial_{\mu}\psi_{\phi}^{\mu}=0$, to solve the Proca equation becomes equivalent to solving the Klein-Gordon equation $$\label{eqn:kg}
\left(-\nabla^{2} + \mu^{2} + 2\mu V(\vec{r})\right)\phi(\vec{r})
= \mathcal{E}^{2}\phi(\vec{r}),$$ where $\mu=m_{\phi}m_{A}/(m_{\phi}+m_{A})$ is the reduced mass of the $\phi$-meson-nucleus system with $m_{\phi}$ $(m_{A}$) the mass of the $\phi$-meson (nucleus $A$) in vacuum, and $V(\vec{r})$ is the complex $\phi$-meson-nucleus potential of [Eq. (\[eqn:Vcomplex\])]{}. We solve the Klein-Gordon equation using the momentum space methods developed in Ref. [@Kwan:1978zh]. Here, [Eq. (\[eqn:kg\])]{} is first converted to momentum space representation via a Fourier transform, followed by a partial wave-decomposition of the Fourier-transformed potential. Then, for a given value of angular momentum, the eigenvalues of the resulting equation are found by the inverse iteration eigenvalue algorithm. The calculated bound state energies ($E$) and widths ($\Gamma$), which are related to the complex energy eigenvalue $\mathcal{E}$ by $E=\Re\mathcal{E}-\mu$ and $\Gamma=-2\Im\mathcal{E}$, are listed in [TABLE \[tab:phibse\]]{} for three values of the cutoff parameter $\Lambda_{K}$, with and without the imaginary part of the potential, $W_{\phi}(r)$.
We first discuss the case in which the imaginary part of the $\phi$-nucleus potential, $W_{\phi}(r)$, is set to zero. The results are listed inside brackets in [TABLE \[tab:phibse\]]{}. From the values shown in brackets, we see that the $\phi$-meson is expected to form bound states with all the seven nuclei selected, for all values of the cutoff parameter $\Lambda_{K} = 2000, 3000$ and $4000$ MeV. (For the variation in the potential depths due to the $\Lambda_K$ values, see Fig. \[fig:VphiA\].) However, the bound state energy is obviously dependent on $\Lambda_{K}$, increasing as $\Lambda_K$ increases.
Next, we discuss the results obtained when the imaginary part of the potential is included. Adding the absorptive part of the potential, the situation changes appreciably. From the results presented in [TABLE \[tab:phibse\]]{} we note that for the largest value of the cutoff parameter $\Lambda_K = 4000$ MeV, which yields the deepest attractive potentials, the $\phi$-meson is expected to form bound states in all the nuclei selected, including the lightest $^4$He nucleus. However, in this case, whether or not the bound states can be observed experimentally is sensitive to the value of the cutoff parameter $\Lambda_K$. One also observes that the width of the bound state is insensitive to the values of $\Lambda_{K}$ for all nuclei. Furthermore, since the so-called dispersive effect of the absorptive potentialis repulsive, the bound states disappear completely in some cases, even though they were found when the absorptive part was set to zero. This feature is obvious for the $^4$He nucleus, making it especially relevant to the future experiments, planned at J-PARC and JLab using light and medium-heavy nuclei [@Buhler:2010zz; @Ohnishi:2014xla; @Csorgo:2014sat; @JLabphi].
We here comment that we have also solved the Schröedinger equation with the potential [Eq. (\[eqn:Vcomplex\])]{} with and without its imaginary part for the single-particle energies and widths, and compared with those given in [TABLE \[tab:phibse\]]{}. The results found in both cases are essentially the same.
\[sec:summary\] Summary and discussion
======================================
We have calculated the $\phi$-meson–nucleus bound state energies and absorption widths for various nuclei. The $\phi$-meson–nuclear potentials were calculated using a local density approximation, with the inclusion of the $K\Kbar$ meson loop in the $\phi$-meson self-energy. The nuclear density distributions, as well as the in-medium $K$ and $\Kbar$ meson masses, were consistently calculated by employing the quark-meson coupling model. Using the $\phi$-meson–nuclear complex potentials, we have solved the Klein-Gordon equation in momentum space, and obtained $\phi$-meson–nucleus bound state energies and absorption widths. Furthermore, we have studied the sensitivity of the results to the cutoff parameter $\Lambda_{K}$ in the form factor at the $\phi-K\Kbar$ vertex appearing in the $\phi$-meson self-energy. We expect that the $\phi$-meson should form bound states for all seven nuclei selected, provided that the $\phi$-meson is produced in (nearly) recoilless kinematics. This feature, is even more obvious in the (artificial) case where the absorptive part of the potential is ignored. Given the similarity of the binding energies and widths reported here, the signal for the formation of the $\phi$-nucleus bound states may be difficult to identify experimentally. Therefore, the feasibility of observation of the $\phi$-meson–nucleus bound states needs further investigation, including explicit reaction cross section estimates.\
Acknowledgements {#acknowledgements .unnumbered}
================
This work was partially supported by Conselho Nacional de Desenvolvimento Científico e Tecnológico - CNPq, Grant Nos. 152348/2016-6 (J.J.C-M.), 400826/2014-3 and 308088/2015-8 (K.T.), 305894/2009-9 (G.K.), and 313800/2014-6 (A.W.T.), and Funda[ç]{}ão de Amparo à Pesquisa do Estado de São Paulo-FAPESP, Grant Nos. 2015/17234-0 (K.T.) and 2013/01907-0 (G.K.). This research was also supported by the University of Adelaide and by the Australian Research Council through the ARC Centre of Excellence for Particle Physics at the Terascale (CE110001104), and through Grant No. DP151103101 (A.W.T.).
[150]{}
T. Hatsuda and T. Kunihiro, Phys. Rept. [**247**]{}, 221 (1994) \[hep-ph/9401310\].
S. Leupold, V. Metag and U. Mosel, Int. J. Mod. Phys. E [**19**]{}, 147 (2010) \[arXiv:0907.2388 \[nucl-th\]\]. R. S. Hayano and T. Hatsuda, Rev. Mod. Phys. [**82**]{}, 2949 (2010) \[arXiv:0812.1702 \[nucl-ex\]\]. R. Muto [*et al.*]{} \[KEK-PS-E325 Collaboration\], Phys. Rev. Lett. [**98**]{}, 042501 (2007) \[nucl-ex/0511019\]. T. Ishikawa [*et al.*]{}, Phys. Lett. B [**608**]{}, 215 (2005) \[nucl-ex/0411016\]. T. Mibe [*et al.*]{} \[CLAS Collaboration\], Phys. Rev. C [**76**]{}, 052202 (2007) \[nucl-ex/0703013 \[NUCL-EX\]\]. X. Qian [*et al.*]{} \[CLAS Collaboration\], Phys. Lett. B [**680**]{}, 417 (2009) \[arXiv:0907.2668 \[nucl-ex\]\]. M. H. Wood [*et al.*]{} \[CLAS Collaboration\], Phys. Rev. Lett. [**105**]{}, 112301 (2010) \[arXiv:1006.3361 \[nucl-ex\]\]. A. Polyanskiy [*et al.*]{}, Phys. Lett. B [**695**]{}, 74 (2011) \[arXiv:1008.0232 \[nucl-ex\]\]. <http://rarfaxp.riken.go.jp/~yokkaich/paper/jparc-proposal-0604.pdf> T. Csörgő, M. Csanád and T. Novák, Proceedings, 10th Workshop on Particle Correlations and Femtoscopy (WPCF 2014) : Gyöngyös, Hungary, August 25-29, 2014, SLAC-econf-C140825.8. <http://j-parc.jp/researcher/Hadron/en/pac_0907/pdf/Ohnishi.pdf>
<http://j-parc.jp/researcher/Hadron/en/pac_1007/pdf/KEK_J-PARC-PAC2010-02.pdf>
P. Buhler [*et al.*]{}, Prog. Theor. Phys. Suppl. [**186**]{}, 337 (2010). H. Ohnishi [*et al.*]{}, Acta Phys. Polon. B [**45**]{}, 819 (2014). <https://www.jlab.org/exp_prog/PACpage/PAC42/PAC42_FINAL_Report.pdf>
G. Krein, A. W. Thomas and K. Tsushima, arXiv:1706.02688 \[hep-ph\]. V. Metag, M. Nanova and E. Y. Paryev, arXiv:1706.09654 \[nucl-ex\]. T. Appelquist and W. Fischler, Phys. Lett. [**77B**]{}, 405 (1978). S. J. Brodsky, I. A. Schmidt and G. F. de Teramond, Phys. Rev. Lett. [**64**]{}, 1011 (1990). M. E. Luke, A. V. Manohar and M. J. Savage, Phys. Lett. B [**288**]{}, 355 (1992) \[hep-ph/9204219\]. A. Sibirtsev, K. Tsushima, K. Saito and A. W. Thomas, Phys. Lett. B [**484**]{} (2000) 23 \[nucl-th/9904015\]. H. Gao, T. S. H. Lee and V. Marinov, Phys. Rev. C [**63**]{}, 022201 (2001) \[nucl-th/0010042\].
S. R. Beane, E. Chang, S. D. Cohen, W. Detmold, H.-W. Lin, K. Orginos, A. Parreño and M. J. Savage, Phys. Rev. D [**91**]{}, no. 11, 114503 (2015) \[arXiv:1410.7069 \[hep-lat\]\]. N. Brambilla, G. Krein, J. Tarr�s Castell� and A. Vairo, Phys. Rev. D [**93**]{}, 054002 (2016) \[arXiv:1510.05895 \[hep-ph\]\]. H. Gao, H. Huang, T. Liu, J. Ping, F. Wang and Z. Zhao, Phys. Rev. C [**95**]{}, no. 5, 055202 (2017) \[arXiv:1701.03210 \[hep-ph\]\].
D. Kawama \[J-PARC E16 Collaboration\], PoS Hadron [**2013**]{}, 178 (2013).
D. Kawama [*et al.*]{} (J-PARC E16 Collaboration), Proceedings, 15th International Conference on Hadron Spectroscopy (Hadron 2013) : Nara, Japan, November 4-8, 2013, PoS Hadron [**2013**]{} (2013) Y. Morino [*et al.*]{} (J-PARC E16 Collaboration), Proceedings, 12th Asia Pacific Physics Conference (APPC12) : Makuhari Messe, Japan, July 14-19, 2013, JPS Conf. Proc. [**1**]{} (2014) J. R. Stone, P. A. M. Guichon, P. G. Reinhard and A. W. Thomas, Phys. Rev. Lett. [**116**]{}, no. 9, 092501 (2016) \[arXiv:1601.08131 \[nucl-th\]\]. P. A. M. Guichon, H. H. Matevosyan, N. Sandulescu and A. W. Thomas, Nucl. Phys. A [**772**]{}, 1 (2006) \[nucl-th/0603044\]. C. M. Ko, P. Levai, X. J. Qiu and C. T. Li, Phys. Rev. C [**45**]{}, 1400 (1992). T. Hatsuda and S. H. Lee, Phys. Rev. C [**46**]{}, no. 1, R34 (1992). T. Hatsuda, H. Shiomi and H. Kuwabara, Prog. Theor. Phys. [**95**]{}, 1009 (1996) \[nucl-th/9603043\]. F. Klingl, T. Waas and W. Weise, Phys. Lett. B [**431**]{}, 254 (1998) \[hep-ph/9709210\]. E. Oset and A. Ramos, Nucl. Phys. A [**679**]{}, 616 (2001) \[nucl-th/0005046\]. D. Cabrera and M. J. Vicente Vacas, Phys. Rev. C [**67**]{}, 045203 (2003) \[nucl-th/0205075\]. P. Gubler and W. Weise, Phys. Lett. B [**751**]{}, 396 (2015) \[arXiv:1507.03769 \[hep-ph\]\]. D. Cabrera, A. N. Hiller Blin and M. J. Vicente Vacas, Phys. Rev. C [**95**]{}, no. 1, 015201 (2017) \[arXiv:1609.03880 \[nucl-th\]\]. J. Yamagata-Sekihara, D. Cabrera, M. J. Vicente Vacas and S. Hirenzaki, Prog. Theor. Phys. [**124**]{}, 147 (2010) \[arXiv:1001.2235 \[nucl-th\]\]. J. J. Cobos-Martínez, K. Tsushima, G. Krein and A. W. Thomas, Phys. Lett. B [**771**]{}, 113 (2017) \[arXiv:1703.05367 \[nucl-th\]\].
K. Saito, K. Tsushima and A. W. Thomas, Prog. Part. Nucl. Phys. [**58**]{}, 1 (2007) \[hep-ph/0506314\]. K. Saito, K. Tsushima and A. W. Thomas, Nucl. Phys. A [**609**]{}, 339 (1996) \[nucl-th/9606020\].
F. Klingl, N. Kaiser and W. Weise, Z. Phys. A [**356**]{}, 193 (1996) \[hep-ph/9607431\]. Z. w. Lin, C. M. Ko and B. Zhang, Phys. Rev. C [**61**]{}, 024904 (2000). S. H. Lee, C. Song and H. Yabu, Phys. Lett. B [**341**]{}, 407 (1995) \[hep-ph/9408266\]. K. Yamawaki, Phys. Rev. D [**35**]{}, 412 (1987).
U. G. Meissner and I. Zahed, Z. Phys. A [**327**]{}, 5 (1987).
U. G. Meissner, Phys. Rept. [**161**]{}, 213 (1988). S. Saito and K. Yamawaki, NAGOYA, JAPAN: UNIV., PHYS. DEPT. (1987) 225p
G. Krein, A. W. Thomas and K. Tsushima, Phys. Lett. B [**697**]{}, 136 (2011) \[arXiv:1007.2220 \[nucl-th\]\]. The Review of Particle Physics (2015), K.A. Olive et al. (Particle Data Group), Chin. Phys. C, 38, 090001 (2014) and 2015 update, <http://pdg.lbl.gov/>
K. Tsushima, K. Saito, A. W. Thomas and S. V. Wright, Phys. Lett. B [**429**]{}, 239 (1998) Erratum: \[Phys. Lett. B [**436**]{}, 453 (1998)\] \[nucl-th/9712044\]. P. A. M. Guichon, Nucl. Phys. A [**497**]{}, 265C (1989). K. Saito, K. Tsushima and A. W. Thomas, Phys. Rev. C [**56**]{}, 566 (1997) \[nucl-th/9703011\].
Y. R. Kwan and F. Tabakin, Phys. Rev. C [**18**]{}, 932 (1978).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Given a torus bundle $Y$ over the circle and a cohomology class $[\omega]\in H^2(Y;\mathbb{Z})$ which evaluates non-trivially on the fiber, we compute the Heegaard Floer homology of $Y$ with twisted coefficients in the universal Novikov ring.'
address:
- 'School of Mathematical Science, Peking University, Beijing 100871, P.R.China, and Department of Mathematics, Columbia University, MC 4406 2990 Broadway, New York, NY 10027'
- |
Department of Mathematics, Columbia University, MC 4406\
2990 Broadway, New York, NY 10027
author:
- Yinghua Ai
- Thomas Peters
title: '**the Twisted Floer homology of torus bundles**'
---
\[section\] \[theorem\][Lemma]{} \[theorem\][Definition]{} \[theorem\][Proposition]{} \[theorem\][Remark]{}
Introduction {#section:intro}
============
Heegaard Floer homology was introduced by Ozsváth and Szabó in [@OSzAnn1; @OSzAnn2], it provides powerful invariants for closed oriented $3$–manifolds. There is also a filtered version, called knot Floer homology, for null-homologous knots [@OSzKnot; @Ra]. It turns out that Heegaard Floer homology provides much geometric information. For instance, it can detect fibrations. Work of Ghiggini [@Gh] and Ni [@Ni1] shows that knot Floer homology detects fiberedness in knots.
Turning to closed fibered $3$–manifolds, note that a $3$–manifold which admits a fibration $\pi \co Y\to S^1$ has a canonical Spin$^c$ structure, $\ell$, obtained as the tangents to the fibers of $\pi$. In [@OSzSympl], Ozsváth and Szabó prove
\[theorem:fiber\] Let $Y$ be a closed $3$–manifold which fibers over the circle, with fiber $F$ of genus $g>1$, and let ${\bf \mathfrak{t}}$ be a Spin$^c$ structure over $Y$ with $$\langle c_1({\bf \mathfrak{t}}),[F]\rangle=2-2g.$$ Then for ${\bf \mathfrak{t}}\neq \ell$, we have that $$HF^+(Y,{\bf \mathfrak{t}})=0;$$ while $$HF^+(Y,\ell)\cong \mathbb{Z}$$
In fact, Yi Ni proved a converse to the above statement in [@Ni2] which states:
\[Theorem:Ni\] Suppose $Y$ is a closed irreducible $3$–manifold, $F\subset Y$ is a closed connected surface of genus $g>1$. Let $HF^+(Y,[F],1-g)$ denote the group $$\bigoplus_{ {\bf \mathfrak{s}}\in Spin^c(Y), \langle c_1({\bf \mathfrak{s}}),[F]\rangle=2-2g}HF^+(Y,{\bf \mathfrak{s}}).$$ If $HF^+(Y,[F],1-g)\cong \mathbb{Z}$, then $Y$ fibers over the circle with $F$ as a fiber.
For $g=1$, the methods in the above proofs fail. In his paper [@Ni2], Ni wrote that Ozsváth and Szabó suggested to him to use Heegaard Floer homology with twisted coefficients in some Novikov ring in order to extend to the genus $1$ case. Investigating the $g=1$ case of the aforementioned theorems formed the motivation for this paper. Much is known, however, about the Floer homology of torus bundles. For instance, John Baldwin has already computed the untwisted Heegaard Floer homologies of torus bundles with $b_1(Y)=1$ in [@Baldwin].
In this paper, we use Heegaard Floer homology with twisted coefficients in the universal Novikov ring, $\Lambda$, of all formal power series of the form $$\lambda = \sum_{r \in\mathbb{R}}a_r t^r,\,\#\{r\in\mathbb{R}|a_r \neq0,r\leq c\}<\infty \text{ for all } c\in\mathbb{R}\$$ where the coefficients $a_r\in\mathbb{R}$, see [@MD Section 11.1]. Given a cohomology class $[\omega]\in H^2(Y;\mathbb{Z})$, $\Lambda$ can be given a $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module structure, and this gives rise to a twisted Heegaard Floer homology $\underline{HF}^+(Y;\Lambda_\omega)$. This version was first defined in [@OSzAnn1 section 10] and can also be derived from the definition of general twisted Heegaard Floer homology in [@OSzAnn2]. We will describe this group explicitly in Section \[subsection:omegatwist\]. It is worth noting that Heegaard Floer homology with twisted coefficients in a certain Novikov ring has already been studied extensively in [@JM]. The main theorem we prove in this paper is the following.
\[theorem:main\] Suppose $Y$ is a closed oriented $3$–manifold which fibers over the circle with torus fiber $F$, $[\omega] \in H^2(Y;\mathbb{Z})$ is a cohomology class such that $\omega(F)\neq0$. Then we have an isomorphism of $\Lambda$–modules $$\underline{HF}^+(Y;\Lambda_{\omega}) \cong \Lambda.$$
In the setting of Monopole Floer homology, a corresponding version of this theorem was proved in [@KM Theorem 42.7.1]. In an upcoming paper [@AiN], it is proved that the converse of the above theorem also holds, i.e the twisted Heegaard Floer homology determines whether an irreducible $3$–manifold is a torus bundle over the circle.
This paper is organized as follows. We provide a review of Heegaard Floer homology with twisted coefficients in section \[section:review\], including the most pertinent example, $S^1\times S^2$. In section \[section:exact\] we prove a relevant exact triangle for $\omega$–twisted Heegaard Floer homology and prove Theorem \[theorem:main\].
**Acknowledgements.** The authors would like to thank Peter Ozsváth for continued guidance and support. The first author would also like to thank Yi Ni for suggesting the problem and providing some key ideas.
This work was carried out while the first author was an exchange graduate student at Columbia University, supported by the China Scholarship Council. He is grateful to the Columbia math department for its hospitality.
Review of Twisted Coefficients {#section:review}
==============================
We recall the construction of Heegaard Floer homology with twisted coefficients, see [@OSzAnn2] and [@OSzGenus] for more details. To a closed oriented $3$–manifold $Y$ we associate a pointed Heegaard diagram $(\Sigma,\mathbf{\alpha}, {\bf \beta}, z)$, where $\Sigma$ is an an oriented surface of genus $g\geq1$ and ${\bf \alpha} = \{\alpha_1,...,\alpha_g\}$ and ${\bf \beta} = \{\beta_1,...\beta_g\}$ are sets of attaching circles (assumed to intersect transversely) for the two handlebodies in the Heegaard decomposition. These give a pair of transversely intersecting $g$-dimensional tori $\mathbb{T}_\alpha = \alpha_1\times\cdots\times\alpha_g$ and $\mathbb{T}_\beta=\beta_1\times\cdots\times\beta_g$ in the symmetric product $Sym^g(\Sigma)$. Recall that the basepoint $z$ gives a map $\mathfrak{s}_z:\mathbb{T}_\alpha\cap \mathbb{T}_\beta\to Spin^c(Y)$. Given a Spin$^c$ structure ${\bf \mathfrak{s}}$ on Y, let $\mathfrak{S}\subset \mathbb{T}_\alpha\cap \mathbb{T}_\beta$ be the set of intersection points ${\bf x}\in \mathbb{T}_\alpha\cap \mathbb{T}_\beta$ such that $\mathfrak{s}_z({\bf x})={\bf \mathfrak{s}}$.
Given intersection points ${\bf x}$ and ${\bf y}$ in $\mathbb{T}_\alpha\cap \mathbb{T}_\beta$, let $\pi_2({\bf x},{\bf y})$ denote the set of homotopy classes of Whitney disks from ${\bf x}$ to ${\bf y}$. There is always a natural map from $\pi_2({\bf x},{\bf x})$ to $H^1(Y;\mathbb{Z})$ obtained as follows: each $\phi\in \pi_2({\bf x},{\bf x})$ naturally gives rise to an associated two-chain in $\Sigma$ whose boundary is a collection of circles among the ${\bf \alpha}$ and ${\bf \beta}$ curves. We then close off this two-chain by gluing copies of the attaching disks for the handlebodies in the Heegaard diagram for $Y$. The Poincaré dual of this two-cycle is the associated element of $H^1(Y;\mathbb{Z})$.
Given a Spin$^c$ structure ${\bf \mathfrak{s}}$ on $Y$ and an ${\bf \mathfrak{s}}$-admissible Heegaard diagram $(\Sigma,{\bf \alpha}, {\bf \beta},z)$ for $Y$, an *additive assignment* is a collection of maps $$A=\{A_{{\bf x},{\bf y}}:\pi_2({\bf x},{\bf y})\to H^1(Y;\mathbb{Z})\}_{{\bf x,y}\in \mathfrak{S}}$$ so that:
- when ${\bf x}={\bf y},$ $A_{{\bf x},{\bf x}}$ is the canonical map from $\pi_2({\bf x},{\bf x})$ onto $H^1(Y;\mathbb{Z})$ defined above.
- A is compatible with splicing in the sense that if ${\bf x,y,u}\in \mathfrak{S}$ then for each $\phi_1\in\pi_2({\bf x},{\bf y})$ and $\phi_2\in\pi_2({\bf y},{\bf u})$, we have that $A(\phi_1\ast\phi_2) = A(\phi_1)+A(\phi_2)$.
- $A_{\mathbf{x},\mathbf{y}}(S*\phi)=A_{\mathbf{x},\mathbf{y}}(\phi)$ for $S\in\pi_2(Sym^g(\Sigma_g))$.
Additive assignments may be constructed with the help of a *complete system of paths* as described in [@OSzAnn2].
We write elements in the group-ring $\mathbb{Z}[H^1(Y;\mathbb{Z})]$ as finite formal sums\
$\sum _{g\in H^1(Y;\mathbb{Z})} n_g\cdot e^g$ for $n_g\in\mathbb{Z}$. The *universally twisted Heegaard Floer complex*, $\underline{CF}^{\infty}(Y,{\bf \mathfrak{s}};\mathbb{Z}[H^1(Y;\mathbb{Z})],A)$, is the free $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module on generators $[{\bf x},i]$ for ${\bf x}\in \mathfrak{S}$ and $i\in\mathbb{Z}$. The differential, $\underline{\partial}^{\infty}$, is given by $$\underline{\partial}^{\infty}[x,i] = \sum_{ {\bf y}\in\mathfrak{S} }\sum_{ \genfrac{}{}{0pt}{}{ \phi\in\pi_2({\bf x},{\bf y})}{\mu(\phi)=1} } \#\widehat{\mathcal{M}}(\phi)\cdot e^{A(\phi)}\otimes[{\bf y},i-n_z(\phi)].$$ Here $\mu(\phi)$ denotes the Maslov index of $\phi$, the formal dimension of the space $\mathcal{M}(\phi)$ of holomorphic representatives, $n_z(\phi)$ denotes the intersection number of $\phi$ with the subvariety $\{z\}\times Sym^{g-1}(\Sigma) \subset Sym^g(\Sigma)$, and $\widehat{\mathcal{M}}(\phi)$ denotes the quotient of $\mathcal{M}(\phi)$ under the natural action of $\mathbb{R}$. Just as in the untwisted setting, this complex admits a $\mathbb{Z}[U]$–action via $U:[\mathbf{x},i]\mapsto[\mathbf{x},i-1]$. This gives rise to variants $\underline{CF}^+,\underline{CF}^-$, and $\underline{\widehat{CF}}$, denoted collectively as $\underline{CF}^\circ$. The homology groups of this complexes are the universally twisted Heegaard Floer homology groups $\underline{HF}^\circ$.
More generally, given any $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module $M$, we may form Floer homology groups with coefficients in $M$ by taking $\underline{HF}^\circ(Y,{\bf \mathfrak{s}};M)$ as the homology of the complex $$\underline{CF}^\circ(Y,{\bf \mathfrak{s}};M,A):=\underline{CF}^\circ(Y,{\bf \mathfrak{s}};\mathbb{Z}[H^1(Y;\mathbb{Z})],A)\otimes_{\mathbb{Z}[H^1(Y;\mathbb{Z})]}M.$$ For instance, by taking $M=\mathbb{Z}$, thought of as being a trivial $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module, one recovers the ordinary untwisted Heegaard Floer homology, $CF^\circ(Y,\mathfrak{s})$.
In [@OSzAnn1], it is proved that the homologies defined above are independent of the choice of additive assignment $A$ and are topological invariants of the pair $(Y,\mathfrak{s})$. As in the untwisted setting, these groups are related by long exact sequences
$$\xymatrix{ \cdots \ar[r] & \underline{\widehat{HF}}(Y,\mathfrak{s};M) \ar[r] & \underline{HF}^+(Y,\mathfrak{s};M) \ar[r]^U & \underline{HF}^+(Y,\mathfrak{s};M) \ar[r] & \cdots }$$
and $$\xymatrix{ \cdots \ar[r] & \underline{HF}^-(Y,\mathfrak{s};M) \ar[r]^\iota & \underline{HF}^\infty(Y,\mathfrak{s};M) \ar[r]^\pi & \underline{HF}^+(Y,\mathfrak{s};M) \ar[r] & \cdots }$$
Of course, the chain complex $\underline{CF}^\circ(Y,\mathfrak{s};M)$ is obtained from the chain complex in the universally twisted case, $\underline{CF}^\circ(Y,\mathfrak{s};\mathbb{Z}[H^1(Y;\mathbb{Z})])$, by a change of coefficients and hence the corresponding homology groups are related by the universal coefficients spectral sequence ([@CE]).
$\omega$–twisted Heegaard Floer homology {#subsection:omegatwist}
----------------------------------------
In this section we briefly recall the notion of $\omega$–twisted Heegaard Floer homology, following [@OSzAnn1; @OSzGenus].
The universal Novikov ring is defined to be $$\Lambda = \left\{\sum_{r \in \mathbb{R}}a_r t^r\bigg|a_r\in\mathbb{R},
\;\#\{a_r|a_r\ne0, r \leq c\}<\infty\quad \text{\rm for all $c\in\mathbb R$}
\right\},$$ endowed with the following multiplication law which makes it into a field: $$(\sum_{r \in \mathbb{R}}a_rt^r ) \cdot (\sum_{r \in \mathbb{R}}b_rt^r)=\sum_{r \in \mathbb{R}}(\sum_{s \in \mathbb{R}}a_sb_{r-s})t^r.$$ Furthermore, by fixing a cohomology class $[\omega]\in H^2(Y;\mathbb{R})$ we can give $\Lambda$ a\
$\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module structure via the ring homomorphism induced by: $$\begin{array}{ccl}
H^1(Y;\mathbb{Z})&\to &\mathbb{R}\\
\gamma &\mapsto&
\int_{Y} \gamma \wedge \omega
\end{array}.$$ When we are interested in its $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module structure, we denote it as $\Lambda_{\omega}$. This $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module structure gives rise to a twisted Heegaard Floer homology $\underline{HF}^+(Y;\Lambda_{\omega})$, which we refer to as $\omega$–*twisted Heegaard Floer homology*. More concretely, it can be defined as follows. Choose a weakly admissible pointed Heegaard diagram $(\Sigma,\mbox{\boldmath${\alpha}$}, \mbox{\boldmath${\beta}$},z)$ for $Y$ and fix a 2–cocycle representative $\omega\in[\omega]$. Every Whitney disk $\phi$ in $\mathrm{Sym}^g(\Sigma)$ (for $\mathbb{T}_\alpha$ and $\mathbb{T}_\beta$) gives rise to a two-chain $[\phi]$ in $Y$ by coning off partial $\alpha$ and $\beta$ circles with gradient trajectories in the $\alpha$ and $\beta$ handlebodies. The evaluation of $\omega$ on $[\phi]$ depends only on the homotopy class of $\phi$ and is denoted $\int_{[\phi]}\omega$ (or sometimes $\omega([\phi]))$. The $\omega$–twisted chain complex $\underline{CF}^+(Y; \Lambda_{\omega})$ is the free $\Lambda$–module generated by $[x,i]$ with $x \in
\mathbb{T}_\alpha \cap \mathbb{T}_\beta$ and integers $i \geq 0$, endowed with the following differential: $$\underline{\partial}^+[{\bf x},i] = \sum_{{\bf y}\in \mathbb{T}_\alpha\cap \mathbb{T}_\beta}\sum_{ \{\phi\in\pi_2({\bf x}, {\bf y})|\mu(\phi)=1\}} \#\widehat{\mathcal{M}}(\phi)[{\bf y},i-n_z(\phi)]\cdot t^{\int_{[\phi]}\omega}.$$ Its homology is the $\omega$-twisted Heegaard Floer homology $\underline{HF}^+(Y;\Lambda_{\omega})$. Notice that this group is both a module for $\mathbb{Z}[H^1(Y;\mathbb{Z})]$ and a module for $\Lambda$. Notice that although the differential depends on the choice of 2–cocycle representative $\omega\in[\omega]$, the isomorphism class of the chain complex only depends on the cohomology class. The advantage of using this viewpoint is that we avoid altogether the notion of an “additive assignment". It is easy to see, however, that the complex defined above is isomorphic to one obtained by chosing an additive assignment and then tensoring with the $\mathbb{Z}[H^1(Y;\mathbb{Z})]$-module $\Lambda_{\omega}$.
Suppose $W:Y_1 \to Y_2$ is a $4$–dimensional cobordism from $Y_1$ to $Y_2$ given by a single $2$–handle addition and we have a cohomology class $[\omega]\in H^2(W;\mathbb{R})$. Then there is an associated Heegaard triple $(\Sigma,{\bf \alpha}, {\bf \beta}, {\bf \gamma}, z)$ and $4$–manifold $X_{\alpha\beta\gamma}$ representing $W$ minus a one complex. Similar to before, a Whitney triangle $\psi\in\pi_2({\bf x},{\bf y},{\bf w})$ determines a two-chain in $X_{\alpha\beta\gamma}$ on which we may evaluate a representative, $\omega\in[\omega]$. As before, this evaluation depends only on the homotopy class of $\psi$ and is denoted by $\int_{[\psi]}\omega$. This gives rise to a $\Lambda$–equivariant map $$\underline{F}^+_{W;\omega} \co \underline{HF}^+(Y_1;\Lambda_{\omega|_{Y_1}})\to \underline{HF}^+(Y_2;\Lambda_{\omega|_{Y_2}})$$defined on the chain level by $$\underline{f}^+_{W;\omega}[{\bf x},i] = \sum_{{\bf y}\in \mathbb{T}_\alpha \cap \mathbb{T}_\gamma} \sum_{ \genfrac{}{}{0pt}{}{\psi\in\pi_2({\bf x},\Theta,{\bf y})}{\mu(\psi)=0} } \#\mathcal{M}(\psi) [{\bf y},i-n_z(\psi)]\cdot t^{\int_{[\psi]}\omega}.$$ where $\Theta \in \mathbb{T}_\beta \cap \mathbb{T}_\gamma$ represents a top dimensional generator for the Floer homology $HF^{\leq 0}(Y_{\beta \gamma}) \cong \wedge^{\ast} H^1(Y_{\beta \gamma})\otimes\mathbb{Z}[U]$ of the $3$–manifold determined by the Heegaard diagram $(\Sigma,{\bf \beta},{\bf \gamma},z)$, which is a connected sum $\#^{g-1}(S^1 \times S^2)$. These maps may be decomposed as a sum of maps, $$\underline{F}^+_{W;\omega} = \sum_{\mathfrak{s}\in Spin^c(W)}\underline{F}^+_{W,\mathfrak{s};\omega},$$ according to Spin$^c$ equivlence classes of triangles, just as in the untwisted setting. This can be extended to arbitrary (smooth, connected) cobordisms from $Y_1$ to $Y_2$ as in [@OSzFour]. These maps also satisfy a *composition law*: if $W_1$ is a cobordism from $Y_1$ to $Y_2$ and $W_2$ is a cobordism from $Y_2$ to $Y_3$, and we equip $W_1$ and $W_2$ with Spin$^c$ structures $\mathfrak{s}_1$ and $\mathfrak{s}_2$ respectively (whose restrictions agree over $Y_2$), then putting $W=W_1\#_{Y_2}W_2$, for any $[\omega]\in H^2(W;\mathbb{R})$ we have $$\underline{F}^+_{W_2,\mathfrak{s}_2;\omega|_{W_2}} \circ \underline{F}^+_{W_1,\mathfrak{s}_1;\omega|_{W_1}} = \sum_{ \{\mathfrak{s}\in Spin^c(W):\mathfrak{s}|_{W_i}=\mathfrak{s}_i\}} \pm \underline{F}^+_{W,\mathfrak{s};\omega}.$$
Example: $S^1\times S^2$ {#subsection:s1xs2}
------------------------
In this section we calculate twisted Heegaard Floer homologies of $S^1\times S^2$. We start with the universally twisted version $\underline{\widehat{HF}}(S^1\times S^2; \mathbb{Z}[t,t^{-1}])$, where we have identified $\mathbb{Z}[H^1(S^1\times S^2;\mathbb{Z})] \cong \mathbb{Z}[t,t^{-1}]$, the ring of Laurent polynomials. $S^1\times S^2$ has a standard genus one Heegaard decomposition $(\Sigma,\alpha,\beta)$ where $\alpha$ is a homotopically nontrivial embedded curve and $\beta$ is an isotopic translate. For simplicity, we only compute $\underline{\widehat{HF}}$. We make the diagram weakly admissible for the unique torsion Spin$^c$ structure $\mathfrak{s}_0$ by introducing cancelling pairs of intersection points between $\alpha$ and $\beta$. This gives a pair of intersection points $x^+$ and $x^-$. We next need an additive assignment. Notice there is an obvious periodic domain consisting of a pair of (nonhomotopic) disks $D_1$ and $D_2$ connecting $x^+$ and $x^-$. When capped off, the periodic domain gives a sphere representing a generator of $H_2(S^1\times S^2;\mathbb{Z}) \cong \mathbb{Z}$. Hence taking Poincaré dual we recover a generator of $H^1(S^1\times S^2;\mathbb{Z}) \cong \mathbb{Z}$. Identifying $H^1(S^1\times S^2;\mathbb{Z}) \cong \mathbb{Z}$, an additive assignment must assign $1$ to this domain. One way this can be done is by assigning $1$ to $D_1$ and $0$ to $D_2$. Hence we see that the complex $\underline{\widehat{CF}}(S^1\times S^2;\mathbb{Z}[t,t^{-1}])$ is just $$\xymatrix{0 \ar[r] & \mathbb{Z}[t,t^{-1}] \ar[r]^{1-t} & \mathbb{Z}[t,t^{-1}] \ar[r] & 0}.$$ Here, the first copy of $\mathbb{Z}[t,t^{-1}]$ corresponds to $x^+$ and the second corresponds to $x^-$. This complex has homology $\mathbb{Z}$, with trivial $\mathbb{Z}[t,t^{-1}]$–action. If we keep track of gradings, we get the universally twisted Floer homology $$\underline{\widehat{HF}}(S^1\times S^2;\mathbb{Z}[t,t^{-1}]) \cong \mathbb{Z}_{(-\frac{1}{2})}$$ supported only in the torsion Spin$^c$ structure $\mathfrak{s}_0$.
Now let’s turn to an $\omega$-twisted example. We can view $S^1\times S^2$ as $0$-surgery on the unknot in $S^3$. Put $\mu$ a meridian for the unknot. Then $\mu$ defines a curve, also denoted $\mu$, in $S^1\times S^2$. Put $[\omega]=d\cdot PD[\mu]$ for an integer $d$. The complex $\underline{\widehat{CF}}(S^1\times S^2;\Lambda_{\omega})$ is $$\xymatrix{0 \ar[r] &\Lambda \ar[r]^{t^c(1-t^d)} & \Lambda \ar[r] &0}.$$ for some $c\in\mathbb{R}$. Notice when $d \neq 0$, $(1-t^d)$ is invertible in $\Lambda$. Hence $$\underline{\widehat{HF}}(S^1\times S^2;\Lambda_{\omega})=\begin{cases} 0 &\text{when $d \neq 0$}\\
\Lambda \oplus \Lambda&\text{when $d=0$}
\end{cases}.$$
As a final example, we prove a proposition regarding embedded $2$–spheres in $3$–manifolds and $\omega$-twisted coefficients. We begin with the following
\[lemma:connectsum\] Let $Y_1$ and $Y_2$ be a pair of closed oriented $3$–manifolds and fix cohomology classes $[\omega_i]\in H^2(Y_i;\mathbb{Z})$. By the Mayer-Vietoris sequence we get a corresponding cohomology class $\omega_{1}\#\omega_{2}\in H^2(Y_1\#Y_2;\mathbb{Z}) \cong H^2(Y_1;\mathbb{Z})\oplus H^2(Y_2;\mathbb{Z})$. Then we have an isomorphism of $\Lambda$–modules $$\underline{\widehat{HF}}(Y_1\#Y_2;\Lambda_{\omega_1\#\omega_2}) \cong \underline{\widehat{HF}}(Y_1;\Lambda_{\omega_{1}})\otimes_{\Lambda} \underline{\widehat{HF}}(Y_2;\Lambda_{\omega_{2}}).$$
This follows readily from the methods of proof of [@OSzAnn2 Proposition 6.1] and the fact that $\Lambda$ is a field.
This allows us to prove
\[proposition:spheres\] Let $S \subset Y^3$ be an embedded nonseparating $2$–sphere in a $3$–manifold $Y$. Suppose $[\omega] \in H^2(Y;\mathbb{Z})$ is a cohomology class such that $\omega([S])\neq0$. Then $\underline{HF}^+(Y;\Lambda_\omega)=0$.
Just as in the untwisted theory, $\underline{HF}^+(Y;M)$ vanishes if and only if $\underline{\widehat{HF}}(Y;M)$ vanishes, so it suffices to show that $\underline{\widehat{HF}}(Y;\Lambda_\omega)=0$. Notice that $Y$ contains an $S^1\times S^2$ summand in its prime decomposition. Hence $Y\cong S^1\times S^2\#Y'$ for some $3$–manifold $Y'$. Now $\omega\in H^2(Y;\mathbb{Z})\cong H^2(S^1\times S^2;\mathbb{Z})\oplus H^2(Y';\mathbb{Z})$ corresponds to classes $\omega_1\in H^2(S^1\times S^2;\mathbb{Z})$ and $\omega_2\in H^2(Y';\mathbb{Z})$ with $\omega_1([S])\neq0$. We already know that $\underline{\widehat{HF}}(S^1\times S^2;\Lambda_{\omega_{1}})=0$ from the above calculation, so the proposition follows from Lemma \[lemma:connectsum\].
Exact sequence for $\omega$–twisted Floer homology {#section:exact}
==================================================
In this section we first prove a long exact sequence for the $\omega$–twisted Heegaard Floer homologies and then use it to prove Theorem \[theorem:main\]. It is interesting to notice that there is a similar exact sequence in Monopole Floer homology with local coefficients, see [@KMOS Section 5]. Our proof is a slight modification of the proof of the usual surgery exact sequence in Heegaard Floer homology.
Let $K \subset Y$ be framed knot in a $3$–manifold $Y$ with framing $\lambda$ and meridian $\mu$. Given an integer $r$, let $Y_r(K)$ denote the $3$–manifold obtained from $Y$ by doing Dehn surgery along the knot $K$ with framing $\lambda+r\mu$. Let $\eta \subset Y-nbd(K)$ be a closed curve in the knot complement. Then for any integer $r$, $\eta \subset Y-nbd(K) \subset Y_r(K)$ is a closed curve in the surgered manifold $Y_r(K)$, we denote its Poincaré dual by $[\omega_r] \in H^2(Y_r(K);\mathbb{Z})$. Also note that $\eta \times I$ represents a relative homology class in the corbordisms $W_0 \co Y \to Y_0(K)$, $W_1 \co Y_0(K) \to Y_1(K)$ and $W_2 \co Y_1(K) \to Y$. So as in Section \[subsection:omegatwist\] it gives rise to homomorphisms between $\omega$–twisted Floer homologies: $$\underline{F}^+_{W_0;PD(\eta \times I)} \co \underline{HF}^+(Y;\Lambda_{\omega}) \to \underline{HF}^+(Y_0(K);\Lambda_{\omega_0}),$$ $$\underline{F}^+_{W_1;PD(\eta \times I)} \co \underline{HF}^+(Y_0(K);\Lambda_{\omega_0}) \to \underline{HF}^+(Y_1(K);\Lambda_{\omega_1}),$$ and $$\underline{F}^+_{W_2;PD(\eta \times I)} \co \underline{HF}^+(Y_1(K);\Lambda_{\omega_1}) \to \underline{HF}^+(Y;\Lambda_{\omega}),$$ where $\omega=PD(\eta) \in H^2(Y;\mathbb{Z})$. We denote the corresponding maps on the chain level by $\underline{f}^+_{W_0;PD(\eta \times I)}$, $\underline{f}^+_{W_1;PD(\eta \times I)}$ and $\underline{f}^+_{W_2;PD(\eta \times I)}$ respectively.
\[theorem:exact\] The maps above form an exact sequence of $\Lambda$-modules:
$$\label{Seq:01}
\begin{xymatrix}{ \underline{HF}^+(Y;\Lambda_\omega) \ar[rr]
&&\underline{HF}^+(Y_0(K);\Lambda_{\omega_0}) \ar[dl]\\
&\underline{HF}^+(Y_1(K);\Lambda_{\omega_1})\ar[ul]&}
\end{xymatrix}$$
Furthermore, analogous exact sequences hold for “hat" versions as well.
Find a Heegaard diagram $(\Sigma, \{\alpha_1,\cdots, \alpha_g\},\{\beta_1,\cdots, \beta_g\},z)$ compatible with the knot $K$, i.e $K$ lies in the handlebody specified by the $\beta$ curves and $\beta_1$ is a meridian for $K$. For each $i \geq 2$ let $\gamma_i,\delta_i$ be exact Hamiltonian isotopies of $\beta_i$. Let $\gamma_1=\lambda$, $\delta_1=\lambda+\mu$ be the $0$-framed and $1$-framed longitude of the knot $K$, respectively. We assume the Heegaard quadruple $(\Sigma, \alpha,\beta,\gamma,\delta,z)$ is weakly admissible in the sense of [@OSzAnn1]. It is easy to see that $Y_{\alpha \beta}=Y$, $Y_{\alpha \gamma}=Y_0(K)$, $Y_{\alpha \delta}=Y_1(K)$, and $Y_{\beta \gamma} \cong Y_{\gamma \delta} \cong Y_{\beta \delta} \cong \#^{g-1}S^2 \times S^1$.
Following [@OSzDouble], we define a map $$h_1\co \underline{CF}^+(Y;\Lambda_{\omega}) \to \underline{CF}^+(Y_1(K);\Lambda_{\omega_1})$$ by counting holomorphic rectangles: $$h_1([\mathbf{x},i])=\sum_{\mathbf{w} \in \mathbb{T}_\alpha \cap \mathbb{T}_\delta}\sum_{ \genfrac{}{}{}{0pt}{\varphi \in \pi_2(\mathbf{x},\Theta_{\beta \gamma},\Theta_{ \gamma \delta},\mathbf{w})}\mu(\varphi)=-1}\# \mathcal{M}(\varphi)[\mathbf{w},i-n_z(\varphi)]t^{\int_{[\varphi]}PD(\eta \times I)}$$ Similarly we define $h_2\co \underline{CF}^+(Y_0(K);\Lambda_{\omega_0}) \to \underline{CF}^+(Y;\Lambda_{\omega})$ and\
$h_3\co \underline{CF}^+(Y_1(K);\Lambda_{\omega_1}) \to \underline{CF}^+(Y_0(K);\Lambda_{\omega_0})$.
We claim that $h_1$ is a nullhomotopy of $\underline{f}^+_{W_1;PD(\eta \times I)}\circ\underline{f}^+_{W_0;PD(\eta \times I)}$. To see this, we consider the moduli spaces of holomorphic rectangles of Maslov index $0$. This moduli spaces can have 6 kinds of ends:
1. splicing holomorphic discs at one of its 4 corners, and
2. splicing two holomorphic triangles. Triangles may be spliced in two ways: one triangle for $X_{\alpha \beta\gamma}$ and one triangle for $X_{\alpha \gamma \delta}$, or one triangle for $X_{\alpha \beta\delta}$ and one triangle for $X _{\beta \gamma \delta}$
Notice $PD(\eta \times I)$ is $0$ when restricted to the corners $Y_{\beta \gamma}$ and $Y_{\gamma \delta}$: in fact, we can make $\eta\times I$ disjoint from these manifolds since $\eta$ may be pushed completely into the $\alpha$-handlebody, $U_\alpha$, by cellular approximation (see figure \[fig:4m\]).
This imples that $$\underline{CF}^+(Y_{\beta\gamma};\Lambda_{PD(\eta \times I)|_{Y_{\beta \gamma}}})\cong CF^+(Y_{\beta\gamma})\otimes_\mathbb{Z}\Lambda$$ and all differentials are trivial (informally, we are using an “untwisted" count). For the end coming from splicing two holomorphic triangles, one for $X_{\alpha \beta\delta}$ and one for $X _{\beta \gamma \delta}$, it is also true that $PD(\eta \times I)$ is 0 when restricted to the 4-manifold $X_{\beta \gamma \delta}$ (again, since $\eta$ may be pushed completely into $U_\alpha$). Therefore we are counting holomorphic triangles in $X_{\beta \gamma \delta}$ without twisting. In [@OSzAnn2] it is shown that the untwisted counting of holomorphic triangles in $X_{\beta \gamma \delta}$ is zero. This leaves three terms remaining:
1. splicing a disc at corner $Y_{\alpha \beta}$ counted with twisting by $PD(\eta \times I)|_{Y_{\alpha \beta}}=[\omega]$. This corresponds to $h_1 \circ \partial$.
2. splicing a disc at corner $Y_{\alpha \delta}$ counted with twisting by $PD(\eta \times I)|_{Y_{\alpha \delta}}=[\omega_1]$. This corresponds to $\partial \circ h_1$, and
3. splicing two holomorphic triangles from $X_{\alpha \beta\gamma}$ and $X_{\alpha \gamma \delta}$ counted with twisting by $PD(\eta \times I)$. This corresponds to $\underline{f}^+_{W_1;PD(\eta \times I)}\circ\underline{f}^+_{W_0;PD(\eta \times I)}$.
From the fact that the moduli space must have total end zero, it is clear that the sum of the above $3$ terms are zero, i.e $h_1$ is a homotopy connecting $\underline{f}^+_ {W_1;PD(\eta \times I)}\circ\underline{f}^+_{W_0;PD(\eta \times I)}$ to the zero map. This shows that $\underline{F}^+_{W_1;PD(\eta \times I)}\circ\underline{F}^+_{W_0;PD(\eta \times I)}=0$ on the homology level. The same argument shows that $\underline{F}^+_{W_2;PD(\eta \times I)}\circ\underline{F}^+_{W_1;PD(\eta \times I)}=0$ and $\underline{F}^+_{W_0;PD(\eta \times I)}\circ\underline{F}^+_{W_2;PD(\eta \times I)}=0$ as well.
At last we prove that Sequence (\[Seq:01\]) is exact. Using the homological algebra method in [@OSzDouble] we need to show that $h \circ\underline{f}^{+}+\underline{f}^{+} \circ h$ is homotopic to the identity map. This can be done by counting holomorphic pentagons and noticing that $PD(\eta \times I)$ is zero when restricting to $Y_{\beta \gamma}$,$Y_{ \gamma \delta}$,$Y_{\delta \beta'}$,$X_{\beta \gamma \delta}$, $X_{\gamma \delta \beta'}$ and $X_{\beta \gamma \delta \beta'}$ (this follows since we may assume that $\eta\subset U_\alpha$). This shows that the counts there are “untwisted". From this observation one can easily see that everything in the proof of exactness in [@OSzDouble] can go through to our twisted version.
In the above theorem, the cohomology classes $[\omega_r]$ are integral. In practice one may need to use real cohomology class as well. In that situation, for a given homology class $[\omega] \in H^2(Y;\mathbb{R})$ we can express it as finite sum $$[\omega]=\sum a_iPD(\eta_i)$$ where $\eta_i$ are closed curves in the knot complement and $a_i \in \mathbb{R}$. Each $\eta_i$ can be viewed as a closed curve in $Y_r(K)$, so the expression $\sum a_iPD(\eta_i)$ also gives a real cohomology class in $Y_r(K)$, denote by $[\omega_r] \in H^2(Y_r(K);\mathbb{R})$. In the coborsim $W_r$, $$\sum a_i PD(\eta_i \times I)$$ is a real cohomology class in $H^2(W_r;\mathbb R)$, hence gives rise to homomorphism between $\omega$–twisted Floer homologies. With this understood, it is easy to see that Theorem \[theorem:exact\] still holds.
The exact sequence in Theorem \[theorem:exact\] depends on $\eta$, not just its Poincaré dual $[\omega] \in H^2(Y;\mathbb{Z})$. In fact if we take another closed curve $\eta'=\eta +k \cdot \mu$ (where $\mu$ is a meridian of $K$), this doesn’t change $[\omega]$, but may change $[\omega_0], [\omega_1]$ and the exact sequence. For example, take $K \subset S^3$ to be the unknot and $\eta=k \cdot \mu$ in the knot complement, then $[\omega_0]$ is $k$ times the generator of $H^2(S^2 \times S^1;\mathbb{Z})$. When $k \neq 0$, the corresponding exact sequence for the hat version is $$\begin{xymatrix}{
\cdots \to \widehat{\underline{HF}}(S^3_0(K);\Lambda_{\omega_0})\ar[r] \ar[d]^\cong & \widehat{\underline{HF}}(S^3_1(K);\Lambda_{\omega_1}) \ar[r]\ar[d]^\cong &\widehat{\underline{HF}}(S^3;\Lambda_{\omega}) \ar[d]^\cong \to \cdots \\
0\ar[r] & \Lambda \ar[r]^{1-t^k} & \Lambda }
\end{xymatrix}$$ Clearly it depends on $k$. When $k=0$, the exact sequence is obtained from the corresponding exact sequence for untwisted Heegaard floer homology by tensoring with $\Lambda$.
In [@OSzGenus], Ozsváth and Szabó used another version of twisted Floer homology $\underline{HF}^+(Y;[\omega])$, which is defined by using the $\mathbb{Z}[H^1(Y;\mathbb{Z})]$–module $\mathbb{Z}[\mathbb{R}]$. The $\omega$–twisted Floer homology we used in this paper can be viewed as a completion of $\underline{HF}^+(Y;[\omega])$. It is easy to see that there is a similar exact sequence in their context. More precisely, we have the following exact sequence:
$$\begin{xymatrix}{ \underline{HF}^+(Y;[\omega]) \ar[rr]
&&\underline{HF}^+(Y_0(K);[\omega_0]) \ar[dl]\\
&\underline{HF}^+(Y_1(K);[\omega_1])\ar[ul]&}
\end{xymatrix}$$
With the above exact sequences in place, we can now prove Theorem \[theorem:main\]. We merely mimic the proof of Theorem 5.2 in [@OSzSympl].
For a given cohomology class $[\omega]\in H^2(Y;\mathbb{Z})$ with $\omega(F)=d \neq 0$, choose a closed curve $\eta \subset Y$ such that its Poincaré dual $PD(\eta)$ equals $[\omega]$. Since the mapping class group of a torus is generated as a monoid by right-handed Dehn twists along non-seperating curves, we can connect $Y$ to the three-manifold $S_0^3(T)$, which is obtained from $S^3$ by performing $0$–surgery on the right-handed trefoil, by a sequence of torus bundles $$\pi_i:Y^i \to S^1$$ and cobordisms $$\xymatrix{ Y=Y^0 \ar[r]^<<<<{W_0} & Y^1 \ar[r]^<<<<{W_1} & \cdots \ar[r]^<<<<{W_{n-1}} & Y^n = S_0^3(T) }$$ such that the monodromy of $Y^{i+1}$ differs from that of $Y^i$ by a single right-handed Dehn twist along a nonseparating knot $K_i$ which lies in a fiber $F_i$ of $\pi_i$. The curve $\eta\subset Y$ induces curves $\eta_i\subset Y^i$ which can be assumed disjoint from $K_i$. In this way, we get a sequence of cohomology classes $\omega_i=PD(\eta_i)\in H^2(Y^i;\mathbb{Z})$ such that $\omega_i(F_i)=d\neq0$. The cobordism $W_i$ is obtained by attaching a single 2-handle to $Y^i \times I$ along the knot $K_i$ with framing $-1$ (with respect to the framing $K_i$ inherits from the fiber $F_i$). Since $\eta_i$ is disjoint from $K_i$, $\eta_i\times I$ defines a relative homology class $[\eta_i\times I]\in H_2(W_i,\partial W_i;\mathbb{Z})$ and hence its Poincaré dual gives rise to homomorphisms between $\omega$–twisted Floer homologies: $$\underline{F}^+_{W_i;PD(\eta_i \times I)}: \underline{HF}^+(Y^i;\Lambda_{\omega_i}) \to \underline{HF}^+(Y^{i+1};\Lambda_{\omega_{i+1}}).$$ We claim it is an isomorphism. Notice that $Y^{i+1} = (Y^i)_{-1}(K_i)$ where the 0-framing of $K_i$ is defined to be the framing $K_i$ inherits from the fiber, $F_i$. Now consider $(Y^i)_0(K_i)$. This manifold contains a $2$–sphere $S_i$ (which is obtained from $F_i$ by surgering along $K_i$) and also an induced curve $\eta_i$ such that $\eta_i\cdot S_i=d\neq0$, therefore $\underline{HF}^+((Y^i)_0(K_i);\Lambda_{PD(\eta_i)})=0$ by proposition \[proposition:spheres\]. The exact sequence (\[Seq:01\]) now proves the claim.
This shows that $$\underline{HF}^+(Y;\Lambda_{\omega}) \cong \underline{HF}^+(S^3_0(T);\Lambda_{PD(\eta)})$$ where $\eta$ is the induced curve in $S_0^3(T)$. We now identify the latter group. For simplicity we write $\omega=PD(\eta)$. Identifying $\mathbb{Z}[H^1(S^3_0(T);\mathbb{Z})]$ with $\mathbb{Z}[t,t^{-1}]$, Ozsváth and Szabó show in [@OSzAbsGr] that there is an identification of $\mathbb{Z}[t,t^{-1}]$–modules: $$\underline{HF}^+_k(S_0^3(T);\mathbb{Z}[t,t^{-1}]) \cong \left\{
\begin{array}{rl}
\mathbb{Z} & \text{if } k\equiv-1/2 \,(\text{mod }2)\,\text{and } k\geq-1/2\\
\mathbb{Z}[t,t^{-1}] & \text{if } k=-3/2\\
0 & \text{otherwise}
\end{array}\right.$$ Where the lefthand group is the universally twisted Heegaard Floer homology of $S^3_0(T)$, the $\mathbb{Z}$’s on the right are trivial $\mathbb{Z}[H^1(S^3_0(T);\mathbb{Z})]$–modules, and $\mathbb{Z}[t,t^{-1}]$ is a module over itself in the natural way. By definition, $$\underline{CF}^+(S_0^3(T);\Lambda_{\omega}) = \underline{CF}^+(S_0^3(T);\mathbb{Z}[t,t^{-1}])\otimes_{\mathbb{Z}[t,t^{-1}]}\Lambda_{\omega}.$$ We now apply the universal coefficients spectral sequence. We need only compute $Tor^{\mathbb{Z}[t,t^{-1}]}_q(\mathbb{Z},\Lambda_{\omega})$. We start with the free $\mathbb{Z}[t,t^{-1}]$-resolution of $\mathbb{Z}$: $$\xymatrix{
0 \ar[r] & \mathbb{Z}[t,t^{-1}] \ar[r]^{1-t} & \mathbb{Z}[t,t^{-1}]\ar[r] & \mathbb{Z} \ar[r] &0 }.$$ Tensoring this complex over $\mathbb{Z}[t,t^{-1}]$ with $\Lambda_{\omega}$ and agumenting gives the complex $$\xymatrix{
0 \ar[r] & \Lambda \ar[r]^{1-t^d} & \Lambda \ar[r] & 0 }.$$ Since we’re working over $\Lambda$ and $d\neq0$, the middle map is an isomorphism and we see that $Tor^{\mathbb{Z}[t,t^{-1}]}_q(\mathbb{Z},\Lambda_{\omega})=0$ for all $q$. Applying the universal coefficients spectral sequence, we obtain an isomorphism of $\Lambda$-modules $\underline{HF}^+(S_0^3(T);\Lambda_{\omega})\cong \Lambda$. Hence we have a $\Lambda$-module isomorphism $$\underline{HF}^+(Y;\Lambda_{\omega})\cong \Lambda.$$ Clearly $\underline{HF}^+(Y;\Lambda_{\omega})$ is supported in a single Spin$^c$ structure, since $\Lambda$ is a field. Notice also that this Spin$^c$ structure must be torsion by the conjugation symmetry of $\underline{HF}^\circ$.
It is worth noting that alternate proofs of this theorem and proposition \[proposition:spheres\] are possible through the use of inadmissible diagrams, which have been explored by Zhong Tao Wu.
[H]{}
, [*Two applications of twisted Floer homology*]{}, in preparation
, [*Heegard Floer homology and genus one, one boundary component open books*]{}, preprint (2008), available at arXiv:0804.3624v1
, [*Homological algebra*]{}, Reprint of the 1956 original. Princeton Landmarks in Mathematics. Princeton University Press, Princeton, NJ, 1999. xvi+390 pp.
, [*Knot Floer homology detects genus-one fibered knots*]{}, to appear in Amer. J. Math., available at arXiv:math.GT/0603445
, [*Product formulae for Ozsváth–Szabó $4$–manifolds invariants*]{}, to appear in Geom. Topol., available at arXiv:0706.0339
, [*Monopoles and three-manifolds*]{}, New Mathematical Monographs 10, Cambridge University Press, Cambridge (2007)
, [*Monopoles and lens space surgeries*]{}, Ann. of Math. (2) 165 (2007), no. 2, 457–546
, [*$J$-holomorphic curves and symplectic topology*]{}, American Mathematical Society Colloquium Publications, 52. American Mathematical Society, Providence, RI, 2004. xii+669 pp
, [*Knot Floer homology detects fibred knots*]{}, Invent. Math. 170 (2007), no. 3, 577–608
, [*Heegaard Floer homology and fibred $3$–manifolds*]{}, preprint (2007), available at arXiv:0706.2031
, [*Holomorphic disks and topological invariants for closed three-manifolds*]{}, Ann. of Math.(2), 159 (2004), no. 3, 1027–1158
, [*Holomorphic disks and three-manifold invariants: properties and applications*]{}, Ann. of Math.(2), 159 (2004), no. 3, 1159–1245
, [*Holomorphic triangle invariants and the topology of symplectic four-manifolds*]{}, Duke Math. J. 121 (2004), no. 1, 1–34
, [*Absolutely Graded Floer homologies and intersection forms for four-manifolds with boundary*]{}, Adv. Math. 173 (2003), no. 2, 179–261
, [*Holomorphic disks and knot invariants*]{}, Adv. Math. 186 (2004), no. 1, 58–116
, [*Holomorphic disks and genus bounds*]{}, Geom. Topol. 8 (2004), 311–334 (electronic)
, [*On the Heegaard Floer homology of branched double-covers*]{}. Adv. Math. 194 (2005), no. 1, 1–33
, [*Holomorphic triangles and invariants for smooth four-manifolds*]{}, Adv. Math. 202 (2006), no. 2, 326–400.
, [*Floer homology and knot complements*]{}, PhD Thesis, Harvard University (2003), available at arXiv:math.GT/0306378
| {
"pile_set_name": "ArXiv"
} |
---
bibliography:
- 'helical.bib'
---
addtoreset[equation]{}[section]{}
1.5cm Aristomenis Donos\
.6cm
*Blackett Laboratory, Imperial College\
London, SW7 2AZ, U.K.*
\
**Abstract**
> We discuss new types of second order phase transitions in holography by constructing striped black holes in $D=4$ with $AdS_{4}$ asymptotics. In the context of $AdS/CFT$, they provide the gravity duals to field theory phases in which translational symmetry is spontaneously broken due to the formation of current density waves. These black holes are associated to three dimensional CFTs at finite temperature and deformed by a uniform chemical potential. We numerically solve a non-linear system of PDEs in order to construct the black hole geometries and extract some of their thermodynamic properties.
Introduction
============
The AdS/CFT correspondence provides a powerful tool for studying strongly coupled systems. Its possible application to problems relevant to condensed matter physics [@Hartnoll:2011fn] has attracted a lot of attention over the last few years. One aspect is the phase structure of CFTs at finite temperature and deformed by a chemical potential. Thermal phase transitions of CFTs can be studied in the context of holography via phase transitions of black holes with $AdS$ asymptotics. A canonical example for the bulk dual of the normal phase is the electrically charged AdS-Reissner-Nordström (AdS-RN) black hole. At low temperatures, these solutions can develop different types of instabilities and the system will be described by a new branch of black hole solutions in general.
A prototype example of such a phase transition is provided by bulk theories which contain scalars that are minimally coupled to the $U(1)$ gauge field used to deform the boundary theory by the chemical potential [@Gubser:2008px]. In this case, when the normal phase becomes unstable against fluctuations of a charged scalar it leads to low temperature black holes where the global $U(1)$ symmetry is broken. The gauge field becomes massive and we are naturally lead to a holographic description of a superconductor/superfluid [@Gubser:2008px; @Hartnoll:2008vx; @Hartnoll:2008kx].
Another class of instabilities leads to spontaneous breaking of the Euclidian group of symmetries preserved after deforming the theory by a uniform chemical potential. Such phenomena are common in strongly coupled condensed matter systems [@vojta] and are worth investigating in the context of holography. The first examples of spatially modulated unstable modes were realized in $D=5$ bulk dimensions in [@Ooguri:2010xs; @Donos:2011ff] (see also [@Domokos:2007kt] for a probe setup). In these five dimensional examples, the unstable modes break translations but they still preserve a Bianchi $VII_{0}$ subgroup giving the dual field theory a helical structure. The slices of constant bulk radius are still homogeneous and the problem of finding the backreacted black holes was reduced to solving a system of ODEs in [@Donos:2011ff; @Donos:2012wi]. The transition was found to be of second order, in general, while the spatial modulation persists all the way down to zero temperature and the corresponding symmetry breaking ground states were identified in [@Donos:2011ff; @Donos:2012js]. Similar zero temperature solutions have appeared in a classification of homogeneous branes-like solutions in [@Iizuka:2012iv; @Iizuka:2012pn].
Modulated instabilities of electrically charged black holes have also been found in $D=4$ [@Donos:2011bh] [^1] and more recently in [@Donos:2013gda] for any $D\geq 4$. Unlike the homogeneous $D=5$ examples, in this case the backreacted geometry is expected to be inhomogeneous and the solution of PDEs seems necessary. A first construction of such backreacted geometries appeared recently in [@Rozali:2012es] for a particular model of the classes studied in [@Donos:2011bh]. The authors employed a numerical method developed in [@Wiseman:2002zc] to integrate the PDEs resulting from the equations of motion and presented solutions for temperatures below $T\approx 0.9\,T_{c}$. It was found that for a range of temperatures down to $T\approx 0.1\,T_{c}$ the new inhomogeneous solutions where not thermodynamically preferred over the AdS-RN and the system remained dynamically unstable.
Here we will consider two different models, including the one investigated in [@Rozali:2012es], and we will solve the system of PDEs by employing the DeTurck method [@Headrick:2009pv; @Figueras:2011va] in order to solve Einstein’s equations. This method was recently used in [@Horowitz:2012ky; @Horowitz:2012gs] in order to study transport properties of backgrounds with explicitly broken translations. For both models we are considering, we find a second order transition with the striped phase being thermodynamically favored for all $T<T_{c}$. We present solutions starting from $T\approx 0.999\,T_{c}$ but it is possible to construct solutions arbitrarily close to the critical temperature.
The remaining sections of this paper are organized as follows. In section \[sec:the\_model\] we will introduce the $D=4$ models in which we will be interested and review the spatially modulated instabilities of the AdS-RN black hole. In section \[sec:solutions\], after describing the method we chose to use, we will present the resulting solutions along with their thermodynamic properties. We conclude in section \[sec:discussion\] with a summary and discussion.
The Einstein-Maxwell-pseudoscalar model {#sec:the_model}
=======================================
We consider the $D=4$ bulk theory which is described by a metric $g_{\mu\nu}$, a gauge field $A_{\mu}$ and a pseudoscalar field $\varphi$ $$\begin{aligned}
\label{eq:action}
S=\int d^{4}x\sqrt{-g}\,\left(\frac{1}{2}R-\frac{1}{2}\,\left(\partial\varphi\right)^{2}-\frac{\tau(\varphi)}{4}\,F^{2}-V(\varphi)\right)-\frac{1}{2}\int \theta (\varphi)\,F\wedge F\end{aligned}$$ with $F=dA$ being the field strength. The action yields the equations of motion $$\begin{aligned}
\label{eomi}
R_{\mu\nu}-\partial_\mu \varphi\partial_\nu \varphi-g_{\mu\nu}\,V+\tau\,\left(\frac{1}{4}g_{\mu\nu}\,F_{\lambda\rho}F^{\lambda\rho} -F_{\mu\rho}F_{\nu}{}^{\rho}\right)&=0{\notag \\}d\left(\tau\ast F+\vartheta F\right)&=0{\notag \\}d\ast d\varphi+V'*1+\frac{1}{2}\tau'\,F\wedge\ast F+\frac{1}{2}\vartheta'\,F\wedge F&=0\, .\end{aligned}$$ As discussed in [@Donos:2011bh], when the functions $V$, $\tau$ and $\theta$ appearing in the action admit the expansions $$\begin{aligned}
\label{expans}
V=-6+\frac{1}{2}m_{s}^{2}\,\varphi^{2}+\dots,\qquad
\tau=1-\frac{n}{12}\,\varphi^{2}+\dots,\qquad
\vartheta=\frac{c_{1}}{2\sqrt{3}}\,\varphi+\dots\, .\end{aligned}$$ around $\varphi=0$, the equations of motion admit the purely electric AdS-RN black hole solution $$\begin{aligned}
\label{eq:RN}
ds_{4}^{2}=&\frac{1}{z^{2}}\,\left(-f(z)\,dt^{2}+\frac{dz^{2}}{f(z)}+dx_{1}^{2}+dx_{2}^{2} \right){\notag \\}A=&\mu \left(1- z\right),\qquad \varphi=0{\notag \\}f(z)=&\frac{1}{2}\,\left(1-z \right)\,\left(-\mu^{2}z^{3}+4\,z^{2}+4\,z+4 \right)\end{aligned}$$ with $\mu$ being the chemical potential at which the dual field theory is held. In these coordinates, the black hole has a regular horizon located at $z=1$ while the $AdS_{4}$ boundary is at $z=0$. A simple surface gravity calculation gives the temperature $T=\left(12-\mu^{2} \right)/\left(8\pi\right)$ showing that the system reaches its zero temperature limit when $\mu=\sqrt{12}$. The near horizon limit of the extremal solution is simply $AdS_{2}\times \mathbb{R}^{2}$ with a non-trivial background gauge field $$\begin{aligned}
\label{eq:AdS2limit}
ds_{4}^{2}=&\frac{1}{12}\,ds^{2}\left(AdS_{2}\right)+dx_{1}^{2}+dx_{2}^{2}{\notag \\}F=&\sqrt{12}\,\mathrm{Vol}\left(AdS_{2}\right)\end{aligned}$$
The perturbative analysis of [@Donos:2011bh] showed that when the, wavenumber $k$ dependent, matrix $$\begin{aligned}
\label{mmfirstmod}
M^{2}&=\left(\begin{array}{ccc}k^{2} & \frac{1}{\sqrt{3}}k & 0 \\24\sqrt{3}k & 24+k^{2} & -c_{1}k \\0 & -c_{1}k & k^{2}+\tilde{m}_{s}^{2}\end{array}\right), {\notag \\}\tilde{m}_{s}^{2}&=m_{s}^{2}+n\end{aligned}$$ has eigenvalues that violate the $AdS_{2}$ Breitenlohner-Freedman (BF) bound $m^{2}<-3$, the near horizon geometry has tachyonic instabilities. Interestingly, for large enough values of the coupling $c_{1}$, it was found that the lightest tachyonic modes appear at finite values of $k$. In such a situation, it is natural to expect that there will be a critical temperature, at the onset of the instability, at which a new inhomogeneous branch of black hole solutions will appear. When these solutions exist at lower temperature and are also thermodynamically preferred, the system will undergo a phase transition. These solutions will break the translational symmetries of the normal phase down to a discrete group corresponding to the periodicity $2\pi/k$ of the unstable mode. The mode involves the current and momentum densities of the dual field theory and since it is vectorial, parity will be broken at the same time.
Envisaging a continuous transition, its critical temperature can be determined by finding the temperatures $T(k)$ at which the corresponding static modes will appear. The details of these static modes were studied in [@Donos:2011bh] and the profile of the temperature as a function of the wavenumber $T(k)$ confirmed that for the cases where the lightest tachyons existed at finite values of $k$, the maximum of $T(k)$ was also at $k\neq 0$. The critical temperature of the system is precisely at this maximum $T_{c}$.
In this paper we will consider two model choices, the first one given by $$\begin{aligned}
\label{eq:vtq_choice}
V(\varphi)=-6\,\cosh\left(\sqrt{\frac{2}{3}}\,\varphi \right),\qquad \tau=\frac{1}{\cosh\left(\sqrt{6}\,\varphi \right)},\qquad \theta=\gamma\,\tanh\left( \sqrt{6}\varphi\right)\end{aligned}$$ which has $m_{s}^{2}=-4$, $n=36$ and $c_{1}=6\sqrt{2}\,\gamma$. For values $\gamma>\gamma_{c}=0.9988\ldots$ there exists a wavenumber interval for which one of the mass matrix eigenvalues violates the $AdS_{2}$ BF bound $m^{2}(k)<-3$. The model corresponding to the choice $\gamma=1$ is of particular importance since it can be derived as a consistent truncation of $D=11$ supergravity compactified on $SE^{7}$ manifolds [@Gauntlett:2009zw; @Gauntlett:2009bh; @Donos:2012yu]. As the critical value $\gamma_{c}$ is approached, the critical temperature $T_{c}$ gets close to zero and a numerical treatment becomes less accurate. We will take a larger value for $\gamma=1.2$ for which we give a plot of the static mode curve $T(k)$ in figure \[fig:Tk\]. From this, we deduce that the critical wavenumber is $k_{c}\approx 1.33\,\mu$ while the critical temperature is $T_{c}\approx 0.0598\,\mu$. As we will explain later, the plot in figure \[fig:Tk\] was made for the quantization choice of the pseudoscalar corresponding to dimension $\Delta=2$.
In order to compare the results of the numerical technique we will use with the findings of [@Rozali:2012es], the second model we will consider has $$\begin{aligned}
\label{eq:vtq_choice2}
V(\varphi)=-6-2\,\varphi^{2},\qquad \tau=1,\qquad \theta=\frac{3\,\sqrt{3}}{4}\,\varphi.\end{aligned}$$ A similar plot with the one in figure \[fig:Tk\] yields $T_{c}\approx 0.012\, \mu$ and $k_{c}\approx 0.53\,\mu$ [^2]. In the next section we will focus on these two models and construct the corresponding broken phase black holes for both of them.
Striped black holes {#sec:solutions}
===================
The setup
---------
In order to find the backreacted black hole solutions, we need to make a consistent ansatz for the metric and the gauge field. The natural requirement is that it should capture both the normal phase solution as well as the perturbative static modes of [@Donos:2011bh] when expanded around . A suitable ansatz for this purpose is $$\begin{aligned}
\label{eq:ansatz}
ds_{4}^{2}&=\frac{1}{z^{2}}\,\left[-f(z)\,Q_{tt}\,dt^{2}+\frac{Q_{zz}}{f(z)}\,dz^{2}+Q_{11}\,\left(dx_{1}+z^{2}Q_{z1}\,dz\right)^{2}+Q_{22}\,\left(dx_{2}+(1-z)\,Q_{t2}\,dt\right)^{2} \right]{\notag \\}A&=\mu\,(1-z)\,a_{t}\,dt+a_{2}\,dx_{2}{\notag \\}\varphi&=z\,h\end{aligned}$$ with all the nine functions appearing in the ansatz $\mathcal{F}=\left\{Q_{tt},\,Q_{zz},\,Q_{ii},\,Q_{z1},\,Q_{t2},\,a_{t},\,a_{2},\,h \right\}$ depending on both $z$ and $x_{1}$ while $f(z)$ is identical with the one that appears in the AdS-RN black hole solution . Since we wish to deal with the constraints of Einstein’s equations by employing the DeTurck method, we have only partially fixed our coordinates in . On the other hand, the gauge field ansatz in leads to second order equations for $a_{t}$ and $a_{2}$ with no further constrains. This method [@Headrick:2009pv; @Figueras:2011va] and an appropriate, slightly less general ansatz than , was recently used in [@Horowitz:2012ky; @Horowitz:2012gs] in order to explicitly break translations by introducing a lattice deformation of the AdS-RN black hole .
Instead of solving Einstein’s equations in , we modify it by shifting its left hand side by $-\nabla_{\left(\mu\right.}\xi_{\left.\nu\right)}$ with $\xi^{\mu}=g^{\nu\lambda}\,\left(\Gamma_{\nu\lambda}^{\mu}\left(g\right)-\bar{\Gamma}^{\mu}_{\nu\lambda}\left(\bar{g} \right)\right)$ and $\bar{g}$ being a reference metric which should have the same asymptotic and near horizon structure with the black holes we wish to construct [@Headrick:2009pv; @Figueras:2011va]. This modification turns Einstein’s equations to the so called Einstein - DeTurck equations. For a metric ansatz like our they are elliptic and after imposing appropriate boundary conditions they should lead to a countable[^3] set of solutions. We will choose $\bar{g}$ to be the AdS-RN black hole metric which is recovered after setting $Q_{tt}=Q_{zz}=Q_{ii}=1$ and $Q_{z1}=Q_{t1}=0$ in giving a space-like $\xi$. Our strategy, as in [@Horowitz:2012ky; @Horowitz:2012gs], will be to check that the solution generated by solving these equations will have $\xi^{2}\approx 0$ at machine precision. For the solutions that we will present in this work we will have $\xi^{2}<10^{-14}$.
In order to numerically solve the system of the PDEs plus boundary conditions we will discretize the coordinates $z$ and $x_{1}$ on a $N_{z}\times N_{x}$ grid. The aim is to transform the system of differential equations to an algebraic one. In order to approximate the derivatives of our functions at the grid points we will use the pseudospectral collocation method. More specifically we will represent our functions as a finite sum of Chebyshev polynomials for the $z$ coordinate dependence and as a Fourier sum for the periodic $x_{1}$ coordinate. The method we chose to solve the resulting non-linear system of equations is the Newton-Raphson method. Typically, we will be taking $N_{x}=N_{z}=35$ grid points for the first model . For the second model , we will be mostly interested in getting accurate results for the thermodynamic quantities close to the transition. In that case we will choose $N_{x}=28$ and $N_{z}=70$ while checking that the highest Fourier mode we resolve stays small down to the temperatures we will construct solutions for.
We now turn to the question of appropriate boundary conditions. In order to avoid the introduction of further deformations to the boundary theory, apart from the chemical potential $\mu$, we have to set the sources of all other operators equal to zero. In particular, for the pseudoscalar $\varphi$ and the spatial component of the gauge field $a_{2}$, the general fall off near the boundary $z=0$ reads $$\begin{aligned}
\label{eq:sc_falloff}
\varphi &=z\,\varphi_{(1)}(x_{1})+ z^{2}\,\varphi_{(2)}(x_{1})+\cdots{\notag \\}a_{2}&=\mu_{j}+j(x_{1})\,z+\mathcal{O}(z^{2})\end{aligned}$$ For the pseudoscalar we may choose between two different quantization conditions corresponding to a dimension $\Delta=1$ or $\Delta=2$ boundary theory operator. We will choose to work with the $\Delta=2$ case in which the function $\varphi_{(2)}$ corresponds to its vev while $\varphi_{(1)}$ corresponds to its source and we should set $\varphi_{(1)}=0$. Imposing the absence of current sources we also require that $\mu_{j}=0$. Under these conditions, the falloff for the remaining fields is $$\begin{aligned}
\label{eq:falloff_2}
Q_{tt}&=1+q_{tt}(x_{1})\,z^{3}+\mathcal{O}(z^{4}),\quad Q_{zz}=\mathcal{O}(z^{4}),\quad{\notag \\}Q_{ii}&=1+q_{ii}(x_{1})\,z^{3}+\mathcal{O}(z^{4}),\quad i=1,2\,,{\notag \\}Q_{t2}&=q_{t2}(x_{1})\,z^{3}+\mathcal{O}(z^{4}),\quad Q_{z1}=\mathcal{O}(z^{2}){\notag \\}a_{t}&=1-a(x_{1})\,z+\mathcal{O}(z^{2}),\quad\end{aligned}$$ for which the equations of motion also impose $$\begin{aligned}
\label{eq:tr_src_free}
q_{tt}(x_{1})+q_{11}(x_{1})+q_{22}(x_{1})&=0{\notag \\}\partial_{x_{1}}q_{11}(x_{1})&=0.\end{aligned}$$ The conditions of equation reflect the fact the energy momentum tensor $T_{\mu\nu}$ of the dual theory, which can be found in appendix \[sec:app\_thermo\], is traceless and source free i.e. $\left<T^{\mu}{}_{\mu} \right>=0$ and $\partial_{\mu}\left<T^{\mu}{}_{\nu} \right>=0$. We are therefore lead to imposing the Dirichlet boundary conditions $$\begin{aligned}
\label{eq:AdS4bc}
Q_{tt}(0,x_{1})&=Q_{zz}(0,x_{1})=Q_{11}(0,x_{1})=Q_{22}(0,x_{1})=a_{t}(0,x_{1})=1,{\notag \\}Q_{z1}(0,x_{1})&=Q_{t2}(0,x_{1})=a_{2}(0,x_{1})=h(0,x_{1})=0\end{aligned}$$ on the $AdS_{4}$ boundary.
Regularity of the solution at the horizon, which is at $z=1$, requires that all our functions take finite values and admit an analytic expansion in $(z-1)$ $$\begin{aligned}
\label{eq:nh_exp}
\mathcal{F}=\mathcal{F}(1,x_{1})+\partial_{z}\mathcal{F}(1,x_{1})\,(z-1)+\cdots\end{aligned}$$ Plugging the near horizon expansion in the equations of motion and expanding around $z=1$ we find a condition of constancy of the surface gravity on the horizon $Q_{zz}(1,x_{1})=Q_{tt}(1,x_{1})$ along with eight relations $\mathcal{M}\left(\mathcal{F},\partial_{z}\mathcal{F},\partial_{x_{1}}\mathcal{F},\partial^{2}_{x_{1}}\mathcal{F}\right)=0$ for our horizon data. These nine equations will serve as our boundary conditions on the horizon. It is easy to check that under these conditions the temperature of the black hole will simply be given by $T=\left(12-\mu^{2}\right)/8\pi$.
We also note that both our equations of motion and our boundary conditions are invariant under translations in $x_{1}$, as they should since are are interesting in breaking this symmetry spontaneously. As a result, if $\mathcal{F}(z,x_{1})$ is a solution of our boundary value problem so is $\mathcal{F}(z,x_{1}+b)$ for constant $b$. Since we will be searching for periodic configurations in $x_{1}$ it is natural to expect that within a period, there will be points for which the $\partial_{x_{1}}$ derivatives of our functions will separately vanish. In order to fix this mode we will in addition require that $\partial_{x_{1}}a_{2}(1,0)=0$. In the following section we report on the solutions we find to the posed problem.
The solutions
-------------
In order to study the thermodynamics we choose to work in the grand canonical ensemble where the chemical potential $\mu$ is held fixed. In practice, we will be generating solutions by varying $\mu$ in and by using the scaling symmetry of the boundary theory we will evaluate all the quantities in units of $\mu$. As previously mentioned, we will be focusing on a particular periodicity in $x_{1}$ for our solutions corresponding to the wavelength $2\pi/k$ of the critical mode with $k=k_{c}\approx 1.33\, \mu$ for e.g. the first model . For temperatures higher than the critical $T_{c}/\mu\approx 0.0598$, the only solution we find to our equations is the RN-black hole . In accordance with our expectations, for temperatures $T<T_{c}\approx 0.0598\,\mu$, we find additional solutions for which the functions in our ansatz have the anticipated periodic profile in $x_{1}$.
For two such solution, with $T=0.6\,T_{c}$ and $T=0.17\,T_{c}$, we present the profiles of the functions $h$ and $a_{t}$ in figure \[fig:3dfunctions\] for a single period in $x_{1}$. In [@Donos:2011bh] it was shown, using a perturbative series argument, that the electric component of the gauge field $a_{t}$ would be modulated with a period equal to $\pi/k$ close to the transition. For both models and we chose to study, a similar argument can be made to all orders in perturbation theory in agreement with the periodicity of the right plots in figure \[fig:3dfunctions\]. We also observe that translation breaking effects become stronger in the IR as we lower the temperature.
\
[{width="7cm"}]{}[{width="7cm"}]{}
\[fig:3dfunctions\]
The averaged entropy and charge densities of our system can be found through $$\begin{aligned}
\label{eq:entropy_FE}
S&=k\,\int_{0}^{2\pi / k}\,\sqrt{Q_{11}(1,x_{1})\,Q_{22}(1,x_{1}) }\,dx_{1}{\notag \\}Q&=\mu+\frac{k\mu}{2\pi}\,\int_{0}^{2\pi / k}\,a(x_{1})\,dx_{1}\end{aligned}$$ while the averaged free energy density is $$\begin{aligned}
\Omega&=M-\mu\,Q-T\,S{\notag \\}M&=2+\frac{\mu^{2}}{2}-\frac{3 k}{2\pi}\,\int_{0}^{2\pi/k}\,q_{tt}(x_{1})\,dx_{1}\end{aligned}$$ with $M$ being the averaged mass density of the black hole, as we explain in Appendix \[sec:app\_thermo\]. In figure \[fig:fe\_entropy\] we give a plot of the averaged entropy density difference between the striped black hole and the RN black hole solution as a function of the temperature. We do the same for the averaged free energy density difference in figure \[fig:fe\_entropy\]. We see that the new phase of black holes leads to lower entropy as well as free energy. More specifically, we find that near the critical temperature $\Delta\left(\frac{\Omega}{\mu^{3}}\right)\propto -\left(1-\frac{T}{T_{c}}\right)^{2}$ and $\Delta\left(\frac{S}{\mu^{2}}\right)\propto \frac{T}{T_{c}}-1$, compatible with a second order phase transition and the first law of thermodynamics[^4] $$\begin{aligned}
\delta \Omega=-S\,\delta T-Q\,\delta\mu\end{aligned}$$ for $\delta\mu=0$.
Another set of quantities we will now consider is the Fourier modes of the pseudoscalar operator vev $$\begin{aligned}
\phi_{(2)}^{(n)}=\frac{k}{2\pi}\int_{0}^{2\pi/k}e^{-\imath nkx_{1}}\phi_{(2)}(x_{1})\,dx_{1}.\end{aligned}$$ Close to the critical temperature $T_{c}$, the mode with $n=1$ is expected to have the mean field behavior $\frac{\phi_{(2)}^{(1)}}{\mu^{2}}\propto (1-\frac{T}{T_{c}})^{1/2}$ and the left plot shown in figure \[fig:scalarVeV\] is compatible with such a power law behavior. In more detail, for the specific model under consideration with even $V(\varphi)$, $\tau(\varphi)$ and odd $\theta(\varphi)$, only modes with odd $n$ will be involved in a perturbative expansion of the pseudoscalar. With the $n$-th mode being sourced at $n$-th order in perturbation theory we expect that we should have the dependence $\frac{\phi_{(2)}^{(n)}}{\mu^{2}}\propto (1-\frac{T}{T_{c}})^{n/2}$. In figure \[fig:scalarVeV\] we present a log-log plot of $\left|\phi_{(2)}^{(n)}\right|$ as a function of $1-\frac{T}{T_{c}}$ for the first three non-trivial modes corresponding to $n=1,3,5$. A linear fit $$\begin{aligned}
\log\left(\left|\phi_{(2)}^{(n)}\right|\right)=-\alpha_{n}+\gamma_{n}\,\ln\left(1-\frac{T}{T_{c}} \right)\end{aligned}$$ based on our near transition data determines the slopes $\gamma_{1}\approx0.501$, $\gamma_{3}\approx 1.508$ and $\gamma_{5}\approx 2.505$, in good agreement with our previous argument. Similar behavior is found for the momentum $q_{t2}(x_{1})$ and current $j(x_{1})$ density waves.
We will now examine the charge density wave which forms in our striped phase and whose density profile can be read off from $a(x_{1})$ in . In figure \[fig:charge\] we plot the difference of the averaged charge density $\Delta Q$ from the normal phase as a function of temperature. In the same figure we also plot its Fourier mode $$\begin{aligned}
q^{(n)}=\frac{\mu k}{2\pi}\,\int_{0}^{2\pi/k}\,e^{-\imath n k x_{1}}a(x_{1})\,dx_{1}\end{aligned}$$ with $n=2$ as a function of temperature. For both cases we observe the expected linear dependence $q^{n=0,2}\propto 1-\frac{T}{T_{c}}$ close to the transition as the electric component of the gauge field $a_{t}$ backreacts at second order.
We now turn our attention to the model and repeat the same procedure for $k_{c}\approx 0.53\,\mu$ and $T_{c}\approx 0.012\,\mu$. We don’t find any qualitative difference between the two models close to the transition, as we would expect. The transition is again second order and in contrast with the results of [@Rozali:2012es], we find solutions with $T<T_{c}$ which have lower free energy and entropy than the AdS-RN black hole for a given temperature. We plot the differences of the free energy and entropy densities between the two phases in figure \[fig:fe\_entropy2\].
In order to calculate the free energy we need to extract the function $q_{tt}(x_{1})$ from the falloff of the function $Q_{tt}$ that parametrizes our ansatz e.g. by taking a third order derivative at $z=0$. Even though the pseudospectral technique we have chosen makes the calculation of derivatives of functions quite accurate, it still is less accurate than the calculation of e.g. the entropy which is using directly the values of the functions. The fact that our data satisfies the first law of thermodynamics at a good accuracy is quite reassuring. In Appendix \[sec:alt\_param\] we outline an equivalent way to phrase the boundary value problem in which the calculation of the normalizable modes of our functions is more direct. We have also solved our PDEs in the way described there and reproduced the plots of figure \[fig:fe\_entropy2\].
Discussion {#sec:discussion}
==========
We numerically constructed four dimensional black hole geometries with inhomogeneous horizons by backreacting on the instabilities of [@Donos:2011bh]. These geometries are the bulk duals of CFT phases held at finite temperature and deformed by a uniform chemical potential. The numerical method we chose to follow [@Headrick:2009pv; @Figueras:2011va] gives results compatible with a second order phase transition. We only reported on the choices and but since the order of the transition is decided at third order in perturbation theory [@Gubser:2001ac] we expect the character of the transition to remain the same for a general class of models. In this work we focused on a particular periodicity for the modulation of our solutions but it is clear from the five dimensional homogeneous analogs [@Donos:2012wi; @Donos:2012gg] that the configurations of minimum free energy will have a temperature dependent periodicity. The order of the transition is unlikely to change due to this phenomenon in general. An open question is that of the minimum free energy configuration, it is possible that the thermodynamically preferred mode is a superposition in $k-$space. To answer this question one would have to solve PDEs in three dimensions or follow a perturbative approach [@Bu:2012mq; @Bao:2013fda].
It would also be interesting to explore the phase transition in theories where the near horizon limit of the normal phase black holes at zero temperature is hyperscaling violating instead of $AdS_{2}\times \mathbb{R}^{2}$. It was recently shown, in [@Iizuka:2013ag], that the near horizon limit of such black holes can have instabilities of the same nature. Based on the general arguments of [@Hartnoll:2012pp] the phase transition in those theories is expected to exhibit qualitative differences reflecting the fact that the normal phase black hole branch has zero entropy at its zero temperature limit.
An important question is that of the zero temperature ground state. As we lower the temperature and the non-linearities of the functions $V$, $\tau$ and $\theta$ become important, we expect a rich class of ground states to emerge in the IR. In the absence of modulation and for a purely electric ansatz for the gauge field, the models reduce to Einstein-Maxwell-Dilaton. In that case, a thorough classification of possible zero temperature IR behavior has been carried out in [@Taylor:2008yq; @Goldstein:2009cv; @Charmousis:2010zz]. It would certainly be interesting to carry out as a similar classification in an inhomogeneous setting. Based on the numerics in [@Rozali:2012es], it was argued by the authors that new IR behavior emerges also in $D=4$. We plan to further explore the low temperature behavior of the striped phases, based on the techniques used in this paper, in the future.
Acknowledgements {#acknowledgements .unnumbered}
================
We would like to thank P. Figueras, J. P. Gauntlett, S. Hartnoll, T. Wiseman and B. Withers for inspiring and valuable discussions.
Holographic renormalization {#sec:app_thermo}
===========================
In order to have a well defined variational problem and a finite total action, we need to supply the action with appropriate boundary counterterms $$\begin{aligned}
S_{T}=S+\int_{\partial M}\,\sqrt{-\gamma_{\infty}}\,\left( \mathcal{K}-2\sqrt{2}-\frac{1}{\sqrt{2}}\,\varphi^{2}\right)\end{aligned}$$ where the last term ensures that our pseudoscalar has dimension $\Delta=2$. We also have that $\mathcal{K}=\lim_{r\rightarrow0}g^{\mu\nu}\,\nabla_{\mu}n_{\nu}$ with $n^{\mu}$ an outward pointing unit normal vector and $\gamma$ is the determinant of the induced metric on the boundary at $r\rightarrow 0$. Following [@Balasubramanian:1999re] and given our asymptotic expansion along with the condition , we obtain the expression $$\begin{aligned}
\langle T_{tt}\rangle &=\sqrt{2}\,\left(2+\frac{\mu^{2}}{2}-3\,q_{tt}(x_{1}) \right),\quad \langle T_{x_{i}x_{i}}\rangle=\frac{1}{\sqrt{2}}\,\left(1+\frac{\mu^{2}}{4}+3\,q_{ii}(x_{1}) \right){\notag \\}\langle T_{tx_{2}}\rangle &=\frac{3}{\sqrt{2}}\,q_{t2}(x_{1})\end{aligned}$$ for the non-trivial component of the boundary stress energy tensor.
The metric of the boundary theory is $$\begin{aligned}
ds_{3}^{2}=\gamma_{(3)}{}_{\mu\nu}dx^{\mu}dx^{\nu}=-2\,dt^{2}+dx_{1}^{2}+dx_{2}^{2}.\end{aligned}$$ After defining the timelike unitary vector $u=2^{-1/2}\,\partial_{t}$, the mass and momentum densities read $$\begin{aligned}
m=&\sqrt{-\gamma_{(3)}}\, u^{\mu}u^{\nu}\,\langle T_{\mu\nu}\rangle=2+\frac{\mu^{2}}{2}-3\,q_{tt}(x_{1}){\notag \\}p_{x_{2}}=&u^{\mu}\langle T_{\mu x_{2}}\rangle=\frac{3}{2}\,q_{t2}(x_{1}).\end{aligned}$$
An equivalent boundary value problem {#sec:alt_param}
====================================
We will now take advantage of the asymptotic falloff to make a better choice of our functions in order to extract the asymptotic data. Instead of using the functions we introduced in the ansatz , we make the change of variables $$\begin{aligned}
\label{eq:redef}
Q_{mm}\left(z,x_{1}\right)&=1+z^{3}\,\hat{q}_{mm}\left(z,x_{1}\right),\quad m=t,\,z,\,1,\,2{\notag \\}Q_{z1}\left(z,x_{1}\right)&=z^{2}\,\hat{q}_{z1}\left(z,x_{1}\right),\quad Q_{t2}\left(z,x_{1}\right)=z^{3}\,\hat{q}_{t2}\left(z,x_{1}\right){\notag \\}a_{t}\left(z,x_{1}\right)&=1-z\,\hat{a}\left(z,x_{1}\right),\quad a_{2}\left(z,x_{1}\right)=z\,\hat{j}\left(z,x_{1}\right){\notag \\}\varphi&=z^{2}\,\hat{\varphi}_{(2)}\left(z,x_{1}\right).\end{aligned}$$ The aim is to solve the problem in such a way that the values of the functions we are solving for are directly related to the values of the vevs we wish to calculate. Our requirement for a regular horizon at $z=1$ leads us again to imposing very similar boundary conditions with the ones described in section \[sec:solutions\] for the functions $\hat{\mathcal{F}}=\left\{ \hat{q}_{mm},\,\hat{q}_{rx_{1}},\,\hat{q}_{tx_{2}},\,\hat{a},\,\hat{j},\,\hat{\varphi}_{(2)}\right\}$. For example we find that we should impose the zeroth law of thermodynamics guaranteed by $\hat{q}_{tt}\left(1,x_{1}\right)=\hat{q}_{rr}\left(1,x_{1}\right)$.
The difference in the nature of the boundary conditions we need to impose comes from the $AdS_{4}$ boundary at $z=0$. Plugging the ansatz in the equations of motion we find the simple first order relations at $z=0$ $$\begin{aligned}
\label{eq:ads4_bc_2}
\partial_{z}\hat{q}_{tt}&=-\hat{\varphi}_{(2)}^{2}+\frac{1}{4}\mu^{2}\,\hat{a}\,\left(\hat{a}+2 \right),\quad
\partial_{z}\hat{q}_{rr}=\frac{1}{2}\,\hat{j}^{2}+3\,\hat{\varphi}_{(2)}^{2} -\frac{1}{4}\mu^{2}\,\hat{a}\,\left(\hat{a}+2 \right){\notag \\}\partial_{z}\hat{q}_{11}&=-\hat{\varphi}_{(2)}^{2},\quad
\partial_{z}\hat{q}_{22}=-\frac{1}{2}\,\hat{j}^{2}-\hat{\varphi}_{(2)}^{2},\quad
\partial_{z}\hat{q}_{z1}=\hat{\varphi}_{(2)}\,\partial_{x_{1}}\hat{\varphi}_{(2)}{\notag \\}\partial_{z}\hat{q}_{t2}&=\hat{q}_{t2}+\frac{1}{2}\mu\,\hat{j}\,\left(\hat{a}+1\right),\quad
\partial_{z}\hat{a}=\hat{a},\quad
\partial_{z}\hat{j}=0,\quad
\partial_{z}\hat{\varphi}_{(2)}=0\end{aligned}$$ which we can use as our boundary conditions. We also find that that we should satisfy the conditions $$\begin{aligned}
\label{eq:ads4_const}
\hat{q}_{tt}\left(0,x_{1}\right)+\hat{q}_{x_{1}x_{1}}\left(0,x_{1}\right)+\hat{q}_{x_{2}x_{2}}\left(0,x_{1}\right)=0,\quad \partial_{x_{1}}\hat{q}_{x_{1}x_{1}}\left(0,x_{1}\right)=0,\quad \hat{q}_{zz}\left(0,x_{1}\right)=0\end{aligned}$$ Given the above boundary conditions for the functions $\hat{\mathcal{F}}$, we can reproduce the plots of figure \[fig:fe\_entropy2\] at relatively low grid resolutions e.g. $N_{x}=25$ and $N_{z}=45$. We also check that the resulting solutions satisfy the constraints .
[^1]: See also [@Bergman:2011rf] for a probe brane setup
[^2]: Our value for $k_{c}$ differs from the one of [@Rozali:2012es] by a factor of $\sqrt{2}$ due to the different scale of $x_{1}$ on the boundary. Notice also a typo for the mass of the pseudoscalar $\psi$ in v1 of [@Rozali:2012es].
[^3]: When a phase transition happens there are at least two solutions with the same boundary conditions.
[^4]: We keep the period of the solutions in the $x_{1}$ direction fixed.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'An effective action is obtained of a Bose gas in the bulk separated into two regions by a strong external potential depending on the single coordinate. The main attention is focused on the relaxation of the difference between phases of the weakly coupling condensates of the different bulk domains separated from each other by the external potential. The cases of low and high temperatures are considered.'
address: 'R.S.C. ”Kurchatov Institute” Kurchatov sq. 46, Moscow 123182, Russia.'
author:
- 'V. S. Babichenko'
title: Low Temperature Relaxation of the Phase In an Inhomogeneous Bose Gas
---
\[theorem\][Acknowledgement]{} \[theorem\][Algorithm]{} \[theorem\][Axiom]{} \[theorem\][Claim]{} \[theorem\][Conclusion]{} \[theorem\][Condition]{} \[theorem\][Conjecture]{} \[theorem\][Corollary]{} \[theorem\][Criterion]{} \[theorem\][Definition]{} \[theorem\][Example]{} \[theorem\][Exercise]{} \[theorem\][Lemma]{} \[theorem\][Notation]{} \[theorem\][Problem]{} \[theorem\][Proposition]{} \[theorem\][Remark]{} \[theorem\][Solution]{} \[theorem\][Summary]{}
The experimental realization of the Bose-Einstein condensation at ultralow temperatures in atomic vapors \[1\] provides an example of the systems in which the approximation of the weakly non-ideal gas is well applicable. This fact is connected with the small density of particles in the systems concerned. The realization of such systems gives the possibility of the experimental investigations of the macroscopic quantum phenomenons of different types. Recently, the manifestations of the macroscopic quantum phase and its behavior are studied actively \[2-5\], \[8\].
The investigation of the kinetic phenomena due to the relaxation of the order parameter is of a doubtless interest for the study of macroscopic quantum phenomena. In the present work we study the spatial relaxation of the phase of the order parameter in the inhomogeneous Bose gas with the weak coupling between different spatial regions due to a barrier tunneling.
The system concerned in the present work is a Bose gas of the small density in the bulk separated into two regions by a strong external potential $%
U\left( \overrightarrow{r}\right) $ which depends on the x coordinate and does not depend on $\overrightarrow{r}_{\perp }=\left( y,z\right) $, i.e., $%
U\left( \overrightarrow{r}\right) =U\left( x\right) $. For simplicity, we assume that the external potential has a rectangular shape $U\left( x\right)
=U_{0}$ for $-d<x<d$ and $U\left( x\right) =0$ beyond this region of the x coordinate. The height of the external potential $U_{0}$ is supposed to be the largest energy parameter in the system, in particular, $U_{0}>>\mu $ where $\mu $ is the chemical potential $\mu =n\lambda $, $\lambda $ is the scattering amplitude of Bose particles, and $n$ is the density of the Bose gas. Due to this assumption the interaction between particles of the Bose gas in the region of the influence of the external potential, i.e., in the region $-d<x<d$, can be neglected.
The left and the right domains of the bulk are supposed to have the same temperature $T$ smaller than the Bose-condensation temperature $T_{c}$. We assume that the densities of the right and the left domains of the bulk have the values $n_{1}=n-\frac{1}{2}\Delta n$ and $n_{2}=n+\frac{1}{2}\Delta n$ correspondingly and the difference between densities $\Delta n$ is much smaller than the average density n. We will consider both the case of the nonzero $\Delta n$ and the case of $\Delta n=0$. At the same time, in these two cases the phases of the left and the right Bose condensates are supposed to be different at the initial time moment. This assumption means that at the initial time moment there is a nonzero current from one side of the bulk to another. This non-equilibrium initial state will relax to the equilibrium state having the same densities and the same phases of the condensates for both sides of the bulk if $\Delta n=0$ or if $\Delta n$ can change in time, and will relax to the stationary state if the density difference $\Delta n$ keeps constant and nonzero. This relaxation process is studied in the present work.
Note that in the case of superconductors the similar initial non-equilibrium state results in the Josephson oscillations with a small damping \[6\], \[7\]. The essential distinction of the Bose gas consisting of neutral atoms from superconductors in which the Cooper pairs represent the charged objects is the presence of the gapless excitation spectrum. The presence of low energy excitations results in the change of the character of the initial state relaxation making the relaxation essentially faster. In the case of the homogeneous in y-z plane difference between the phases of the condensates the excitations radiated during the relaxation process have the one-dimensional character. The one-dimensional character of the radiated excitations results in a divergency of the small momentum correlators of these excitations. Hence, the consideration of the relaxation process to the second order in the tunneling amplitude is not correct in contrast to the case of superconductors \[7\] and requires more accurate analysis \[9\], \[14\], \[15\].
The effective action for the difference of phases.
==================================================
Below the approach to the problem of the phase difference relaxation without using the perturbation theory in the tunneling amplitude is developed. For this purpose the effective action for the difference between phases of the Bose field at the right-hand and the left-hand boundaries of the potential barrier is derived. This effective action is obtained by the integration over the bulk components of the Bose field with the fixed values of this field at the potential barrier boundaries. The obtained effective action contains the relaxation part which is proportional to the first power of the frequency of the phase field. For the small frequencies the relaxation part is found to be much larger than the usual kinetic part of the phase dynamics which is proportional to the second power of the frequency. In order to describe the phase difference relaxation process the consideration is developed in the framework of the Schwinger-Keldysh technique \[10\], \[11\] which is very convenient for the description of quantum kinetic processes.
The generating functional of the system can be written in the form $Z=\int
D\psi D\overline{\psi }e\exp \left\{ i\left( S_{U}+S_{j}\right) \right\} $. The action $S_{U}$ of the inhomogeneous non-equilibrium Bose gas is given by
$$S_{U}=\oint dt\int d^{3}r\left\{ \overline{\psi }\left[ i\partial _{t}+\mu
\left( x\right) -U\left( x\right) \right] \psi -\frac{1}{2}\left(
\overrightarrow{\nabla }\overline{\psi }\right) \left( \overrightarrow{%
\nabla }\psi \right) -\frac{\lambda }{2}\left( \overline{\psi }\psi \right)
^{2}\right\}$$ and the term $S_{j}$ reads as $S_{j}=\oint dt\int d^{3}r\left( \overline{%
\psi }j+\overline{j}\psi \right) $, where $j$ and $\overline{j}$ are the infinitely small sources. The integration over the time t in the action is realized over the time reversing contour and is denoted as $\oint dt$. The Planck constant $\hbar $ and mass m of a Bose particle are set equal to unity $\hbar =m=1$. The chemical potential $\mu \left( x\right) $ is supposed to have the constant value $\mu _{1}=\mu -\frac{1}{2}\Delta \mu $ for the coordinate xd and $\mu _{2}=\mu +\frac{1}{2}\Delta \mu $ for x-d where the chemical potential $\mu $ obeys the equality $\mu =\lambda n$ and the chemical potential difference obeys the equality $\Delta \mu
=\lambda \Delta n$. The small value of $\Delta n$ results in the small value of the chemical potential difference $\Delta \mu <<\mu $.
Due to the large magnitude of potential $U_{0}$ in the region x$\in \left[
-d,d\right] $ the interaction between Bose particles can be neglected and the integral for Z over the fields $\psi ,\overline{\psi }$ takes the Gaussian form in this region of the x-coordinate. In this connection the integral for Z over the fields $\psi \left( x,\overrightarrow{r}_{\perp
};t\right) ,\overline{\psi }\left( x,\overrightarrow{r}_{\perp };t\right) $ for x$\in \left[ -d,d\right] $ with the fixed magnitudes of these fields at the boundary of the region of the nonzero external potential can be calculated. Moreover, the characteristic scale of the $\psi $-field variation in the time and in the $\overrightarrow{r}_{\perp }$ space is supposed to be much larger than $1/U_{0}$ and $1/\sqrt{U_{0}}$, respectively. It is these fields that describe the slow relaxation process possessing the characteristic frequency proportional to the small tunneling coefficient. Due to this assumption the tunneling amplitude of the $\psi $ fields has the local character for the time variable t and space variables $%
\overrightarrow{r}_{\perp }$. The magnitudes of the $\psi $-fields at the boundary of the region of the nonzero external potential are denoted as $%
\psi \left( d,\overrightarrow{r}_{\perp },t\right) =\psi _{s1}\left(
\overrightarrow{r}_{\perp },t\right) $; $\psi \left( -d,\overrightarrow{r}%
_{\perp },t\right) =\psi _{s2}\left( \overrightarrow{r}_{\perp },t\right) $. The calculation of the integral for Z over the $\psi $-fields in the region x$\in \left[ -d,d\right] $ gives the action in the form of the sum of four terms $S=S_{vol}+S_{surf}^{\left( \rho \right) }+S_{surf}^{\left(
tunn\right) }+S_{j}$. The part of the action $S_{vol}$ is
$$S_{vol}=\oint dt\int\limits_{U\left( x\right) =0}dx\int d^{2}r_{\perp
}\left\{ \overline{\psi }\left( i\partial _{t}+\mu \right) \psi -\frac{1}{2}%
\left( \overrightarrow{\nabla }\overline{\psi }\right) \left(
\overrightarrow{\nabla }\psi \right) -\frac{\lambda }{2}\left( \overline{%
\psi }\psi \right) ^{2}\right\} \eqnum{1}$$
where the region of the integration over the x-coordinate is the sum of the regions $\left( -L,-d\right) \cup \left( d,L\right) $. The parts of the action $S_{surf}^{\left( \rho \right) }$ and $S_{surf}^{\left( tunn\right) }$ can be represented in the form \[17\] $\ $
$$S_{surf}^{\left( \rho \right) }=-\frac{\varkappa }{2}\coth \left( 2\varkappa
d\right) \oint dt\int d^{2}r_{\perp }\left\{ \mid \psi _{s1}\left(
\overrightarrow{r}_{\perp },t\right) \mid ^{2}+\mid \psi _{s2}\left(
\overrightarrow{r}_{\perp },t\right) \mid ^{2}\right\} \eqnum{2}$$
$$S_{surf}^{\left( tunn\right) }=\frac{\varkappa }{2\sinh \left( 2\varkappa
d\right) }\oint dt\int d^{2}r_{\perp }\left\{ \overline{\psi }_{s1}\left(
\overrightarrow{r}_{\perp },t\right) \psi _{s2}\left( \overrightarrow{r}%
_{\perp },t\right) e^{i\Delta \mu t}+\overline{\psi }_{s2}\left(
\overrightarrow{r}_{\perp },t\right) \psi _{s1}\left( \overrightarrow{r}%
_{\perp },t\right) e^{-i\Delta \mu t}\right\} \eqnum{3}$$
where the magnitude $\varkappa $ is $\varkappa =\sqrt{2U_{0}}$, the index 1 denotes the right-hand side of the bulk and the index 2 denotes the left-hand side of the bulk. The term $S_{j}$ can be written as $S_{j}=\oint
dt\sum\limits_{\alpha =1,2}\left( \overline{\psi }_{s\alpha }j+\overline{j}%
\psi _{s\alpha }\right) $ where $j$ and $\overline{j}$ are the infinitely small sources. Later on we assume that the width d of the potential barrier is sufficiency large so that the inequality $\varkappa d>>1$ takes place and, thus, $\sinh \left( 2\varkappa d\right) \thickapprox \cosh \left(
2\varkappa d\right) \thickapprox \frac{1}{2}e^{2\varkappa d}>>1$. The term $%
S_{surf}^{\left( tunn\right) }$ describes the tunneling between the right-hand and left-hand sides of the bulk. From the form of this tunneling part of the action it can easily be seen that the tunneling amplitude for the condensate particles and for the low energy non-condensate particles has the same nonzero magnitude in contrast with the work \[8\].
Our goal is the calculation of the functional integral for Z over the $\psi $-fields in the bulk, i.e., in the region x$\in \left( -L,-d\right) \cup
\left( d,L\right) $ with the fixed values at the potential barrier boundary, and, thus, obtain the effective action for the fields at the boundary $\psi
_{s1}$, $\psi _{s2}$.
Below the onedimensional character of the system and its excitations is supposed. In this supposition the considering fields depend on the x coordinate only. Later on, it is convenient to introduce the dimensionless space and time coordinates via the following change of these variables $%
x\rightarrow \xi x;$ $t\rightarrow t/\mu $. Thus, the system of units which we use later measures the quantities of the dimension of length in units of $\xi $ and the quantities of the dimension of energy in units of $%
\mu $. Moreover, we translate the x coordinate for the right-hand side of the bulk on the value -d and for the left-hand side on +d, so as the right-hand and left-hand boundaries of the potential barrier take the zero coordinates x=0.
It is convenient to represent the fields $\psi $ in the modulus-phase form $%
\psi =Re^{i\varphi }$; $\overline{\psi }=Re^{-i\varphi }$. Below the modulus $R$ of the field $\psi $ is represented as the sum of the saddle-point configuration $F\left( x\right) $ and the fluctuations of the modulus field $\rho $ which characterize the density fluctuations
$$R\left( x,t\right) =F\left( x\right) +\rho \left( x,t\right) \eqnum{4}$$
The saddle-point configuration $F\left( x\right) $ obeys the equation
$$\left( 1+\frac{1}{2}\nabla _{x}^{2}-F^{2}\left( x\right) \right) F\left(
x\right) =\varkappa F\left( x\right) \delta \left( x\right) \eqnum{5}$$
and has the form $F\left( x\right) =\tanh \left( \mid x\mid +X_{0}\right) $, and the constant $X_{0}$ is the small value $X_{0}=1/\varkappa $. The fluctuations $\rho $ can be represented as the sum of two summands
$$\rho \left( x,t\right) =\delta \rho \left( x,t\right) +\frac{\nabla
_{x}F\left( x\right) }{\nabla _{x}F\left( 0\right) }\rho _{s}\left( t\right)
\eqnum{6}$$
where the fluctuations $\delta \rho \left( x,t\right) $ obey the zero boundary conditions $\delta \rho \left( 0,t\right) =0$. Note, that the second summand of (6) is the zero mode of the Hamiltonian $\widehat{H}_{1}=-%
\frac{1}{2}\nabla _{x}^{2}-1+3F^{2}$ which describe the dynamics of the fluctuations $\rho \left( x,t\right) $. To describe the slow process of the tunneling through the potential barrier we are interesting for the fields $%
\delta \rho $ and $\varphi $ which are slow varying values at the space scale $\xi $ and the time scale $1/\mu $. For the derivation of the slow field effective action we substitute the field $\psi $ into the action (1-3) in the modulus-phase representation where the field $R\left( x,t\right) $ has the form (4) and where $\rho $ is determined by (6). At the same time, in certain terms of the action due to the slowness of $\delta \rho $ and $%
\varphi $ the function like $F^{2}\left( x\right) $ can be replaced by the function $1-\delta \left( x\right) $, where $\delta \left( x\right) $ is the $\delta $-function. Integrating over the fields $\delta \rho \left(
x,t\right) $ and $\rho _{s}\left( t\right) $ in the generation functional we obtain the effective action for the slow phase field
$$S^{\left( slow\right) }\left[ \varphi \right] =K\oint dt\int\limits_{U\left(
x\right) =0}dx\left\{ \frac{1}{2}\left( \partial _{t}\varphi \right) ^{2}-%
\frac{1}{2}\left( \nabla _{x}\varphi \right) ^{2}+\left[ \frac{1}{2}\left(
\partial _{t}\varphi _{s}\right) ^{2}+\gamma \cos \left( \Delta \varphi
_{s}\left( t\right) +\Delta \mu t\right) \right] \delta \left( x\right)
\right\}$$
where $\gamma =\frac{2}{\varkappa }\exp \left( -2\varkappa d\right) $, the value $\Delta \varphi _{s}\left( t\right) $ is the difference of phases at the right-hand and the left-hand boundaries of the potential barrier $\Delta
\varphi _{s}\left( t\right) =\varphi _{s2}\left( t\right) -\varphi
_{s1}\left( t\right) $, the constant $K$ is $K=n\xi S_{\perp }>>1$ and $%
\delta \left( x\right) $ is the Dirac $\delta $-function. The value $%
S_{\perp }$ is the square of the bulk section which is supposed to obey the inequality $S_{\perp }\lesssim \xi ^{2}$ to ensure the 1D character of the system. The dimensionless constant $K=n\xi S_{\perp }$ has the large value $%
K>>1$ due to the large value of the parameter $n\xi ^{3}>>1$. The last inequality is ensured by the application of the gas approximation.
The last step to the derivation of the effective action for the phase difference $\Delta \varphi _{s}\left( t\right) $ is the integration in Z over the bulk components of the phase $\varphi \left( x,t\right) $ with the fixing boundary values $\varphi _{s}\left( t\right) $. For this purpose the field $\varphi \left( x,t\right) $ is represented in the form of the Fourier expansion $\varphi \left( x,t\right) =\sum\limits_{k}\varphi _{k}\left(
t\right) e^{ikx}$ with the Fourier components $\varphi _{k}$ obeying the condition $\varphi _{s}\left( t\right) =\sum\limits_{k}\varphi _{k}\left(
t\right) $ where the coordinate of the right-hand boundary of the potential barrier we choose as x=0. The integration over $\varphi _{k}$ gives
$$S_{eff}\left[ \varphi _{s}\right] =\oint dtdt^{\prime }\left\{ \frac{1}{2}%
\left( \partial _{t}\varphi _{s}\right) ^{2}\delta \left( t-t^{\prime
}\right) +\frac{1}{2}\varphi _{s}\left( t\right) \left[ <\widehat{D}>\right]
_{t,t^{\prime }}^{-1}\varphi _{s}\left( t^{\prime }\right) +\gamma \cos
\left( \Delta \varphi _{s}\left( t\right) -\Delta \mu t\right) \delta \left(
t-t^{\prime }\right) \right\} \eqnum{8}$$
where we denote $<\widehat{D}>=\int \frac{dk}{2\pi }\widehat{D}_{k}$. In the frequency-momentum representation in the ”triangular” form \[10\], \[11\] at the temperature T the phonon propagator $\widehat{D}\left( \omega ,k\right) $ can be written
$$\widehat{D}\left( \omega ,k\right) =\left(
\begin{array}{cc}
0 & D^{A}\left( \omega ,k\right) \\
D^{R}\left( \omega ,k\right) & D^{K}\left( \omega ,k\right)
\end{array}
\right) \eqnum{9}$$
where $D^{R,A}\left( \omega ,k\right) =\left[ \left( \omega \pm i\delta
\right) ^{2}-k^{2}\right] ^{-1}$ and $D^{K}\left( \omega ,k\right) =\coth
\left( \frac{\omega }{2T}\right) \left( D^{R}-D^{A}\right) $. The simple calculations of $<\widehat{D}>_{\omega }$ give
$$<\widehat{D}>_{\omega }=\frac{1}{2}\left(
\begin{array}{cc}
0 & <D^{A}>_{\omega } \\
<D^{R}>_{\omega } & <D^{K}>_{\omega }
\end{array}
\right) \text{\ } \eqnum{10}$$
where $<D^{R}>_{\omega }=\frac{1}{i\left( \omega +i\delta \right) }$; $%
<D^{A}>_{\omega }=\frac{1}{-i\left( \omega -i\delta \right) }$; $\
<D^{K}>_{\omega }=2\coth \left( \frac{\omega }{2T}\right) \frac{1}{i\omega }$.
The effective action for small frequencies and the Fokker-Planck equation.
==========================================================================
Introducing the half-sum and the difference of the fields instead of the fields $\varphi _{s1}$ and $\varphi _{s2}$ we obtain the effective action for the difference of phases at the boundary of the potential barrier. In the ”triangle” representation it can be written as
$$S\left[ \varphi \right] =2K\int dt\left\{ \varphi \left[ \frac{1}{2}\widehat{%
\omega }^{2}+i\widehat{\omega }\right] \Phi +\Phi \left[ \frac{1}{2}\widehat{%
\omega }^{2}-i\widehat{\omega }\right] \varphi +i\varphi \left[ 2\widehat{%
\omega }\coth \left( \frac{\widehat{\omega }}{2T}\right) \right] \varphi -V%
\left[ \Phi ,\varphi \right] \right\} \eqnum{11}$$
where $\widehat{\omega }=i\partial _{t}$, the function $V\left[ \Phi
,\varphi \right] $ is $V\left[ \Phi ,\varphi \right] =\gamma \sin \left(
\Phi +\Delta \mu t\right) \sin \left( \varphi \right) $. In this representation the fields $\Phi $ and $\varphi $ are expressed via the fields belonging to the upper and the lower time contour branches as $\Phi =%
\frac{1}{2}\left( \varphi _{+}+\varphi _{-}\right) $ and $\varphi =\frac{1}{2%
}\left( \varphi _{+}-\varphi _{-}\right) $. Note that as usually in the ”triangle” representation of the Keldysh-Schwinger technique the field $%
\Phi $ describes the kinetics of the phase difference and the field $\varphi
$ describes the quantum noise of this value.
The fields $\Phi $ and $\varphi $ can be fragmented into fast and slow components. The slow fields are the fields having the frequencies $\omega $ obeying the inequality $\omega <\omega _{0}$ and the fast fields are the fields with the frequencies $\omega >\omega _{0}$ where $\gamma <<\omega
_{0}<<1$. The part of the action for the fast fields is obtained by the expansion of the potential $V$ up to the second order of the fast components $\delta \Phi $, $\delta \varphi $ and has the form
$$S^{\left( fast\right) }\left[ \delta \varphi ,\delta \Phi \right] =2K\int
dt\left\{
\begin{array}{c}
\delta \varphi \left[ \frac{1}{2}\widehat{\omega }^{2}+i\widehat{\omega }%
\right] \delta \Phi +\delta \Phi \left[ \frac{1}{2}\widehat{\omega }^{2}-i%
\widehat{\omega }\right] \delta \varphi +i\delta \varphi \left[ 2\widehat{%
\omega }\coth \left( \frac{\widehat{\omega }}{2T}\right) \right] \delta
\varphi - \\
-V_{\varphi \Phi }^{\prime \prime }\delta \varphi \delta \Phi -\frac{1}{2}%
V_{\Phi \Phi }^{\prime \prime }\left( \delta \Phi \delta \Phi +\delta
\varphi \delta \varphi \right)
\end{array}
\right\} \eqnum{12}$$
$$V_{\varphi \Phi }^{\prime \prime }=\gamma \cos \left( \Phi +\Delta \mu
t\right) \cos \left( \varphi \right) \text{; \ \ \ }V_{\Phi \Phi }^{\prime
\prime }=V_{\varphi \varphi }^{\prime \prime }=-\gamma \sin \left( \Phi
+\Delta \mu t\right) \sin \left( \varphi \right) \text{\ }$$ where $\delta \varphi ,\delta \Phi $ are the fast components of fields and $%
\varphi ,\Phi $ are the slow components. The part of the action for the slow fields can be written as
$$S^{\left( slow\right) }\left[ \varphi \right] =2K\int dt\left\{ -2\varphi
\partial _{t}\Phi -V\left[ \Phi ,\varphi \right] +i\varphi \left[ 2\widehat{%
\omega }\coth \left( \frac{\widehat{\omega }}{2T}\right) \right] \varphi
\right\} \eqnum{13}$$
Due to the large value of the constant $K>>1$, so as the value $\frac{1}{K}%
\ln \frac{1}{\gamma }$ obeys the inequality $\frac{1}{K}\ln \frac{1}{\gamma }%
<<1$, the action $S^{\left( fast\right) }\left[ \delta \varphi ,\delta \Phi %
\right] $ can be approximated as
$$S^{\left( fast\right) }\left[ \delta \varphi ,\delta \Phi \right] =2K\int
dt\left\{
\begin{array}{c}
\delta \varphi \left[ \frac{1}{2}\widehat{\omega }^{2}+i\widehat{\omega }%
\right] \delta \Phi +\delta \Phi \left[ \frac{1}{2}\widehat{\omega }^{2}-i%
\widehat{\omega }\right] \delta \varphi +i\delta \varphi \left[ 2\widehat{%
\omega }\coth \left( \frac{\widehat{\omega }}{2T}\right) \right] \delta
\varphi - \\
-V_{\varphi \Phi }^{\prime \prime }<\delta \varphi \delta \Phi >_{0}-\frac{1%
}{2}V_{\Phi \Phi }^{\prime \prime }<\delta \Phi \delta \Phi >_{0}-\frac{1}{2}%
V_{\varphi \varphi }^{\prime \prime }<\delta \varphi \delta \varphi >_{0}
\end{array}
\right\} \eqnum{14}$$
where
$$\begin{aligned}
&<&\delta \varphi _{t}\delta \varphi _{t}>_{0}=0\text{; }<\delta \Phi
_{t}\delta \Phi _{t}>_{0}=\frac{i}{2K}\int\limits_{\mid \omega \mid >\omega
_{0}}\frac{d\omega }{2\pi }D^{\left( K\right) }\left( \omega \right) =\frac{1%
}{4\pi K}\ln \left( \frac{1}{\omega _{0}}\right) \text{; } \eqnum{15} \\
&<&\delta \varphi _{t}\delta \Phi _{t}>_{0}=i\int\limits_{\mid \omega \mid
>\omega _{0}}\frac{d\omega }{2\pi }D_{0}^{\left( R\right) }\left( \omega
\right) =\frac{i}{8K} \nonumber\end{aligned}$$
The correlator $\widehat{D}_{0}$ is defined by the equality $\widehat{D}%
_{0}=\left( \frac{\omega ^{2}}{2}\widehat{\sigma }_{x}+\frac{1}{2}<\widehat{D%
}>^{-1}\right) ^{-1}$, where $\widehat{\sigma }_{x}$ is the $\widehat{\sigma
}_{x}$- Pauli matrix. Thus, taking into account the renormalizations resulting from the fast fluctuations the effective action for the slow fields can be written as
$$S^{\left( slow\right) }\left[ \varphi ,\Phi \right] =2K\int dt\left\{
-2\varphi \left( \partial _{t}\Phi \right) -V\left[ \Phi ,\varphi \right]
+i\varphi \left[ 2\widehat{\omega }\coth \left( \frac{\widehat{\omega }}{2T}%
\right) \right] \varphi -\frac{i}{8K}V_{\varphi \Phi }^{\prime \prime
}\right\} \eqnum{16}$$
In this expression the renormalizations proportional to the correlators $%
<\delta \Phi _{t}\delta \Phi _{t}>_{0}$ give the small renormalization of the constant $\gamma $ in the potential $V$. They can be neglected due to the smallness of the value $\frac{1}{K}\ln \left( \frac{1}{\omega _{0}}%
\right) <<1$. On the other side the renormalizations proportional to the correlators $<\delta \varphi _{t}\delta \Phi _{t}>_{0}$ are taken into account because they can become comparable with the relaxation term $\varphi %
\left[ 2\omega \coth \left( \frac{\omega }{2T}\right) \right] \varphi $ in the case of the low temperatures and small frequencies $T\thicksim \frac{1}{K%
}$, $\omega \thicksim \frac{1}{K}$.
It is convenient to shift the field $\Phi \rightarrow \Phi -\Delta \mu t$. As a result of this shift the action (16) takes the form
$$S^{\left( slow\right) }\left[ \varphi ,\Phi \right] =2K\int dt\left\{
-2\varphi \left( \partial _{t}\Phi \right) -V^{\left( 0\right) }\left[ \Phi
,\varphi \right] +i\varphi \left[ 2\widehat{\omega }\coth \left( \frac{%
\widehat{\omega }}{2T}\right) \right] \varphi -\frac{i}{8K}V_{\varphi \Phi
}^{\left( 0\right) \prime \prime }\right\}$$ where $V^{\left( 0\right) }\left[ \Phi ,\varphi \right] =\gamma \sin \Phi
\sin \left( \varphi \right) -v_{0}\varphi $, at that $v_{0}=2\Delta \mu $. Taking into account the smallness of the quantum noise $\varphi $ due to the large value of the constant $K>>1$ the action can be expanded in the powers of the field $\varphi $. This expansion gives
$$S_{eff}^{\left( slow\right) }\left[ \varphi ,\Phi \right] =2K\int dt\left\{
-2\varphi \left( \partial _{t}\Phi \right) +i\varphi \left[ 2\widehat{\omega
}\coth \left( \frac{\widehat{\omega }}{2T}\right) +\frac{1}{16K}\gamma \cos
\Phi \right] \varphi -A\left( \Phi \right) \varphi -i\frac{1}{8K}A^{\prime
}\left( \Phi \right) \right\} \eqnum{17}$$
where
$$A\left( \Phi \right) =\gamma \sin \Phi -v_{0}\text{; \ \ \ }A^{\prime
}\left( \Phi \right) =\gamma \cos \Phi \text{\ } \eqnum{18}$$
The first correction to the renormalized amplitude of the potential $V\left[
\Phi ,\varphi \right] $, i.e., to the constant $\gamma $ with respect to the value $1/K$ has the form $\gamma ^{\ast }=\gamma \left( 1-\frac{1}{4\pi K}%
\ln \left( \frac{K}{\gamma }\right) \right) $. Integrating over $\varphi $ in the generation functional Z we obtain
$$S_{eff}^{\left( slow\right) }\left[ \Phi \right] =-\frac{1}{2}Sp\ln \left(
\widehat{\Gamma }\right) +2K\int dtdt^{\prime }\left\{ \frac{i}{4}\left[
-2\left( \partial _{t}\Phi \right) -A_{t}\right] \widehat{\Gamma }%
_{t,t^{\prime }}\left[ -2\left( \partial _{t^{\prime }}\Phi \right)
-A_{t^{\prime }}\right] -i\frac{1}{8K}A_{t}^{\prime }\delta \left(
t-t^{\prime }\right) \right\}$$ where the correlator $\widehat{\Gamma }_{t,t^{\prime }}$ is
$$\widehat{\Gamma }=\left[ 2\widehat{\omega }\coth \left( \frac{\widehat{%
\omega }}{2T}\right) +\frac{1}{16K}\gamma ^{\ast }\cos \Phi \right]
^{-1}\approx \left[ 2\widehat{\omega }\coth \left( \frac{\widehat{\omega }}{%
2T}\right) +\frac{1}{16K}\gamma ^{\ast }\right] ^{-1} \eqnum{19}$$
In the approximation of the local in time action we have
$$S_{eff}^{\left( slow\right) }\left[ \Phi \right] =i2K\int dt\left\{ \frac{1}{%
4}\left[ 2\left( \partial _{t}\Phi _{t}\right) +A\left( \Phi _{t}\right) %
\right] ^{2}\widehat{\Gamma }_{0}-\frac{1}{8K}A^{\prime }\left( \Phi
_{t}\right) \right\} \eqnum{20}$$
$$\widehat{\Gamma }_{0}=\frac{1}{4T+\frac{1}{16K}\gamma ^{\ast }}=\frac{1}{%
4T^{\ast }} \eqnum{21}$$
Finally, the action can be represented in the form
$$S_{eff}^{\left( slow\right) }\left[ \Phi \right] =i2K\int dt\left\{ \frac{M}{%
2}\left( \partial _{t}\Phi \right) ^{2}+\frac{M}{8}A^{2}-\frac{1}{8K}%
A^{\prime }\right\} \eqnum{22}$$
where $M=2\widehat{\Gamma }_{0}=\frac{1}{2T^{\ast }}$. Note that the generation potential Z with the action $S_{eff}^{\left( slow\right) }\left[
\Phi \right] $ (22) has the form corresponding to the quantum mechanics with the ”imaginary time” in which the role of the ”imaginary time” plays the usual real time. This action corresponds to the action which is obtained for the classical random process describing by the Langevin equation with the white noise \[14\] and having the form of the supersymmentric quantum mechanics. The Schredinger equation and the Hamiltonian corresponding to the action $S_{eff}^{\left( slow\right) }\left[ \Phi \right] $ are
$$-\frac{1}{2K}\partial _{t}\Psi =\widehat{H}_{FP}\Psi \eqnum{23}$$
$$\widehat{H}_{FP}=-\frac{1}{\left( 2K\right) ^{2}}\frac{d^{2}}{2Md^{2}\Phi }+%
\frac{M}{8}A^{2}-\frac{1}{8K}A^{\prime } \eqnum{24}$$
This Hamiltonian $\widehat{H}_{FP}$ can be represented in the factorized form
$\widehat{H}_{FP}=-\frac{1}{2M\left( 2K\right) ^{2}}\left( \frac{d}{d\Phi }%
-MKA\right) \left( \frac{d}{d\Phi }+MKA\right) $. The transformation $\Psi
=\exp \left\{ MKV\left( \Phi \right) \right\} P\left( \Phi \right) $, where $%
V\left( \Phi \right) =\gamma ^{\ast }\left( 1-\cos \Phi \right) -v_{0}\Phi $ and $A=\frac{d}{d\Phi }\left( V\left( \Phi \right) \right) $ transfer the Hamiltonian (18) to the form of the Fokker-Planck operator $\widehat{L}%
_{FP}=-\frac{1}{2M\left( 2K\right) ^{2}}\frac{d}{d\Phi }\left( \frac{d}{%
d\Phi }+2MKA\right) $. The stationary solution of the corresponding Fokker-Planck equation in the case of the zero velocity $v_{0}=0$ has the form $P\left( \Phi \right) =C_{0}\exp \left( -\frac{KV\left( \Phi \right) }{%
T^{\ast }}\right) $, where $V\left( \Phi \right) =\gamma ^{\ast }\left(
1-\cos \Phi \right) $ and the constant $C_{0}$ is $C_{0}=\left( \frac{%
K\gamma ^{\ast }}{2\pi T^{\ast }}\right) ^{\frac{1}{2}}$. In the case of the nonzero velocity $v_{0}$ the stationary solution of the Fokker-Planck equation is determined by the equation
$$\left( \frac{d}{d\Phi }+2MKA\right) P\left( \Phi \right) =j$$ and can be found taking into account the condition of the periodicity of the probability distribution $P\left( 0\right) =P\left( \Phi \right) $ and the normalization condition $\int\limits_{0}^{2\pi }d\Phi P\left( \Phi \right)
=1 $ \[13\], \[18\]. The constant j is found from the normalization condition for $P\left( \Phi \right) $ and for the case of the small $\Delta n$ we obtain $j=\frac{v_{0}}{\gamma ^{\ast }}\exp \left\{ -2\left( \frac{K\gamma
^{\ast }}{T^{\ast }}\right) \right\} $. The probability of the phase slip process is determined by the value $\left( \partial _{t}<\Phi >\right) $ $%
\sim j$. For the zero temperature and small velocities $v_{0}$, so as $%
v_{0}=\Delta \mu <<T^{\ast }$, the value $T^{\ast }$ can be considered as the effective temperature and for the large time t, when the action can be considered in the local in time form, determines the diffusion coefficient. In the opposite case $v_{0}=\Delta \mu >>T^{\ast }$ the diffusion coefficient is defined by the value $\widehat{\Gamma }_{t,t}=\ln \left(
\frac{1}{v_{0}}\right) $ and this value should be substituted instead of the value $\left( \gamma ^{\ast }/T^{\ast }\right) $. This can be seen from the expression (20) for $\widehat{\Gamma }$ with taking into account that in the case of the large value of $\Delta \mu $ compared with $T^{\ast }$ the frequencies $\omega $ should be cut of for the small values by the chemical potential difference $\Delta \mu $. Note, that the result for the probability of the phase slip $\left( 1/\tau _{ps}\right) \thicksim
v_{0}\exp \left\{ -2K\ln \left( \frac{1}{v_{0}}\right) \right\} $ coincides with the result of the works \[15-17\].
The Fokker-Planck equation describing the relaxation process of the difference of phases has been obtained both for high temperatures and for low temperatures. For high temperatures $T\gtrsim \gamma $ the obtained Fokker-Planck equation is equal to the classical Langevin equation with the white noise with the correlator proportional to 1/T \[18\]. For the low temperatures $T<<\gamma $ and the small density difference $\Delta n$, so as the inequality $v_{0}<<1/K$ takes place, the relaxation kinetics of the difference of phases is described by the Fokker-Planck equation with the effective temperature $T^{\ast }$ and the diffusion coefficient determined by both the usual temperature and the quantum fluctuations of the Bose gas. For the small temperatures and the large value of the density difference $%
\Delta n$, so as $v_{0}>>\gamma /K$, the diffusion coefficient is determined by the value $v_{0}$. In the last case the result is analogues to the result of the works \[15-17\].
The author thanks S. Burmistrov, A. Kozlov and Yu. Kagan for helpful discussions. This work was supported by the Russian Foundation for Basic Research, by the Netherlands Organization for Scientific Research (NWO) and by INTAS -2001-2344.
\[1\] M. H. Anderson, J. H. Ensher, M. B. Matthews, C. E. Weiman, and E. A. Cornell, Science 269, 198 (1995); K. B. Davis et al., Phys. Rev. Lett. 75, 3969 (1995); C. C. Bradley, C. A. Sacklet, J. T. Tollett, and R. G. Hulet, Phys. Rev. Lett. 75, 1687 (1995).
\[2\] D. S. Hall, M. R. Matthews, C. E. Weiman, and E. A. Cornell, Phys. Rev. Lett. 81, 1543 (1998).
\[3\] F. Dalfovo, L. P. Pitaevskii, and S. Stringari, Phys. Rev. A 54, 4213 (1996).
\[4\] A. Smerzi, S. Fantoni, S. Giovanazzi, and S. R. Shenoi, Phys. Rev. Lett. 79, 4950 (1997).
\[5\] I. Zapata, F. Sols, and Leggett, Phys. Rev. A 57, R28 (1998).
\[6\] B. D. Josephson, Phys. Lett. 1, 251 (1962).
\[7\] V. Ambegaokar, U. Eckern, and G. Schen, Phys. Rev. Lett. 48, 1745 (1982),U. Eckern, and G. Schen, V. Ambegaokar, Phys. Rev. B 30, 6419 (1984).
\[8\] F. Meier, and W. Zwerger, cond-mat/ 9904147; F. Meier, and W. Zwerger, Phys. Rev. A 64, 033610 (2001).
\[9\] V. S. Babichenko, cond-mat/ 0109248.
\[10\] L. V. Keldysh, JETP, 47, 1515, (1964).
\[11\] J. Schwinger, J. Math. Phys. 2, 407 (1961).
\[12\] R. P. Feinman, A. R. Hibbs, ”Quantum Mechanics and Path Integrals”, N.Y., 1965.
\[13\] H. Risken, Z. Phys., 251, 231 (1972); K. Kawasaki, Progr. Theor. Phys. 51, 1064 (1974); R. Graham Z. Phys. B40, 149, (1980).
\[14\] M. V. Feigelman, A. M. Tsvelik, JETP, 83, 1430 (1982).
\[15\] H. P. Buchler, V. B. Geshkenbein, and G. Blatter, Phys. Rev. Lett. 87, 100403 (1982).
\[16\] C. L. Kane and M. R. A. Fisher, Phys. Rev. Lett. 68, 1220 (1992).
\[17\] A. Schmid, Phys.Rev.Lett. 51,1506 (1983); S. A. Bulgadaev JETP Lett. 39, 315 (1984); S. E. Korshunov JETP Lett. 65, 1025 (1987).
\[18\] V. Ambegaokar and B. I. Halperin, Phys. Rev. Lett. 22, 1364 (1969).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We develop the density matrix renormalization group approach to systematically identify the topological order of the quantum spin liquid (QSL) through adiabatically obtaining different topological degenerate sectors of the QSL on an infinite cylinder. As an application, we study the anisotropic kagome Heisenberg model known for hosting a Z$_2$ QSL, however no numerical simulations have been able to access all four sectors before. We obtain the complete set of four topological degenerate ground states distinguished by the presence or absence of the spinon and vison quasiparticle line, which fully characterizes the topological nature of the quantum phase. We have also studied the kagome Heisenberg model, which has recently attracted a lot of attention. We find two topological sectors accurately and also estimate various properties of the other topological sectors, where the larger correlation length is found indicating the possible proximity to another phase.'
author:
- 'Yin-Chen He'
- 'D. N. Sheng'
- Yan Chen
title: Obtaining topological degenerate ground states by the density matrix renormalization group
---
Introduction
============
The quantum spin liquid (QSL), a state which does not break any lattice or spin-rotational symmetry at zero temperature, has attracted much attention in the past twenty years. [@Balents2010] Different from a trivial disordered state, the QSL possesses topological order [@Wen1990; @Wen2004] with the deconfined and fractionalized quasiparticles obeying the anyonic braiding statistics. The physics of the QSL may also have an implication for understanding the high temperature superconductivity. [@Anderson1987] Since Anderson first proposed the resonating valence bond (RVB) state for the triangular Heisenberg magnet, [@Anderson1973] the debate on whether a QSL is a realistic quantum state in two dimensional (2D) systems has never ceased. In recent years, there is growing experimental [@Kagawa2005; @Kurosaki2005; @Helton2007; @Lee2007; @Mendels2007; @Okamoto2007; @Yamashita2008; @Olariu2008; @Wulferding2010; @Imai2011; @Jeong2011; @Clark2013; @Han2012] and theoretical [@Misguich1999; @Moessner2001; @Balents2002; @Sheng2005a; @Isakov2006; @Isakov2011; @Jiang2008; @Meng2010; @Yan2011; @Wang2011; @Jiang2012a; @Jiang2012b; @Depenbrock2012; @Nishimoto2013] evidence supporting the existence of the QSL in realistic materials or contrived model systems.
Theoretically, frustrated magnetic interactions may lead to a QSL. Those systems impose serious difficulties for theoretical studies, where analytical methods are under development and the quantum Monte Carlo method is usually not applicable due to the sign problem. To tackle these problems, the slave particle formalism [@Read1991; @Wen1991; @Lee2005; @Wang2006; @Ran2007; @Iqbal2011; @Clark2011; @Ruegg2012; @Senthil2000] has been developed and different model Hamiltonians have been constructed, [@Moessner2001; @Rokhsar1988; @Misguich2002a; @Balents2002; @Motrunich2002; @Levin2003; @Kitaev2003; @Hermele2004; @Ruegg2012] which give many insights for the properties of the QSL. However, these studies still can not provide concrete predictions regarding the existence of the QSL in more realistic quantum systems. The development of the density matrix renormalization group (DMRG) [@DMRG] and the tensor network approach [@PEPS; @Wang2011; @MERA; @Evenbly2012; @Wang2013] has opened a new route to study the QSL in general magnetic systems; in particular, the accurate DMRG studies have provided extensive evidence for a possible gapped Z$_2$ QSL for the kagome lattice Heisenberg model. [@Yan2011; @Jiang2012b; @Depenbrock2012] However, the variational Monte Carlo [@Ran2007; @Iqbal2011] study of the same system suggests a possible gapless QSL, which appears to be more consistent with experimental observations. [@Helton2007; @Wulferding2010; @Imai2011; @Jeong2011; @Clark2013] Recently, it has been suggested [@Wang2013] that both a small correlation length of one topological sector [@Yan2011; @Depenbrock2012] and the positively quantized topological entanglement entropy used in identifying the $Z_2$ QSL [@Jiang2012b; @ssg2013h] may not be sufficient to fully establish the nature of the quantum state limited by the range of system sizes being studied. Therefore, it is crucial to find other topological sectors, which may lead to a full understanding of the topological nature of QSL through extracting the modular matrix. [@Wen1990; @Zhang2012; @Cincio2013; @Zaletel2013; @Zhu2013]
DMRG has been proven powerful in solving the ground state of quasi-one dimensional frustrated magnets, [@Jiang2008; @Meng2010; @Yan2011; @Jiang2012a; @Jiang2012b; @Depenbrock2012; @Nishimoto2013] however it can not directly obtain excited state accurately for larger systems, especially on a torus geometry where the topological degeneracies exist. In a recent work, [@Cincio2013] Cincio and Vidal showed that the topological degeneracy can be studied in an infinite cylinder, and they can obtain topological degenerate ground states by using random initial conditions in the infinite DMRG simulation. [@McCulloch2008] In this paper, we propose a systematical and controlled approach based on the DMRG calculations to find different topological degenerate sectors of the quantum system on an infinite cylinder. In Sec. \[sec:top\_deg\], we briefly review the origin of topological degeneracy and derive topological degeneracies of a Z$_2$ as well as double-semion QSL phases. We show that different topological degenerate ground states differ from each other by certain type of quasiparticle (spinon or vison for the QSL) lines threaded in the system, and these states can be tuned into each other by inserting flux. In Sec. \[sec:algorithm\], we outline the general numerical scheme of the algorithm, which are based on the origin of topological degeneracy. We show that, to obtain topological degenerate ground states in QSL, two operations can be implemented in the DMRG simulation: (1) creating edge spinon. (2) adiabatically inserting $2\pi$ flux. In Sec. \[sec:winding\_number\], we discuss the relation between winding number and spinon line. We also propose a simple method to determine the presence of spinon line by observing the entanglement spectrum. In Sec. \[sec:easy\_axis\], we apply the method to the anisotropic easy axis kagome Heisenberg model (EAKM), which is shown to host a Z$_2$ QSL theoretically [@Balents2002] and numerically. [@Sheng2005a; @Isakov2006; @Isakov2011] We successfully find four topological degenerate states and calculate various quantities to show that they are four distinct states. Further, in Sec. \[sec:KHM\], we apply our method to nearest neighbor kagome Heisenberg model (KHM). For the KHM, we have only identified two topological sectors, however, our results may provide a good approximation on the properties of other topological sectors. We find that the states in the new topological sectors have a much larger correlation length, which demands future study.
Topological degeneracy: Z$_2$ and double-semion quantum spin liquid\[sec:top\_deg\]
===================================================================================
The topological degeneracy originates from the existence of the fractionalized quasiparticles and their anyonic braiding statistics. Following Ref. , we will give an exact derivation of the topological degeneracy in Z$_2$ and double-semion QSL. [@Levin2003] Either Z$_2$ or double-semion QSL supports two kinds of fractionalized excitations, spinon and vison. To deduce the topological degeneracies, we should first define the Wilson loop operator $\mathcal{T}_{s}^{x(y)}$ ($\mathcal{T}_{v}^{x(y)})$ (Fig. \[fig:top\_deg\]), which creates a pair of spinons (visons), then winds them along $x$ ($y$) direction, and finally annihilates them. If the system has a gap and we drag the quasiparticles slowly enough, the system will get back into the ground state after the annihilation of quasiparticles. Therefore, no matter whether there is ground state degeneracy, we can always find a ground state $|\psi_0\rangle$ which is the eigenstate of the $\mathcal{T}_{s}^y$ and $\mathcal T_v^y$, satisfying $\mathcal{T}_{s}^y|\psi_0\rangle=\alpha_s|\psi_0\rangle$ and $\mathcal{T}_{v}^y|\psi_0\rangle=\alpha_v|\psi_0\rangle$. Here $\alpha_{s(v)}$ are non-universal numbers, for simplicity we just take them to be $1$. In the following, we will use two ways to derive that the system also has three other topological degenerate ground states $|\psi_s\rangle$, $|\psi_s\rangle$, $|\psi_{sv}\rangle$, whose corresponding eigenvalues are $(-1,1)$, $(1,-1)$ and $(-1,-1)$: $$\begin{aligned}
\psi_0&= |\mathcal{T}_s^y= 1, \mathcal T_v^y= 1\rangle, \quad \psi_s= |\mathcal{T}_s^y= -1, \mathcal T_v^y= 1\rangle \nonumber \\ \psi_v&= |\mathcal{T}_s^y= 1, \mathcal T_v^y= -1\rangle, \,\psi_{sv}= |\mathcal{T}_{s}^y= -1, \mathcal T_v^y= -1\rangle.\end{aligned}$$ These topological degenerate ground states, defined by eigenstates of the Wilson loop operator along $y$ direction, are the minimal entangled states introduced in Ref. .
 Topological sectors created by threading flux. (b) Topological sectors created by dragging out an anyon line.](top_deg.eps){width="35.00000%"}
One way to derive the topological degeneracy is using the existence of the deconfined fractionalized quasiparticles. Similar to the fractional quantum Hall effect, one can insert flux $\theta$ in the torus along $x$ direction. We call the corresponding operators $\mathcal F_{s(v)}^x(\theta)$. Then, the spinon (vison) circles around $y$ direction will acquire an Aharonov–Bohm phase $\exp (i\theta c)$, where the charge of the quasiparticle is $c=1/2$. Thus, $$\mathcal T_s^y \mathcal F_s^x(\theta)|\psi_0\rangle=\exp (i\theta c) \mathcal F_s^x(\theta)|\psi_0\rangle.$$ If one inserts $2\pi$ flux, the system will be brought back into the original system, however, due to the fractional charge, the Aharonov–Bohm phase from the flux will be $-1$. Therefore, $\mathcal F_s^x(2\pi)|\psi_0\rangle$ is just the state $|\psi_s\rangle$, $\mathcal T_s^y \mathcal F_s^x(2\pi)|\psi_0\rangle=-\mathcal F_s^x(2\pi)|\psi_0\rangle$. In general, we have $$\begin{aligned}
|\psi_s\rangle&=\mathcal F_s^x(2\pi)|\psi_0\rangle, \quad |\psi_v\rangle=\mathcal F_v^x(2\pi)|\psi_0\rangle, \nonumber \\ |\psi_{sv}\rangle&=\mathcal F_s^x(2\pi)\mathcal F_v^x(2\pi)|\psi_0\rangle. \label{eq:top_flux}\end{aligned}$$ The topological degenerate ground states obtained by inserting flux only rely on the fact of fractionalized particles, there is no difference between Z$_2$ spin liquid and double-semion spin liquid.
On the other hand, we can deduce the topological degeneracy from the braiding statistics of quasiparticles. The anyonic braiding statistics of Z$_2$ spin liquid can be summarized as the following. Firstly, either spinon or vison has trivial braiding statistics to itself. Secondly, spinon (vison) obeys semionic statistics relative to vison (spinon), which means if a spinon (vison) encircles around a vison (spinon), there will be a phase factor $-1$. The braiding statistics between spinon and vison in Z$_2$ QSL can be written as: $$\begin{aligned}
\mathcal{T}_{s}^{x} \mathcal{T}_{v}^{y}&=-\mathcal{T}_{v}^{y}\mathcal{T}_{s}^{x}, \quad \mathcal{T}_{v}^{x} \mathcal{T}_{s}^{y}=-\mathcal{T}_{s}^{y}\mathcal{T}_{v}^{x}, \nonumber \\ \quad \mathcal T^x_s \mathcal T^y_s&=\mathcal T^y_s \mathcal T^x_s , \quad \mathcal T^x_v \mathcal T^y_v=\mathcal T^y_v \mathcal T^x_v.\end{aligned}$$ Then, $\mathcal{T}_{s}^y(\mathcal{T}_{s}^x|\psi_0\rangle)=\mathcal{T}_{s}^x\mathcal{T}_{s}^y|\psi_0\rangle=( \mathcal{T}_{s}^x|\psi_0\rangle)$ and $\mathcal{T}_{v}^y(\mathcal{T}_{s}^x|\psi_0\rangle)=-\mathcal{T}_{s}^x\mathcal{T}_{v}^y|\psi_0\rangle=-( \mathcal{T}_{s}^x|\psi_0\rangle)$. Therefore, $|\psi_v\rangle=\mathcal{T}_{s}^x|\psi_0\rangle$ is the ground state in the sector-$(1,-1)$. This sector can be simply understood as a state with a spinon line in the $x$ direction, as in Fig. \[fig:top\_deg\](b). Similarly, we can have two other ground states. In general, in a Z$_2$ QSL we have: $$|\psi_v\rangle=\mathcal T_s^x |\psi_0\rangle, \,\, |\psi_{s}\rangle=\mathcal T_v^x|\psi_0\rangle, \,\, |\psi_{sv}\rangle=\mathcal T_s^x \mathcal T_v^x|\psi_0\rangle,$$ and, $$\mathcal T_s^x=\mathcal F_v^x(2\pi), \quad \mathcal T_v^x=\mathcal F_s^x(2\pi). \label{eq:z2_flux_anyon}$$
On the contrary, the double-semion QSL has quite different braiding statistics. Spinon (vison) obeys semionic braiding statistics to itself, but trivial braiding statistics to vison (spinon). The braiding statistics between spinon and vison in double-semion QSL can be written as: $$\begin{aligned}
\mathcal{T}_{s}^{x} \mathcal{T}_{v}^{y}&=\mathcal{T}_{v}^{y}\mathcal{T}_{s}^{x}, \quad \mathcal{T}_{v}^{x} \mathcal{T}_{s}^{y}=\mathcal{T}_{s}^{y}\mathcal{T}_{v}^{x}\nonumber \\ \mathcal T^x_s \mathcal T^y_s&=-\mathcal T^y_s \mathcal T^x_s , \quad \mathcal T^x_v \mathcal T^y_v=-\mathcal T^y_v \mathcal T^x_v.\end{aligned}$$ Therefore, in double-semion QSL we have: $$|\psi_s\rangle=\mathcal T_s^x |\psi_0\rangle, \,\, |\psi_{v}\rangle=\mathcal T_v^x|\psi_0\rangle, \,\, |\psi_{sv}\rangle=\mathcal T_s^x \mathcal T_v^x|\psi_0\rangle,$$ and, $$\mathcal T_s^x=\mathcal F_s^x(2\pi), \quad \mathcal T_v^x=\mathcal F_v^x(2\pi). \label{eq:ds_flux_anyon}$$
Different topological sectors from infinite DMRG \[sec:algorithm\]
==================================================================
The topological degeneracy is accurately defined on a torus, however one can also have topological degenerate sectors on an infinite cylinder. [@Cincio2013; @Poilblanc2012; @Schuch2013] This can be understood by cutting the torus into a cylinder as shown in Fig. \[fig:deg\](a). The subtle difference for the cylinder is that a pair of anyons will be standing at the two ends of the cylinder instead of annihilating each other as in the torus case. This difference will not bring any distinct behaviors to the local wave function, as long as one measures the system far away from the ends. However, the existence of edge quasiparticles will bring large energy punishment to the system, as a result, different topological sectors only have the same energies in the bulk of the cylinder. This will not result in any important effect in the infinite DMRG simulation, [@McCulloch2008] since one only optimizes the energy in the center of the infinite cylinder. A nice feature is that simulations on an infinite cylinder will collapse into one topological sector, [@Cincio2013; @Zaletel2013] which is the key to the controlled method we develop.
 Cutting a torus into a cylinder. (b) Illustration of an infinite DMRG algorithm. (c) $\psi_0$, $\psi_v$, $\psi_s$ and $\psi_{sv}$ of a Z$_2$ QSL from two operations $\mathcal T_s^x$ and $\mathcal F_s^x$.](top_infDMRG.eps){width="48.00000%"}
To obtain topological sectors systematically, we should find out how to realize those $\mathcal T^x_\alpha$ and $\mathcal F^{x}_\alpha$ operations. To drag out an anyon line, one can apply a string operator on the system, however the string operator is usually difficult to identify. Here we use an equivalent way, which does not drag out the anyon line directly, but helps the system to develop an anyon line. This can be done by creating edge quasiparticles on the cylinder at the beginning of the infinite DMRG simulation as illustrated in Fig. \[fig:deg\](b). In the simulation, one cuts the cylinder into two halves, inserts one column of sites and optimizes the energy within the newly added sites. This procedure is repeated until the convergence is reached. It appears that there are two options for the newly inserted sites, either with or without an anyon line attaching to them. If there is an anyon line, there will be an energy cost $\Delta E$, which comes from the energy splitting of different topological sectors due to finite size effect. On the contrary, if there is no anyon line, two quasiparticle domain walls will appear at the interface of newly added sites and the older system, which brings in an energy punishment $2E_q$ (the energy of two quasiparticles). Therefore, if $2E_q$ is much bigger than $\Delta E$, the simulation will collapse into the topological sector with an anyon line in it as desired.
To apply the flux insertion operation $\mathcal F_\alpha^x$, we begin with DMRG simulation and obtain one topological sector first. Now we can adiabatically turn on the flux characterized by a boundary twist parameter $\theta$ from $0$ to $2\pi$ gradually, and the starting topological sector will evolve into another topological sector. The key point here is how to ensure the adiabaticity during the flux insertion. We discretize the flux $\theta\in (0,2\pi)$ and slowly increase $\theta$ to obtain new wave function by using the state from previous $\theta$ as the initial wave function. Under this procedure, the system should evolve adiabatically into the new topological sector to avoid having two quasiparticle domain walls at the interface between the inserted sites and the older system.
In a QSL, there exists at least one type of quasiparticles, the spinon, which can be considered as an unpaired spin in the RVB representation. Thus, to get a topological sector with a spinon line threaded in, we can simply put unpaired impurity spins (pin or remove one site) on the left and right boundaries similar to the pinning in the earlier work. [@Yan2011] Meanwhile, the spinon flux insertion can be realized by the twist boundary condition.[@Niu1985; @Misguich2010] In the hard-core boson representation of the spin system, the spinon is the fractionalized hard-core boson with charge number $1/2$. Then flux $\mathcal F_s$ is just the “magnetic” flux seen by hard-core bosons, which can be realized by the twist boundary condition along the $y$ direction: $$S_1^+S_N^-+S_1^-S_N^+ \rightarrow e^{i\theta} S_1^+S_N^-+e^{-i\theta}S_1^-S_N^+. \label{eq:twist}$$ A similar twist technique has been used in the 2D extension of the LSM theorem, [@Hastings2004] however there is a difference which was also mentioned in Ref. . The flux insertion in 2D LSM theorem relies on the momentum counting, while our method applies to a topologically fractionalized phase relying on the adiabatic evolution of the topological sector.
In a QSL, we can apply these two operations $\mathcal T_s^x$ and $\mathcal F_s^x$ to get different topological sectors. Since only spinon is sensitive to the spinon flux $\mathcal F_s^x$, the topological sector obtained by spinon flux insertion should always have the Wilson loop operator $\mathcal T_s^y=-1$, as shown in Eq. (\[eq:top\_flux\]). On the contrary, which topological sector we obtain through developing spinon line will depend on the details of the quasiparticle braiding statistics in the QSL. In a Z$_2$ QSL, the spinons (visons) obey the mutual semionic braiding statistics relative to the visons (spinons), but trivial statistics to themselves. Then dragging out a spinon line equals to inserting a vison flux (Eq. \[eq:z2\_flux\_anyon\]), since only vison is sensitive to the spinon line. Therefore, the ground states in four sectors of a Z$_2$ QSL can be obtained by those two operations $\mathcal T_s^x$ and $\mathcal F_s^x$ as shown in Fig. \[fig:deg\](c). In contrast, for the double-semion QSL, [@Levin2003] the spinons (visons) obey semionic braiding statistics to themselves but trivial statistics relative to visons (spinons). Then, only spinon is sensitive to spinon line, so that dragging out a spinon line is equivalent to inserting a spinon flux (Eq. \[eq:ds\_flux\_anyon\]). Therefore, these two operations will lead to exactly the same state and we need other operations to get new topological sectors.
The winding number, spinon line and entanglement spectrum \[sec:winding\_number\]
=================================================================================
In literatures, the winding number is often used to label the topological sectors in the RVB type QSL. The winding number $W_{x(y)}$ is defined by the parity of the number of the singlets cut by a loop along the $x$ ($y$) direction. In fact, a topological sector with a spinon line threaded in $x$ direction just has winding number $W_y=-1$. [@Misguich2010] To understand this, we consider a cylinder with even number of sites one column, and we do a cut along $y$ (vertical) direction. For $2N$ spins on the half cylinder, $2N_p$ of which are forming singlet pairs within the half cylinder, $N_c$ of which are forming singlet with spins from the other half cylinder. If there is no unpaired spinon in the cylinder, we have $2N=2N_p+N_c$, then $N_c$ is an even number, $W_y=1$. If there is a spinon line in the cylinder, leaving unpaired spinon on the left and right edge of the cylinder, then $2N=2N_p+N_c+1$, which means $W_y=-1$.
Although we have an exact correspondence between spinon line and winding number, the form of the Wilson loop operator corresponding to the winding number actually depends on the details of quasiparticles braiding statistics. For example, in a Z$_2$ QSL, the winding number $W_y$ is the eigenvalue of Wilson loop operator $\mathcal T_v^y$, while in the double-semion QSL, $W_y=\mathcal T_s^y$. In a Z$_2$ QSL, the topological sectors labeled in winding number $(W_x=\pm 1, W_y=\pm 1)$ can be represented by the superposition of the topological sectors we used in the paper:
$$\begin{aligned}
|\pm , +\rangle&= |\mathcal T_s^y=1, \mathcal T_v^y=1\rangle \pm |\mathcal T_s^y=-1, \mathcal T_v^y=1\rangle \\ |\pm ,-\rangle&= |\mathcal T_s^y=1, \mathcal T_v^y=-1\rangle \pm |\mathcal T_s^y=-1, \mathcal T_v^y=-1\rangle. \nonumber\end{aligned}$$
However, for some QSLs those two winding numbers are not enough to label all the topological sectors. On the other hand, for some QSLs like the double-semion state, where spinons have the nontrivial braiding statistics relative to themselves, these two winding numbers can not be used simultaneously to label the topological sectors since the corresponding operators do not commute with each other.
There is a simple way to judge whether a state has a spinon line in it by observing the symmetry properties of the entanglement spectrum (ES) corresponding to a vertical-loop cut. For the sector with no spinon line, the ES is symmetric about total $S^z=0$; while for the sector with a spinon line, the ES is symmetric about $S^z=1/2$. The reason for this different symmetry behavior simply comes from the even or odd number of singlets cut by the loop. In the RVB states, only the singlet pair being cut contributes a $\pm 1/2$ to the total $S_z$ with equal probability. As a result, ES of the state with the different winding number will have different symmetry (Fig. \[fig:entanglement\]).
Easy axis kagome model \[sec:easy\_axis\]
=========================================
We apply our method to the easy axis kagome system, which has a Hamiltonian in the following form: $$H=\sum J _{ij}^z S_i^z S_j^z+ \sum \frac{J_{ij}^{xy}}{2} (S_i^+ S_j^-+S_i^-S_j^+),$$ where the spin $z$ components have the first, second and third nearest neighbor coupling terms with the same magnitude (Fig. \[fig:easy-axis\](a)), while the spin $x$ and $y$ components have only the first nearest neighbor couplings. This model is shown to host a Z$_2$ QSL both theoretically [@Balents2002] and numerically [@Sheng2005a; @Isakov2006; @Isakov2011] for $J^z_1=J^z_2=J^z_3=J^z=1$ and FM $J_1^{xy} \geq -0.14$. [@Isakov2011] The system we study has $4$ unit cells (8 lattice sites) one column.
![(color online)\[fig:easy-axis\] (a) Easy axis kagome with second and third nearest interaction. The loop operators $\mathcal L_v$ and $\mathcal L_s$ are defined as $\mathcal L_v=\prod 2S^i_z$ and $\mathcal L_s=\prod 2S^i_x$. Energy (b) and entropy (c) evolution of four topological sectors under flux insertion. The parameter we take here is $J^{xy}=-0.1$.](easy_kagome.eps){width="48.00000%"}
We obtain the topological sector $\psi_0$ naturally in the conventional DMRG simulation, and find the $\psi_v$ by creating the edge spinon as discussed before. Then we apply the adiabatic flux insertion to get the other two sectors. The evolution of the energy and entropy with flux insertion is demonstrated in Fig. \[fig:easy-axis\]. Apparently, by threading the $2\pi$ flux, the two starting states $\psi_0$ and $\psi_v$ both evolve into the new ground states ($\psi_s$ and $\psi_{sv}$) in other topological sectors, and the states evolve back to the original states after inserting the $4\pi$ flux.
Entanglement spectrum (ES) of four states are shown in Fig. \[fig:entanglement\]: the state ($\psi_0$, $\psi_s$) without spinon line has ES symmetric about $S^z=0$; the state ($\psi_v$, $\psi_{sv}$) with spinon line has ES symmetric about $S^z=1/2$.
![\[fig:entanglement\] Entanglement spectrum of four Z$_2$ sectors. (a) $\psi_0$ (b) $\psi_s$ (c) $\psi_v$ (d) $\psi_{sv}$. Since we are using a $U(1)$ symmetry matrix product state representation, each entanglement spectrum has an $S^z$ charge. Here $J^{xy}=-0.1$.](ES_easy.eps){width="48.00000%"}
To verify that these four states are distinct states from different Z$_2$ topological sectors, we define and measure the loop operators $\mathcal L_v$ and $\mathcal L_s$ as shown in Fig. \[fig:easy-axis\](a). At the RK exactly solvable point, [@Balents2002] $\mathcal L_v$ and $\mathcal L_s$ will be identical to the Wilson loop operators $\mathcal T_v$ and $\mathcal T_s$, respectively. The obtained expectation values of the $\mathcal L_v$ and $\mathcal L_s$ for different states are shown in Table \[table:wilson\]. Clearly, in our parameter region, $\mathcal L_v\approx \pm 1$ is still a very good operator to approximate $\mathcal T_v$. On the other hand, the expectation values of $\mathcal L_s$ for the different states are much smaller than $1$, but they still are reasonably large values compared to the zero. More importantly, the two loop operators take distinct signs in four sectors, which fit well with the Z$_2$ QSL theory illustrated in Fig. \[fig:deg\](c).
$J^{xy}$ state $\mathcal L_v$ $\mathcal L_s$ $E$ $S$ $\xi_{\textrm{TM}}$ $\langle\psi_0|$ $\langle\psi_s|$ $\langle\psi_v|$
---------- ---------------------- ---------------- ---------------- ----------- ------ --------------------- ------------------ ------------------ ------------------
$|\psi_0\rangle$ 0.90 0.21 -0.25522 2.94 0.63
$|\psi_s\rangle$ 0.91 -0.18 -0.25512 3.09 0.79 0.49
$|\psi_v\rangle$ -0.90 0.13 -0.25519 3.24 0.74 0.39 0.19
$ |\psi_{sv}\rangle$ -0.90 -0.12 -0.25511 3.37 0.9 0.18 0.29 0.54
$|\psi_0\rangle$ 0.98 0.20 -0.251199 2.94 0.65
$|\psi_s\rangle$ 0.98 -0.20 -0.251197 2.97 0.66 0.43
$|\psi_v\rangle$ -0.98 0.12 -0.251194 3.30 0.82 0.14 0.06
$ |\psi_{sv}\rangle$ -0.98 -0.11 -0.251192 3.32 0.84 0.06 0.12 0.52
: The loop operators, energy, entropy, correlation length $\xi_{\textrm{TM}}$ and one column overlap $\chi_{ij}$ of four topological sectors. The unit of correlation length is unit cell.[]{data-label="table:wilson"}
The correlation length $\xi_{\textrm{TM}}$ measured from the transfer matrix (Append. \[sec:num\]) for each of these states is much smaller than the cylinder width indicating a small finite size effect. We also show the one column overlap $\chi_{ij}$ between states in different sectors (Append. \[sec:num\]) in Table \[table:wilson\]. The overlap we calculate here is conceptually different from the overlap in a finite system. If we consider the overlap between different sectors on a cylinder with length $L$, then we will obtain $(\chi_{ij})^L$, which vanishes exponentially fast. On the other hand, since the different topological sectors only differ from each other due to the different types of the quasiparticle lines threaded in them, the overlap between these sectors on a cylinder is just the spinon-spinon or vison-vison correlation: $$\begin{aligned}
\langle \psi_0|\mathcal S(0) \mathcal S(L)|\psi_0\rangle &=\langle \psi_0 | \psi_v\rangle=(\chi_{0v})^L, \,\xi_s=-1/\ln \chi_{0v}, \nonumber \\
\langle \psi_0|\mathcal V(0) \mathcal V(L)|\psi_0\rangle& =\langle \psi_0 | \psi_s\rangle=(\chi_{0s})^L, \, \xi_v=-1/\ln \chi_{0s}.\end{aligned}$$ Here $\mathcal S(0)$ and $\mathcal S(L)$ ($\mathcal V (0)$ and $\mathcal V(L)$) represent the spinons (visons) located at the left and right edge of the cylinder, respectively. The $\xi_s$ and $\xi_v$ are the spinon-spinon and the vison-vison correlation lengths, respectively. We can also extract the vison-vison correlation length approximately from $\langle \psi_0|\widetilde {\mathcal L_v^x}|\psi_0\rangle$, where $\widetilde {\mathcal L_v^x}=\prod 2S_i^z$ is a string operator defined in the $x$ direction, similar to the loop operator $\mathcal L_v$ shown in Fig. \[fig:easy-axis\](a). Extrapolating $\langle \psi_0|\widetilde {\mathcal L_v^x}|\psi_0\rangle$ (Fig. \[fig:correlation\](a)), we get the vison-vison correlation length, $\widetilde \xi_v=1.141$ for $J^{xy}=-0.1$, and $\widetilde \xi_v=1.176$ for $J^{xy}=-0.05$. On the other hand, the vison-vison correlation length from the overlap $\chi_{0s}$ is $\xi_v=1.402$ for $J^{xy}=-0.1$ and $\xi_v=1.185$ for $J^{xy}=-0.05$. At $J^{xy}=-0.1$, two correlation lengths have a bigger difference because the string operator $\widetilde {\mathcal L_v^x}$ starts to have a considerable deviation from the real vison-vison correlation operator for more negative $J^{xy}$. The spinon (vison) correlation can also be extracted from the energy splitting $\Delta E$ of different topological degenerate states by $\Delta E_{s(v)}\sim \exp (-L_y/\xi_{s(v)})$. [@Schuch2013]
![(color online) \[fig:correlation\] (a) The vison-vison correlation length $\widetilde \xi_v$ from $\langle \psi_0|\widetilde W_v^x |\psi_0\rangle$. (b) $J^{xy}$ dependent behavior of four topological sectors’ correlation lengths. For $J^{xy}\lesssim -0.14$, the system is in the superfluid (SF) phase, for $J^{xy}\gtrsim -0.14$, the system is in the Z$_2$ spin liquid phase, where one has four topological degenerate ground states.](correlation.eps){width="48.00000%"}
To see the system’s behavior under different $J_{xy}$, we plot the correlation length in Fig. \[fig:correlation\](b). For $J^{xy}\lesssim-0.14$, we only get one ground state and it has a very large correlation length indicating a gapless superfluid state; $J^{xy}\gtrsim -0.14$, the four topological degenerate ground states with degenerating bulk energies always exist.
Kagome Heisenberg model \[sec:KHM\]
===================================
In this section, we will discuss the kagome Heisenberg model with the following Hamiltonian: $$H=\sum_{\langle i, j\rangle} \bm S_{i}\cdot \bm S_j+J_2\sum_{\langle \langle i, j\rangle\rangle} \bm S_{i}\cdot \bm S_j.$$ Here the spins are coupled by the isotropic nearest neighbor $J_1$ and the next nearest neighbor $J_2$ interactions. In the following, the simulation is mainly done on systems with a width of $4$ unit cells ($8$ lattice sites) and for convenience we stick to the notation of a Z$_2$ QSL used before, which we find is more consistent with what we have observed for this system.
Without and with creating edge spinons we can get the two states $\psi_0$ and $\psi_v$, and the energy splitting between these two states for $J_2=0$ is $0.00070(4)$ per site, which is consistent with the results of the Ref. . To get the other two topological sectors, we apply the twist boundary phase (flux insertion) adiabatically, but we find that the adiabaticity of the evolution for the ground state breaks down around $\theta=240^\circ$ as shown in Fig. \[fig:kagome\](a) [@footnote2]. However one observation from our results is that the energy and the entropy continue to increase as the twist angle passes the time reversal invariant point $\theta=\pi$, which may indicate the existence of a different topological sector $\psi_s$. The breakdown of the adiabaticity may result from the instability of the $\psi_s$ as a higher energy state with larger splitting $\Delta E$.
![(color online)\[fig:kagome\] Energy (a) and entropy (b) evolution of $\psi_0$ at $J_2=0$. (c) Correlation length $\xi_{\textrm{TM}}$ of different topological sectors versus the states kept. Here the unit of the correlation length is the unit cell.](kagome_evolve.eps){width="48.00000%"}
Although we can not fully access $\psi_s$, we think that the state $\psi_0(\theta=240^\circ)$ is close to $\psi_s$ at $\theta=360^\circ$, from which we can have some estimate of the properties of $\psi_s$. For $J_2=0$, the energy splitting between $\psi_s$ and $\psi_0$ should be larger than $0.00108$ and $\psi_s$ has an entropy larger than $3.6$. We also measure the correlation length from the transfer matrix, as shown in Fig. \[fig:kagome\] (b). The $\psi_0$ has a small correlation length, which is consistent with the results obtained in Ref. . However, we find that the correlation lengths of other topological sectors, $\psi_v$ at $J_2=0$, $0.05$ and $\psi_0(\theta=240^\circ)$, are much larger. In these three states, the nearest bond spin-spin correlations are very homogeneous (Append. \[sec:bond\]), so we think they are also the QSL states with no symmetry breaking. Recently, a gapless QSL has been explicitly constructed in Ref. , and one finds that in a cylinder, some topological sector has a small correlation length while other topological sectors have a very large correlation length. This scenario is similar to what we have obtained here. However, based on our numerical results, we can not draw a conclusion about the nature of QSL on the KHM due to the strong finite size effect associated with longer correlation length, which we leave for the future study.
Summary
=======
A controlled method for DMRG approach to find different topological sectors of the QSL on an infinite cylinder is proposed and applied to the EAKM and KHM. In EAKM, the complete set of four Z$_2$ topological sectors are obtained. In KHM, we can get two topological sectors exactly, and estimate the properties of other topological sectors. We find larger correlation lengths for other topological sectors. Our numerical scheme based on creating boundary quasiparticles or threading the flux adiabatically, may also be applied to other numerical algorithms like the tensor network. Our method may help to understand the topological nature of different QSL systems and bring new excitement to the study of QSLs.
Acknowledgments
===============
This work was supported by the State Key Programs of China (Grant Nos. 2012CB921604 and 2009CB929204) and the National Natural Science Foundation of China (Grant Nos. 11074043 and 11274069) (YCH and YC), and the US National Science Foundation under grant DMR-0906816 (DNS).
Numerical Algorithm \[sec:num\]
===============================
The numerical algorithm we use is the infinite density matrix renormalization group (iDMRG) invented by McCulloch. [@McCulloch2008] Similar to the finite DMRG, we use a one dimensional path to cover all the sites of the 2D cylinder. In the iDMRG calculation, we first get left and right Hamiltonian ($L, R$ in Fig. \[fig:iDMRG\](b)) from small size simulation (or choosing random initial state [@Cincio2013]). Then we insert one column in the center and optimize the energy only within the inserted column by sweeping. After the optimization, we cut one column into two halves, absorb them into the left and right Hamiltonian respectively to get the new boundary Hamiltonians $\widetilde L, \widetilde R$, respectively. The inserting, optimizing and cutting procedure is repeated until the convergence is achieved. With the converged results for the column, one can represent the translational invariant wave function of the infinite cylinder or mimic the wave function on a torus. [@Cincio2013]
![(color online)\[fig:iDMRG\] (a) Covering a cylinder with one dimensional path. (b) Illustration of iDMRG algorithm with 4 sites in the column.](iDMRG.eps){width="48.00000%"}
In the simulation of different topological sectors, there are two important quantities to calculate, the correlation length $\xi_{\textrm{TM}}$ and overlap between different topological sectors. To calculate the correlation length $\xi_{\textrm{TM}}$, we should calculate the first and second largest eigenvalue $\lambda_{1,2}$ of the transfer matrix $T$ defined in Fig. \[fig:transfer\](a). For a normalized wave function, the largest eigenvalue $\lambda_1=1$. Then the correlation length $\xi_{\textrm{TM}}=-1/\ln \lambda_2$. This correlation length determines the largest correlation in the infinite cylinder [@McCulloch2008; @Cincio2013]. Therefore, instead of calculating various correlation functions, one can simply calculate this single quantity $\xi_{\textrm{TM}}$ to know the length scale of the largest possible correlations.
Similarly, the overlap of different states is the largest eigenvalue ($\chi$) of $F$ matrix defined in Fig. \[fig:transfer\](b). This overlap is slightly different from the overlap on a finite system. However, if one puts the wave function from iDMRG simulation on cylinder or torus with length $L$, then the overlap between two different states is $\langle \psi_1|\psi_2\rangle\approx\chi^L$.
![(color online)\[fig:transfer\] (a) Calculating the correlation length $\xi_{\textrm{TM}}$. (b) Calculating the overlap between different states.](transfer.eps){width="48.00000%"}
The bond spin correlation of the Kagaome Heisenberg Model\[sec:bond\]
=====================================================================
To know whether the obtained states have lattice symmetry breaking, we need to check if the bond correlation is uniformly distributed in the whole system, which is shown in Fig. \[fig:bond\_energy\].
![(color online)\[fig:bond\_energy\] The bond spin correlation $\langle \bm S_i \cdot \bm S_j\rangle-e_\alpha$ of different sectors, $e_\alpha$ is the average of bond spin correlations. All the results plotted are obtained by keeping 8000 states in DMRG simulations. The truncation error of $\psi_0$ is smaller than $10^{-6}$, while other states has a truncation error smaller than $5\times 10^{-6}$.](bond_cor.eps){width="48.00000%"}
For the state with nonzero twist, one should transform the twisted Hamiltonian into a translational invariant one by a gauge transformation, $$U(\theta)=\prod_{x} \prod_{y=1}^N \exp(i\frac{y }{N} \theta S_{x,y}^z).$$ Under the gauge transformation, the Hamiltonian becomes $$\begin{aligned}
J_z S_{x_1,y_1}^z S_{x_2,y_2}^z&+\frac{J}{2} (S_{x_1,y_1}^+S_{x_2,y_2}^-+S_{x_1,y_1}^-S_{x_2,y_2}^+) \nonumber \rightarrow \\ J_z S_{x_1,y_1}^z S_{x_2,y_2}^z&+\frac{J}{2} \left[e^{i(y_1-y_2)\theta/N} S_{x_1,y_1}^+S_{x_2,y_2}^- \right. \nonumber \\ & \left.
+e^{-i(y_1-y_2)\theta/N}S_{x_1,y_1}^-S_{x_2,y_2}^+ \right].
\label{eq:hoppingphase}\end{aligned}$$
The bond spin correlations in all the states are very homogeneous with fluctuations around or smaller than $1\%$.
[99]{} L. Balents, Nature (London), [**464**]{}, 199 (2010). X.-G.Wen, Int. J. Mod. Phys. B 4, 239 (1990). X.-G. Wen, *Quantum Field Theory of Many-Body Systems* (Oxford University Press, Oxford, 2004). P. W. Anderson, Science, [**235**]{}, 1196 (1987). P. W. Anderson, Mater. Res. Bull. [**8**]{}, 153 (1973). F. Kagawa, K. Miyagawa, and K. Kanoda, Nature (London) [**436**]{}, 534 (2005). Y. Kurosaki, Y. Shimizu, K. Miyagawa, K. Kanoda, and G. Saito, Phys. Rev. Lett. [**95**]{}, 177001 (2005). S.-H. Lee, H. Kikuchi, Y. Qiu, B. Lake, Q. Huang, K. Habicht, and K. Kiefer, Nature Mater. [**6**]{}, 853 (2007). P. Mendels, F. Bert, M.A. de Vries, A. Olariu, A. Harrison, F. Duc, et al., Phys. Rev. Lett. [**98**]{}, 077204 (2007). Y. Okamoto, M. Nohara, H. Aruga-Katori, and H. Takagi, Phys. Rev. Lett. [**99**]{}, 137207 (2007). S. Yamashita, Y. Nakazawa, M. Oguni, Y. Oshima, H. Nojiri, Y. Shimizu, et al., Nat. Phys. [**4**]{}, 459 (2008). A. Olariu, P. Mendels, F. Bert, F. Duc, J.C. Trombe, et al., Phys. Rev. Lett. [**100**]{}, 087202 (2008). D. Wulferding, P. Lemmens, P. Scheib, J. Röder, P. Mendels, S. Chu, et al., Phys. Rev. B [**82**]{}, 144412 (2010). J.S. Helton et al., Phys. Rev. Lett. [**98**]{}, 107204 (2007). T. Imai, M. Fu, T.H. Han, and Y.S. Lee, Phys. Rev. B [**84**]{}, 020411 (2011). M. Jeong, F. Bert, P. Mendels, F. Duc, J.C. Trombe, et al., Phys. Rev. Lett. [**107**]{}, 237201 (2011). L. Clark, J. C. Orain, F. Bert, M. A. De Vries, F. H. Aidoudi, et al., Phys. Rev. Lett. [**110**]{}, 207208 (2013). T.-H. Han, J. S. Helton, S. Chu, D. G. Nocera, J. A. Rodriguez-Rivera, C. Broholm, and Y. S. Lee, Nature (London) [**492**]{}, 406 (2012). G. Misguich, C. Lhuillier, B. Bernu, and C. Waldtmann, Phys. Rev. B [**60**]{}, 1064 (1999). R. Moessner and S. L. Sondhi, Phys. Rev. Lett. [**86**]{}, 1881 (2001). L. Balents, M. P. A. Fisher, and S. M. Girvin, Phys. Rev. B [**65**]{}, 224412 (2002). D. N. Sheng and L. Balents, Phys. Rev. Lett. [**94**]{}, 146805 (2005). S. V. Isakov, Y. B. Kim, and A. Paramekanti, Phys. Rev. Lett. [**97**]{}, 207204 (2006). S. V. Isakov, M. B. Hastings, and R. G. Melko, Nat. Phys. [**7**]{}, 772 (2011). H.-C. Jiang, Z. Y. Weng, and D. N. Sheng, Phys. Rev. Lett. [**101**]{}, 117203 (2008). Z.Y. Meng, T. C. Lang, S. Wessel, F. F. Assaad, and A. Muramatsu, Nature (London) [**464**]{}, 847 (2010). S. Yan, D. Huse, S. White, Science [**332**]{}, 1173 (2011). H.-C. Jiang, Z. Wang, and L. Balents, Nat. Phys. [**8**]{}, 902 (2012). S. Depenbrock, I. P. McCulloch, and U. Schollwöck, Phys. Rev. Lett. [**109**]{}, 067201 (2012). H.-C. Jiang, H. Yao, and L. Balents, Phys. Rev. B [**86**]{}, 024424 (2012). S. Nishimoto, N. Shibata, and C. Hotta, Nature Comm. [**4**]{}, 2287. L. Wang, Z.-C. Gu, F. Verstraete, and X.-G. Wen, arxiv: 1112.3331. N. Read, S. Sachdev, Phys. Rev. Lett. [**66**]{}, 1773 (1991). X.G. Wen, Phys. Rev. B [**44**]{}, 2664 (1991). T. Senthil and M. P. A. Fisher, Phys. Rev. B [**62**]{}, 7850 (2000). S.-S Lee and P. A. Lee, Phys. Rev. Lett. [**95**]{}, 036403 (2005) F. Wang and A. Vishwanath, Phys. Rev. B [**74**]{}, 174423 (2006). Y. Ran, M. Hermele, P. A. Lee, and X. -G. Wen, Phys. Rev. Lett. [**98**]{}, 117205 (2007). Y. Iqbal, F. Becca, and D. Poilblanc, Phys. Rev. B [**84**]{}, 020407 (2011). B. K. Clark, D. A. Abanin, and S. L. Sondhi, Phys. Rev. Lett. [**107**]{}, 087204 (2011). Y. Huh, M. Punk, and S. Sachdev, Phys. Rev. B [**84**]{}, 094419 (2011). D. S. Rokhsar and S. A. Kivelson, Phys. Rev. Lett. [**61**]{}, 2376 (1988). G. Misguich, D. Serban, and V. Pasquier, Phys. Rev. Lett. [**89**]{}, 137202 (2002). O. I. Motrunich and T. Senthil, Phys. Rev. Lett. [**89**]{}, 277004 (2002). M. A. Levin, and X.-G. Wen, Phys. Rev. B [**71**]{}, 045110 (2005). A. Kitaev, Ann. Phys. [**303**]{}, 2 (2003). M. Hermele, M. P. A. Fisher, and L. Balents, Phys. Rev. B [**69**]{}, 064404 (2004). A. Rüegg and G.A. Fiete, Phys. Rev. Lett. [**108**]{}, 046401 (2012) S. R. White, Phys. Rev. Lett. [**69**]{}, 2863 (1992); Phys. Rev. B [**48**]{}, 10345 (1993). F. Verstraete and J. I. Cirac, arXiv:cond-mat/0407066; J. I. Cirac and F. Verstraete, J. Phys. A: Math. Theor. [**42**]{}, 504004 (2009). G. Vidal, Phys. Rev. Lett. [**101**]{}, 110501 (2008). G. Evenbly and G. Vidal, Phys. Rev. Lett. [**104**]{}, 187203 (2010). L. Wang, D. Poilblanc, Z.-C. Gu, X.-G. Wen, and F. Verstraete, Phys. Rev. Lett. [**111**]{}, 037202 (2013). S.-S. Gong, D. N. Sheng, O. I. Motrunich, and M.P. A. Fisher, Phys. Rev. B [**88**]{}, 165138 (2013). Y. Zhang, T. Grover, A. Turner, M. Oshikawa, and A. Vishwanath, Phys. Rev. B [**85**]{}, 235151 (2012). L. Cincio and G. Vidal, Phys. Rev. Lett. [**110**]{}, 067208 (2013). M. P. Zaletel, R. S. K. Mong, and F. Pollmann, Phys. Rev. Lett. [**110**]{}, 236801 (2013). W. Zhu, D. N. Sheng, and F. D. M. Haldane, Phys. Rev. B [**88**]{}, 035122 (2013). I. P. McCulloch, arxiv: 0804.2509. M. Oshikawa and T. Senthil, Phys. Rev. Lett. [**96**]{}, 060601 (2006). D. Poilblanc, N. Schuch, D. Pérez-García, and J. I. Cirac, Phys. Rev. B [**86**]{}, 014404 (2012). D. Poilblanc and N. Schuch, Phys. Rev. B [**87**]{}, 140407(R) (2013), Q. Niu, D. J. Thouless and Y.-S. Wu, Phys. Rev. B [**31**]{}, 3372 (1985). G. Misguich, in *Introduction to Frustrated Magnetism*, edited by C. Lacroix, P. Mendels and F. Mila, Springer (2010). M. B. Hastings, Phys. Rev. B [**69**]{}, 104431 (2004). We have tried many tricks (including using a very small step $2^\circ$) to ensure the adiabaticity, but all failed. The state flows into $\psi_v$ after the failure of adiabaticity is a random phenomenon. When the $J_2$ interaction is large $J_2\sim 0.1$, the state will flow back into $\psi_0$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We have investigated the stability limits of small spin-polarized clusters consisting of up to ten spin-polarized tritium T$\downarrow$ atoms and the mixtures of T$\downarrow$ with spin-polarized deuterium D$\downarrow$ and hydrogen H$\downarrow$ atoms. All of our calculations have been performed using the variational and diffusion Monte Carlo methods. For clusters with D$\downarrow$ atoms, the released node procedure is used in cases where the wave function has nodes. In addition to the energy, we have also calculated the structure of small clusters using unbiased estimators. Results obtained for pure T$\downarrow$ clusters are in good accordance with previous calculations, confirming that the trimer is the smallest spin-polarized tritium cluster. Our results show that mixed T$\downarrow$-H$\downarrow$ clusters having up to ten atoms are unstable and that it takes at least three tritium atoms to bind one, two or three D$\downarrow$ atoms. Among all the considered clusters, we have found no other Borromean states except the ground state of the T$\downarrow$ trimer.'
author:
- 'I. Bešli[ć]{}'
- 'L. Vranješ Marki[ć]{}'
- 'J. Boronat'
title: 'Quantum Monte Carlo study of small pure and mixed spin-polarized tritium clusters'
---
Introduction
============
Electron spin-polarized hydrogen (H$\downarrow$) has extreme quantum nature due to its very small mass and weakly attractive potential. It remains in gas phase even in the limit of zero temperature and up to pressures of about 170 bar, [@hydrogen] after which it solidifies. Stwalley and Nosanow [@StwaleyNosanow] suggested in 1976 the use of H$\downarrow$ for achieving a Bose-Einstein condensate (BEC) state, which was finally achieved for hydrogen in 1998 by Fried *et al.*. [@fried] Heavier isotopes of hydrogen, spin-polarized deuterium (D$\downarrow$) and tritium (T$\downarrow$) also show a remarkable quantum behavior. D$\downarrow$ atoms, which obey Fermi statistics, have nuclear spin one and thus the zero-pressure state of bulk D$\downarrow$ depends on the number of occupied nuclear spin states. Previous theoretical estimations [@panoff; @flynn; @skjetne] have shown that (D$\downarrow_1$) with only one occupied nuclear spin state is gas at zero pressure, while bulk D$\downarrow$ with two (D$\downarrow_2$) and three (D$\downarrow_3$) equally occupied nuclear spin states remains liquid at zero pressure and zero temperature. Spin-polarized tritium, which obeys Bose statistics, is expected to be liquid [@etters; @miller; @tritium] due to its larger mass.
Recently, Blume *et al.* [@blume] have studied microscopic properties of tritium (T$\downarrow$)$_N$ clusters with up to $N$=40 atoms using the diffusion Monte Carlo (DMC) method and compared them to bosonic $^4$He$_N$ clusters. In that work, it is shown that tritium clusters (T$\downarrow$)$_N$ are even more weakly bound than helium clusters $^4$He$_N$ with the same number of atoms. Furthermore, their results suggest the use of T$\downarrow$ as a new BEC gas with the advantage of a nearly exact knowledge of its interatomic potential and the possibility of manipulating the strength of the interactions via Feshbach resonances. In addition, it has been shown [@blume] that the trimer (T$\downarrow$)$_3$ is the smallest spin-polarized tritium cluster with a ground-state energy of only $-4.2(7)$ mK. It is thus an example of Borromean or halo state, [@fedorov; @jensen] because the two-body T$\downarrow$ system does not have a bound state. The prediction of a total zero angular momentum bound state of the spin-polarized atomic tritium trimer has been recently confirmed by Salci *et al.* [@salci] using the finite element method. Importantly, the use of this accurate few-body approach leads to conclude that there are not any other bound or shape resonant states with zero and nonzero angular momentum. [@salci]
Unlike the case of pure $^3$He, $^4$He and mixed $^3$He-$^4$He clusters, which have been the subject of both experimental and theoretical study, [@expmix; @review; @fantoni; @mixhel; @ester] nothing is known about the stability of mixed or other pure spin-polarized hydrogen clusters different from tritium ones. $^3$He has almost the same mass as T$\downarrow$, but because of its small mass and Fermi statistics it takes at least 30 $^3$He atoms to form a bound state. [@ester] Recently, Hanna and Blume [@hanna] have studied the near-threshold behavior of weekly bound three-dimensional bosonic clusters with up to 40 atoms interacting additively through two-body van der Waals potentials modeled after the T$\downarrow$-T$\downarrow$ interaction. Among other results, they found a four-parameter fit for the critical mass needed to bind a bosonic cluster of $N$ atoms. From this, it follows that if D$\downarrow$ were a boson, due to its smaller mass compared to T$\downarrow$, it would take at least 13 atoms to form a bound cluster (D$\downarrow)_N$. Considering that D$\downarrow$ obeys Fermi statistics we expect that, similarly to the case of $^3$He, [@ester] it will take more than 30 D$\downarrow$ atoms to form a self-bound cluster. On the other hand, although H$\downarrow$ obeys Bose statistics, its approximately three times larger zero-point energy (with respect to T$\downarrow$) implies that clusters of H$\downarrow$ are not stable. [@hanna] This statement is in agreement with the gas nature of bulk H$\downarrow$ at zero temperature. [@hydrogen] Therefore, hydrogen-tritium and deuterium-tritium clusters are reasonable choices for getting stable self-bound spin-polarized clusters.
If tritium could be experimentally prepared to achieve a BEC state, as suggested in Ref. , the presence of a finite fraction of deuterium would be a new example of a Bose-Fermi mixture with BEC. In fact, Bose-Fermi mixtures composed with alkalines are nowadays studied in several laboratories and the main physics underlying them is well known. [@stefano_rev] Tritium-deuterium mixtures would be even more interesting since they are isotopic and with a very well known interaction. Dilute vapors are also candidates for searching the elusive Efimov states, [@efimov] and recently experimental evidence of its existence has been observed in an ultracold gas of cesium atoms. [@cesium] In the case of pure tritium, it has been proved that such state does not exist [@blume] but mixed isotopic hydrogen systems could offer new possibilities for continuing the search of trimers with Efimov character.
In this article, we report energetic and structural properties of small spin-polarized tritium-deuterium and tritium-hydrogen clusters, obtained using the diffusion Monte Carlo method. It is our main goal to determine which of these clusters are stable. As many of them extend beyond the classically forbidden region they can be considered as quantum halo systems. We are particularly interested in finding among these halo clusters mixed Borromean or "super-Borromean” [@hanna] systems for which all subsystems are unbound. So far, $^3$He$_2$K is the only mixed molecular system for which the Borromean state has been predicted theoretically. [@li]
In Sec. II, we briefly describe the DMC method and the trial wave functions used for importance sampling. Due to the fermionic nature of D$\downarrow$ , the fixed-node and released-node methods are used in cases where the wave function has nodes. Sec. III reports the results obtained, and finally, Sec. IV comprises the summary of the results and main conclusions.
Method
======
We consider that all the H$\downarrow$, D$\downarrow$, and T$\downarrow$ atoms interact with the spin-independent central triplet pair potential $b$$^3\Sigma_u^+$. This interaction has been determined in an essentially exact way by Kolos and Wolniewicz, [@kolos] and recently extended to larger interparticle distances by Jamieson *et al.* (JDW). [@jamieson] We have used a cubic spline interpolation between JDW data and smoothly connected the points to the long-range behavior as calculated by Yan *et al.*. [@yan] The same form has been used in the recent DMC calculation of bulk H$\downarrow$,[@hydrogen] work in which a comparison between different potentials employed in the past is also reported. On the other hand, we have checked that the addition of mass-dependent adiabatic corrections (as calculated by Kolos and Rychlewski [@kolos2]) to the JDW potential does not modify the energy of spin-polarized hydrogen clusters.
The starting point of the DMC method is the Schrödinger equation written in imaginary time, $$-\hbar \frac{\partial \Psi(\bm{R},t)}{\partial t} = (H- E) \Psi(\bm{R},t) \ ,
\label{srodin}$$ where $E$ is a constant acting as a reference energy, $\bm{R} \equiv (\bm{r}_1,\ldots,\bm{r}_N)$ is a *walker* in Monte Carlo terminology, and $H$ is the $N$-particle Hamiltonian $$H = -\sum_{i=1}^{N}\frac{\hbar^2}{2m_i} \bm{\nabla}_i^2 +
\sum_{i<j}^{N} V(r_{ij}) \ .
\label{hamilto}$$
As usual in the method, a trial wave function $\psi(\bm{R})$ is introduced for importance sampling and then the Schrödinger equation (\[srodin\]) is rewritten in terms of $\Phi(\bm{R},t)= \Psi(\bm{R},t)
\psi(\bm{R})$. The resulting equation is solved stochastically by considering a short-time approximation for the Green’s function. In the limit $t\rightarrow\infty$ (for long simulation times), only the lowest energy eigenfunction not orthogonal to $\psi(\mathbf{R})$ survives and then the sampling of the ground state is effectively achieved.
The trial wave function used for the simulation of the pure T$\downarrow$ clusters is of Jastrow form, $\psi_{\text J}(\bm{R}) = \prod_{i<j}^{N}
f_T(r_{ij})$, where $f_T(r_{ij})$ describes correlations between pairs of tritium atoms. Similarly, wave functions for the study of mixed tritium-hydrogen clusters, containing $N_1$ H$\downarrow$ atoms and $N_2$ T$\downarrow$ atoms ($N_1+N_2=N$), are constructed as a product of two-body correlation functions between all the pairs, $$\psi(\mathbf{R};N_1,N_2)=\prod_{\stackrel{{\scriptstyle
i,j=1}}{i<j}}^{N_1}f_H(r_{ij})\prod_{\stackrel{{\scriptstyle i,j=N_1+1}}{i<j}}^{N}f_{T}(r_{ij})
\prod_{i=1}^{N_1}\prod_{j=N_1+1}^{N}f_{HT}(r_{ij}) \ ,$$ where $f_{H}(r)$ describes the two-body correlations between H$\downarrow$ atoms and $f_{HT}(r)$ accounts for the H$\downarrow$-T$\downarrow$ pairs. In case of mixed (D$\downarrow$)$_{N_1}$(T$\downarrow$)$_{N_2}$ clusters, the total wave-function is constructed as a product of an antisymmetric function corresponding to deuterium ($\psi_A$) and Jastrow factors, $$\psi(\mathbf{R};N_1,N_2)=\psi_A
\prod_{\stackrel{{\scriptstyle i,j=1}}{i<j}}^{N_1}f_D(r_{ij})
\prod_{\stackrel{{\scriptstyle i,j=N_1+1}}{i<j}}^{N}
f_{T}(r_{ij})\prod_{i=1}^{N_1}\prod_{j=N_1+1}^{N}f_{DT}(r_{ij}) \ ,$$ where $f_{D}(r)$ describes two-body correlations between D$\downarrow$ atoms and $f_{DT}(r)$ two-body correlations for the D$\downarrow$-T$\downarrow$ pairs. For the function $\psi_A$ it is enough to consider the product of Slater determinants for each one of the nuclear spin states. Consequently, $\psi_A$ has different form in the case of D$\downarrow_2$ and D$\downarrow_3$ mixed D$\downarrow$-T$\downarrow$ clusters. For example, in the case of (D$\downarrow_2)_{N_1}$(T$\downarrow)_{N_2}$ clusters, $\psi_A=1$ for $N_1=1,2$, $\psi_A$=$x_1-x_2$ for $N_1=3$, and $\psi_A$=$(\bf{r}_1-\bf{r}_2)(\bf{r}_3-\bf{r}_4)$ for $N_1$=4, while for (D$\downarrow_3)_{N_1}$($T\downarrow)_{N_2}$ clusters $\psi_A=1$ for $N_1=1,2,3$, $\psi_A=x_1-x_2$ for $N_1$=4 and $\psi_A$=$(\bf{r}_1-\bf{r}_2)(\bf{r}_3-\bf{r}_4)$ for $N_1$=5.
We have worked with three different models for the two-body correlation function $f_T(r)$, $$f_T(r)=\exp [-b_1 \exp(-b_2r) -b_3r] \ ,
\label{trial1}$$ $$f_T(r)=\frac{1}{r} \,
\exp\left[-\left(\frac{\alpha}{r}\right)^{\gamma}-sr\right] \ ,
\label{trial2}$$ $$f_T(r)=\exp\left[-\left(\frac{b}{r}\right)^{5}-sr\right] \ ,
\label{trial3}$$ where $b_1$, $b_2$, $b_3$, $\alpha$, $\gamma$, $s$ and $b$ are variational parameters. They have been obtained by optimizing the variational energy calculated with the VMC method. Finally, we found the form (\[trial2\]) to be optimal for clusters from three to five T$\downarrow$ atoms, while the other two (\[trial1\],\[trial3\]) suit better for larger clusters. The same function, with different variational parameters, has then been used for all the other two-body correlation functions ($f_H(r)$, $f_D(r)$, $f_{HT}(r)$, $f_{DT}(r)$). For example, in case of pure T$\downarrow$ clusters $(\alpha, \gamma, s)$ range from (3.9,4.0,0.0012) for trimer to (4.0,3.7,0.001) for (T$\downarrow)_5$ and $(b,s)$ range from (3.55,0.065) for (T$\downarrow)_6$ to (3.6,0.05) for (T$\downarrow)_{10}$. The VMC recovers from 70% to 90% of the DMC energy, except in case of (T$\downarrow)_3$ where the VMC energy is only -0.85(8) mK. In case of mixed clusters the parameters in the $f_T(r)$ function do not change significantly, so for example in case of the form (\[trial3\]) $b$ assumes values from 3.5 to 3.62 and $s$ from 0.07 to 0.05, going from smaller to larger clusters, respectively. At the same time, in the function $f_{TH}(r)$, $b$ goes from 3.58 to 3.7 and $s$ is around 0.003. Similar behavior of variational parameter $b$ is obtained for the other correlation functions, while $s$ is around 0.01 in $f_{DT}(r)$ and around 0.003 in $f_{D}(r)$. In most cases of mixed clusters VMC recovers from 60% to 90% of the DMC energy, exceptions being smaller clusters with more than two D$\downarrow$ atoms and the clusters having only three T$\downarrow$ atoms, where the VMC energies are further from the DMC ones.
The DMC method we have used is accurate to second order in the time step $\Delta t$. [@boro] The ground-state energies have been calculated for several time-steps and then extrapolated to zero $\Delta t$ to remove any possible time-step bias. We have also studied the optimal mean walker population and finally chosen $N_w=1000$ in order to eliminate any bias coming from it.
Apart from statistical uncertainties, the energy of the bosonic clusters is exactly calculated. In the case of mixed clusters with more than three D$\downarrow_3$ or two D$\downarrow_2$ atoms, the sign problem appears because the trial wave function changes sign. Since the Monte Carlo method requires that $f(\mathbf{R},t)=\Psi(\mathbf{R},t) \psi(\mathbf{R})\geq 0$, we have first used the so-called fixed-node approximation which allows only the moves in which $\psi$ and $\Psi$ change sign together, thus fixing the nodes. In this way, an upper bound to the energy is obtained. [@boundfn] In a subsequent step, we have removed the nodal constraint imposed by the trial wave function by using the released-node method. We have adopted the methodology used previously in liquid helium calculations, [@boro2] and recently employed in the study of small mixed helium clusters. [@mixhel] In this approach, walkers are allowed to cross the nodes imposed by the trial wave function and survive for a finite lifetime. In order to achieve an effective crossing of the nodal surface an auxiliary guiding wave function $\psi_{g}$ is introduced in the DMC calculation. This function is positive everywhere, different from zero in the nodes, and approaches $|\psi|$ away from the nodes. Like in Refs. , we have taken $\psi_{g}=\psi_{J}(\bf{R})(\psi_{A}^2(\bf{R})+ \textit{a}^2)^{1/2}$, which satisfies the above condition for suitable choices of $a$. The released-node energy is estimated through an exponential fit $E(t)=E_{r} +
A e^{-(t/\tau)}$ to the DMC data, with $t$ the released time. In all cases where the RN method is used in this work the difference between the last calculated point in released time and $E_{r}$ is of the same order as the statistical noise. The behavior of $a$ and the maximum released time needed to achieve the asymptotic limit of the released energy has been similar to the case studied in Ref. . Also, like in Ref. , we find the RN energies to be very close to the FN ones. In addition, in the case of $(T\downarrow)_4(D\downarrow_2)_3$ we have verified that the inclusion of backflow correlations in $\psi_A$ does not change the final FN results.
Results
=======
DMC results for the ground state energy of pure and mixed T$\downarrow$-H$\downarrow$ clusters are given in Table \[tab:energiesTH\]. Our results for pure tritium T$\downarrow$ clusters are compared to the results of Blume *et. al.* [@blume] in Fig. \[fig:blume\]. We obtain a good agreement with these published data, although our results are slightly lower for all $N$. Underlying this difference is the fact that in Ref. a damped three-body Axilrod-Teller potential term [@ATterm] is introduced in the Hamiltonian, raising the trimer energy 1.6% and the energy of the cluster with 40 particles 6%. We also find that the trimer is the smallest spin-polarized tritium cluster with an energy of only $-4.8(2)$ mK, confirming it to be a halo state. In order to test the sensitivity of the present results to the details in the interaction potential, we have also calculated the energy with the potential that Silvera and Goldman [@silvera] constructed as a fit to the older Kolos and Wolniewicz data. [@kolos] It is worth mentioning that a full comparison between the different potentials used in the literature for studying bulk H$\downarrow$ can be found in the Ref. . Using the Silvera model, we obtain a trimer binding energy significantly smaller, $-1.9(4)$ mK. The trimer energy is therefore sizably affected by the interatomic potential and the reason for that lies on the huge cancellation between the kinetic and potential energies. For instance, using the JDW potential and the pure estimator for the potential energy, we obtain $E_p=-355(9)$ mK and $E_k=350(9)$ mK. For larger clusters, the influence of the potential form on the binding strength is reduced: for the tetramer and with the Silvera potential, we obtain an energy $E =-107(2)$ mK which is only about 15% weaker than the one for the JDW potential, $E=-126(2)$ mK. The potential and kinetic energies of the tetramer using the JDW potential are $E_p=-1871(11)$ mK and $E_k=1745(12)$ mK.
The addition of one H$\downarrow$ atom to the (T$\downarrow)_N$ cluster creates a system that appears to be at the threshold of stability or unstable, that is the ground-state energy of the (T$\downarrow)_N$H$\downarrow$ cluster is within the errorbars equal to or higher than the energy of the (T$\downarrow)_N$ cluster, respectively. In order to be certain about the conclusion concerning the stability of the (T$\downarrow)_N$H$\downarrow$ clusters, we have repeated the calculations using two different types of trial wave functions. In the case where VMC energies are further away from the DMC ones, a lot of simulation time is needed for the system to reach the equilibrium state, and in some cases with the trial function (\[trial2\]) the lowering of the energy with imaginary time going on is almost imperceptible. On the other hand, with both (\[trial1\]) and (\[trial3\]), and for small values of the parameter $s$ or $b_3$ in the $f(r_{TH})$ function, the system rather quickly reaches the equilibrium energy, which is within the errorbars equal to the energy of the system without the H$\downarrow$ atom. Furthermore, if we continue the simulation after the system has reached the ground-state energy, we can observe the surplus H$\downarrow$ atom leaving the cluster. This indicates that these clusters are effectively unstable. We have also calculated the ground-state energies of several clusters with more hydrogen atoms (T$\downarrow)_N$(H$\downarrow)_M$, such that $N+M \le 10$. Our results indicate that these clusters are also unstable, because although in some cases negative energies are obtained they are above those for (T$\downarrow)_N$H$\downarrow$.
Table \[tab:energiesTD\] presents the results for the ground-state energy of mixed T$\downarrow$-D$\downarrow$ clusters. Due to the approximately twice larger mass of D$\downarrow$ compared to H$\downarrow$, clusters with one or two D$\downarrow$ atoms and at least three T$\downarrow$ atoms are stable. In the clusters (T$\downarrow)_N$(D$\downarrow)_2$, we have assumed that D$\downarrow$ atoms occupy two different nuclear spin states. For clusters (T$\downarrow)_N$(D$\downarrow)_M$ with $M>$2, we have considered both the case where two different nuclear spin states are occupied (T$\downarrow)_N$(D$\downarrow_2)_M$ and the case with three different occupied nuclear spin states (T$\downarrow)_N$(D$\downarrow_3)_M$. Similar to the case of the bulk system, [@panoff; @flynn; @skjetne] we find that the three-component spin clusters are more strongly bound than the ones where D$\downarrow$ occupies two nuclear spin states, $E(($T$\downarrow)_N($D$\downarrow_3)_M) <
E(($T$\downarrow)_N($D$\downarrow_2)_M)$. Furthermore, (T$\downarrow)_N$(D$\downarrow_3)_3$, whose spatial wave function has no nodes, are all stable for $N \ge 3$, while it takes 6 T$\downarrow$ atoms to form a stable (T$\downarrow)_N$(D$\downarrow_2)_3$ cluster. The smaller (T$\downarrow)_N$(D$\downarrow_2)_3$ clusters have the same energy as the (T$\downarrow)_N$(D$\downarrow)_2$ clusters within the errorbars, but the analysis of the distributions shows one of the D$\downarrow$ atoms leaving the system. The clusters (T$\downarrow)_N$(D$\downarrow_3)_4$ are similarly at the threshold of binding for $N=3,4,5,6$, just like the clusters (T$\downarrow)_N$(D$\downarrow_2)_4$, and again it is necessary to consider the structure to better determine their stability. Our analysis indicates that only (T$\downarrow)_6$(D$\downarrow_3)_4$ is stable. The same behavior, as far as the energy is concerned, is reproduced for (T$\downarrow)_N$(D$\downarrow_3)_5$ with $N\le 5$: as the separation of the D$\downarrow$ atoms is growing along the simulation we are led to consider them unstable. We expect that all other clusters (T$\downarrow)_N$(D$\downarrow)_M$, with $N+M\le 10$ and $M>5$ are unstable because the exchange of T$\downarrow$ with the D$\downarrow$ atoms does not change the interaction but raises the kinetic energy. The evolution of the energy of mixed clusters formed by $N$ tritium atoms and one hydrogen atom or up to four deuterium atoms is shown in Fig. \[fig:allen\].
In addition to the energy, DMC simulations allow also for exact estimations of other relevant magnitudes such as the distribution of interparticle distances $P(r)$ or the distribution of particles with respect to the center of mass of the cluster $\rho(r)$. In both cases it is possible to eliminate the bias coming from the trial wave function by using pure estimators [@pures] and arrive to exact results.
Our calculations confirm that the pure T$\downarrow$ clusters are spatially very diffuse. [@blume] From the DMC results for $P(r)$ one can see that the average separation between particles $\langle r_{TT} \rangle$ ranges between 34 Å in the case of (T$\downarrow)_3$ to 10.9 Å for (T$\downarrow)_{6}$ and 10.5 Å for (T$\downarrow)_{10}$. The addition of one D$\downarrow$ or one H$\downarrow$ reduces the average separation between T$\downarrow$ atoms in the case of clusters with $N=3$ or 4, as can be seen in Fig. \[fig:pair3T1D1H\], which compares the pair distribution functions of (T$\downarrow)_3$, (T$\downarrow)_3$D$\downarrow$ and (T$\downarrow)_3$H$\downarrow$. In contrast, for clusters with more T$\downarrow$ atoms the T$\downarrow$-T$\downarrow$ separation remains almost unaffected by the addition of one D$\downarrow$ or one H$\downarrow$ atom. Due to the larger zero-point motion of D$\downarrow$ and H$\downarrow$ atoms, $\langle r_{TT} \rangle ~>~ \langle r_{TD} \rangle
~>~ \langle r_{TH} \rangle$. Moreover, we obtain $\langle r_{TH} \rangle$ greater than 100 Å indicating that the cluster (T$\downarrow)_3$H$\downarrow$ is unstable. Similar behavior is noticed for larger (T$\downarrow)_N$H$\downarrow$ clusters.
The effect of adding D$\downarrow$ atoms to the core of (T$\downarrow)_6$ cluster is shown in Fig. \[fig:pair7TnD\]. The separation between pairs of T$\downarrow$ atoms remains almost unchanged, while the T$\downarrow$-D$\downarrow$ and D$\downarrow$-D$\downarrow$ separations grow with the addition of D$\downarrow$ atoms. There is a noticeable difference between clusters with D$\downarrow_2$ and D$\downarrow_3$ spin-polarized deuterium. Former clusters with more than three particles are much more extended, i.e., the average separation between D$\downarrow$ atoms is larger and the tail of the distribution decays slowly. Despite the large size of the cluster, the average separation between particles does not appear to grow in the case of (T$\downarrow)_6$(D$\downarrow_2)_3$ so we are led to consider this cluster as stable; the same conclusion applies to (T$\downarrow)_6$(D$\downarrow_3)_{4}$. On the other hand, the average D$\downarrow$-D$\downarrow$ distance in the (T$\downarrow)_6$(D$\downarrow_2)_{4}$ cluster is slightly above 40 Å but grows very slowly in the course of the simulation, indicating that it could be unstable.
Finally, Fig. \[fig:TnD\] presents the density distribution of the clusters (T$\downarrow)_N$(D$\downarrow)$, for $N=$3,6,9. With the increase in the number of bosons, deuterium is pushed to the surface of the cluster, although even for $N=9$ there is still an appreciable probability of finding it inside the cluster. Similar behavior is found for (T$\downarrow)_N$(H$\downarrow)$ clusters. For larger clusters, one can expect that both D$\downarrow$ and H$\downarrow$ will be pushed to the surface forming the so-called Andreev states, [@Andreev] similarly to the well known behavior of $^3$He in mixed $^3$He-$^4$He clusters. [@review]
conclusions
===========
The ground-state properties of spin-polarized pure and mixed tritium clusters have been accurately determined using the DMC method. Our results show that the trimer is a Borromean or halo state in agreement with the results of Ref. . The most promising candidates for the super-Borromean states in mixed clusters have been (T$\downarrow)_2$(D$\downarrow)_{1,2,3,4,5}$, but after a careful analysis we have found no bound states in either of these clusters. On the other hand, we conclude that it takes at least three T$\downarrow$ atoms in order to bind 1-3 D$\downarrow$ atoms, the case of (T$\downarrow)_3$(D$\downarrow)_3$ being bound only when the deuterium atoms are of D$\downarrow_3$ type. The DMC results show that 6 T$\downarrow$ atoms are needed in order to bind 4 D$\downarrow_3$ atoms, while clusters with up to total 10 atoms and more than 4 D$\downarrow$ atoms seem all to be unstable. We have not considered mixed clusters with the D$\downarrow$ atoms in only one nuclear spin-state D$\downarrow_1$ since, from analogy with the bulk, we expect them to be less bound than D$\downarrow_2$ and D$\downarrow_3$.
Finally, and concerning mixed tritium-hydrogen clusters, our results show that all the (T$\downarrow)_N$(H$\downarrow)_M$ clusters with $N+M \le 11$ are unstable. Work is in progress to find out how many T$\downarrow$ atoms are needed to bind one H$\downarrow$ atom.
J. B. acknowledges support from DGI (Spain) Grant No. FIS2005-04181 and Generalitat de Catalunya Grant No. 2005SGR-00779. I.B and L.V.M. acknowledge support from MSES (Croatia) Grant No. 177-1770508-0493. We also acknowledge the support of the Central Computing Services at the Johannes Kepler University in Linz, where part of the computations was performed.
[21]{}
L. Vranješ Markić, J. Boronat and J. Casulleras, Phys. Rev. B **75**, 064506 (2007).
W. C. Stwaley and L. H. Nosanow, Phys. Rev. Lett. **36**, 910 (1976).
D. G. Fried, T. C. Killian, L. Wilmann, D. Landhuis, S. C. Moss, P. Kleppner, and T. J. Greytak, Phys. Rev. Lett. **81**, 3811 (1998).
R. M. Panoff and J. W. Clark, Phys. Rev. B **36** 5527 (1987).
M. F. Flynn, J. W. Clark, E. Krotscheck, R. A. Smith, and R. M. Panoff, Phys. Rev. B **32**, 2945 (1985).
B. Skjetne and E. Østgaard, J. Phys.: Condens. Matter **11** 8017 (1999).
R. D. Etters, J. V. Dugan, Jr., and R. W. Palmer, J. Chem. Phys. **62**, 313 (1975).
M. D. Miller, L. H. Nosanow, Phys. Rev. B **15**, 4376 (1976).
B. R. Joudeh, M. K. Al-Sugheir, H. B. Ghassib, Physica B **388**, 237 (2007).
D. Blume, B. D. Esry, Chris H. Greene, N. N. Klausen, and G. J. Hanna, Phys. Rev. Lett. **89**, 163402 (2002).
D. V. Fedorov, A. S. Jensen and K. Riisager, Phys. Rev. C **49**, 201 (1994); **50**, 2372 (1994).
A. S. Jensen, K. Riisager, D. V. Fedorov and E. Garrido, Rev. Mod. Phys. **76**, 215 (2004).
M. Salci, Sergey B. Levin and Nils Elander, Phys. Rev. A **69**, 044501(2004).
A. Kalinin, O. Kornilov, W. Schölkopf and J. P. Toennies, Phys. Rev. Lett. **95**, 113402 (2005).
M. Barranco, R. Guardiola, S. Hernández, R. Mayol, and M. Pi, J. Low Temp. Phys. **142**,1 (2006), and references therein.
S. Fantoni, R. Guardiola, J. Navarro, A. Zuker, J. Chem. Phys. **123**, 054503 (2005).
I. Bešlić, L. Vranješ Markić, S. Kilić, J. Low Temp. Phys. **143**, 257 (2006).
E. Sola, J. Casulleras and J. Boronat, Phys. Rev. B **73**, 092515 (2006).
G. J. Hanna and D. Blume, Phys. Rev. A **74**, 063604 (2006).
See, e.g., S. Giorgini, L. P. Pitaevskii, and S. Stringari, preprint, cond-mat/0706.3360.
V. Efimov, Phys. Lett. **33B**, 563 (1970).
T. Kraemer, M. Mark, P. Waldburger, J. G. Danzl, C. Chin, B. Engeser, A. D. Lange, K. Pilch, A. Jaakkola, H.-C. Nägerl, and R. Grimm, Nature (London) **440**, 315 (2006).
Yong Li, Qungdong Gou and Tingyun Shi, Phys. Rev. A **74**, 032502 (2006).
W. Kolos and L. Wolniewicz, J. Chem. Phys. **43**, 2429 (1965); Chem. Phys. Lett. **24**, 457 (1974).
M. J. Jamieson, A. Dalgarno, and L. Wolniewicz, Phys. Rev. A **61**, 042705 (2000).
Zong-Chao Yan, James F. Babb, A. Dalgarno, and G. W. F. Drake, Phys. Rev A **54**, 2824(1996).
W. Kolos and J. Rychlewski, J. Mol. Spectrosc. **143**, 237 (1990).
J. Boronat and J. Casulleras, Phys. Rev. B **49** 8920 (1994).
P. J. Reynolds, D. M. Ceperley, B. J. Alder, and W. A. Lester Jr., J. Chem. Phys. **77**, 5593 (1982).
J. Boronat and J. Casulleras, Europhys. Lett. **38**, 291 (1997).
T. I. Sachse, K.T. Tang, and J. P. Toennies, Chem. Phys. Lett. **317**, 346 (2000).
I. F. Silvera and V.V. Goldman, Phys. Rev. Lett. **45** 915 (1980).
J. Casulleras and J. Boronat, Phys. Rev. B **52**, 3654 (1995).
A. F. Andreev, Sov. Phys. JETP **23**, 939 (1966).
------------------------------------------------------------------------
N (T$\downarrow)_N$ (T$\downarrow)_N$H$\downarrow$
------- ------------------------- --------------------------------------
2 - -
3 -4.8 (0.2) *[-4.7(0.7)]{}\
4 &-126(2) & *[-126(1)]{}\
5 & -398(1) & *[-398(2)]{}\
6 & -810(2) & *[-807(3)]{}\
7 & -1348(4) &*[-1339(6)]{}\
8 & -1991(5) & *[-1982(7)]{}\
9 & -2727(7) & *[-2720(9)]{}\
10 &-3553(8) &*******
------------------------------------------------------------------------
: The ground state energy (in mK) of pure spin-polarized tritium clusters with $N$ atoms and tritium clusters with an additional hydrogen atom. Clusters which appear to be unstable are written in italic. Figures in parenthesis are the statistical errors.[]{data-label="tab:energiesTH"}
[c|ccccccc]{}\
N& D$\downarrow$ & (D$\downarrow)_2$ & (D$\downarrow_2)_3$ & (D$\downarrow_2)_4$ & (D$\downarrow_3)_3$ & (D$\downarrow_3)_4$ & (D$\downarrow_3)_5$\
2 & - & - & - & -& - & -& -\
3 & -12.2(0.9) & -31(3)& *[-27(3)]{}&*[-21(2)]{}&-56(2) & *[-58(5)]{} & *[-46(6)]{}\
4 & -182 (3)& -256(3)& *[-253(4)]{}&*[-241(3)]{}& -336(3) &*[-327(6)]{} & *[-313(6)]{}\
5 & -510 (4)&-629(4) &*[-633(6)]{} &*[-610(6)]{}&-769(4) &*[-755(7)]{} & *[-732(6)]{}\
6 &-972(4) & -1139(3)& *[-1145(9)]{}&*[-1135(5)]{}& -1322(5)&*[-1326(10)]{} &\
7 &-1552(4) & -1763(10 )&*[-1747(15)]{}& &-1989(7)& &\
8 & -2237(6)&-2492(12) & -2554(15)&& -2755(8) & &\
9 & -3013(7)& & & & &****************
**Figure captions**\
FIG. 1: Comparison of our results for spin-polarized tritium clusters (circles) with the results of Blume *et. al.* [@blume] (crosses). The difference between our results and the ones from Blume *et. al.* [@blume] is mainly due to the use of a slightly different interaction (see text). The error bars of the DMC energies are smaller than the size of the symbols.
FIG. 2: Energies of the spin-polarized pure and mixed tritium clusters in K as a function of the number of T$\downarrow$ atoms.
FIG. 3: Distribution of interparticle distances in the (T$\downarrow)_3$, (T$\downarrow)_3$D$\downarrow$ and (T$\downarrow)_3$H$\downarrow$ clusters.
FIG. 4: Distribution of interparticle distances of the (T$\downarrow)_6$(D$\downarrow)_N$ clusters. Full line denotes T$\downarrow$-T$\downarrow$, dashed line T$\downarrow$-D$\downarrow$, and dotted line D$\downarrow$-D$\downarrow$ separation. $N=3, 4$ distributions with lower maxima in D$\downarrow$-D$\downarrow$ separations correspond to D$\downarrow_2$ and those with higher maxima to D$\downarrow_3$.
FIG. 5: Density distributions of T$\downarrow$ (upper curves) and D$\downarrow$ (lower curves) for (T$\downarrow)_N$D$\downarrow$ clusters.
![I. Bešli[ć]{} [*et al.*]{} []{data-label="fig:blume"}](Tritium.eps){width="8.5cm"}
![I. Bešli[ć]{} [*et al.*]{}[]{data-label="fig:allen"}](Allen.eps){width="8.5cm"}
![ I. Bešli[ć]{} [*et al.*]{} []{data-label="fig:pair3T1D1H"}](pair3T1D1H.eps){width="8.5cm"}
![ I. Bešli[ć]{} [*et al.*]{}[]{data-label="fig:pair7TnD"}](pair6TnD.eps){width="8.5cm"}
![I. Bešli[ć]{} [*et al.*]{} []{data-label="fig:TnD"}](rhoTnDnew.eps){width="8.5cm"}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The notion of soft sets is introduced as a general mathematical tool for dealing with uncertainty. In this paper, we consider the concepts of soft compactness, countably soft compactness and obtain some results. We study some soft separation axioms that have been studied by Min and Shabir-Naz. By constructing a special soft topological space, show that some classical results in general topology are not true about soft topological spaces, for instance every compact Housdorff spaces need not be normal.\
\
[**[Keywords]{}**]{}: Soft closed, Soft compact space, Soft open, Soft topological spaces.[^1]
author:
- 'E. Peyghan , B. Samadi and A. Tayebi'
title: Some Results Related to Soft Topological Spaces
---
Introduction
============
During recent years General Topology was developed by many mathematicians. The theory of generalized topological spaces, which was founded by Á. Császár is one of these developments [@CS]. Recently, in [@SN] Shabir-Naz introduced and studied the concepts of soft topological spaces and some related concepts. The generalized topology is different from topology by its axioms ( A collection of subsets of X is a generalized topology on X if and only if it contains empty set and arbitrary union of its elements). But the soft topology is based on soft sets theory and not sets.
Some notions in Mathematics can be considered as mathematical tools for dealing with uncertainties, namely theory of fuzzy sets, theory of intuitionistic fuzzy sets, theory of vague sets, theory of rough sets and etc. But all of these theories have their own difficulties. In [@M], Molodtsov introduced the concept of a soft set in order to solve complicated problems in the economics, engineering, and environmental areas because no mathematical tools can successfully deal with the various kinds of uncertainties in these problems. He successfully applied the soft theory in several directions, such as game theory, probability, Perron integration, Riemann integration and theory of measurement [@M; @MLK].
In [@MBR], Maji-Biswas-Roy defined and studied operations of soft sets. Then Pei-Miao [@PM] and Chen [@C] improved the work of Maji-Biswas-Roy [@MBR0; @MBR]. The properties and applications of soft set theory have been studied increasingly in [@Ali]. In [@CE], Çağman-Enginoglu redefined the operations of the soft sets and constructed a uniint decision making method by using these new operations, and developed soft set theory. Then to make easy compaction with the operations of soft sets, they presented the soft matrix theory and set up the soft max–min decision making method [@CE2]. These decision making methods can be successfully applied to many problems that contain uncertainties. In [@SN], the authors studied some concepts related to soft spaces such as soft interior, soft subspace and soft separation axioms. Recently, Aygunoglu-Aygun introduced the soft product topology and defined the version of compactness in soft spaces named soft compactness [@AA].
In this paper, we consider the concepts of soft compactness and countably soft compact and get some results. Then, we study some soft separation axioms that have studied by Min and Shabir-Naz. By constructing some examples we show that some classical results in general topology are not true about soft topological spaces, for instance every compact Housdorff spaces need not be normal.
Preliminaries
=============
In this section, we recall some definitions and concepts discussed in [@HA; @WKM; @SN; @ZAMA]. Let $U$ be an initial universe and $E$ be a set of parameters. Let $\mathbb{P}(U)$ denotes the power set of $U$ and $A$ be a nonempty subset of $E$. A pair $(F, A)$ is called a [*soft set*]{} over $U$, where $F$ is a mapping given by $F: A\rightarrow\mathbb{P}(U)$. For two soft sets $(F, A)$ and $(G, B)$ over common universe $U$, we say that $(F, A)$ is a [*soft subset*]{} $(G, B)$ if $A\subseteq B$ and $F(e)\subseteq G(e)$, for all $e\in A$. In this case, we write $(F,
A)\widetilde{\subseteq}(G, B)$ and $(G, B)$ is said to be a [*soft super set*]{} of $(F, A)$. Two soft sets $(F, A)$ and $(G, B)$ over a common universe $U$ are said to be [*soft equal*]{} if $(F,
A)\widetilde{\subseteq}(G, B)$ and $(G, B)\widetilde{\subseteq}(F,
A)$. A soft set $(F, A)$ over $U$ is called a [*null soft set*]{}, denoted by $\Phi_A$, if for each $e\in A$, $F(e)=\emptyset$. Similarly, it is called [*absolute soft set*]{}, denoted by $\widetilde{U}$, if for each $e\in A$, $F(e)=U$.
The [*union*]{} of two soft sets $(F, A)$ and $(G, B)$ over the common universe $U$ is the soft set $(H, C)$, where $C=A\cup B$ and for each $e\in
C$, $$H(e)=\left\{
\begin{array}{ccc}
F(e)&e\in A-B\\
G(e)&e\in B-A\\
F(e)\cup G(e)&e\in A\cap B
\end{array}
\right.$$ We write $(F, A)\cup(G, B)=(H, C)$. Moreover, the [*intersection*]{} $(H, C)$ of two soft sets $(F, A)$ and $(G,
B)$ over a common universe $U$, denoted by $(F, A)\cap(G, B)$, is defined as $C=A\cap B$ and $H(e)=F(e)\cap G(e)$ for each $e\in C$. The [*difference*]{} $(H, E)$ of two soft sets $(F, E)$ and $(G, E)$ over $X$, denoted by $(F, E)\backslash(G, E)$, is defined as $H(e)=F(e)\backslash G(e)$, for each $e\in E$. Let $Y$ be a nonempty subset of $X$. Then $\widetilde{Y}$ denotes the soft set $(Y, E)$ over $X$ where $Y(e)=Y$, for each $e\in
E$. In particular, $(X, E)$ will be denoted by $\widetilde{X}$. Let $(F, E)$ be a soft set over $X$ and $x\in X$. We say that $x\in(F, E)$, whenever $x\in F(e)$, for each $e\in E$ [@PST].
The relative complement of a soft set $(F, A)$ is denoted by $(F,
A)'$ and is defined by $(F, A)'=(F', A)$ where $F: A\rightarrow
\mathbb{P}(U)$ is defined by following $$F'(e)=U-F(e), \ \ \ \forall e\in A.$$
Let $\tau$ be the collection of soft sets over $X$. Then $\tau$ is called a soft topology on $X$ if $\tau$ satisfies the following axioms:
(i) $\Phi_E$, $\widetilde{X}$ belong to $\tau$.
(ii) The union of any number of soft sets in $\tau$ belongs to $\tau$.
(iii) The intersection of any two soft sets in $\tau$ belongs to $\tau$.\
The triple $(X, \tau, E)$ is called a soft topological space over $X$. The members of $\tau$ are said to be soft open in $X$, and the soft set $(F, E)$ is called soft closed in $X$ if its relative component $(F, E)'$ belongs to $\tau$.
The proof of the following proposition is an easy application of De Morgan’s lows with the definition of a soft topology on $X$ (see Proposition 3.3 of [@ZAMA]).
\[prop21\] *Let $(X, \tau, E)$ be a soft space over $X$. Then*
1)
: *$\Phi_E$, $\widetilde{X}$ are closed soft set over $X$;*
2)
: *The intersection of any number of soft closed sets is a soft closed set over $X$;*
3)
: *The union of any two soft closed sets is a soft closed set over $X$.*
Soft Compactness
================
In this section, we are going to introduce the concept of soft compactness about soft topological spaces and study some properties related to these spaces (also, see [@ZAMA]).
A family $\mathcal{A}=\{(F_\alpha, E)\}_{\alpha\in J}$ of soft sets is a cover of a soft set $ (F,E) $ if $$(F, E)\widetilde{\subseteq}\bigcup_{\alpha\in J}(F_\alpha, E).$$ It is a soft open cover if each member of $\mathcal{A}$ is a soft open set. A subcover of $\mathcal{A}$ is a subfamily of $\mathcal{A}$ which is also a cover. A soft topological space $(X, \tau, E)$ is said to be soft compact if each soft open cover of $(X, E)$ has a finite subcover.
Let $(X, \tau_1, E)$ and $(X, \tau_2, E)$ be soft topological spaces. If $\tau_1\subseteq\tau_2$, then $\tau_2$ is soft finer than $\tau_1$. If $\tau_1\subseteq\tau_2$ or $\tau_2\subseteq\tau_1$, then $\tau_1$ is soft comparable with $\tau_2$. Then, we have the following.
Let $(X, \tau_2, E)$ be a soft compact space and $\tau_1\subseteq\tau_2$. Then $(X, \tau_1, E)$ is soft compact.
Let $\{(F_\alpha, E)\}_{\alpha\in J}$ be a soft open cover of $\widetilde{X}$ by soft open sets of $(X, \tau_1, E)$. Since $\tau_1\subseteq\tau_2$, then $\{(F_\alpha, E)\}_{\alpha\in J}$ is a soft open cover of $\widetilde{X}$ by soft open sets of $(X, \tau_2, E)$. But $(X, \tau_2, E)$ is soft compact. Therefore $$(X, E)\widetilde{\subseteq}(F_{\alpha_1}, E)\cup\ldots\cup(F_{\alpha_n}, E),$$ for some $\alpha_1,\ldots,\alpha_n\in J$. Hence $(X, \tau_1, E)$ is soft compact.
In this paper, for convenience, let $SS(X)_E$ be the family of soft sets over $X$ with set of parameters $E$. We will apply two next propositions so much in the proofs.
\[3.2\] Let $(F, E)$, $(G, E)$, $(H, E)$ and $(I, E)$ be soft sets in $SS(X)_E$. Then the following hold.
(i)
: $(F, E)\widetilde{\subseteq}(G, E)$ if and only if $(F, E)\cap(G, E)=(F, E)$;
(ii)
: $(F, E)\widetilde{\subseteq}(G, E), (H, E)$ if and only if $(F, E)\widetilde{\subseteq}(G, E)\cap(H, E)$;
(iii)
: If $(F, E)\widetilde{\subseteq}(H, E)$ and $(G, E)\widetilde{\subseteq}(I, E)$, then $(F, E)\cup(G, E)\widetilde{\subseteq}(H, E)\cup(I, E)$;
(iv)
: $(F, E)\cap(F, E)'=\Phi_E$;
(v)
: $(F, E)\cap(G, E)=\Phi_E$ if and only if $(F, E)\widetilde{\subseteq}(G, E)'$;
(vi)
: $(F,E)\widetilde{\subseteq}(G,E)$ if and only if $(G,E)'\widetilde{\subseteq}(F,E)'$.
Here, we only prove the (iii). Let $ (F,E)\cup (G,E)=(J,E)$ and $(H,E)\cup (I,E)=(K,E)$. Since $(F,E)\widetilde{\subseteq}(H,E)$ and $(G,E)\widetilde{\subseteq}(I,E) $; then $$F(e){\subseteq}H(e) \ \ and \ \ G(e){\subseteq}I(e), \ \ \forall e\in E.$$ Therefore $$J(e)=F(e)\cup G(e)\subseteq H(e)\cup I(e)=K(e).$$ Hence $(J,E)\widetilde{\subseteq}(K,E)$.
Also we can obtain the following easily.
\[3.3\] Let $(F,E)$ be a soft set and $\{(F_{\alpha},E)\}_{\alpha\in J} $ be a family of soft sets in $SS(X)_E$. Then the following hold.
(i)
: $(F, E)\cap(F, E)'=\Phi_E$;
(ii)
: $(F, E)\cup\Phi_E=(F, E)$;
(iii)
: $(F, E)\cap (\cup_{\alpha\in J}(F_{\alpha},E))=\cup_{\alpha\in J}((F,E)\cap(F_{\alpha},E))$;
(iv)
: $\Phi_E'=\widetilde{X}$;
(v)
: $\widetilde{X}'=\Phi_E$.
Let $(F,E)$ be a soft set over $X$ and $Y$ be a nonempty subset of $X$. Then the sub-soft set of $(F,E)$ over $Y$ denoted by $(^YF,E)$ is defined as follows $$^YF(e)=Y\cap F(e),$$ for each $e\in E$. In other word $(^YF, E)=\widetilde{Y}\cap(F, E)$. Now, suppose that $(X, \tau, E)$ be a soft topological space over $X$ and $Y$ be a nonempty subset of $X$. Then $$\tau_Y=\{(^YF, E)|(F, E)\in\tau\},$$ is said to be soft relative topology on $Y$ and $(Y, \tau_Y, E)$ is called a soft subspace of $(X, \tau, E)$. Here, we exhibit a criterion that applies $\widetilde{Y}$ is soft compact by soft open covers of $\widetilde{Y}$, that all of members are soft open sets in $X$.
\[babi\] Let $(Y,\tau_Y, E)$ be a soft subspace of a soft space $(X,\tau,E)$. Then $(Y,\tau_Y, E)$ is soft compact if and only if every cover of $\widetilde{Y}$ by soft open sets in $X$ contains a finite subcover.
Let $(Y,\tau_Y, E)$ be soft compact and $\{(F_\alpha,E)\}_{\alpha\in J}$ be a cover of $\widetilde{Y}$ by soft open sets in $X$. By Propositions \[3.2\] and \[3.3\], we can see that $\{^YF_\alpha,E\}_{\alpha\in J}$ is a soft open cover of $\widetilde{Y}$. Therefore $$(Y,E)\widetilde{\subseteq}(^YF_{\alpha_1},E)\cup\ldots\cup (^YF_{\alpha_n},E),$$ for some $\alpha_1,\ldots,\alpha_n\in J$. This implies that $\{(F_{\alpha_i},E)\}_{i=1}^n$ is a subcover of $\widetilde{Y}$ by soft open sets in $X$. Conversely, let $\{(^YF_\alpha,E)\}_{\alpha\in J}$ be a soft open cover of $\widetilde{Y}$. It is easy to see that $\{(F_\alpha,E)\}_{\alpha\in J}$ is a cover of $\widetilde{Y}$ by soft open sets in $X$. Then we can write $$\widetilde{Y}\widetilde{\subseteq}(F_{\alpha_1},E)\cup,\ldots,\cup(F_{\alpha_n},E),$$ for some $\alpha_1,\ldots,\alpha_n\in J$. Therefore $\{(^YF_{\alpha_i},E)\}_{i=1}^n$ is a subcover of $\widetilde{Y}$. Hence $(Y, \tau_Y, E)$ is soft compact.
Every soft compact subspace of a soft Hausdorff space is soft closed.
Let $(Y, \tau_Y, E)$ be a soft compact subspace of soft Hausdorff space $(X, \tau, E)$. Let $x\in(X, E)-(Y, E)$. Then for all $y\in(Y, E)$, $x\neq y$. Therefore, there exist soft open sets $(U_y, E)$ and $(U_{xy}, E)$ containing $x$ and $y$, respectively such that $(U_y, E)\cap(U_{xy}, E)=\Phi_E$. Obviously, $\{(U_{xy}, E)\}_{y\in Y}$ is a cover of $\widetilde{Y}$ by soft open sets in $X$. By Theorem \[babi\], we have $(Y, E)=(U_{xy_1}, E)\cup\ldots\cup(U_{xy_n}, E)$ for some $y_1,\ldots,y_n\in Y$. Now, $x\in(U_{y_1}, E)\cap\ldots\cap(U_{y_n}, E)=(U_x, E)$ and Proposition \[3.3\] implies that $(U_x, E)\cap(Y, E)=\Phi_E$. Hence $x\in(U_x, E)\subseteq(X, E)-(Y, E)$. Then $(X, E)-(Y, E)=\bigcup_{x\in X-Y}(U_x, E)$. Therefore $(X, E)-(Y, E)$ is soft open. Hence $(Y, E)$ is soft closed.
Using Propositions \[3.2\] and \[3.3\], we are going to prove that every soft closed subspace of a soft compact space is soft compact.
Every soft closed subset of a soft compact space is soft compact.
Let $(Y,\tau_Y, E)$ be a soft subspace of a soft compact space $(X, \tau, E)$ such that $(Y,E)$ is a soft closed in $X$. Let $\{(F_\alpha,E)\}_{\alpha\in J}$ be a cover of $\widetilde{Y}$ by soft open sets in $X$. $(Y,E)'$ is a soft open set in $X$. Propositions \[3.2\] and \[3.3\] show that $\{(F_\alpha, E)\}_{\alpha\in J}\cup\{(Y', E)\}$ form a soft open cover of $\widetilde{X}$. Therefore $$(X, E)\widetilde{\subseteq}(F_{\alpha_1},E)\cup\ldots\cup(F_{\alpha_n,E})\cup(Y',E),$$ for some $\alpha_1,\ldots,\alpha_n\in J$. Applying the previous proposition we can see that $\{(^YF_{\alpha_i},E)\}_{i=1}^n$ is a subcover of $\widetilde{Y}$. This completes the proof.
Let $(X, \tau, E)$ be a soft topological spaces and $\mathcal{B}\subseteq\tau$. If every element of $\tau$ can be written as a union of elements of $\mathcal{B}$, then $\mathcal{B}$ is called a soft basis for the soft topology $\tau$. Each element of $\mathcal{B}$ is called a soft basis element.
We can characterize soft compact spaces in term of basis elements as follows:
A soft topological space $(X, \tau, E)$ is soft compact if and only if there is a soft basis $\mathcal{B}$ for $\tau$ such that every cover of $\widetilde{X}$ by elements of $\mathcal{B}$ has a finite subcover.
Let $(X, \tau, E)$ be soft compact. Obviousely, $\tau$ is a soft basis for $\tau$. Therefore, every cover of $\widetilde{X}$ by elements of $\tau$ has finite subcover. Conversely, let $\{(U_\alpha, E)\}_{\alpha\in J}$ be a soft open cover of $\widetilde{X}$. We can write $(U_\alpha, E)$ as a union of basis elements, for each $\alpha\in J$. These elements form a soft open cover of $\widetilde{X}$ such as $\{(F_{\mathcal{B}}, E)\}_{\mathcal{B}\in I}$. Therefore $\widetilde{X}=(F_{\mathcal{B}_1}, E)\cup\ldots\cup(F_{\mathcal{B}_n}, E)$, for some $\mathcal{B}_1,\ldots,\mathcal{B}_n\in I$. Let $(F_{\mathcal{B}_i}, E)\widetilde{\subseteq}(U_{\alpha_i}, E)$, for each $1\leq i\leq n$. This implies that $\{(U_{\alpha_i}, E)\}_{i=1}^n$ is a finite subcover of $\widetilde{X}$. Hence, $(X, \tau, E)$ is soft compact.
\[rem36\] Clearly, a soft set is not a set. Indeed, the differences between soft topological spaces and topological spaces arise from this fact. In a sense, when $|E|=1$, a soft set $(F,E)$ behaves similar to a set. In fact, in this case the soft set $(F,E)$ is the same as the set $F(e)$, where $E=\{e\}$. Therefore when $|E|=1$, the soft topological spaces are the same as topological spaces. Nevertheless, in this paper we will see some differences between these two concepts when $|E|\geq 2$.
Now, we consider a countably soft compact space constructed around a soft topology. A soft topological space $(X, \tau, E)$ is said to be *countably soft compact* if every countable soft open cover of $\widetilde{X}$ contains a finite subcover of $\widetilde{X}$. Obviously, every soft compact space is countably soft compact but the following example shows that the converse is not true in general.
We consider the (topological) space $S_{\Omega}$, the minimal uncountable well-ordered set with order topology (see [@Ma]). Let $X=S_\Omega$, $E=\{e\}$ and $\tau=\{(F,E) | F(e)$ is open in $S_\Omega\}$. Considering Remark \[rem36\], the soft topological space $(X, \tau,
E)$ is countably soft compact but not soft compact.
There is a criterion for a soft space to be countable soft compact in term of soft closed sets rather than soft open sets. First we have a definition.
A collection $\mathcal{A}$ of soft set is said to have the finite intersection property if for every finite sub-collection $\{(A_1,E)\cap\ldots\cap(A_n,E)\}$ of $\mathcal{A}$, the intersection $(A_1,E)\cap\ldots\cap(A_n,E)$ is non-null.
A soft topological space is countably soft compact if and only if every countable family of soft closed sets with the finite intersection property has a nonnull intersection.
Let the soft space $(X, \tau, E)$ be countably soft compact. Let the family $\{(F_n,E)\}_{n=1}^\infty$ of soft closed sets have the finite intersection property. If $\cap_{n=1}^\infty(F_n,E)=\phi_E$ by Proposition \[3.3\], $\{(F_n,E)'\}_{n=1}^\infty$ is a countable soft open cover of $\widetilde{X}$. Therefore $\widetilde{X}=(F_{n_1},E)\cup\ldots\cup(F_{n_k},E)$, for some $n_1,\ldots,n_k\in N$. Now, De Morgan’s lows and Proposition \[3.3\] imply that $(F_{n_1},E)\cap\ldots\cap(F_{n_k},E)=\phi_E$. This is a contradiction. Conversely, Let $\{(F_n,E)\}_{n=1}^\infty$ be a countable soft open cover of $\widetilde{X}$ without any subcover. Then $\{(F_n,E)'\}_{n=1}^\infty$ is a family of soft closed sets over $X$ such that $\cap_{n=1}^\infty(F_n,E)'=\phi_E$. Let $n_1,\ldots,n_k$ be arbitrary positive integers. If $(F_{n_1},E)'\cap\ldots\cap(F_{n_k},E)'=\phi_E$ then $\widetilde{X}=(F_{n_1},E)\cup\ldots\cup(F_{n_k},E)$, that is impossible. Therefore $(F_{n_1},E)'\cap\ldots\cap(F_{n_k},E)'\neq\phi_E$, for each $n_1,\ldots,n_k\in N$. This shows that $\{(F_n,E)'\}_{n=1}^\infty$ have the finite intersection property. Therefore $\cap_{n=1}^\infty(F_n,E)'\neq\phi_E$. This is a contradiction.
An immediate result of previous theorem is the following.
A soft space $(X, \tau, E)$ is countably soft compact if and only if every nested sequence $(F_1,E)\widetilde{\supseteq}(F_2,E)\widetilde{\supseteq}\ldots$ of nonnull soft closed sets over $X$ has a nonnull intersection.
Let $(X, \tau, E)$ is countably soft compact. The collection $\{(F_n,E)\}_{n=1}^\infty$ have the finite intersection property. Therefore $\cap_{n=1}^\infty(F_n,E)\neq\phi_E$. Conversely, let $\{(C_n,E)\}_{n=1}^\infty$ be a collection of soft closed sets with the finite intersection property. By Proposition \[prop21\], we construct nested sequence $(F_1,E)\widetilde{\supseteq}(F_2,E)\widetilde{\supseteq}\ldots$ of nonnull soft closed sets by setting $(F_n,E)=(C_1,E)\cap\ldots\cap (C_n,E)$, for each positive integer $n$. By the hypothesis $\cap_{n=1}^\infty(F_n,E)=\cap_{n=1}^\infty(C_n,E)\neq\phi_E$. Now, Theorem 3.8 implies that $(X, \tau, E)$ is countably soft compact.
Soft Separation Axioms
======================
In this section, we will study some soft separation axioms that have studied in [@WKM; @SN]. First, we recall the definitions.
A soft topological space $(X, \tau, E)$ over $X$ is called a soft $T_0$-space if for each pair of distinct points, at least one has neighborhood not containing the other, and a soft $T_1$-space if for each pair of distinct points, each one has a neighborhood not containing the other. Also, the soft space $(X, \tau, E)$ is said to be soft $T_2$- space (or soft Hausdorff) if for each pair $x,y$ of distinct points of $X$, there exist disjoint soft open sets containing $x$ and $y$, respectively.
Obviously, every soft $T_i$-space $(i=1,2)$ is a soft $T_{i-1}$-space. But by Remark 3.6 and general topology the converse is not true. In [@SN], the authors have shown that if $(x, E)$ is a soft closed set in soft set $(X, \tau, E)$, for all $x\in X$, then $(X, \tau, E)$ is soft $T_1$, but the converse does not hold in general.
The soft space $(X, \tau, E)$ over $X$ is called soft regular if for each soft closed set $(G,E)$ and $x\in X$ such that $x \notin (G,E)$ there exist soft open sets $(F_1,E)$ and $(F_2,E)$ such that $x\in (F_1,E),(G,E)\widetilde{\subseteq}(F_2,E)$ and $(F_1,E)\cap (F_2,E)=\phi_E$. The soft space $(X, \tau, E)$ is said to be soft $T_3$-space if it is soft regular and soft $T_1$-space.
Before proceeding, we introduce the concept of soft closure of a soft set (see [@HA]). Let $(X, \tau, E)$ be a soft topological space and $(F,E)$ be a soft set over $X$. Then the soft closure of $(F,E)$, denoted by $(\overline{F,E})$, is the intersection of all soft closed super sets of $(F,E)$. First, we prove the following.
\[lem41\] Let $(X, \tau, E)$ be a soft topological space and $(F,E)$ be a soft set over $X$. If $x\in (\overline{F,E})$, then every soft open set $(G,E)$ containing $x$ intersects $(F,E)$.
Let $x\in (\overline{F,E})$. Let there is a soft open set $(G,E)$ containing $x$ such that $(F,E)\cap (G,E)=\phi_E$. By Proposition \[3.2\], we have $(F,E)\widetilde{\subseteq}(G,E)'$. Therefore $(\overline{F,E})\widetilde{\subseteq}(G,E)'$. Hence $x\in
(G,E)\cap(G,E)'$. This is a contradiction. Therefore $(F,E)\cap(G,E)\neq \phi_E$, for each soft open set $(G,E)$ containing $x$.
The following example shows that the converse of Lemma \[lem41\] is not true.
Suppose that the following sets are given: $X=\{h_1,h_2,h_3\}$, $E=\{e_1,e_2\}$ and $\tau=\{\phi_E,\widetilde{X},(F_1,E),(F_2,E),\ldots,(F_{30},E)\}$ where $F_1,F_2,\ldots,F_{30}$ are given in Example 9 of [@SN]. Then $(X,\tau)$ is a soft topological space over $X$. We consider the soft set $(F_{25},E)$, where $$F_{25}(e_1)=\{h_2\},\ \ F_{25}(e_2)=X.$$ It is easy to see that the following hold $$(\overline{F_{25},E})=(F_{25},E),\ h_1\notin (\overline{F_2,E}).$$ But for every soft open set $(F,E)$ containing $h_1$, we have $(F,E)\cap (F_{25},E)\neq\phi_E$.
\[prop43\] Let $(X, \tau, E)$ be a soft regular space. Then, for each point $x$ of $X$ and a soft open set $(F,E)$ containing $x$, there is a soft open set $(G,E)$ containing $x$ such that $(\overline{G,E})\widetilde{\subseteq}(F,E)$.
$(F,E)'$ is a soft closed set not containing $x$. Therefore, there exist soft open sets $(G,E)$ and $(H,E)$ such that $x\in
(G,E)$, $(F,E)'\widetilde{\subseteq}(H,E)$ and $(G,E)\cap
(H,E)=\phi_E$. Proposition \[3.2\] implies that $(G,E)\widetilde{\subseteq}(H,E)'$. Therefore $(\overline{G,E})\widetilde{\subseteq}(H,E)'\widetilde{\subseteq}((F,E)')'=(F,E)$.
The following example shows that the converse of Proposition \[prop43\] does not hold in general.
\[ex44\] Let $X=\{h\}$, $E=\{e_1,e_2\}$ and $\tau=\{\phi_E, \widetilde{X},
(F_1,E),(F_2,E) \}$, where $$F_1(e_1)=\{h\},\ F_1(e_2)=\emptyset\ \ \&\ \ F_2(e_1)=\emptyset,
F_2(e_2)=\{h\}$$
It is easy to see that $(X, \tau, E)$ is not soft regular Nevertheless, for $h\in X$ and soft open set $\widetilde{X}$ containing $h, \widetilde{X}$ itself is a soft open set containing $h$ such that $h\in
\overline{{\widetilde{X}}}\widetilde{\subseteq} \widetilde{X}$.
Now, we exhibit a necessary and sufficient condition for a soft space to be a soft regular space.
A soft space $(X, \tau, E)$ is soft regular if and only if for each $x\in X$ and soft closed set $(F,E)$ not containing $x$, there is a soft a open set $(G,E)$ containing $x$ such that $(\overline{G,E})\cap(F,E)=\phi_E$.
Let $(X, \tau, E)$ be soft regular. There exist soft open sets $(G,E)$ and $(H,E)$ such that $x\in(G,E)$, $(F,E)\widetilde{\subseteq}(H,E)$ and $(G,E)\cap(H,E)=\phi_E$. Then $(G,E)\widetilde{\subseteq}(H,E)'\widetilde{\subseteq}(F,E)'$. This implies that $(\overline{G,E})\widetilde{\subseteq}(H,E)'\widetilde{\subseteq}(F,E)'$. Therefore $(\overline{G,E})\cap(F,E)=\phi_E$.
Conversely, Proposition \[3.2\] implies that $(F,E)\widetilde{\subseteq}(\overline{G,E})'$. Therefore there is a soft open set $(\overline{G,E})'$ containing $(F,E)$ such that $(G,E)\cap(\overline{G,E})'=\phi_E$. This completes the proof.
A soft space topological space $(X, \tau, E)$ is said to be soft normal if for each soft closed sets $(F,E)$ and $(G,E)$ over $X$ with null intersection there exist soft open sets $(F_1,E)$ and $(F_2,E)$ containing $(F,E)$ and $(G,E)$ respectively, such that $(F_1,E)\cap(F_2,E)=\phi_E$. Also, a soft topological space $(X,
\tau, E)$ is said to be a soft $T_4$-space if it is soft normal and soft $T_1$-space.
Let $(X, \tau, E)$ be a soft space. Let for each soft closed set $(F,E)$ and soft open set $(G,E)$ containing $(F,E)$ there is a soft open set $(H,E)$ containing $(F,E)$ such that $(\overline{H,E})\widetilde{\subseteq}(G,E)$. Then $(X, \tau, E)$ is soft normal.
For each soft closed sets $(F,E)$ and $(I,E)$ with null intersection $(I,E)'$ is a soft open set containing $(F,E)$. Therefore there is a soft open set $(H,E)$ containing $(F,E)$ such that $(\overline{H,E})\widetilde{\subseteq}(I,E)'$. By Proposition \[3.2\], $(I,E)\widetilde{\subseteq}(\overline{H,E})'$. Since $(H,E)\widetilde{\subseteq}((\overline{H,E})')'$, we have $(H,E)\cap(\overline{H,E})'=\Phi_E$. Hence $(X, \tau, E)$ is soft normal.
There is an obvious question to ask at this point. Is a soft $T_4$-space a soft $T_3$-space? The soft space $(X, \tau, E)$ in Example \[ex44\], shows that the answer is ”NO”. In fact it is easy to see that $(X, \tau, E)$ is a soft $T_4$-space and not a soft $T_3$-space.
In Theorem 3.17 of [@WKM], the following is proved:\
\
Theorem. ([@WKM]) Let $(X, \tau, E)$ be a soft topological space over $X$ and $x\in X$. Then the following are equivalent:
(1)
: $(X, \tau, E)$ is a soft regular space;
(2)
: For each soft closed set $(G,E)$ such that $(x,E)\cap(G,E)=\phi_E$.
There exit soft two open sets $(F_1,E)$ and $(F_2,E)$ such that $(x,E)\widetilde{\subseteq}(F_1,E)$, $(G,E)\widetilde{\subseteq}(F_2,E)$ and $(F_1,E)\cap(F_2,E)=\phi_E$.
By Example \[ex44\], we can see that this theorem is incorrect. In fact the soft space $(X, \tau, E)$ in this example satisfies in (2), but it is not soft regular. We note that $(x,E)\cap(G,E)=\phi_E$ is not equivalent to $x\notin (G,E)$. But $(x,E)\widetilde{\nsubseteq}(G,E)$ is. Therefore, we must replace the condition $(x,E)\widetilde{\nsubseteq}(G,E)$ instead of $(x,E)\cap(G,E)=\phi_E$ in Theorem 3.17 of [@WKM].
In Theorem 3.25 of [@WKM], the following is proved:\
\
Theorem. ([@WKM]) Let $(X, \tau, E)$ be a soft topological space over $X$. If $(X, \tau, E)$ is a soft normal space and if $(x,E)$ is a soft closed set for each $x\in X$, then $(X, \tau,
E)$ is a soft $T_3$-space.
This theorem is incorrect. The soft space $(X, \tau, E)$ in Example \[ex44\] satisfies in the conditions of the theorem, but it is not a soft $T_3$-space.
There are some familiar results on the applications of compactness in separation axioms in General Topology such as: *Every compact Hausdorff space is normal*. But it is not true about soft topological spaces. Consider the following example.
Let $X=\{h\}$, $E=\{e_i\}_{i=1}^5$ and $\tau=\{\phi_E,
\widetilde{X}, (F_1,E),(F_2,E), (F_3,E) \}$, where $$F_1(e_1)=\emptyset,\ F_1(e_2)=X,\ F_1(e_3)=\emptyset,\
F_1(e_4)=X,\ F_1(e_5)=\emptyset;$$ $$F_2(e_1)=X,\ F_2(e_2)=X,\ F_2(e_3)=X,\ F_2(e_4)=\emptyset,\
F_2(e_5)=X;$$ $$F_3(e_1)=\emptyset,\ F_3(e_2)=X,\ F_3(e_3)=\emptyset,\
F_3(e_4)=\emptyset,\ F_3(e_5)=\emptyset.$$ It is easy to see that $(X, \tau, E)$ is not soft normal. Nevertheless, it is soft compact.
It is remarkable that every compact Hausdorff space is not normal, even if we consider $(X, \tau, E)$ as a soft regular space. Indeed, the Example \[ex44\] is a counterexample.
[MaHo]{} M.I. Ali, F. Feng, X. Liu, W.K. Min and M. Shabir, [*On some new operations in soft set theory*]{}, Comput. Math. Appl., [**57**]{} (2009), 1547-1553. A. Aygunoglu and H. Aygun, [*Some notes on soft topological spaces*]{}, Neural. Comput. Appl., (2011), 1-7. N. Çağman and S. Enginoglu, [*Soft set theory and uni–int decision making*]{}, Europ. J. Operat. Res., [**207**]{} (2010), 848-855. N. Çağman and S. Enginoglu, [*Soft matrix theory and its decision making*]{}, Comput. Math. Appl., [**59**]{} (2010), 3308-3314. D. Chen, [*The parametrization reduction of soft sets and its applications*]{}, Computers and Mathematics with Applications, [**49**]{} (2005), 757-763. Á. Császár, [*Generalized topology, generalized countiniuty*]{}, Acta. Math. Hungar., [**96**]{} (2002), 351-357. S. Hussain and B. Ahmad, [*Some properties of soft topological spaces*]{}, Comput. Math. Appl., [**62**]{} (2011), 4058-4067. P. K. Maji, R. Biswas and A. R. Roy, [*An application of soft sets in a decision making problem*]{}, Computers and Mathematics with Applications, [**44**]{} (2002), 1077-1083 P. K. Maji, R. Biswas and A. R. Roy, [*Soft set theory*]{}, Comput. Math. Appl., [**45**]{} (2003), 555-562. W. K. Min, [*A note on soft topological spaces*]{}, Comput. Math. Appl., [**62**]{} (2011), 3524-3528. D. A. Molodtsov, [*Soft set theory-first results*]{}, Comput. Math. Appl., [**37**]{} (1999), 19-31. D. A. Molodtsov, V. Y. Leonov and D. V. Kovkov, [*Soft sets technique and its application*]{}, Nechetkie Sistemy i Myagkie Vychisleniya, 1 (1)(2006), 8-39. J. Munkres, [*Topology*]{}, Prentice-Hall, Englewood Cliffs, NJ, 2000. D. Pie and D. Miao, [*From soft sets to information systems*]{}, in: Granular computing, IEEE Inter. Conf., [**2**]{} (2005), 617-621. E. Peyghan, B. Samadi and A. Tayebi, [*About soft topological spaces*]{}, preprint. M. Shabir and M. Naz, [*On soft topological spaces*]{}, Comput. Math. Appl., [**61**]{} (2011), 1786-1799. I. Zorlutuna, M. Akdag, W. K. Min and S. Atmaca, [*Remarks on soft topological spaces*]{}, Annal. Fuzzy. Math. Inform., [**3**]{} (2012), 171- 185.
Esmaeil Peyghan and Babak Samadi\
Department of Mathematics, Faculty of Science\
Arak University\
Arak 38156-8-8349, Iran\
Email: [email protected]
Akbar Tayebi\
Department of Mathematics, Faculty of Science\
University of Qom\
Qom. Iran\
Email: [email protected]
[^1]: 2010 Mathematics subject Classification: 06D72, 54A40.
| {
"pile_set_name": "ArXiv"
} |
---
address: |
Institute of Theoretical Physics\
50-205 Wrocław,\
pl. Maxa Borna 9, Poland\
E-mail: lukier,[email protected]
author:
- 'J. LUKIERSKI and M. WORONOWICZ'
title: 'Twisted Space-Time Symmetry, Non-Commutativity and Particle Dynamics'
---
Introduction
============
Since the work of Doplicher et all. (see e.g.[@luwo1; @luwo2]) there is a strong indication that due to quantum gravity effects the space-time coordinates are becoming noncommutative. In general case one can write[^1] $$\begin{aligned}
\label{luwoeq1}
[ \widehat{x}_{\mu},\widehat{x}_{\nu}] & = &
\frac{i}{\kappa^2} \, \theta_{\mu\nu} (\kappa \widehat{x}_{\rho})
\cr
& = &
\frac{i}{\kappa^2} \, \theta_{\mu\nu}^{(0)}
+
\frac{i}{\kappa} \, \theta_{\mu\nu}^{(1) \rho } \,
\widehat{x}_\rho
+ i
\theta_{\mu\nu}^{(2) \rho \tau}
\widehat{x}_{\rho}\widehat{x}_{\tau}\, ,
\end{aligned}$$ where the fundamental mass parameter $\kappa$ has been introduced in order to exhibit the mass dimensions of respective terms and have the constant tensors $ \theta_{\mu\nu}^{(0)}$, $ \theta_{\mu\nu}^{(1) \rho}$, $ \theta_{\mu\nu}^{(2) \rho \tau}$ as dimensionless. If we link (\[luwoeq1\]) with quantum gravity one can put $\kappa = m_{\rm pl}$ ($m_{\rm pl}$ - Planck mass). Further we add that the relation (\[luwoeq1\]) describes in D=10 first-quantized open string theory the noncommutative coordinates on D-branes providing the localizations of the ends of the strings [@sw; @chs].
There are two important problems related with the application of formula (\[luwoeq1\]) to physical models:
[**i)**]{} In standard relativistic theory, with classical Poincaré symmetries, the first term on rhs of (\[luwoeq1\]) breaks the Lorentz invariance, and further two terms break both Lorentz and translational invariance. One can ask how looks the deformation of classical Poincaré invariance which permits to consider relations (\[luwoeq1\]) as covariant under deformed Poincaré transformations, i.e. the same in any deformed Poincaré frame.
[**ii)**]{} There should be given prescriptions how to formulate the classical mechanics and field theory models with noncommutative space-time coordinates (\[luwoeq1\]), covariant under the twisted Poincaré symmetries.
If the time coordinate remains classical (i.e. in formula (\[luwoeq1\]) $\theta_{0\mu} = 0$) both points i) and ii) can be applied to the nonrelativistic noncommutative theories with classical Galilean invariance broken by relation (\[luwoeq1\]).
Twisted Space-Time Symmetries
=============================
We shall look for the quantum relativistic symmetries implying the covariance of noncommutative Minkowski spaces. In systematic study firstly one should consider all possible quantum relativistic symmetries (quantum Poincaré algebras) in the form of noncommutative Hopf algebras, and then derive corresponding quantum Minkowski spaces as deformed Hopf algebra modules. An example of such a construction which is already more than ten years old is the $\kappa$-deformed Minkowski space [@zak; @mr; @lrz] $$\label{kappa}
[\widehat{x}_0,\widehat{x}_i]=\frac{i}{\kappa}\widehat{x}_i\,,\qquad\qquad [\widehat{x}_i,\widehat{x}_j]=0\,,$$ corresponding in (\[luwoeq1\]) to the choice $\theta^{(0)}_{\mu\nu}=\theta^{(2)\rho\tau}_{\mu\nu}=0$ and $\theta^{(1)\rho}_{\mu\nu}=\eta_{\nu 0}\delta_\mu^{\ \rho}-\eta_{\mu 0}\delta_{\nu}^{\ \rho}$. Using the Hopf-algebraic formulae of $\kappa$-deformed Poincaré algebra in bicrossproduct basis one can show [@mr] that the relations (\[kappa\]) are covariant under the Hopf-algebraic action of $\kappa$-deformed Poincaré algebra.
It appears that the most effective way of describing the noncommutative space-times covariant under quantum relativistic symmetries is to consider twisted symmetry algebras. In such a case the classical Poincaré-Hopf algebra is modified only in the coalgebraic sector, with all the algebraic relations preserved. We change the classical Poincaré Hopf algebra $\mathcal{H}^{(0)}=(\mathcal{U}(\mathcal{P}_4), m, \Delta_0, S_0, \epsilon)$ into twisted Poincaré Hopf algebra $\mathcal{H}=(\mathcal{U}(\mathcal{P}_4), m, \Delta, S, \epsilon)$ by means of the twist factor $\mathcal{F}\in\mathcal{U}(\mathcal{P}_4)\otimes\mathcal{U}(\mathcal{P}_4)$ as follows ($\mathcal{P}_4\ni\hat{g}=(P_\mu, M_{\mu\nu})$) $$\Delta(\hat{g})=\mathcal{F}\circ\Delta_0(\hat{g})\circ\mathcal{F}^{-1}\,,\qquad S(\hat{g})=US_0(\hat{g})U^{-1}\,,$$ $$\Delta_0(\hat{g})=\hat{g}\otimes 1+1\otimes\hat{g}\,,\qquad S_0(\hat{g})=-\hat{g}\,,\qquad \epsilon(\hat{g})=0\,,$$ where $(a\otimes b)\circ(c\otimes d)=ac\otimes bd$. The twist $\mathcal{F}$ satisfies the cocycle and normalization conditions [@dr] $$\label{cocy}
\mathcal{F}_{12}\,\left(\Delta_0\otimes 1\right)\,\mathcal{F}=\mathcal{F}_{23}\,\left(1\otimes\Delta_0 \right)\,\mathcal{F}\,,\qquad (\epsilon\otimes 1)\mathcal{F}=(1\otimes\epsilon)\mathcal{F}=1\,,$$ where $\mathcal{F}_{12}=\textup{f}_{(1)}\otimes \textup{f}_{(2)}\otimes 1$ etc. $(\mathcal{F}=\textup{f}_{(1)}\otimes \textup{f}_{(2)})$ and $U= \textup{f}_{(1)}S(\textup{f}_{(2)})$.
The advantage of using twisted Poincaré algebra is the explicit formula for the multiplication in twisted Hopf algebra module $\mathcal{A}$ which should satisfy the condition (see e.g. [@majid], $h\in\mathcal{U}(\mathcal{P}_4),\ a,b\in\mathcal{A}$) $$\label{module}
h\rhd(a\bullet b)=(h_1 a)\bullet(h_2 b)\,,$$ where $\Delta(h)=h_1\otimes h_2$. We see from (\[module\]) that if $h_1\neq h_2$ then $a\bullet b\neq b\bullet a$, i.e. from quantum-deformed relativistic symmetry follow necessarily the noncommutative Minkowski space as its Hopf-algebraic module.
One can show that the multiplication in $\mathcal{A}$ for twisted Hopf algebra $\mathcal{H}$ which is consistent with the relation (\[module\]) ($h\in\mathcal{H}$) provides the formula [@km; @bloh; @cknt] $$\label{modtw}
a\bullet b=(\overline{\textup{f}}_{(1)} a)(\overline{\textup{f}}_{(2)} b)\,,\qquad \mathcal{F}^{-1}=\overline{\textup{f}}_{(1)}\otimes\overline{\textup{f}}_{(2)}\,.$$ In the case of relativistic symmetries one can use the classical space-time representation for the Poincaré generators $P_\mu,\ M_{\mu\nu}$ $$\label{repgen}
P_\mu=i\partial_\mu\,,\qquad M_{\mu\nu}=i(x_\nu\partial_\mu-x_\mu\partial_\nu)\,.$$ Subsequently in the formula (\[modtw\]) one can assume that $a,\ b$ are classical functions on commutative Minkowski space $x_\mu$, and define $\overline{\textup{f}}_{(i)}(P_\mu,\ M_{\mu\nu})\equiv \overline{\mathfrak{f}}_{(i)}(x,\ \partial),\ i=1,2$. One gets the following star product multiplication which is a particular representation of algebraic formula (\[modtw\]) $$\xi(x)\star\zeta(x)=(\bar{\mathfrak{f}}_{(1)}(x,\partial)\xi(x)) (\bar{\mathfrak{f}}_{(2)}(x,\partial)\zeta(x))\,.$$
The important application of twisted Poincaré algebras to the covariant description of noncommutative Minkowski spaces, namely describing the quantum covariance of (\[luwoeq1\]) for the case $\theta_{\mu\nu}=\theta^{(0)}_{\mu\nu}$ is quite recent[^2]. The quantum symmetry which leaves invariant the simplest form of (\[luwoeq1\])[^3] $$\label{thetacom}
[\widehat{x}_\mu,\widehat{x}_\nu]_\bullet=\frac{i}{\kappa^2}\theta^{(0)}_{\mu\nu}\,,$$ (where $[a,b]_\bullet=a\bullet b-b\bullet a$) is generated by the following Abelian twist $$\label{twist0}
\mathcal{F}_{\theta }=\exp \,\frac{i}{2\kappa^2}(\,\theta ^{\mu \nu }_{(0)}\,P_{\mu }\wedge P_{\nu
}\,)\,.$$ We obtain the twisted Poincaré-Hopf structure with classical Poincaré algebra relations and modified coproducts of Lorentz generators $M_{\mu\nu}$ $$\begin{aligned}
\label{cop0}
\Delta_\theta(P_\mu)&=&\Delta_0(P_\mu), \\
\Delta _{\theta }(M_{\mu \nu })&=&\mathcal{F}_{\theta }\circ \,\Delta _{0}(M_{\mu \nu
})\circ \mathcal{F}_{\theta }^{-1}\nonumber\\
&=&\Delta _{0}(M_{\mu \nu })-\frac{1}{\kappa ^{2}}%
\theta ^{\rho \sigma }_{(0)}[(\eta _{\rho \mu }P_{\nu }-\eta _{\rho \nu
}\,P_{\mu })\otimes P_{\sigma }\label{cop00}\\
&&\qquad\qquad\qquad\qquad\qquad+P_{\rho}\otimes (\eta_{\sigma \mu}P_{\nu}-\eta_{\sigma \nu}P_{\mu})]\,.\nonumber\end{aligned}$$
One can consider however also other Abelian twists of Poincaré symmetries, depending on the Lorentz generators $M_{\mu\nu}$ (see [@lnrt; @asch96; @asch99; @luwo]). It appears that only subclass of general commutator (\[luwoeq1\]) with linear and quadratic terms can be covariantized by twisted Poincaré algebras. In the following section we shall consider the quantum Poincaré symmetries corresponding to the following two twist functions [@luwo]:
- Lie-algebraic relations for noncommutative Minkowski space $$\label{twist1}
\mathcal{F}_{(\alpha\beta)}= \exp \,\frac{i}{2\kappa}( \zeta^\lambda\,P_{\lambda }\wedge M_{\alpha \beta }) \,,$$ where $\alpha,\ \beta=0,1,2,3$ are fixed and the vector $\zeta^\lambda=\theta^{\lambda\alpha\beta}_{(1)}$ has vanishing components $\zeta^\alpha,\ \zeta^\beta$.
- Quadratic deformations of Minkowski space $$\label{twist2}
\mathcal{F}_{(\alpha\beta\gamma\delta)}=\exp\,\frac{i}{2}\zeta\, M_{\alpha\beta}\wedge M_{\gamma\delta}\,,$$ where $\zeta=\theta_{(2)}^{\alpha\beta\delta\gamma}$ is a numerical parameter, all the four indices $\alpha,\beta,\gamma,\delta$ are fixed and different.
Lie-algebraic and Quadratic Quantum-Covariant Noncommutative Minkowski Spaces
=============================================================================
In this Section we shall report on results presented in [@luwo], which we supplement by the proof of quantum translational invariance.
In the formalism of quantum-deformed Hopf-algebraic symmetries the quantum-covariant noncommutative Minkowski space can be introduced in two ways:
- as the translation sector of quantum Poincaré group,
- as the quantum representation space (a Hopf algebra module) for quantum Poincaré algebra with the action of the deformed symmetry generators satisfying suitably deformed Leibnitz rule (\[module\]).
In the case of constant tensor $\theta_{\mu\nu}=\theta_{\mu\nu}^{(0)}$ the quantum Poincaré group algebra dual to the coproducts (\[cop0\]), (\[cop00\]) is known[@oeckl; @koma; @luwo], and the quantum translations do not satisfy the relation (\[thetacom\]). It appears that the relation (\[thetacom\]) as describing quantum-covariant noncommutative Minkowski space can be obtained only as the Hopf algebra module. To the contrary, in the case of twisted relativistic symmetries generated by the twist factors (\[twist1\], \[twist2\]) it can be shown that both definitions i) and ii) coincide [@luwo].
- Lie-algebraic noncommutative Minkowski space.
The commutator algebra following from (\[twist1\]) and the formula (\[modtw\]) has the form [@luwo] $$\label{lie}
\left[ \widehat{x}_{\mu },\widehat{x}_{\nu }\right] _{\bullet}=C_{\ \mu \nu }^{\rho }\widehat{x}_{\rho }\,,$$ where $$\label{const}
C_{\ \mu \nu }^{\rho }=\frac{i}{\kappa}\zeta_\mu( \eta _{\beta \nu }\delta
_{\ \alpha }^{\rho }-\eta _{\alpha \nu }\delta _{\ \beta }^{\rho })+\frac{i}{\kappa}\zeta_\nu( \eta _{\alpha \mu }\delta _{\ \beta }^{\rho }-\eta _{\beta \mu }\delta _{\ \alpha }^{\rho
}) \,.$$ The relations (\[lie\]) can be written in more transparent way as follows ($\alpha,\ \beta$ are fixed by the choice of twist function) $$\label{komlie}
\left[\widehat{x}_\alpha,\widehat{x}_\lambda\right]_\bullet=\frac{i}{\kappa}\zeta_\lambda \eta_{\alpha\alpha}\widehat{x}_\beta\,,\qquad\qquad
\left[\widehat{x}_\beta,\widehat{x}_\lambda\right]_\bullet=-\frac{i}{\kappa}\zeta_\lambda \eta_{\beta\beta}\widehat{x}_\alpha\,,$$ where $\zeta_\alpha=\zeta_\beta=0$.
The quantum Lorentz covariance of (\[lie\]) under the Hopf action of the Lorentz generators $M_{\mu\nu}$ has been shown in [@luwo]. We shall show the quantum translational invariance of (\[lie\]) using the differential realization (\[repgen\]). The fourmomentum coproduct generated by twist (\[twist1\]) has the form [@luwo] $$\Delta(P_\mu)=\Delta_0(P_\mu)+\frac{1}{2\kappa}\xi^\lambda P_\lambda\wedge(\eta_{\alpha\mu} P_\beta-\eta_{\beta\mu} P_\alpha)+\mathcal{O}(P^3)\,.$$ Putting in (\[module\]) $h\equiv P_\mu$, $a\equiv x_\rho$, $b\equiv x_\sigma$ and using (\[const\]) we obtain $$\begin{aligned}
P_{\mu}\rhd\left(x_\rho\bullet x_\sigma\right) &=&ix_{\{\rho}\eta_{\sigma\}\mu}+\eta_{\alpha\mu}\xi_{[\sigma}\eta_{\rho]\beta}-\eta_{\beta\mu}\xi_{[\sigma}\eta_{\rho]\alpha}\,,\\
&=&ix_{\{\rho}\eta_{\sigma\}\mu}+\frac{1}{2}P_{\mu}\rhd C_{\ \rho\sigma}^{\lambda}x_\lambda\,.\nonumber\end{aligned}$$ Finally we get $$P_\mu\rhd\left[x_\rho,x_\sigma\right]_\bullet=P_\mu\rhd C_{\ \rho\sigma}^{\lambda}x_\lambda\,,$$ i.e. the relation (\[lie\]) is covariant.
- Quadratic noncommutativity of Minkowski space coordinates.
After using the formula (\[modtw\]) with inserted twist (\[twist2\]) one gets the following commutation relations of space-time coordinates ($\{a,b\}_\bullet=a\bullet b+b\bullet a$) $$\begin{aligned}
[\widehat{x}_\mu,\widehat{x}_\nu]_\bullet&=&i\sinh\frac{\zeta}{2}\cosh\frac{\zeta}{2}
(\eta_{\alpha[\mu}\eta_{\gamma\nu]}\{\widehat{x}_\beta,\widehat{x}_\delta\}_\bullet
-\eta_{\alpha[\mu}\eta_{\delta\nu]}\{\widehat{x}_\beta,\widehat{x}_\gamma\}_\bullet\label{quadcom}\\
&&\qquad\qquad\qquad\qquad-
\eta_{\beta[\mu}\eta_{\gamma\nu]}\{\widehat{x}_\alpha,\widehat{x}_\delta\}_\bullet
+\eta_{\beta[\mu}\eta_{\delta\nu]}\{\widehat{x}_\alpha,\widehat{x}_\gamma\}_\bullet)\nonumber\\&-&\sinh^2\frac{\zeta}{2}(\sum_{\genfrac{}{}{0pt}{}{k=\alpha,\beta}{l=\gamma,\delta}}^{}\delta^k_{\ [\mu}\delta^l_{\ \nu]}[\widehat{x}_k,\widehat{x}_l]_\bullet)\,,\nonumber\end{aligned}$$ or in more explicit form ($k=\alpha,\beta$ and $l=\gamma,\delta$) $$\begin{aligned}
\label{komquad}
&&[\widehat{x}_k,\widehat{x}_l]_\bullet=i\tanh\frac{\zeta}{2}
(\eta_{\alpha k}\eta_{\gamma l}\{\widehat{x}_\beta,\widehat{x}_\delta\}_\bullet
-\eta_{\alpha k}\eta_{\delta l}\{\widehat{x}_\beta,\widehat{x}_\gamma\}_\bullet\label{star1q}\\
&&\qquad\qquad\qquad\qquad-
\eta_{\beta k}\eta_{\gamma l}\{\widehat{x}_\alpha,\widehat{x}_\delta\}_\bullet
+\eta_{\beta k}\eta_{\delta l}\{\widehat{x}_\alpha,\widehat{x}_\gamma\}_\bullet)\,,\nonumber\end{aligned}$$ and $[\widehat{x}_\alpha, \widehat{x}_\beta]_\bullet=[\widehat{x}_\gamma,\widehat{x}_\delta]_\bullet=0$.
We conjecture that the relations (\[quadcom\]) are covariant under the action of quantum Poincaré symmetries, generated by twist (\[twist2\]).
The linear and quadratic relations (\[komlie\]) and (\[komquad\]) provide special choices of the constant parameters $\theta^{(1)\rho}_{\mu\nu}$, $\theta^{(2)\rho\tau}_{\mu\nu}$ for which the quantum covariance group was found in [@luwo].
Particle Dynamics Invariant Under Twisted Relativistic and Galilean Symmetries
==============================================================================
The discussion of the noncommutative dynamical theories one begins naturally with the consideration of classical mechanics models. We shall restrict our considerations here to the case $\theta_{\mu\nu}=\theta^{(0)}_{\mu\nu}$, i.e. the noncommutative space-time described by (\[thetacom\]). One can introduce the Lagrangian models describing free point particles moving in noncommutative space-time in the following two ways:
- If $\theta_{\mu 0}=0$, i.e. we have the relations $$\begin{aligned}
&&[\widehat{x}_i,\widehat{x}_j]=i\theta_{ij}\,,\label{dir}\\
&&[\widehat{x}_0,\widehat{x}_i]=0\,,
\end{aligned}$$ we deal with classical time variable $t$, where $\hat{x}_0=ct$ and noncommutative space coordinates $\hat{x}_i$. In such a case one can look for the non-relativistic Lagrangian models with constraints, which provide the relation (\[dir\]) as the quantized Dirac bracket. Such a first model was constructed in [@lsz1] in $D=(2+1)$ dimensions with the following Lagrangian $$\label{lag}
\mathcal{L}=\frac{m\dot{x}^2_i}{2}-k\epsilon_{ij}\dot{x}_i \ddot{x}_j\,.$$ The higher order Lagrangian (\[lag\]) can be expressed if first order form in six-dimensional phase space $(x_i, p_i, \tilde{p}_i)$[^4] and after introducing the linear transformations $$\begin{aligned}
&&X_i=x_i-\frac{2}{m}\tilde{p}_i\,,\nonumber\\
&&P_i=p_i\,,\\
&&\tilde{P}_i=\epsilon_{ij}\tilde{p}_j+\frac{k}{m}p_i\,,\nonumber
\end{aligned}$$ one obtains the following symplectic structure for the variables $Y_A=(X_i,P_i,\tilde{P}_i),\ (A=1...6)$ $$\{Y_A,Y_B\}=\Omega_{AB}\,,\qquad\qquad \Omega=
\left(\begin{array}{ccc}
\frac{2k}{m^2}\epsilon & 1_2 & 0 \\
-1_2 & 0 &0 \\
0 & 0 & \frac{k}{2}\epsilon
\end{array}\right)\,.$$ One can identify (\[dir\]) with quantized PB for the space variables $X_i$ if we put in (\[dir\]) $\theta_{ij}=\frac{2k}{m^2}\epsilon_{ij}$.
In [@lsz1] the dimension $D=2+1$ was chosen because in two space dimensions one can put $\theta_{ij}=\theta\epsilon_{ij}$, i.e. the relation (\[dir\]) does not break the classical Galilean invariance. However if $k\neq 0$ the Galilean algebra is centrally extended by second *exotic* central charge [@levy].\
- For general constant $\theta_{\mu\nu}$ one obtains the noncommutative action describing free particle motion if we introduce in the first order action for classical massive relativistic particle $$\label{act1}
S=\int d\tau[\dot{y}_\mu p^\mu-e(p^2-m^2)]\,,$$ the following change of variables (we recall that $\theta_{\mu\nu}=\theta_{\mu\nu}^{(0)}$) $$\label{change}
y_\mu=x_\mu+\frac{1}{a}\theta_{\mu\nu}p^\nu\,.$$ It is easy to check that if we introduce CCR following from (\[act1\]) $$[y_\mu,y_\nu]=0\,,\qquad [y_\mu, p^\nu]=i\delta_\mu^{\ \nu}\,, \qquad [p^\mu,p^\nu]=0\,,$$ then the variables $x_\mu$ in (\[change\]) satisfy the relation (\[thetacom\]) if we put $a=2\kappa^2$. Using the relation (\[change\]) one can rewrite the action (\[act1\]) as follows $$\label{act2}
S=\int d\tau[\dot{x}_\mu p^\mu-e(p^2-m^2)+\frac{1}{a}\theta^{\mu\nu}\dot{p}_\mu p_\nu]\,.$$
The variables $y_\mu,p_\mu$ in (\[act1\]) are classical, i.e. transform under Lorentz rotations in standard way $$\label{lorentz}
y'_\mu=\Lambda_\mu^{\ \nu}y_\nu\,,\qquad\qquad p'_\mu=p_\mu\,.$$ Using (\[change\]) and (\[lorentz\]) one gets however $$\begin{aligned}
\label{deflor}
x'_\mu&=&y'_\mu+\frac{1}{a}\theta_{\mu\nu}\Lambda^\nu_{\ \rho}p^\rho\\
&=&\Lambda_\mu^{\ \nu}x'_\nu+\frac{1}{a}(\Lambda_\mu^{\ \rho}\theta_{\rho\nu}+\theta_{\mu\rho}\Lambda^\rho_{\ \nu})p^\nu\nonumber\,.
\end{aligned}$$ Interestingly enough, the transformations (\[deflor\]) describe exactly the twisted Lorentz transformations, generated by the coproduct (\[cop00\]), which leave invariant the action (\[act2\]) for the noncommutative relativistic particle.
The model (\[act2\]) has been firstly obtained without reference to twisted Lorentz symmetries by Deriglazov [@der] and its non-relativistic version $$\label{actnr}
S_{NR}=\int dt[\dot{x}_i\dot{p}_i-\frac{1}{2m}\vec{p}^{\ 2}+\frac{1}{a}\theta_{ij}\dot{p}_ip_j]\,,$$ in $D=2+1$, when $\theta_{ij}=\epsilon_{ij}$, it was proposed by Duval and Horvathy [@dhor]. It is well-known however that the model (\[actnr\]) can be also derived from the model (\[lag\]). Indeed, the first order formulation of the model (\[lag\]) in Faddeev-Jackiw approach [@fj] to higher order Lagrangians provides the action [@hp; @lsz2] $$\label{lag3}
\mathcal{L}=p_i(\dot{x}_i-y_i)+\frac{\vec{y}^{\ 2}}{2m}+\frac{1}{a}\epsilon_{ij}\dot{p}_i p_j\,.$$ The Lagrangian (\[lag3\]) after introducing the new coordinates $$X_{i} = x_{i} + \frac{1}{a} \epsilon_{ij}(y_{j} - p_{j})\,,$$ provides the Lagrangian (\[actnr\]) (with $x_{i}$ replaced by $X_{i}$) and additional term which depends on auxiliary internal variables commuting with $(X_i, p_i)$ [@hp].
The nonrelativistic model (\[actnr\]) can be considered in any space dimension $d$ . If $d = 2$ the action (\[actnr\]) is, similarly as (\[lag\]), invariant under the transformations of exotic $(2+1)$ - dimensional Galilean group. If $d >2$ the invariance of the nonrelativistic model (\[actnr\]) can be achieved by considering quantum Galilean symmetries, with twisted space rotations generated by the following nonrelativistic twist $$\mathcal{F}^{NR}_{\theta} = \exp \frac{i}{a} \theta_{ij}P_i \wedge P_j\,.$$ The formulation of twisted quantum mechanics invariant under twisted quantum Galilei group is now under our consideration.
Final Remarks
=============
We presented in this paper some selected aspects of the theory of noncommutative space-times, with new results on quantum Poincaré covariance of a class of linearly and quadratically deformed Minkowski spaces. We also considered the non-relativistic and relativistic particle models on noncommutative space-time with numerical value of the noncommutativity function $\theta_{\mu\nu}=\theta_{\mu\nu}^{(0)}$ and have pointed out their twisted quantum covariance. We see that the role of quantum deformations is to introduce in place of broken classical symmetries a modified transformations which imply the quantum covariance. Such a possibility selects only particular class of tensors $\theta^{(1)\rho}_{\mu\nu}$ and $\theta^{(2)\rho\tau}_{\mu\nu}$ in formula (\[luwoeq1\]).
Most of the applications of the noncommutative space-times in the literature assume the choice $\theta_{\mu\nu}=\theta^{(0)}_{\mu\nu}$ (see (\[thetacom\])). In this talk we presented also the results for linear ($\theta^{(1)\rho}_{\mu\nu}\neq 0$) and quadratic ($\theta^{(2)\rho\tau}_{\mu\nu}\neq 0$) deformations of Minkowski space. The extension of particle models on noncommutative space-times to linearly and quadratically deformed Minkowski spaces is now studied.
#### Acknowledgements
One of the authors (JL) would like to thank Prof. Ge Mo-Lin and Dr. Cheng-Ming Bai for their fantastic hospitality during the conference at Nankai Mathematical Institute in Tianjin. We also acknowledge the support of KBN grant 1P03B 01828 and the EPSRC.
[9]{}
S. Doplicher, K. Fredenhagen and J.E. Roberts, Phys. Lett. **B331**, 39 (1994); Commun. Math. Phys. 172, 187 (1995); hep-th/0303037.
A. Kempf and G. Mangano, Phys. Rev. **D55**, 7909 (1997); hep-th/9612084.
N. Seiberg and E. Witten, JHEP **09**, 032 (1999); hep-th/9908142.
Chong-Sun Chu, *Noncommutative geometry from strings*; hep-th/0502167.
S. Zakrzewski, Journ. of Phys. **A27**, 2075 (1994).
S. Majid, H. Ruegg, Phys. Lett **B334**, 338 (1994).
J. Lukierski, H. Ruegg, W. Zakrzewski, Ann. Phys. **243**, 90 (1995).
V.G. Drinfeld, Leningrad Math. J. 1, 1419 (1990).
S. Majid, *Foundation of Quantum Groups*, Cambridge University Press (1994).
P.P. Kulish, A.I. Mudrov, Proc. Stekl. Inst. Math. **226**, 97 (1999); q-alg/9901019.
C. Blohmann, Jour. Math. Phys. **44**, 4736 (2003).
M. Chaichian, P.P Kulish, K. Nishijima and A. Tureanu, Phys. Lett. **B604**, 98 (2004); hep-th/0408069.
P. Aschieri and L. Castellani, Int. J. Mod. Phys. [**A11**]{}, 4513 (1996); q-alg/9601006.
P. Aschieri, L. Castellani and A.M. Scarfone, Eur. Phys. J. [**C7**]{}, 159 (1999).
R. Oeckl, Nucl. Phys. **B581**, 559 (2000); hep-th/0003018.
C. Jambor and A. Sykora, *Realization of algebras with the help of $\star$-products*; hep-th/0405268.
J. Wess, *Deformed coordinate spaces: Derivatives*; hep-th/0408080.
F. Koch and E. Tsouchnika, Nucl. Phys. **B717**, 387 (2005); hep-th/0409012.
P. Kosiński and P. Maślanka, *Lorentz - invariant interpretation of noncommutative space-time: Global version*; hep-th/0408100.
J. Lukierski, A. Nowicki, H. Ruegg and V.N. Tolstoy, J. Phys. **A27**, 2389 (1994); hep-th/9312068.
J. Lukierski and M. Woronowicz, *New Lie-algebraic and quadratic deformations of Minkowski space from twisted Poincare symmetries*; hep-th/0508083, PLB (in press).
J. Lukierski, P. Stichel and W.J. Zakrzewski, Ann. of Phys. **260**, 224 (1997).
J.M. Levy-Leblond, *Group Theory and Applications*, vol. 2, ed. Loeble, Acad. Press, New York (1972), p.222.
A.A. Deriglazov, *Noncommutative relativistic particle*; hep-th/0207274.
C.A. Duval and P. Horvathy, J. Phys. **A34**, 10097 (2001).
L. Faddeev and R. Jackiw, Phys. Rev. Lett **60**, 1968 (1988).
P. Horvathy and M.S. Plyushchay, JHEP **206** 33 (2002).
J. Lukierski, P. Stichel and W.J. Zakrzewski, Ann. Phys. **306**, 78 (2003).
[^1]: Formula (\[luwoeq1\]) is not the most general one. One can assume that the rhs of (\[luwoeq1\]) depends also on momenta (or derivative operators) as well as on other operators, e.g. spin variables. In this note we shall not consider such extensions of (\[luwoeq1\]). The expansion (\[luwoeq1\]) is only up to quadratic term because higher orders do not have classical limit $\kappa \to \infty$.
[^2]: The twisted Poincaré symmetries corresponding to $\theta_{\mu\nu}=\theta^{(0)}_{\mu\nu}$ were earlier discussed in [@asch96; @asch99; @oeckl; @js], but the full consequences of the twisted description were realized in 2004 (see e.g. [@cknt; @jwess; @kt; @koma]).
[^3]: Below, in chapter 2 and 3, we shall use explicitly the *fat dot notation* for the algebra of functions on quantum Minkowski space in order to stress its Hopf algebra module origin.
[^4]: The momenta $p_i,\tilde{p}_i$ are described by the following formulae $$p_i=\frac{\partial\mathcal{L}}{\partial \dot{x}_i}-\frac{d}{dt}\frac{\partial\mathcal{L}}{\partial\ddot{x}_i}\,,\qquad \tilde{p}_i=\frac{\partial\mathcal{L}}{\partial\ddot{x}_i}\,.\nonumber$$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We give a generalization of the duality of a zero-dimensional complete intersection to the case of one-dimensional almost complete intersections, which results in a [*Gorenstein module*]{} $M=I/J$. In the real case the resulting pairing has a signature, which we show to be constant under flat deformations. In the special case of a non-isolated real hypersurface singularity $f$ with a one-dimensional critical locus, we relate the signature on the jacobian module $I/J_f$ to the Euler characteristic of the positive and negative Milnor fibre, generalising the result for isolated critical points. An application to real curves in ${{\mathbb P}}^2({{\mathbb R}})$ of even degree is given.'
author:
- Duco van Straten and Thorsten Warmt
bibliography:
- 'math.bib'
title: 'Gorenstein-duality for one-dimensional almost complete intersections – with an application to non-isolated real singularities'
---
Introduction
============
The algebraic determination of the number of real roots of a polynomial has a long history going back at least to Descartes. Of particular relevance are the methods of Sylvester and Hermite that determine the number of real roots as the [*signature of an associated quadratic form*]{}. For a nice account of the classical approaches we refer to [@weber].
In a similar spirit, the celebrated theorem of Eisenbud-Levine [@eisenbudlevine], and Khimshiashvilli [@khimshiashvilli] provides an algebraic method to determine the local degree of a finite map germ $F:({{\mathbb R}}^n,0){\longrightarrow}({{\mathbb R}}^n,0)$ with component functions $f_1,f_2,\ldots,f_n \in P:={{\mathbb R}}[[x_1,x_2,\ldots,x_n]]$. One considers the local ${{\mathbb R}}$-algebra $${\cal A}_F:=P/(f_1,f_2,\ldots,f_n),$$ which has finite dimension precisely when $f_1,f_2,\ldots,f_n$ form a regular sequence in $P$. Morever, in that case ${\cal A}_F$ is a [*Gorenstein ring*]{}: if we let $$h:=\left| \begin{array}{ccc} \partial_1f_1&\ldots&\partial_n f_1\\
\vdots& &\vdots\\\partial_1 f_n& \ldots&\partial_n f_n \\
\end{array} \right|$$ and choose any linear form $\phi: {\cal M} {\longrightarrow}{{\mathbb R}}$ with $\phi(h) \neq 0$, then the pairing $$B_{\phi}: {\cal A}_F \times {\cal A}_F {\longrightarrow}{{\mathbb R}},\;(a,b) \mapsto \phi(a\cdot b)$$ is non-degenerate.
(Eisenbud-Levine [@eisenbudlevine], Khimshiashvilli [@khimshiashvilli]) If $\phi(h) >0$ then $$Signature(B_{\phi})=Degree(F,0)$$
The theorem is a result of key importance and has been the starting point of many subsequent works. We mention a few of the applications and generalisations.
Consider an isolated complete intersection curve $C=f^{-1}(0)$, where $f:({{\mathbb R}}^n,0) {\longrightarrow}({{\mathbb R}}^{n-1},0)$. According to Aoki, Fukuda and Nishimura [@aokifukudanishimura], one can compute the [*number of real branches of $C$*]{} as follows: consider $g:=x_1^2+x_2^2+\ldots+x_n^2$ and let $$Jac(g):=\left| \begin{array}{ccc} \partial_1 g&\ldots&\partial_n g\\
\partial_1 f_1&\ldots&\partial_nf_1\\\vdots&&\vdots\\\partial_1f_{n-1}&\ldots&\partial_nf_{n-1}\\ \end{array} \right|$$ On the ${{\mathbb R}}$-algebra ${\cal A}_{f,Jac(g)}:=P/(f_1,\ldots,f_{n-1},Jac(g))$ one defines as above a pairing $B_{\phi}$.
(Aoki-Fukuda-Nishimura) $$Signature(B_{\phi})=\textup{Number of real branches of}\;\; C$$
This result was further generalised to the case of arbitrary Gorenstein curve singularities $C$ in [@jamesduco].
In a similar vein, the work [@sza1] associates to a polynomial mapping $f:{{\mathbb R}}^n {\longrightarrow}{{\mathbb R}}^k$ an ${{\mathbb R}}$-algebra ${\cal A}$ with a quadratic form $B_{\phi}$ such that $$Signature(B_{\phi})=\chi(f^{-1}(0)),$$ and further variations can be found in [@sza2], [@sza3], [@dutertre].\
Another type of application is to the topology of real hypersurface singularities. A function $f \in P$ defines an isolated hypersurface singularity precisely when the partial derivatives $f_i:=\partial f/\partial x_i$ form a regular sequence in $P$. In this case the algebra ${\cal A}$ is nothing but the (real) Milnor algebra $P/J_f$, where $J_f$ is the jacobian ideal of $f$. The degree of $F:=(\partial_1f,\ldots,\partial_nf)$ is the Poincaré-Hopf index of the gradient vector field of $f$.
The Milnor ring is the most important algebraic invariant of the singularity $f$. Its dimension $\mu(f):=\dim(P/J_f)$ is the Milnor number, which equals the dimension of the cohomology $H^{n-1}(F)$ of the Milnor fibre $F$, [@agv], [@milnor]. In the real case one can also consider the real Milnor fibres: $$F_{\epsilon,\eta}:=\{ x \in {{\mathbb R}}^n \;|\; \|x\| \le \epsilon, f(x)=\eta\}, \;\; 0 < \epsilon \ll 1,\;\; 0 < \eta \ll \epsilon$$ For $\eta >0$, we put $F_{+}:=F_{\epsilon,\eta}$ and $F_{-}:=F_{\epsilon,-\eta}$ and call them the [*positive*]{} and [*negative*]{} Milnor fibers of $f$.
([@arnold]) Let $f \in {{\mathbb R}}\{x_1,x_2,\ldots,x_n\}$ have an isolated critical point at the origin. Then: $$Signature(f)=-\widetilde{\chi}(F_{-})=(-1)^{n-1}\widetilde{\chi}(F_{+}),$$ where $\widetilde{\chi}$ denotes the reduced Euler characteristic.
Although the positive and negative Milnor fibre have, in general, a quite different topology, it is a simple but remarkable fact that their reduced Euler characteristics are the same, up to a sign. The real $D_4$ surface singularity defined by $x(x^2-y^2)+z^2=0$ may exemplify this.
----------------------- ------------------- -----------------------
negative Milnor fibre $D_4$ singularity positive Milnor fibre
----------------------- ------------------- -----------------------
In this paper we give a partial generalisation of this result to the case where the sequence $f_1,f_2,\ldots,f_n$ defines a [*one-dimensional locus*]{}. In this case one says the the ideal $J=(f_1,\ldots,f_n)$ defines an [*almost complete intersection*]{}. We show that the torsion sub-module $M=I/J$ of $P/J$ is a [*Gorenstein module*]{}. An isomorphism $\phi: P \stackrel{\cong}{{\longrightarrow}} \Omega^n_P$ determines a natural pairing $$B_{\phi}: I/J \times I/J {\longrightarrow}{{\mathbb R}}$$ Its signature is an invariant of the real topology of the situation.
In the case of partial derivatives $f_i=\partial f/ \partial x_i$ of a function $f$ with a one-dimensional critical locus, such modules $I/J_f$ were also considered by Ruud Pellikaan [@pellikaan1] under the name of [*jacobian module*]{}. The case where $I$ is a radical ideal defining a reduced curve, corresponds to $f$ having [*transverse type $A_1$*]{}. We proof the following theorem
Assume that $f \in {{\mathbb R}}\{x_1,x_2,\ldots,x_n\}$ has one-dimensional critical locus and transverse type $A_1$. Assume furthermore that either $f$ has a morsification or that $n=3$. Then
$$2 Signature(B_{\phi})=-\widetilde{\chi}(F_{-})+(-1)^{n-1}\tilde{\chi}(F_{+})$$ wher as before $F_{\pm}$ are the positive and negative Milnor fibres of $f$ and $\widetilde{\chi}$ denotes the reduced Euler characteristic.
Note that contrary to case of isolated singularities, the two Euler characteristics appearing at the right hand side are in general no longer equal up to a sign. It appears that the above statement has a much broader range of validity, and holds for many other transverse types of singularities. However, as stated, it is not true in full generality, but we were not able to identify the precise limits of its validity.\
A nice application arises in the situation of a homogeneous polynomial $f \in {{\mathbb R}}[x,y,z]$ of even degree. It defines a curve $C =\{f=0\} \subset {{\mathbb P}}^2({{\mathbb R}})$, whose complement consists of a part $V_{+}$ where $f>0$ and a part $V_{-}$ where $f<0$.
If $f \in {{\mathbb R}}[x,y,z]$ defines a curve with only ordinary double points, then $$Signature(B_{\phi}(f))=\chi(V_+)-\chi(V_{-})$$
The structure of the paper is as follows. After reviewing the duality in the complete intersection case, we explain the emergence of the Gorenstein module $I/J$ for one-dimensional almost complete intersections and explain how the pairing behaves in a relative situation. Then we revies some material about singularities with one-dimensional singular locus that provide a rich source of examples and explain for special situations the meaning of the resulting signature in the real case,
Duality for zero-dimensional complete intersections
===================================================
Let $P=K[[x_1,x_2,\ldots,x_n]]$ or any other regular local $K$-algebra of dimension $n$. Let there be given a sequence of elements $$f_1,\ldots,f_n \in P$$ and let $J$ be ideal generated by them. We denote the [*Koszul-complex*]{} associated to the sequence ${\bf f}:=f_1,f_2,\ldots,f_n$ by $$K_{\bullet}({\bf f}):=0 \rightarrow K_n \rightarrow K_{n-1} \rightarrow
\cdots \rightarrow K_2 \rightarrow K_1 \rightarrow K_0 \rightarrow 0$$ Its terms are $K_p:= \bigwedge^p P^n$, so that $K_0=P$. The differentials are induced by sending the $i$-th basis vector $e_i$ of $P^n$ to $f_i \in P$. We denote its homology groups by $$H_k({\bf f}):=H_k(K_{\bullet}({\bf f}))$$ The following result is well-known (see e.g. [@brunsherzog], thm. 1.6.17 and thm. 2.1.2).
\[vanishing\] Let $V(J) \subset Spec(P)$ the variety defined by the ideal $J$ and let $d:=dim(V(J))$. Then $H_k({\bf f})=0$ for $k >d$ and $H_d({\bf f}) \neq 0$
Let us first look at the case where $\dim(J)=0$. Then the above result tells us that $H_0({\bf f})=P/J$ and $H_k({\bf f})=0$ for $k>0$. So ${\bf f}$ is a regular sequence and the Koszul complex provides a free resolution of $P/J$ as a $P$-module. As the transpose of the map $ K_n {\longrightarrow}K_{n-1}$ can be identified, up to signs, with the map $K_1 {\longrightarrow}K_0$, we obtain an isomorphism $$Ext^n_P(P/J,P) \cong P/J$$
Recall Grothendiecks [*local duality theorem*]{} (see e.g. [@brunsherzog], thm. 3.5.8):
\[localduality\] Let $P$ be a $n$-dimensional local ring with maximal ideal ${{\mathfrak{m}}}$ and dualizing module $\omega_P$. For any finitely generated $P$-module $M$, there exists a natural non-degenerate pairing $$H^i_{{\mathfrak{m}}}(M) \times Ext^{n-i}_P(M,\omega_P) {\longrightarrow}H^n_{{\mathfrak{m}}}(\omega_P) \stackrel{tr}{{\longrightarrow}} K$$
In particular, for $M=P/J$ and $i=0$ we obtain a non-degenerate pairing $$P/J \times Ext^n_P(P/J,\omega_P) {\longrightarrow}K$$ as $H^0_{{\mathfrak{m}}}(P/J)=P/J$. The choice of an isomorphism $\phi: P \stackrel{\cong}{{\longrightarrow}} \omega_P$ induces isomorphisms $$Ext^n_P(P/J,\omega_P)\cong Ext^n_P(P/J,P)\cong P/J$$ then hence provides us with a perfect pairing $$B_{\phi}: P/J \times P/J {\longrightarrow}K$$ As $B_{\phi}$ is $P$-linear, it follows that it factors over the multiplication map and is of the form $B_{\phi}(a,b)=\beta(a\cdot b)$ for some linear form $$\beta: P/J {\longrightarrow}K$$ The linear space $ker(\beta)$ is the [*socle*]{} of $P/J$, that is, its unique minimal ideal. There is a classical result of Scheja and Storch that states that this socle has a [ canonical generator]{}.
[@schejastorch] If $char(K)=0$ and $x_1,\ldots,x_n$ are generators of the maximal ideal, then the Jacobian determinant
$$h:=\left|\begin{array}{ccc} \frac{\partial f_1}{\partial x_1}& \ldots &\frac{\partial f_1}{\partial x_n}\\
\vdots&&\vdots\\
\frac{\partial f_n}{\partial x_1} &\ldots& \frac{\partial f_n}{\partial x_n}\\
\end{array}\right|$$ is a generator for the socle of $P/J$.
In the case where $f \in P$ and $f_i=\partial f/\partial x_i$, the complex can be identified (up to some signs) with the $df \wedge$-complex $$0 {\longrightarrow}P \stackrel{df \wedge }{{\longrightarrow}}\Omega^1 \stackrel{df\wedge}{{\longrightarrow}} \ldots {\longrightarrow}\Omega^{n-1}
\stackrel{df \wedge}{{\longrightarrow}}\Omega^n {\longrightarrow}0$$ and the natural pairing takes the form $$B: \Omega^n/df\wedge \Omega^{n-1} \times\;\; \Omega^n/df\wedge \Omega^{n-1} {\longrightarrow}K$$ It is usually called [*residue pairing*]{} and has an analytic expression as $$B(P\omega,Q\omega)=\frac{1}{(2\pi i)^n}\int_{T_{\epsilon}} \frac{PQ\omega}{\partial_1f\ldots \partial_n f}$$ where $\omega:=dx_1\wedge dx_2 \wedge \ldots \wedge dx_n$ and $T_{\epsilon}:=\{x\;|\;|\partial_i f|=\epsilon\}$, $0<\epsilon \ll 1$. In the papers [@giventalvarchenko], [@varchenko] the relation between this pairing and the Poincaré-pairing in the cohomology of the Milnor fibre is described. The pairing is also the first in a sequence of [*higher residue pairings*]{}, introduced by K. Saito, which express the self-duality of the Gau[ß]{}-Manin system of the singularity, [@saito].
Homology and Cohomology
=======================
We have seen that in the case of a regular sequence ${\bf f}$, the origin of the pairing on $P/J$ lies in the self-dual nature of the Koszul complex. We now investigate in general what consequence this self-duality has for the Koszul homology groups.
\[selfdual\] (c.f. [@eisenbud], prop. 17.15) Let ${\bf f}=f_1,f_2,\ldots,f_n$ be a sequence of elements in a ring $P$. There are isomorphisms $$\alpha_p: H_p\left({\bf f}\right)\cong H^{n-p}\left({\bf f}\right), \;\;p=0,1,\ldots,n,$$ where $H^p({\bf f}):= H^p(Hom_P\left(K_{\bullet}\left({\bf f}\right),P\right))$ is the Koszul cohomology.
Using the basis $e_1,e_2,\ldots,e_n$ for the free module $P^n$ and the dual basis $\phi_1,\ldots,\phi_n$ for the dual module $Hom_P(P^n,P)$, one obtains basis elements $e_I:=e_{i_1}\wedge e_{i_2} \wedge \ldots \wedge e_{i_p}$ for $\bigwedge^p P^n$ and $\phi_I=\phi_{i_1}\wedge\ldots \wedge\phi_{i_p}$, $(I=(i_1,i_2,\ldots,i_p))$ for the dual module $Hom_P(\bigwedge^p P^n,P)$. Define isomorphisms $\alpha_p:\bigwedge P^p {\longrightarrow}(\bigwedge^pP^n)^*:=Hom_P(\bigwedge^{n-p}P^n,P)$ by setting $\alpha_p(e_I)=sign(\sigma)\phi_J$ where $J$ is the sequence of indices complementary to $I$ and $sign(\sigma)$ is the sign of the permutation that puts the sequence $(I,J)$ into $(1,2,\ldots,n)$. One verifies that in this way one obtains a mapping between complexes $\bigwedge^{\bullet} P^n$ and $(\bigwedge^{\bullet} P^n)^*$.
We refer to [@eisenbud], section 17.4 for more details.
The modules $H_i(\bf f)$ and $H^i(\bf f)$ depend only on the ideal $J$ generated by $f_1,f_2,\ldots,f_n$. Note however, that changing the order changes $\alpha_p: H_p({\bf f}){\longrightarrow}H^{n-p}({\bf f})$ by a sign.
In general cohomology is also [*dual*]{} to homology, in the following sense.
Let $P$ be a ring, $(F_{\bullet},\partial_{\bullet})$ a complex of free $P$-modules and $M$ a $P$-module. Then there exists a spectral sequence with $$E_2^{p,q}=Ext^p_P(H_q(F_{\bullet}),M)\Longrightarrow H^{p+q}(Hom_P(F_{\bullet},M)).$$
This is of course classical, see for example [@eilen section XVI.2]. It can be shown as follows: consider an injective resolution $(G^{\bullet},\delta^{\bullet})$ of $M$ and the double complex $C_{\bullet,\bullet}$ with terms $C_{p,q}=Hom_P(F_{q}, G^{p})$ and differentials induced from $\partial$ (going up. increasing $q$) and $\delta$ (going right, increasing $p$). The spectral sequence obtained by first taking $\delta$ and then $\partial$ degenerates at $E_2$ and has $H^p(Hom_P(F_{\bullet},M))$ at spot $(p,0)$ as homology. The other spectral sequence obtained by first taking $\partial$ and followed by $\delta$ has as $E_2$-term $$\begin{array}{c|cccc}
q&Hom_P(H_q, M)&Ext^1_P(H_q, M)&\ldots&Ext^p_P(H_q, M)\\
\vdots&\vdots&\vdots& &\vdots\\
1&Hom_P(H_1, M)&Ext_P^1(H_1, M)&\ldots&Ext^p_P(H_1, M)\\
0&Hom_P(H_0, M)&Ext_P^1(H_0, M)&\ldots&Ext^p_P(H_0, M)\\
\hline
&0&1&\ldots&p
\end{array}$$ where $H_i:=H_i(F_{\bullet})$.
By taking $M=P$, this spectral sequence can be used to connect the homology and the cohomology of a complex. As a corollary, we note the two special situations which lead to exact sequences.
\[corollaryspek\] (i) If the groups $Ext^k_P(H_i(F_{\bullet}),P)$ vanish for $k>1$, the spectral sequence collapses to exact sequences $$0{\longrightarrow}Ext^1_P(H_k(F_{\bullet}),P){\longrightarrow}H^{k+1}(F^{\bullet}){\longrightarrow}Hom_P(H_{k+1}(F_{\bullet}),P){\longrightarrow}0.$$ (“universal coefficient theorem”)\
(ii) If the homology groups $H_k:=H_k(F_{\bullet})$ vanish for $k>1$, we get a long exact sequence $$\cdots {\rightarrow}Ext^k_P(H_0,P){\rightarrow}H^k {\rightarrow}Ext^{k-1}_P(H_1,P)\stackrel{d_2}{{\rightarrow}} Ext^{k+1}_P(H_0,P){\rightarrow}\cdots$$ where $H^i:=H^i(F^{\bullet})$.
We specialise the above to the case of an almost complete intersection.
A sequence ${\bf f}=f_1,f_2,\ldots,f_n$ of elements in a ring $P$ is said to defines an [*almost complete intersection*]{} if $codim(V(J))=n-1$.
It follows from \[vanishing\] that in this case one has two non-vanishing Koszul homology groups, $H_0({\bf f})=P/J$ and one further module $H_1({\bf f})$.
\[aci\] Let ${\bf f}=f_1,f_2,\ldots,f_n$ define an almost complete intersection in $P$ and put $H_i:=H_i({\bf f})$. Then:\
(i) $$Ext^{n-1}_P(H_0,P)\cong H_1$$ (ii) There is an exact sequence $$0 {\longrightarrow}Ext^{n}_P(H_0,P) {\longrightarrow}H_0 {\longrightarrow}Ext^{n-1}_P(H_1,P) \stackrel{d_2}{{\longrightarrow}} Ext_P^{n+1}(H_0,P) {\longrightarrow}0$$ (iii) For $k \ge n$ there are isomorphisms $$Ext^{k}_P(H_1,P) \stackrel{d_2}{{\longrightarrow}} Ext_P^{k+2}(H_0,P)$$
The self-duality \[selfdual\] of the Koszul complex gives $H^{n-1}\left({\bf f}\right)=H_1\left({\bf f}\right)$ and $H^n\left({\bf f}\right)=H_0\left({\bf f}\right)$ as only non-vanishing Koszul cohomology groups. As both modules are supported on $V(J)$, it follows from Ischebeck’s lemma that $Ext^p_P(H_i({\bf f}),P)=0$, for $p<n-1$. The theorem follows then immediately from the above long exact sequence \[corollaryspek\] (ii).
One-dimensional almost complete intersections
=============================================
We now make the further assumption, that $dim(P)=n$, so that the almost complete intersection ${\bf f}:=f_1,\ldots,f_n$ gives an ideal $J=(f_1,f_2,\ldots,f_n)$ that defines a one-dimensional locus $V(J) \subset Spec(P)$.
\[koszhom\] Let ${\bf f}:=f_1,\ldots,f_n$ define a one-dimensional almost complete intersection. Then:\
(i) $$Ext^{n-1}_P(H_0({\bf f}),P)\cong H_1({\bf f}),\;\;Ext_P^{n}(H_1({\bf f}),P)=0.$$ (ii) There is an exact sequence $$0 {\longrightarrow}Ext^{n}_P(H_0,P) {\longrightarrow}H_0({\bf f}){\longrightarrow}Ext^{n-1}_P(H_1({\bf f}),P) {\longrightarrow}0$$
As $P$ is a assumed to be a regular local ring of dimension $n$, we have $Ext^k_P(-,P)=0$ for $k > n$. So the result follows from the above proposition \[aci\]
\[imodj\] For a one-dimensional almost complete intersection defined by ${\bf f}$ we put $$M:=Ext^{n}_P(H_0({\bf f}),P)$$ As $Ext^{n}_P(H_0({\bf f}),P) \subset H_0({\bf f})=P/J$ we see that $$M=I/J$$ for an ideal $I \subset P$.
In other words, $I$ is defined by having an isomorphism of exact sequences $$\begin{array}{cccccccccc}
0 &{\longrightarrow}&Ext^{n}_P(H_0({\bf f}),P)& {\longrightarrow}&H_0({\bf f})&{\longrightarrow}& Ext^{n-1}_P(H_1({\bf f}),P)& {\longrightarrow}& 0\\
&&\cong\downarrow&&\cong \downarrow&&\cong\downarrow&&\\
0&{\longrightarrow}&I/J&{\longrightarrow}&P/J&{\longrightarrow}&P/I&{\longrightarrow}& 0\\
\end{array}$$
\[saturation\] The ideal $I$ is equal to the [*saturation*]{} of $J$ with respect to the maximal ideal: $$I=\bigcup_k(J:\mathfrak{m}^k) \supset J$$ Thus the module $M=I/J$ is the *${\mathfrak{m}}$-torsion submodule of $P/J$*: $$M=I/J=H^0_{{\mathfrak{m}}}(P/J)$$
Because $Ext^{n}_P(H_1({\bf f}),P)=0$, we see that the module $H_1({\bf f})$ has $Ext^{n-1}_P(H_1({\bf f}),P)=P/I$ as the only non-vanishing $Ext$. Hence it is a Cohen-Macaulay $P$-module of dimension one. It follows that $H^0_{{\mathfrak{m}}}(P/I)=0$ and thus the artinian module $M=Ext^{n}_P(H_0({\bf f}),P)$ is equal to $H^0_{{\mathfrak{m}}}(P/J)$. Hence $I$ is nothing but the saturation of $J$ with respect to ${{\mathfrak{m}}}$.
In fact this provides an easy was to compute $I$ and $I/J$ using a computer algebra system like [Singular]{} or [Macaulay]{}. The ring $P/I$ obtained by dividing out the [**m**]{}-primary torsion of $P/J$ is sometimes called the [*Cohen-Macaulayfication*]{} of $P/J$.
Using $I/J$ and $P/I$ we can give a slight reinterpretation of the $Ext$ of Koszul homology.
\[extiso\] The map $P/J {\longrightarrow}P/I$ induces an isomorphism $$Ext^{n-1}_P(P/I,P) \cong Ext^{n-1}_P(P/J,P)$$ and the map $I/J {\longrightarrow}P/J$ induces an isomorphism $$Ext^{n}_P(P/J,P) \cong Ext^n_P(I/J,P)$$
Apply $Hom_P(\bullet,P)$ to the short exact sequence of $P$-modules $0{\longrightarrow}I/J{\longrightarrow}P/J{\longrightarrow}P/I{\longrightarrow}0$. As $I/J$ is ${\mathfrak{m}}$-torsion and $P/I$ is ${{\mathfrak{m}}}$-torsion free the result follows.
We now have all the ingredients for the following central result.
\[isoextij\] Let the sequence ${\bf f}=f_1,\ldots,f_n$ define a one-dimensional almost complete intersection $J$, and $M=I/J=H^0_{\bf m}(P/J)$. Then there is an isomorphism $$Ext^n_P(M,P) \cong M.$$
Combining theorem \[extiso\] and theorem \[koszhom\](i) we obtain an isomorphism $$Ext^{n-1}_P(P/I,P)\cong Ext^{n-1}_P(P/J,P) \cong H_1({\bf f}),$$ showing that $H_1({\bf f})$ is isomorphic to the dualising module of $P/I$. Because $P/I$ is Cohen-Macaulay, we have $$Ext^{n-1}_P(H_1({\bf f}),P)\cong Ext^{n-1}_P(Ext^{n-1}_P(P/I,P),P) \cong P/I.$$
Combining this with \[koszhom\] (ii) we see that $$Ext^n_P(P/J,P) \cong ker(P/J {\rightarrow}P/I)=I/J.$$
But by \[extiso\] we also have $Ext^n_P(I/J,P) \cong Ext^n_P(P/J,P)$ and as a result we obtain an isomorphism $$Ext^n_P(I/J,P) \cong I/J$$
As a corollary one finds:
\[pairing\] The module $M=I/J=H_{{\mathfrak{m}}}^0(P/J)$ is an artinian Gorenstein module. The choice of an isomorphism $\phi: P \stackrel{\cong}{{\longrightarrow}} \omega_P $ determines a non-degenerate pairing $$B_{\phi}: M \times M {\longrightarrow}K.$$
From local duality \[localduality\] we obtain a non-degenerate pairing $$M \times Ext^n(M,\omega_P) {\longrightarrow}H^n_{{\mathfrak{m}}}(\omega_P)\stackrel{tr}{{\longrightarrow}} K$$ The choice of an isomorphism $\phi: P \stackrel{\cong}{{\longrightarrow}} \omega_P $ determines an isomorphism $Ext^n_P(I/J,P) \cong Ext^n_P(M,\omega_P)$ an thus a non-degenerate pairing $B_{\phi}: M \times M {\longrightarrow}K.$
\(i) The material of this section should be rather well-known. For example, the isomorphism $H_1=Ext_P^{n-1}(P/I,P)$ can be found in [@pellikaan3], but the self-duality of $I/J$ seems to have escaped attention. It is reflected in the readily observed symmetry of the Hilbert-Poincaré polynomial.\
(ii) Using a computer algebra system, the duality pairing can in be calculated by running through the appropriate sequences and isomorphisms. This was implemented in [Singular]{} and is described in some detail in the thesis of the second author, [@egodiss]. As a simple example, for ${\bf f}=(xy,x^2) \subset K[[x,y]]$, $\phi(1)=dx\wedge dy$ one has $I=(x)$, $\dim(I/J)=1$ and $B_{\phi}([x],[x])=1$.\
(iii) The above can be generalised, with almost identical proof, to the case of a one-dimensional complete intersection in an arbitrary Gorenstein ring $P$. If $\omega_P$ denotes the dualising module of $P$, then one obtains a natural pairing $$B: {\cal M} \times {\cal M} {\longrightarrow}K$$ where ${\cal M}=H_{\bf m}^0(\omega_P/J\omega_P)$.\
(iv) In the special case of hypersurfaces singularity with one-dimensional singular locus, these modules $I/J$ were studied in [@pellikaan1] under the name of [*Jacobian modules*]{} and play a rôle that can be compared to that of the Milnor ring in the isolated case. See also section 6.\
(v) In the case of hypersurface singularities with one-dimensional singular locus it is again more natural to consider the cohomology $H^n$ and $H^{n-1}$ of the $df \wedge$-complex. On the ${\mathfrak{m}}$-torsion submodule of $H^n=\Omega^n/df \wedge \Omega^{n-1}$ (isomorphic to $I/J$) there again is a pairing that does not involve any choice. In this situation there is also a map $d:H^{n-1}{\longrightarrow}H^n$ induced by exterior differentiation. That map plays a role in the Gauss-Manin system of $f$, [@vanstraten].\
(vi) Hypersurfaces in projective space with isolated singularities correspond to homogeneous singularities with one-dimensional singular locus. The module $I/J$ plays a role in the Dwork-Griffiths description of the Hodge-pieces of the cohomology. For example, let $f \in {{\mathbb C}}[x_0,x_1,x_2,x_3,x_4]$ be the equation of a projective threefold $X=V(f) \subset {{\mathbb P}}^4$ of degree $d$ with only nodes as singularities, and let $\pi:Y {\longrightarrow}X$ a small resolution. Then the degree $a=2d-5$ part $(I/J)_a$ of $I/J$ can be identified with $H^{2,1}:=H^1(\Omega^2_Y)$ via $$A {\longrightarrow}\pi^*(Res(\frac{A\Omega}{f}))$$ where $\Omega:=\iota_E(dx_0 \wedge \ldots \wedge dx_4)$. Similarly, the degree $b=3d-5$ part of $I/J$ is identified with $H^{1,2}:=H^{2}(\Omega^1_Y)$, making a commutative diagram $$\begin{array}{ccccc}
(I/J)_a & \times & (I/J)_b & {\longrightarrow}&{{\mathbb C}}\\
\downarrow &&\downarrow&&\cong \downarrow\\
H^{2,1}&\times&H^{1,2}&\stackrel{\cup}{{\longrightarrow}}&H^{3,3}\\
\end{array}$$ For details we refer to [@dimcasaitowotzlaw].
Behaviour under flat deformations
=================================
We now study the behaviour of the module $I/J$ under deformation. By this we mean that we let the almost complete interesection $f_1,\ldots,f_n$ depend on additional parameters. If we require $H_0=P/(f_1,f_2,\ldots,f_n)$ to deform in a flat way, then the same will be true for $I/J$.\
Consider a local ring $S$ with maximal ideal ${\mathfrak{m}}_S$ and $K=S/{\mathfrak{m}}_S$ and let $R$ be a flat $S$-algebra such that $P=K \otimes_S R$. A sequence ${\bf F}:=F_1,F_2,\ldots,F_n \in R$ is called relative almost complete intersection if\
1) $$f_i:=F_i\;\; \textup{mod}\;\;{\mathfrak{m}}_S$$ 2) $H_0({\bf F})$ is $S$-flat.
\[flat\] If ${\bf F}$ defines a relative almost complete intersection, then $Ext_R^n(H_0({\bf F}),R)$ is free as $S$-module and $Ext_R^k(H_0({\bf F}),R)=0$ for $k \ge n+1$.
This follows from a cohomology-and-base change argument. We assume for simplicity $S=K[[t]]$, so that we have an exact sequence $0{\longrightarrow}R \stackrel{t\cdot}{{\longrightarrow}} R {\longrightarrow}P {\longrightarrow}0$. As ${\bf F}$ defines an almost complete intersection in $R$, we have as before only two Koszul groups $H_0:=H_0({\bf F})$ and $H_1:=H_1({\bf F})$, and by assumption we have an exact sequence $$0 {\longrightarrow}H_1 \stackrel{t\cdot}{{\longrightarrow}} H_1 {\longrightarrow}\overline{H}_1 \stackrel{0}{{\longrightarrow}} H_0 \stackrel{t\cdot}{{\longrightarrow}} H_0 {\longrightarrow}\overline{H}_0 {\longrightarrow}0$$ where $\overline{H}_k:=H_k({\bf f})$. Furthermore, we have the statements from \[aci\]. We will show that $Ext^{n+1}_R(H_0,R)=0$, as are all further higher $Ext^{k}_R(H_1,R) \cong Ext^{k+2}_R(H_0,R)$, $k \ge n$. For this, note that one obtains a long exact sequence $$\ldots {\longrightarrow}E^{n+1} \stackrel{t\cdot}{{\longrightarrow}} E^{n+1} {\longrightarrow}\overline{E}^{n+1}{\longrightarrow}E^{n+2} \stackrel{t\cdot}{{\longrightarrow}} E^{n+2} {\longrightarrow}\overline{E}^{n+2}{\longrightarrow}\ldots$$ where we put temporarily $E^k:=Ext^k_R(H_0,R)$, $\overline{E}^k:=Ext_P(\overline{H}_0,P)$. As $\overline{E}^k=0$ for $k \ge n+1$ and the modules are $S$-finite, one concludes with Nakayama that all $E^{k}=0,k \ge n+1$. As $H_1=E^{n-1}$ it follows that we have an exact sequence $$0 {\longrightarrow}E^{n-1} \stackrel{t\cdot}{{\longrightarrow}} E^{n-1} {\longrightarrow}\overline{E}^{n-1} \stackrel{0}{{\longrightarrow}} E^{n} \stackrel{t \cdot}{{\longrightarrow}} E^{n} {\longrightarrow}\overline{E}^{n}{\longrightarrow}0$$ hence we find that $E^n=Ext_P^n(H_0({\bf F}),P)$ and $S$-flat, and hence $S$-free.
As in \[koszhom\] we have $$Ext^{n-1}_R(H_0({\bf F},R))=H_1({\bf F}),\;\;\;Ext^n_R(H_1({\bf F}),R)=0$$ $$0 {\longrightarrow}Ext_R^n(H_0({\bf F}),R) {\longrightarrow}H_0({\bf F}){\longrightarrow}Ext^{n-1}_R(H_1({\bf F}),R) {\longrightarrow}0$$
\[relativemodule\] In the above situation we put $$M_S:=Ext^n_R(H_0({\bf F}),R)$$
As $H_0({\bf F})=R/J_S$, we see as before from the exact sequence that there exists an ideal $I_S \subset R$ such that $M_S=I_S/J_S$. Completely analoguous to \[isoextij\] we have
\[isoextrelativ\] $$Ext^n_R(M_S,R)=M_S$$
We omit the proof, that is identical to that of \[isoextij\].
However, we want to understand this duality in terms of a [*family of pairings*]{}, parametrised by $S$. We can not just apply the local duality theorem, but rather we would have to use the duality statement for the morphism $Spec(R) {\longrightarrow}Spec(S)$.\
To treat this in an elementary way, we express higher $Ext$-groups $Ext_R^i(M,R)$ as groups $Hom_{\overline{R}}(M,\overline{R})$ of homomorphisms, where $\overline{R}$ arises from $R$ by dividing out suitable elements. In this way we can reduce to the case of a finite ring extension and use the duality for a finite map, a change-of-rings isomorphism. First we recall
\[exthom\] Let $R$ be a ring and let $M$, $N$ be two non-trivial finite $R$-modules. If $\textup{ann}(M)+\textup{ann}(N)=R$, $Ext_R^i(M,N)$ is zero for all $i\in{{\mathbb N}}$. Otherwise $$d=\textup{depth}(\textup{ann}(M),N)$$ is the smalles number $i$ with $Ext^i_R(M,N)$ not equal to zero. If $t_1,\ldots,t_d$ is a maximal regular $N$-sequence in $\textup{ann}(M)$ and we define $$\overline{R}=R/(t_1,\ldots,t_d), \overline{N}=N/(t_1,\ldots,t_d)N,$$ then there is an isomorphism $$Ext^d_R(M,N) \cong Hom_{\overline{R}}(M,\overline{N}).$$
This is well-known. The first part of the statement is essentially [@brunsherzog], 1.2.10 and the rest follows by induction from the long exact Ext-sequence, obtained by dividing out an element.
\[pairingfamily\] An isomorphism $\phi: R \stackrel{\cong}{{\longrightarrow}} \omega_{R/S}$ defines a bilinear pairinng $$B_{\phi}: M_S \times M_S {\longrightarrow}S$$ This pairing is non-degenerate in the sense that the adjoint map $$M_S \cong Hom_S(M_S,S)$$ is an isomorphism of $S$-modules.
From \[isoextrelativ\] there is is an isomorphism $M_S \cong Ext^n_R(M_S,R)$. We take a maximal regular sequence $t_1,\ldots,t_d$ in the annihilator of $M_S$. We divide out these elements and obtain a factor ring $\overline{R}$ of $R$ and applying \[exthom\] we get an isomorphism $Ext^n_R(M_S,R)= Hom_{\overline{R}}(M_S,\overline{R})$. As $M_S$ is a finite $S$-module, the ring $\overline{R}$ is finite over $S$. From the isomorphism $\phi: R \stackrel{\cong}{{\longrightarrow}} \omega_{R/S}$ we obtain by dividing out $t_1,\ldots,t_d$ an isomorphism $\overline{R} \cong \omega_{\overline{R}/S}$. Duality for the finite map $S {\longrightarrow}\overline{S}$ tells us $\omega_{\overline{R}/S}=Hom_S(\overline{R},S)$. From the change-of-rings isomorphism we get $$Hom_{\overline{R}}\left(M_S,Hom_S\left(\overline{R},S\right)\right) \cong Hom_{S}(M_S,S).$$ Combining these isomorphisms we obtain $$Ext_R^n(M_S,R)\cong Hom_{\overline{R}}(M_S,\overline{R})\cong Hom_S(M_S,S)$$ Hence, in total we obtain a natural isomorphism $M_S\cong Hom_S\left(M_S,S\right)$, which can be seen as a family of non-degenerate pairings $B_{\phi}: M_S\times M_S{\longrightarrow}S.$
In the case $S=K[[t]]$ we obtain a commutative diagram of the following form $$\begin{array}{ccccccccc}
0&{\longrightarrow}&M_S &\stackrel{t\cdot}{{\longrightarrow}}&M_S &{\longrightarrow}& M_0 &{\longrightarrow}&0\\
& &\uparrow\cong& &\uparrow \cong& &\uparrow \cong& &\\
0&{\longrightarrow}&Ext^n_R(M_S,R)&\stackrel{t\cdot}{{\longrightarrow}}&Ext^n_R(M_S,R) &{\longrightarrow}& Ext^n_P(M_0,P)&{\longrightarrow}&0\\
& &\uparrow\cong& &\uparrow \cong& &\uparrow \cong& &\\
0&{\longrightarrow}&Hom_S(M_S,S) &\stackrel{t\cdot}{{\longrightarrow}}&Hom_S(M_S,S) &{\longrightarrow}& Hom_K(M_0,K) &{\longrightarrow}&0\\
\end{array}$$
Application to non-isolated hypersurface singularity with one-dimensional critical locus
========================================================================================
Hypersurfaces with one-dimensional singular locus
-------------------------------------------------
Following [@pellikaan1], the [*primitive*]{} of an ideal $I \subset P:={{\mathbb C}}\{x_1,\ldots,x_n\}$ is the ideal $$\int I:=\{f \in P\;|\;(f,\partial_1f,\ldots,\partial_nf)\subset I\}$$ This ideal arises when studying functions $f$ which contain a specific sub-space $\Sigma:=V(I)$ inside their critical locus. One can pursue the classification program of singularity theory in this context and Siersma [@siersma] started the investigation of [*line-singularities*]{}, which correspond to the case where $I$ is a radical ideal defining a line. The [*extended $I$-codimension*]{} of a function $f \in \int I$ is defined as $$c_{e,I}(f):=dim_{{{\mathbb C}}}(\int I/\int I \cap J_f)$$ and one can try to classify the cases of low codimension. Here $J_f=(\partial_1f,\ldots\partial_nf)$ is the jacobian ideal of $f$, [@pellikaan1]. As $J_f \subset I$, one can consider the associated [*jacobian module*]{} $I/J_f$.
We will assume from now on that $I$ is a radical ideal, defining a curve germ $\Sigma \subset ({{\mathbb C}}^n,0)$. On has the following basic result.
(Pellikaan, [@pellikaan1], prop. 1.7) The following statements about a function $f \in \int I$ with jacobian ideal $J_f$ are equivalent:\
1) $\dim(I/J_f) <\infty$\
2) $c_{e,I}(f) < \infty$\
3) The singular locus of $f$ is $\Sigma$ and $f$ has only $A_1$-singularities transverse to $\Sigma\setminus\{0\}$.
Note that in this situation $I=rad(J_f)$, which is the same as the saturation of $J_f$ with respect to ${\mathfrak{m}}$. So the jacobian module $I/J_f$ is precisely the module considered in \[imodj\] for the sequence $f_1,f_2,\ldots,f_n$, with $f_i=\partial f/\partial x_i$.\
There is a formula expressing $\dim I/J$ in terms of $c_{e,I}$ and some other invariant that we explain now. The two dual exact sequences $$I/I^2 \stackrel{d}{{\longrightarrow}} \Omega \otimes \mathcal{O}_{\Sigma} {\longrightarrow}\Omega_{\Sigma} {\longrightarrow}0$$ $$0 {\longrightarrow}\Theta_{\Sigma} {\longrightarrow}\Theta \otimes \mathcal{O}_{\Sigma} \stackrel{d^*}{{\longrightarrow}} Hom(I/I^2,\mathcal{O}_{\Sigma})$$ provide important invariants of the curve $\Sigma$.\
$\bullet$ The first tangent homology is $T_1(\Sigma)=ker(d)=\int I/I^2$. In case $I$ is a reduced complete intersection, one has $\int I=I^2$. But for $I=(xy,yz,zx) \subset {{\mathbb C}}\{x,y,z\}$ one has $\int I=(xyz,I^2)$, so $\int I/I^2$ is one-dimensional.\
$\bullet$ The first tangent cohomology if $T^1(\Sigma)=Coker(d^*)$. It is the space of first order infinitesimal deformations of $\Sigma$. The normal module $N=Hom_{\Sigma}(I/I^2,{\cal O}_{\Sigma})$ is identified with the space of embedded deformations of $\Sigma$.\
$\bullet$ Dualising once more, we obtain a double duality map $ I/I^2 {\longrightarrow}N^*$, where $N^*:=Hom_{\Sigma}(N,P/I)$. The kernel is again $\int I/I^2$, the cokernel $N^*/I:=N^*/(I/\int I)$ is a further invariant. For $I=(xy,xz,zx)$ one finds $\dim(N^*/I)=3$.\
$\bullet$ In the special where $\Sigma \subset ({{\mathbb C}}^3,0)$ is a space curve, $I$ is Cohen-Macaulay of codimension $2$. Such curves are syzygetic ($T_2(\Sigma)=0$) and unobstructed ($T^2(\Sigma)=0$). Furthermore, one has in that case $$\int I/I^2=Ext^1({\cal O}_{\Sigma},{\omega_{\Sigma}}),\;\;\; N^*/I=Ext^2({\cal O}_{\Sigma},{\omega_{\Sigma}})$$
$\bullet$ In [@dJ] and [@dJdJ] an invariant $VD_{\infty}(f)$ called the virtual number of $D_{\infty}$-points was defined. In case that $f \in I^2$ Pellikaan [@pellikaan2] (and more generally in [@dJvS1]), the expression of $f$ as a quadratic form in the generators of $I$ can be used to define a [*transverse Hessian map*]{} $$H: N \times N {\longrightarrow}{\cal O}_{\Sigma}$$ which defines by transposition a map $$h: N {\longrightarrow}N^*$$ which has a finite cokernel in case $f$ has transverse $A_1$-singularities.
([@pellikaan2], [@dJvS1]) Let the radical ideal define a curve $\Sigma$ and let $f \in \int I$ have transverse type $A_1$. Assume furthermore that $\Sigma$ is smoothable, with $T_2(\Sigma)=T^2(\Sigma)=0$. Then: $$\dim(I/J_f)=c_{e,I}(f)+\dim(T^1(\Sigma))+VD_{\infty}(f)-\dim(\int I/I^2)$$ $$VD_{\infty}(f)=\dim(N/h(N))-\dim(N^*/I)+\dim(\int I/I^2)$$
These invariants have furthermore an interpretation in terms of deformation theory. An [*admissible deformation*]{} of the pair $f, \Sigma$ over a base $S$ consist a flat deformation of $\Xi {\longrightarrow}S$ of $\Sigma$, together with a deformation of the $F:({{\mathbb C}}^n \times S,0){\longrightarrow}({{\mathbb C}}\times S,0)$, such that $\Xi$ is contained in the critical locus of $F$. We refer to [@pellikaan2] and [@dJvS1] for more details. There is a notion of good represenetatives for the germs involved and such a good representative of a one-parameter deformation $F$, $\Xi$ is called a [*morsification*]{}, if for $t \neq 0$ the curve $\Xi_t$ is smooth and the critical points of $F_t: X {\longrightarrow}{{\mathbb C}}$ are of the simplest possible type, namely $A_1$, $A_{\infty}$ or $D_{\infty}$. If the curve $\Sigma$ defined by $I$ is smoothable, and $f \in I^2$, then morsifications do exist ([@pellikaan2], prop. 3.4). But even for functions of three variables morsifications do not alway exist, as triple points $f=xyz$ generically occur. Allowing for these leads to the notion of [*disentanglement*]{}, but even these do not always exists, as projections of non-smoothable normal surfaces singularities to $({{\mathbb C}}^3,0)$ show. The following result can be found in [@pellikaan2].
(Pellikaan, [@pellikaan2], prop. 2.19) If $f$ posesses a morsification, then $$\dim(I/J_f)=\# A_1 +\# D_{\infty}$$
where $\# A_1$ and $\# D_{\infty}$ denote the number of these singularities appearing in a morsification.
(sketch) Consider a morsification over $S$. The construction of $I/J$ in the relative case \[relativemodule\] sheafifies in an obvious way to produce a sheaf ${\cal M}_S={\cal I}_S/{\cal J}_S$. Result \[flat\] implies that $\pi_*({\cal M}_S)$ is a free ${\cal O}_S$-module of finite rank, equal to $\dim(I/J)$. The freeness implies that this number is also equal to sum of the local contributions for the function $F_t$. A local calculation shows that both an $A_1$- and $D_{\infty}$-point give a contribution of $1$, hence the fomula follows.
In a similar vein, T. de Jong has shown that for a disentaglement of $f$ one has $VD_{\infty}(f)=\# D_{\infty}-2\# T$ where $T$ denotes the triple point $f=xyz$.
A signature theorem for functions
---------------------------------
We now consider a real function $f \in {{\mathbb R}}\{x_1,x_2,\ldots,x_n\}$ with jacobian ideal $J_f$ with $\dim(V(J_f))=1$. We let $I$ be the saturation of $J_f$ and consider the jacobian module $I/J_f$. Using the sequence ${\bf f}=f_1,f_2,\ldots,f_n$ and the isomorphism $\phi: P {\longrightarrow}\Omega^n$ given by $1 \mapsto dx_1\wedge dx_2 \wedge
\ldots dx_n$. From theorem \[pairing\] we obtain a non-degenerate pairing $$B_f: I/J_f \times I/J_f {\longrightarrow}{{\mathbb R}}$$
The signature invariant of $f$ is $$\sigma(f):=Signature(B_f) \in {{\mathbb Z}}$$
This is indeed an invariant of $f$; it equals the signature of the canonical pairing $B:M_f \times M_f {\longrightarrow}{{\mathbb R}}$, where $M_f:=H_{\bf m}^0(\Omega^n/df\wedge \Omega^{n-1})$. Note that, in the definition with $B_f$, if we interchange two coordinates, then $dx_1\wedge\ldots \wedge dx_n$ changes sign, but also the order the $f_i$ is changed in a corresponding way. As a result, the signature of $B_f$ does not change.
It is easy to verify that $\sigma$ has the following properties:\
1) $\sigma(-f)=(-1)^n \sigma(f)$.\
2) If $f$ has at most one-dimensional critical locus, and $g$ has an isolated critical point, then $$\sigma(f \oplus g)=\sigma(f) \cdot \sigma(g)$$ Here $\oplus$ denotes the Thom-Sebastiani sum of $f$ and $g$. Furthermore, one computes $\sigma(x^2)=1$, hence $\sigma(-x^2)=-1$, and so $\sigma(x^2+y^2)=\sigma(-(x^2+y^2)=1$, whereas $\sigma(x^2-y^2)=-1$. For $x^2y \in {{\mathbb R}}[[x,y]]$ one finds $$\sigma(x^2y)=1$$ so that $\sigma(x^2y+z^2)=1$, $\sigma(x^2y-z^2)=-1$. Note that $x^2y-z^2=-(x^2(-y)+z^2)$, but indeed $\sigma(-f)=-\sigma(f)$ for a function of three variables.
\[signatureconstant\] Let $f \in {{\mathbb R}}\{x_1,x_2,\ldots,x_n\}$ be a function with a one-dimensional critical locus $\Sigma$ and let $F:X \times S {\longrightarrow}{{\mathbb R}}\times S$ be a good representative of an admissible deformation of $f,\sigma$. Then for a $s \in S$ one has $$\sigma(f)=\sum_{p \in X} \sigma(F_s,p)$$
We denote by $\pi: X \times S {\longrightarrow}S$ the canonical projection and consider as before the sheaf ${\cal M}_S={\cal I}_S/{\cal J}_S$. Recall that $\pi_*({\cal M}_S)$ is a free ${\cal O}_S$-module of finite rank, equal to $\dim(I/J)$. Using \[pairingfamily\] we obtain a family of pairings $$B_S:\pi_*({\cal M}_S) \times \pi_*({\cal M}_S) {\longrightarrow}{\cal O}_S$$ parametrised by $s \in S$. Let $B_{S,s}$ the resulting pairing on the fibre $\pi_*({\cal M}_S)_s$. The function $$Signature: S {\longrightarrow}{{\mathbb Z}},\;\;s \mapsto Signature(B_{S,s})$$ is constant, as in a basis it is described as the signature of a non-singular matrix that depends holomorphically on $s \in S$. For fixed $s \neq 0$, this matrix appears in block diagonal form, corresponding to the singularities of the function $F_s$. Note also that complex singularities appear in complex conjugate pairs, whose contribution to $\sigma$ turn out to cancel each other. The result follows.
\[top\] Let $f \in {{\mathbb R}}\{x_1,x_2,\ldots,x_n\}$ have a one-dimensional critical locus which admits a morsfication. Let $F_{\pm}$ denote the positive and negative Milnor fibre of $f$. Then one has: $$2 \sigma(f)=-\widetilde{\chi}(F_{-})+(-1)^{n-1}\widetilde{\chi}(F_{+})$$
We apply the previous result to a (good representative of a) morsification $F: X \times S {\longrightarrow}{{\mathbb R}}\times S$. For $s \neq 0$ the singularities of the function $F_s$ are of typ $A_1$, $A_{\infty}$ and $D_{\infty}$. For each case there are different real forms to consider. Furthermore, complex singularities appear in complex conjugate pairs, whose contribution to $\sigma$ turn out to cancel each other. The result then follows from the truth for these singularities, an easy cut and paste argument. More details can be found in [@egodiss].
------- -------------- --------------
$A_1$ $A_{\infty}$ $D_{\infty}$
------- -------------- --------------
We conjecture the above formula to hold for all singularities with critical locus a curve $\Sigma$ and with transverse type $A_1$. However, the above proof does not apply, in particular not if the curve $\Sigma$ is not smoothable.
\[threespace\] The formula of theorem \[top\] is valid for all non-isolated hypersurface singularities in three space with one-dimensional critical locus of transverse type $A_1$: $$2 \sigma(f)=\chi(F_{+})-\chi(F_{-})$$
Let $\Sigma \subset ({{\mathbb C}}^3,0)$ be the critical locus of $f$, described by a radical ideal $I$, so $f \in \int I$ and let $g \in I^2$ be generic. The statement holds for $g$, because for these a morsification does exist and thus we can apply the previous result. Also, there is an admissible deformation described by $F=tf+(1-t)g$ and where the curve $\Sigma$ is deformed trivially. By \[signatureconstant\] we have $\sigma(g)=\sigma(f)+\sum \sigma_{i}(f_i,p_i)$, where the $(f_i,p_i$ are the germs of singularties (of type $A_1$ of $D_{\infty}$ that split off in this deformation. As we know the truth for these singularities, we can conclude the truth of the statement for $f$ itself.
\(i) The formula \[threespace\] appears to holds for a much broader class of surface singularities with other transverse types, but it is not true without further assumptions. For example, the jacobian module $I/J_f$ for the sextic $f=(x^2+y^2)^3-4x^2y^2z^2=0$ has $2t^5+3t^6+2t^7$ as Poincaré polynomial. So its dimension is seven, and thus the signature in any case is odd. In fact it can be computed to be equal to three. In contrast, the difference of the real Euler characterisics is eight.\
(ii) From additivity of the Euler number one has $$\chi(F_{+})+\chi(F_{-})+\chi(F_0)=2$$ where $\chi(L_0)$ is the Euler number of the real link $L_0:=S^{n-1} \cap F_0$ of the singularity. In turn, this number is equal to $$\chi(L_0)=\sum_i \sigma(\Sigma_i)$$ where the sum runs over the [*real half-branches*]{} $\Sigma_i$ of the curve $\Sigma$, and where $\sigma(\Sigma_i)$ denotes the $\sigma$ of the singularity transverse to $\Sigma_i$. Formulated differently, $$\sum_i \sigma(\Sigma_i)=N-D$$ where $N$ denotes the nunmber of [*naked half-branches*]{} (transverse type $\pm(x^2+y^2)$) and $D$ the number of [*dressed half-branches*]{} (transverse type $x^2-y^2$).
A signature theorem for projective curves
-----------------------------------------
A homogenous polynomial $f \in {{\mathbb R}}[x,y,z]$ defines a cone $X:=\{f=0\}\subset {{\mathbb R}}^3$ which is the same as a projective curve $ C \subset {{\mathbb P}}^2({{\mathbb R}})$. Furthermore, if $f$ has even degree, $f$ has a well defined [*sign*]{} for each point in the complement of the curve. We put $V_+:=\{(x:y:z)\in{{\mathbb P}}^2({{\mathbb R}}):f(x,y,z)>0\}$ and $V_-:=\{(x:y:z)\in {{\mathbb P}}^2({{\mathbb R}}):f(x,y,z)<0\}$ The singularities of the cone $X$ is a finite union of lines, namely the cone over the singularities of $C$. The curve $C$ has ordinary double points, precisely if the the transverse type outside the origin ist $A_1$.
Let $f\in{{\mathbb R}}[x,y,z]$ be a homogenous polynomial of even degree defining a curve with only double points as singularities. Then $$\sigma_f=\chi(V_+)-\chi(V_-).$$
(see also [@arnold]) Let $F_+$ and $F_-$ denote the positive and negativ Milnor fibre of $f$ in ${{\mathbb R}}^3$. The restriction of the canonical projection ${{\mathbb R}}^3\setminus\{0\}\rightarrow {{\mathbb P}}^2({{\mathbb R}})$ to the Milnor fibres $F_+\rightarrow V_+$ and $F_-\rightarrow V_-$ is an unramified $2:1$ covering. For the Euler characteristics we thus obtain $\chi(F_+)=2\chi(V_+)$, $\chi(F_-)=2\chi({\mathcal{V}}_-)$. hence the result follows from \[threespace\].
The signature of the cone over the nodal quartic defined by $$(x^2+y^2)^2+3x^2y-y^3=0$$ is $\sigma=-3$, we see $\chi_+=0$ and $\chi_-=3$. The signature of the cone over the quartik with a $D_4$-singularity defined by $$2(x^2+y^2)^2+10y(3x^2-y^2)+11(x^2+y^2)-3=0$$
is $\sigma=-4$, the Euler numbers are $\chi_+=0$ and $\chi_-=4$. Figure \[quad\] shows the curves and the defined regions.
Depending on the situation, the signature can be used to find real components of algebraic curves.
$$\begin{array}{ccc}
\epsfig{figure=lage1.eps}&\epsfig{figure=lage2.eps}&\epsfig{figure=lage3.eps}\\
\sigma=2&\sigma=0&\sigma=-2
\end{array}$$
In figure \[test\], the signature of the quartic, consisting of the shown cubic and a moving test line, separates the different topological positions of the line relative to the cubic.
In figure \[zweiqu\] the signature of the shown quartic separates the relative topological configuration of the two quadrics.
$$\begin{array}{ccccc}
\epsfig{figure=klage1.eps}&\epsfig{figure=kreis1.eps,width=1.7cm}&\epsfig{figure=klage2.eps}&\epsfig{figure=klage4.eps}&\epsfig{figure=klage3.eps}\\
\sigma=1&\sigma=0&\sigma=-1&\sigma=-2&\sigma=-3
\end{array}$$
Some open problems
==================
We have shown that the self-duality of the Koszul-complex leads in the almost complete intersection case to a self-duality of $I/J$. In the real case one can define a signature. For the special case of Jacobi-modules of hypersurface singularities with one-dimensional singular locus this signature was related, in special cases, to Euler-characteristics of positive and negative Milnor fibre, which can be seen as a generalisation of the theorem formulated in [@arnold].\
$\bullet$ What is the topological meaning of the signature in general? In other words, proper generalisation of the Eisenbud-Levine theorem? We have seen that for Jacobi-modules of a special class of hypersurfaces with one-dimensional singular locus there is a direct relation with the Euler characteristic of the positive and negative Milnor fibre. But we have also seen that this relation does not hold in all examples. The more special question is: for exactly what class of singularities with one-dimensional singular locus is theorem \[top\] is true?\
$\bullet$ In the complete intersection case the pairing on $P/J$ factors over the multiplication map. For an almost complete intersection the non-degenerate pairing $$I/J \times I/J {\longrightarrow}K$$ is also $P$-linear, but there is no evident ’multiplication map’ to factor over.
In the case where $I$ is a radical ideal one has:
1\) The pairing on $I/J$ factors over the multiplication map $$I/J \times I/J {\longrightarrow}I^2/IJ$$
2\) If $I/J \neq 0$ then The $P$-module $I^2/IJ$ has a one-dimensional socle.\
3) The socle of $I^2/IJ$ is generated by the Jacobian determinant $$h:=\left|\begin{array}{ccc} \frac{\partial f_1}{\partial x_1}& \ldots &\frac{\partial f_1}{\partial x_n}\\
\ldots&\ldots&\ldots\\
\frac{\partial f_n}{\partial x_1} &\ldots& \frac{\partial f_n}{\partial x_n}\\
\end{array}\right|$$
We note that for radical ideals $I$ in three variables one always has $\dim(I/J)=\dim(I^2/IJ)$, by [@dJvS1] For non-radical ideals $I$ the statements of the conjecture often hold, but again the function $f=(x^2+y^2)^3-x^2y^2z^2$ provides a counterexample. Here $I^2/IJ$ has Poincaré polynomial $2t^{10}+2t^{11}$, however, the Hessian $h$ sits in degree $12$. The series $(x^3+y^3)^p-x^py^pz^p$ provides further examples.
The pairing on $I/J$ is [*symmetric*]{}, but we do not know a good algebraic proof of this fact. Of course it would be explained by the above conjecture.\
$\bullet$ Is there a way to exhibit a self-dual resolution resolution of $I/J$ as $P$-module?\
If $$0{\longrightarrow}F_{n-1} {\longrightarrow}\ldots {\longrightarrow}F_1 {\longrightarrow}P {\longrightarrow}P/I$$ is a free resolution over $P$, the map $P/J {\longrightarrow}P/I$ can be covered by a map of the Koszul-complex $\wedge^{\bullet}$ to $F_{\bullet}$ giving a diagram $$\begin{array}{cccccccccc}
\wedge^{n-1} &{\longrightarrow}& \wedge^{n-2}& {\longrightarrow}&\ldots &\wedge^1&{\longrightarrow}&P& {\longrightarrow}&P/J\\
\downarrow&&\downarrow&&&\downarrow&&\downarrow&&\downarrow\\
F_{n-1}& {\longrightarrow}&F_{n-2}& \ldots& {\longrightarrow}& F_1& {\longrightarrow}& P& {\longrightarrow}& P/I\\
\end{array}$$ Applying $Hom_P(-,P)$ to the diagram, and using $Hom(\wedge^p,P)=\wedge^{n-p}$ we obtain also maps $$\begin{array}{ccccccccc}
F_{1}^*& {\longrightarrow}&F_{2}^*&{\longrightarrow}& \ldots& {\longrightarrow}&F_{n-2}^*&{\longrightarrow}& F_{n-1}^*\\
\downarrow&&\downarrow&&&&\downarrow&&\downarrow\\
\wedge^{n-1} &{\longrightarrow}& \wedge^{n-2}& {\longrightarrow}&\ldots&{\longrightarrow}& \wedge^2&{\longrightarrow}&\wedge^1\\
\end{array}$$
These two diagrams can be put on top of each other: $$\begin{array}{ccccccccc}
F_{1}^*& {\longrightarrow}&F_{2}^*&{\longrightarrow}& \ldots& {\longrightarrow}&F_{n-2}^*&{\longrightarrow}& F_{n-1}^*\\
\downarrow&&\downarrow&&&&\downarrow&&\downarrow\\
\wedge^{n-1} &{\longrightarrow}& \wedge^{n-2}& {\longrightarrow}&\ldots&{\longrightarrow}& \wedge^2&{\longrightarrow}&\wedge^1\\
\downarrow&&\downarrow&&&&\downarrow&&\downarrow\\
F_{n-1}& {\longrightarrow}&F_{n-2}&{\longrightarrow}& \ldots& {\longrightarrow}&F_{2}&{\longrightarrow}& F_{1}\\
\end{array}$$ If this were a double complex, the total complex would be a self-dual resolution of $I/J$ of the right length, see [@pellikaan3]. Unfortunately, is is not clear that the maps always can be chosen as to obtain a double complex.
$\bullet$ Furthermore, one might ask for generalisations going further than almost complete intersections and look for self-dual pieces in the Koszul-homology. It is not clear how such a theorem might look like: there are simple examples of sequences $f_1,\ldots,f_n$ in $P$ with $\dim(P/J)=2$, but for which $H^0_{{\mathfrak{m}}}(P/J)$ is not a Gorenstein module. The homogeneous function $f=(x+y)(xu)^2+(u+v)(xv)^2+(x+u+v)(yu)^2+(y+u)(yv)^2$ is singular along the union of two planes defined by the ideal $(x,y) \cap (u,v)$, so the locus defined by $J=J_f$ is two-dimensional. The module $H^0_{{\mathfrak{m}}}(P/J)$ is computed to have $$6t^4+13t^5+15t^6+9t^7$$ as Poincaré series, so it is not Gorenstein.\
[**Acknowledgement:**]{} The first author wants to thank D. Eisenbud for showing interest in the pairing in an early stage of this work and asked the question as to the meaning of the signature in the real case. The work was part of the Ph. D. thesis of the second author. The authors thank further R. Pellikaan exchange of ideas and T. de Jong for suggesting the argument used in \[threespace\].
K. Aoki, T. Fukuda and T. Nishimura, [*On the number of branches of the zero locus of a map germ $({\mathds{R}}^n,0)\rightarrow(\mathds{R}^{n-1},0)$*]{}, In: [*Topology and computer science*]{} (Atami, 1986), 347–-363, (Kinokuniya, Tokyo, 1987).
, V. I. Arnold, S. M. Guzein-Zade, A. N. Varchenko, [*Singularities of differentiable maps. Vol. I and II.*]{} Monographs in Mathematics [**83**]{}. (Birkhäuser Boston, 1988).
V. I. Arnold,[*The index of a singular point of a vector field, the Petrovskiĭ-Oleĭnik inequalities, and mixed Hodge structures*]{}, Funct. Anal. Appl. [**12**]{} no.1 (1978), 1-–14.
, W. Bruns, J. Herzog, [*Cohen-Macaulay rings*]{}, Cambridge studies in adv. math. [**39**]{}, Cambridge University Press, 1993.
H. Cartan and S. Eilenberg, [*Homological Algebra*]{}, (Princeton University Press, 1956).
N. Dutertre, [*On topological invariants associated with a polynomial with isolated critical points*]{}. Glasg. Math. J. [**46**]{} (2004), no. 2, 323–-334.
A. Dimca, M. Saito, L. Wotzlaw, [*A generalization of the Griffiths theorem on rational integrals. II*]{}. Michigan Math. J. [**58**]{} (2009), no. 3, 603-–625.
D. Eisenbud, [*Commutative Algebra with a View Toward Algebraic Geometry*]{}, (Springer, Berlin, 1995).
W. Ebeling, S. M. Gusein-Zade, [*Indices of vector fields and 1-forms on singular varieties. Global aspects of complex geometry*]{}, 129-–169, (Springer, Berlin, 2006).
D. Eisenbud and H. Levine, [*An algebraic formula for the degree of a $C^{\infty}$ map germ*]{}, Annals of Mathematics, [**106**]{} (1977), 19–44.
T. de Jong, [*The virtual number of $D_\infty$ points. I*]{}. Topology [**29**]{} (1990), no. 2, 175–184.
J. de Jong, T. de Jong, [*The virtual number of $D_\infty$ points.*]{} II. Topology [**29**]{} (1990), no. 2, 185–188.
T. de Jong and D. van Straten, [*A deformation theory for nonisolated singularities*]{}. Abh. Math. Sem. Univ. Hamburg [**60**]{} (1990), 177–208.
T. de Jong and D. van Straten, [*Disentanglements*]{}, In: Singularity theory and its applications, Part I (Coventry, 1988/1989), 199–211, Lecture Notes in Math., [**1462**]{}, (Springer, Berlin, 1991).
A. N. Varchenko, A. B. Givental, [*The period mapping and the intersection form.*]{} Funct. Anal. Appl. [**16**]{} (1982), no. 2, 7–-20.
G. M. Khimshiashvilli, [*On the local degree of a smooth map*]{}, [Soobshch. Akad. Nauk. GruzSSR]{}, [**85**]{}(2) (1977), 309–311.
H. Matsumura, [*Commutative ring theory*]{}, (Cambridge Universtiy Press 1986).
J. Milnor, [*Singular points of complex hypersurfaces*]{}, Ann. Math. Stud. [**61**]{} (Princeton University Press, 1968).
J. Montaldi and D. van Straten, D. [*One-forms on singular curves and the topology of real curve singularities*]{}. [Topology]{}[**29**]{} (1990), no.4, 501–510.
R. Pellikaan, [*Finite determinacy of functions with nonisolated singularities*]{}. Proc. London Math. Soc. (3) [**57**]{} (1988), no. 2, 357–-382.
R. Pellikaan, [*Deformations of hypersurfaces with a one-dimensional singular locus.*]{} J. Pure Appl. Algebra [**67**]{} (1990), no. 1, 49–71.
R. Pellikaan, [*Projective resolutions of the quotient of two ideals.*]{} Nederl. Akad. Wetensch. Indag. Math. [**50**]{} (1988), no. 1, 65–84.
K. Saito, [*The higher residue pairings $K_{F}^{(k)}$ for a family of hypersurface singular points*]{}. Singularities, Part 2 (Arcata, Calif., 1981), 441-–463, Proc. Sympos. Pure Math. [**40**]{}, (Amer. Math. Soc., Providence, RI, 1983).
D. Siersma, [*Isolated line singularities*]{}. Singularities, Part 2 (Arcata, Calif., 1981), 485–496, Proc. Sympos. Pure Math. [**40**]{}, (Amer. Math. Soc., Providence, RI, 1983).
D. van Straten, [*On the Betti numbers of the Milnor fibre of a certain class of hypersurface singularities*]{}. In: Singularities, representation of algebras, and vector bundles (Lambrecht, 1985), 203–220, Lecture Notes in Math. [**1273**]{}, (Springer, Berlin, 1987).
G. Scheja und U.Storch, [*Über Spurfunktionen bei vollständigen Durchschnitten.*]{} [ J. Reine und Angew. Math.]{} [**278/279**]{} (1975), 174–189.
Z. Szafraniec, [*A formula for the Euler characteristic of a real algebraic manifold*]{}. Manuscripta Math. [**85**]{} (1994), no. 3-4, 345–360.
Z. Szafraniec, [*Topological degree and quadratic forms*]{}. J. Pure Appl. Algebra [**141**]{} (1999), no. 3, 299–314.
Z. Szafraniec, [*Topological invariants of real Milnor fibres*]{}. Manuscripta Math. [**110**]{} (2003) 2, 159–169.
A. N. Varchenko, [*Local residue and the intersection form in vanishing cohomology*]{}, Izv. Akad. Nauk SSSR Ser. Mat. [**49**]{} (1985), no. 1, 32–54.
T. Warmt, [*Gorenstein-Dualität und topologische Invarianten von Singularitäten*]{}, PhD thesis, Johannes-Gutenberg Universität Mainz (2003).
H. Weber, [*Lehrbuch der Algebra*]{}, Bd. 1, (Verlag Vieweg und Sohn, Braunschweig 1898).
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We compute time-periodic and relative-periodic solutions of the free-surface Euler equations that take the form of overtaking collisions of unidirectional solitary waves of different amplitude on a periodic domain. As a starting guess, we superpose two Stokes waves offset by half the spatial period. Using an overdetermined shooting method, the background radiation generated by collisions of the Stokes waves is tuned to be identical before and after each collision. In some cases, the radiation is effectively eliminated in this procedure, yielding smooth soliton-like solutions that interact elastically forever. We find examples in which the larger wave subsumes the smaller wave each time they collide, and others in which the trailing wave bumps into the leading wave, transferring energy without fully merging. Similarities notwithstanding, these solutions are found quantitatively to lie outside of the Korteweg-de Vries regime. We conclude that quasi-periodic elastic collisions are not unique to integrable model water wave equations when the domain is periodic.'
address: 'Dept of Mathematics, University of California, Berkeley, CA 94720-3840'
author:
- Jon Wilkening
date: 'April 21, 2014'
title: ' Relative-Periodic Elastic Collisions of Water Waves'
---
=1
[^1]
Introduction
============
A striking feature of multiple-soliton solutions of integrable model equations such as the Korteweg-deVries equation, the Benjamin-Ono equation, and the nonlinear Schrödinger equation is that they interact elastically, leading to time-periodic, relative-periodic, or quasi-periodic dynamics. By contrast, the interaction of solitary waves for the free-surface Euler equations is inelastic. However, it has been observed many times in the literature [@chan:street:70; @cooker:97; @maxworthy:76; @su:mirie; @mirie:su; @zou:su; @craig:guyenne:06; @milewski:11] that the residual radiation after a collision of such solitary waves can be remarkably small. In the present paper we explore the possibility of finding nearby time-periodic and relative-periodic solutions of the Euler equations using a collision of unidirectional Stokes waves as a starting guess. Such solutions demonstrate that recurrent elastic collisions of solitary waves in the spatially periodic case do not necessarily indicate that the underlying system is integrable.
A relative-periodic solution is one that returns to a spatial phase shift of its initial condition at a later time. This only makes sense on a periodic domain, where the waves collide repeatedly at regular intervals in both time and space, with the locations of the collisions drifting steadily in time. They are special cases (with $N=2$) of quasi-periodic solutions, which have the form $u(x,t)=U(\vec\kappa
x+\vec \omega t + \vec\alpha)$ with $U$ an $N$-periodic continuous function, i.e. $U\in C\big(\mathbb{T}^N\big)$, and $\vec\kappa$, $\vec\omega$, $\vec\alpha\in\mathbb{R}^N$. Throughout the manuscript, we will use the phrase “solitary waves” in a broad sense to describe waves that, most of the time, remain well-separated from one another and propagate with nearly constant speed and shape. “Stokes waves” will refer to periodic progressive solutions of the free-surface Euler equations of permanent form, or waves that began at $t=0$ as a linear superposition of such traveling waves. They comprise a special class of solitary waves. “Solitons” will refer specifically to superpositions of ${\operatorname}{sech}^2$ solutions of the KdV equation on the whole line, while “cnoidal solutions” will refer to their spatially periodic, multi-phase counterparts; see §\[sec:kdv\] for elaboration.
It was found in [@water2] that decreasing the fluid depth causes standing waves to transition from large-scale symmetric sloshing behavior in deep water to pairs of counter-propagating solitary waves that collide repeatedly in shallow water. In the present work, we consider unidirectional waves of different amplitude that collide due to taller waves moving faster than shorter ones. We present two examples of solutions of this type: one where the resulting dynamics is fully time-periodic; and one where it is relative-periodic, returning to a spatial phase shift of the initial condition at a later time. Both examples exhibit behavior typical of collisions of KdV solitons. In the first, one wave is significantly larger than the other, and completely subsumes it during the interaction. In the second, the waves have similar amplitude, with the trailing wave bumping into the leading wave and transferring energy without fully merging.
Despite these similarities, the amplitude of the waves in our examples are too large for the assumptions in the derivation of the KdV equation to hold. In particular, the larger wave in the first example is more than half the fluid depth in height, and there is significant vertical motion of the fluid when the waves pass by. A detailed comparison of the Euler and KdV equations for waves with these properties is carried out in §\[sec:kdv\]. A review of the literature on water wave collisions and the accuracy of the KdV model of water waves is also given in that section.
Rather than compute such solutions by increasing the amplitude from the linearized regime via numerical continuation, as was done for counter-propagating waves in [@water2], we use collisions of right-moving Stokes waves as starting guesses. The goal is to minimally “tune” the background radiation generated by the Stokes collisions so that the amount coming out of each collision is identical to what went into it. In the first example of §\[sec:num\], we find that the tuned background radiation takes the form of a train of traveling waves of smaller wavelength moving to the right more slowly than either solitary wave. By contrast, in the counter-propagating case studied in [@water2], it consists of an array of smaller-wavelength standing waves oscillating rapidly relative to the time between collisions of the primary waves. In the second example of §\[sec:num\], the background radiation is essentially absent, which is to say that the optimized solution is free from high-frequency, low-amplitude disturbances in the trough, and closely resembles a relative-periodic cnoidal solution of KdV. We call the collisions in this solution “elastic” as they repeat forever, unchanged up to spatial translation, and there are no features to distinguish radiation from the waves themselves. This process of tuning parameters to minimize or eliminate small-amplitude oscillations in the wave troughs is reminiscent of Vanden-Broeck’s work [@vandenBroeck91] in which oscillations at infinity could be eliminated from solitary capillary-gravity waves by choosing the amplitude appropriately.
To search for relative periodic solutions, we use a variant of the overdetermined shooting method developed by the author and collaborators in previous work to study several related problems: time-periodic solutions of the Benjamin-Ono equation [@benj1; @benj2] and the vortex sheet with surface tension [@vtxs1; @vtxs2]; Hopf bifurcation and stability transitions in mode-locked lasers [@lasers]; cyclic steady-states in rolling treaded tires [@tires1]; self-similarity (or lack thereof) at the crests of large-amplitude standing water waves [@water1]; harmonic resonance and spontaneous nucleation of new branches of standing water waves at critical depths [@water2]; and three-dimensional standing water waves [@water3d]. The three approaches developed in these papers are the adjoint continuation method [@benj1; @lasers], a Newton-Krylov shooting method [@tires1], and a trust region shooting method [@water2] based on the Levenberg-Marquardt algorithm [@nocedal]. We adopt the latter method here to exploit an opportunity to consolidate the work in computing the Dirichlet-Neumann operator for many columns of the Jacobian simultaneously, in parallel. One computational novelty of this work is that we search directly for large-amplitude solutions of a nonlinear two-point boundary value problem, without using numerical continuation to get there. This is generally difficult. However, in the present case, numerical continuation is also difficult due to non-smooth bifurcation “curves” riddled with Cantor-like gaps [@plotnikov01], and the long simulation times that occur between collisions in the unidirectional case. Our shooting method has proven robust enough to succeed in finding time-periodic solutions, when they exist, with a poor starting guess. False positives are avoided by resolving the solutions spectrally to machine accuracy and overconstraining the minimization problem. Much of the challenge is in determining the form of the initial condition and the objective function to avoid wandering off in the wrong direction and falling into a nonzero local minimum before locking onto a nearby relative-periodic solution.
Equations of motion {#sec:eqm}
===================
The equations of motion of a free surface $\eta(x,t)$ evolving over an ideal fluid with velocity potential $\phi(x,y,t)$ may be written [@whitham74; @johnson97; @craik04; @craik05] $$\begin{aligned}
\label{eq:ww}
\eta_t &= \phi_y - \eta_x\phi_x, \\[-3pt]
\notag
\varphi_t &= P\left[\phi_y\eta_t - \frac{1}{2}\phi_x^2 -
\frac{1}{2}\phi_y^2 - g\eta\right],
$$ where subscripts denote partial derivatives, $\varphi(x,t) =
\phi(x,\eta(x,t), t)$ is the restriction of $\phi$ to the free surface, $g=1$ is the acceleration of gravity, $\rho=1$ is the fluid density, and $P$ is the projection $$Pf = f - \frac{1}{2\pi}\int_0^{2\pi} f(x)\,dx,$$ where we assume a $2\pi$-periodic domain. The velocity components $u=\phi_x$ and $v=\phi_y$ at the free surface can be computed from $\varphi$ via $$\label{eq:uv:from:G}
\begin{pmatrix} \phi_x \\ \phi_y \end{pmatrix} =
\frac{1}{1+\eta'(x)^2}\begin{pmatrix}
1 & -\eta'(x) \\ \eta'(x) & 1 \end{pmatrix}
\begin{pmatrix}
\varphi'(x) \\
{\mathcal{G}}\varphi(x)
\end{pmatrix},$$ where a prime denotes a derivative and ${\mathcal{G}}$ is the Dirichlet-Neumann operator [@craig:sulem:93] $$\label{eq:DNO:def}
{\mathcal{G}}\varphi(x)
= \sqrt{1+\eta'(x)^2}\,\, {\frac{\partial \phi}{\partial n}}(x+i\eta(x))
= \phi_y - \eta_x\phi_x$$ for the Laplace equation, with periodic boundary conditions in $x$, Dirichlet conditions ($\phi=\varphi$) on the upper boundary, and Neumann conditions ($\phi_y=0$) on the lower boundary, assumed flat. We have suppressed $t$ in the notation since time is frozen in the Laplace equation. We compute ${\mathcal{G}}\varphi$ using a boundary integral collocation method [@lh76; @baker:82; @krasny:86; @mercer:92; @baker10] and advance the solution in time using an 8th order Runge-Kutta scheme [@hairer:I] with 36th order filtering [@hou:li:07]. See [@water2] for details.
Computation of relative-periodic solutions {#sec:method}
==========================================
Traveling waves have the symmetry that $$\label{eq:init}
\eta(x,0) \, \text{ is even}, \qquad \varphi(x,0) \, \text{ is odd.}$$ This remains true if $x$ is replaced by $x-\pi$. As a starting guess for a new class of time-periodic and relative-periodic solutions, we have in mind superposing two traveling waves, one centered at $x=0$ and the other at $x=\pi$. Doing so will preserve the property (\[eq:init\]), but the waves will now interact rather than remain pure traveling waves. A solution will be called *relative periodic* if there exists a time $T$ and phase shift $\theta$ such that $$\label{eq:ts:def}
\eta(x,t+T) = \eta(x-\theta,t), \qquad\quad
\varphi(x,t+T) = \varphi(x-\theta,t)$$ for all $t$ and $x$. Time-periodicity is obtained as a special case, with $\theta\in2\pi\mathbb{Z}$. We can save a factor of 2 in computational work by imposing the alternative condition $$\label{eq:even:odd}
\eta(x+\theta/2,T/2) \, \text{ is even}, \qquad\quad
\varphi(x+\theta/2,T/2) \, \text{ is odd.}$$ From this, it follows that $$\begin{aligned}
\eta(x+\theta/2,T/2) &= \eta(-x+\theta/2,T/2) = \eta(x-\theta/2,-T/2), \\
\varphi(x+\theta/2,T/2) &= -\varphi(-x+\theta/2,T/2) = \varphi(x-\theta/2,-T/2).\end{aligned}$$ But then both sides of each equation in (\[eq:ts:def\]) agree at time $t=-T/2$. Thus, (\[eq:ts:def\]) holds for all time.
In the context of traveling-standing waves in deep water [@trav:stand], it is natural to define $T$ as twice the value above, replacing all factors of $T/2$ by $T/4$. That way a pure standing wave returns to its original configuration in time $T$ instead of shifting in space by $\pi$ in time $T$. In the present work, we consider pairs of solitary waves moving to the right at different speeds, so it is more natural to define $T$ as the first (rather than the second) time there exists a $\theta$ such that (\[eq:ts:def\]) holds.
Objective function {#sec:obj:fun}
------------------
We adapt the overdetermined shooting method of [@water1; @water2] to compute solutions of (\[eq:init\])–(\[eq:even:odd\]). This method employs the Levenberg-Marquardt method [@nocedal] with delayed Jacobian updates [@water2] to solve the nonlinear least squares problem described below.
For (\[eq:init\]), we build the symmetry into the initial conditions over which the shooting method is allowed to search: we choose an integer $n$ and consider initial conditions of the form $$\label{eq:init:trav}
\hat\eta_k(0) = c_{2|k|-1}, \qquad\quad
\hat\varphi_k(0) = \pm ic_{2|k|},$$ where $k\in\{\pm1,\pm2,\dots,\pm \frac{n}{2}\}$ and $\hat\eta_k(t)$, $\hat\varphi_k(t)$ are the Fourier modes of $\eta(x,t)$, $\varphi(x,t)$. The numbers $c_1,\dots,c_n$ are assumed real and all other Fourier modes (except $\hat\eta_0$) are zero. We set $\hat\eta_0$ to the fluid depth so that $y=0$ is a symmetry line corresponding to the bottom wall. This is convenient for computing the Dirichlet-Neumann operator [@water2]. In the formula for $\hat\varphi_k$, the minus sign is taken if $k<0$ so that $\hat\varphi_{-k} =\overline{\hat\varphi_k}$. We also solve for the period, $$\label{eq:T:theta}
T=c_{n+1}.
$$ The phase shift $\theta$ is taken as a prescribed parameter here. Alternatively, in a study of traveling-standing waves [@trav:stand], the author defines a traveling parameter $\beta$ and varies $\theta=c_{n+2}$ as part of the algorithm to obtain the desired value of $\beta$. This parameter $\beta$ is less meaningful for solitary wave collisions in shallow water, so we use $\theta$ itself as the traveling parameter in the present study. We also need to specify the amplitude of the wave. This can be done in various ways, e.g. by specifying the value of the energy, $$E = \frac{1}{2\pi}\int_0^{2\pi} {\textstyle}\frac{1}{2}\varphi{\mathcal{G}}\varphi
+ \frac{1}{2}g\eta^2\,dx,$$ by constraining a Fourier mode such as $\hat\eta_1(0)$, or by specifying the initial height of the wave at $x=0$: $$\eta(0,0) = \hat\eta_0 + \sum_{k=1}^{n/2} 2c_{2k-1}.$$ Thus, to enforce (\[eq:even:odd\]), we can minimize the objective function $$\label{eq:f}
f(c) = \frac{1}{2} r(c)^Tr(c),$$ where $$\begin{aligned}
\label{eq:r:def}
r_1 = \big(\;\text{choose one:} \quad
& E-a \quad,\quad
\hat\eta_1(0)-a \quad,\quad
\eta(0,0)-a \;\big), \\ \notag
r_{2j} = {\operatorname{Im}}\{e^{ij\theta/2}\hat\eta_j(T/2)\}, \qquad
&r_{2j+1} = {\operatorname{Re}}\{e^{ij\theta/2}\hat\varphi_j(T/2)\}, \qquad
(1 \le j\le M/2).\end{aligned}$$ Here $a$ is the desired value of the chosen amplitude parameter. Alternatively, we can impose (\[eq:ts:def\]) directly by minimizing $$\label{eq:f1}
\tilde f = \frac{1}{2}r_1^2 + \frac{1}{4\pi}
\int_0^{2\pi} \left(\big[\eta(x,T)-\eta(x-\theta,0)\big]^2 +
\big[\varphi(x,T)-\varphi(x-\theta,0)\big]^2\right)dx,$$ which also takes the form $\frac{1}{2}r^Tr$ if we define $r_1$ as above and $$\label{eq:r1:def}
\begin{aligned}
r_{4j-2}+ir_{4j-1} &= \sqrt{2}\left[ \hat\eta_j(T) - e^{-ij\theta}\hat\eta_j(0) \right], \\
r_{4j}+ir_{4j+1} &= \sqrt{2}\left[ \hat\varphi_j(T) - e^{-ij\theta}\hat\varphi_j(0) \right],
\end{aligned} \qquad\quad (1\le j\le M/2).$$ Note that $f$ measures deviation from evenness and oddness of $\eta(x+\theta/2,T/2)$ and $\varphi(x+\theta/2,T/2)$, respectively, while $\tilde f$ measures deviation of $\eta(x+\theta,T)$ and $\varphi(x+\theta,T)$ from their initial states. In the first example of §\[sec:num\], we minimize $\tilde f$ directly, while in the second we minimize $f$ and check that $\tilde f$ is also small, as a means of validation. The number of equations, $m=M+1$ for $f$ and $m=2M+1$ for $\tilde f$, is generally larger than the number of unknowns, $n+1$, due to zero-padding of the initial conditions. This adds robustness to the shooting method and causes all Fourier modes varied by the algorithm, namely those in (\[eq:init:trav\]), to be well-resolved on the mesh.
Computation of the Jacobian
---------------------------
To compute the $k$th column of the Jacobian $J=\nabla_c r$, which is needed by the Levenberg-Marquardt method, we solve the linearized equations along with the nonlinear ones: $$\label{eq:q:qdot}
{\frac{\partial }{\partial t}}
\begin{pmatrix} q \\ \dot q \end{pmatrix} =
\begin{pmatrix} F(q) \\ DF(q)\dot q \end{pmatrix}, \quad
\begin{aligned}
q(0) &= q_0 = (\eta_0,\varphi_0), \\
\dot q(0) &= \dot q_0 = \partial q_0/\partial c_k.
\end{aligned}$$ Here $q=(\eta,\varphi)$, $\dot q=(\dot\eta,\dot\varphi)$, $F(q)$ is given in (\[eq:ww\]), $DF$ is its derivative (see [@water2] for explicit formulas), and a dot represents a variational derivative with respect to perturbation of the initial conditions, not a time derivative. To compute $\partial r_i/\partial c_k$ for $i\ge2$ and $k\le n$, one simply puts a dot over each Fourier mode on the right-hand side of (\[eq:r:def\]) or (\[eq:r1:def\]), including $\hat\eta_j(0)$ and $\hat\varphi_j(0)$ in (\[eq:r1:def\]). If $k=n+1$, then $c_k=T$ and $${\frac{\partial r_{2j}}{\partial T}} = {\operatorname{Im}}\{e^{ij\theta/2}(1/2)\partial_t\hat\eta_j(T/2)\}, \qquad
{\frac{\partial (r_{4j}+ir_{4j+1})}{\partial T}} = \sqrt{2}\big[\partial_t\hat\varphi_j(T)\big]$$ in (\[eq:r:def\]) and (\[eq:r1:def\]), respectively, with similar formulas for $\partial(r_{4j-2}+ir_{4j-1})/\partial T$ and $\partial
r_{2j+1}/\partial T$. The three possibilities for $r_1$ are handled as follows: $$\begin{aligned}
&\text{case 1:} \quad {\frac{\partial r_1}{\partial c_k}} = \dot E
= \frac{1}{2\pi}\int_0^{2\pi} \left[\dot\varphi\eta_t - \dot\eta\varphi_t \right]_{t=0}dx,
\quad (k\le n), \qquad
{\frac{\partial r_1}{\partial c_{n+1}}} = 0, \\
&\text{case 2:} \quad {\frac{\partial r_1}{\partial c_k}} = {{\dot\eta}^{\scriptscriptstyle\bm\wedge}}_1(0) = \delta_{k,1},
\quad (k\le n+1),\\
&\text{case 3:} \quad {\frac{\partial r_1}{\partial c_k}} = \dot\eta(0,0) = 2\delta_{k,\text{odd}}, \quad
(k\le n), \qquad {\frac{\partial r_1}{\partial c_{n+1}}} = 0,\end{aligned}$$ where $\delta_{k,j}$ and $\delta_{k,\text{odd}}$ equal 1 if $k=j$ or $k$ is odd, respectively, and equal zero otherwise. The vectors $\dot q$ in (\[eq:q:qdot\]) are computed in batches, each initialized with a different initial perturbation, to consolidate the work in computing the Dirichlet-Neumann operator during each timestep. See [@water2; @trav:stand] for details.
Numerical results {#sec:num}
=================
As mentioned in the introduction, our idea is to use collisions of unidirectional Stokes (i.e. traveling) waves as starting guesses to find time-periodic and relative periodic solutions of the Euler equations. We begin by computing traveling waves of varying wave height and record their periods. This is easily done in the framework of §\[sec:method\]. We set $\theta=\pi/64$ (or any other small number) and minimize $\tilde f$ in (\[eq:f1\]). The resulting “period” $T$ will give the wave speed via $c=\theta/T$. Below we report $T=2\pi c$, i.e. $T$ is rescaled as if $\theta$ were $2\pi$. We control the amplitude by specifying $\hat\eta_1(0)$, which is the second option listed in §\[sec:method\] for defining the first component $r_1$ of the residual. A more conventional approach for computing traveling waves is to substitute $\eta(x-ct)$, $\varphi(x-ct)$ into (\[eq:ww\]) and solve the resulting stationary problem (or an equivalent integral equation) by Newton’s method [@chen80a; @chandler:93; @milewski:11]. Note that the wave speed $c$ here is unrelated to the vector $c$ of unknowns in (\[eq:init:trav\]).
![\[fig:bif:stokes\] Plots of wave height and first Fourier mode versus period for Stokes waves with wavelength $2\pi$ and fluid depth $h=0.05$. The temporal periods are $6T_A=137.843\approx
137.738 = 5T_C$.](figs/bif_stokes){width="3.3in"}
![\[fig:align:stokes\] Collision of two right-moving Stokes waves that nearly return to their initial configuration after the interaction. (left) Solutions A and C were combined via (\[eq:AandC\]) and evolved through one collision to $t=137.738$. (right) Through trial and error, we adjusted the amplitude of the smaller Stokes wave and the simulation time to obtain a nearly time-periodic solution. ](figs/align_stokes){width="\linewidth"}
With traveling waves in hand, out next goal is to collide two of them and search for a nearby time-periodic solution, with $\theta=0$. As shown in Figure \[fig:bif:stokes\], varying $\hat\eta_1(0)$ from 0 to $7.4\times 10^{-4}$ causes the period of a Stokes wave with wavelength $\lambda=2\pi$ and mean fluid depth $h=0.05$ to decrease from $T_O=28.1110$ to $T_A=22.9739$, and the wave height (vertical crest-to-trough distance) to increase from 0 to $0.02892$. Solution C is the closest among the Stokes waves we computed to satisfying $5T_C=6T_A$, where $p=5$ is the smallest integer satisfying $\frac{p+1}{p}T_A<T_O$. We then combine solution A with a spatial phase shift of solution C at $t=0$. The resulting initial conditions are $$\label{eq:AandC}
\begin{aligned}
\eta^{A+C}_0(x) &= h + \big[\eta^A_0(x)-h\big] + \big[\eta^C_0(x-\pi)-h\big], \\
\varphi^{A+C}_0(x) &= \varphi^A_0(x) + \varphi^C_0(x-\pi),
\end{aligned}$$ where $h=0.05$ is the mean fluid depth. Plots of $\eta_0^A(x)$, $\eta_0^C(x-\pi)$, $\varphi_0^A(x)$ and $\varphi_0^C(x-\pi)$ are shown in Figure \[fig:AandC\]. If the waves did not interact, the combined solution would be time-periodic (to the extent that $5T_C=6T_A$, i.e. to about $0.076\%$). But the waves do interact. In addition to the complicated interaction that occurs when they collide, each slows the other down between collisions by introducing a negative gradient in the velocity potential between its own wave crests. Indeed, as shown in the right panel of Figure \[fig:AandC\], the velocity potential increases rapidly across a right-moving solitary wave and decreases elsewhere to achieve spatial periodicity. The decreasing velocity potential induces a background flow opposite to the direction of travel of the other wave. In the left panel of Figure \[fig:align:stokes\], we see that the net effect is that neither of the superposed waves has returned to its starting position at $t=5T_C$, and the smaller wave has experienced a greater net decrease in speed. However, as shown in the right panel, by adjusting the amplitude of the smaller wave (replacing solution C by B) and increasing $T$ slightly to $138.399$, we are able to hand-tune the Stokes waves to achieve $\tilde
f\approx5.5\times10^{-8}$, where $\theta$ is set to zero in (\[eq:f1\]). Note that as $t$ varies from 0 to $T/10$ in the left panel of Figure \[fig:evol:kdv1\], the small wave advances by $\pi$ units to the right while the large wave advances by $1.2\pi$ units. The waves collide at $t=T/2$. This generates a small amount of radiation, which can be seen at $t=T$ in the right panel of Figure \[fig:align:stokes\]. Some radiation behind the large wave is present for all $t>0$, as shown in Figure \[fig:pov:kdv1\].
Before minimizing $\tilde f$, we advance the two Stokes waves to the time of the first collision, $t=T/2$. At this time, the larger solitary wave has traversed the domain 3 times and the smaller one 2.5 times, so their peaks lie on top of each other at $x=0$. The reason to do this is that when the waves merge, the combined wave is shorter, wider, and smoother than at any other time during the evolution. Quantitatively, the Fourier modes of $\hat\eta_k(t)$ and $\hat\varphi_k(t)$ decay below $10^{-15}$ for $k\ge600$ at $t=0$, and $k\ge200$ when $t=T/2$. Thus, the number of columns needed in the Jacobian is reduced by a factor of 3, and the problem becomes more overdetermined, hence more robust. For the calculation of a time-periodic solution, we let $t=0$ correspond to this merged state, which affects the time labels when comparing Figures \[fig:evol:kdv1\] and \[fig:evol:kdv2\]. As a final initialization step, we project onto the space of initial conditions satisfying (\[eq:init:trav\]) by zeroing out the imaginary parts of $\hat\eta_k(0)$ and the real parts of $\hat\varphi_k(0)$, which are already small. Surprisingly, this improves the time-periodicity of the initial guess in (\[eq:f1\]) to $\tilde f = 2.3\times 10^{-8}$.
![\[fig:evol:kdv1\] Evolution of two Stokes waves that collide repeatedly, at times $t\approx T/2+kT$, $k\ge0$. (left) Traveling solutions A and B in Figure \[fig:bif:stokes\] were initialized with wave crests at $x=0$ and $x=\pi$, respectively. The combined solution is approximately time-periodic, with period $T=138.399$. (right) The same solution, at later times, starting with the second collision ($t=3T/2$).](figs/evol_kdv1){width="\linewidth"}
![\[fig:pov:kdv1\] A different view of the solutions in Figure \[fig:evol:kdv1\] shows the generation of background waves. Shown here are the functions $\eta(x+8\pi t/T,t)$, which give the dynamics in a frame moving to the right fast enough to traverse the domain four times in time $T$. In a stationary frame, the smaller and larger solitary waves traverse the domain 5 and 6 times, respectively.](figs/pov_kdv1){height="2in"}
We emphasize that our goal is to find *any* nearby time-periodic solution by adjusting the initial conditions to drive $\tilde f$ to zero. Energy will be conserved as the solution evolves from a given initial condition, but is only imposed as a constraint (in the form of a penalty) on the search for initial conditions when the first component of the residual in (\[eq:r:def\]) is set to $r_1=E-a$. In the present calculation, we use $r_1=\eta(0,0)-a$ instead. In the second example, presented below, we will constrain energy. In either case, projecting onto the space of initial conditions satisfying (\[eq:init:trav\]) can cause $r_1$ to increase, but it will decrease to zero in the course of minimizing $\tilde f$. This projection is essential for the symmetry arguments of §\[sec:obj:fun\] to work.
![\[fig:evol:kdv2\] Time-periodic solutions near the Stokes waves of Figure \[fig:evol:kdv1\]. (left) $h=0.05$, $\eta(0,0) =
0.0707148$, $T=138.387$, $\tilde f=4.26\times10^{-27}$. (right) $h=0.0503$, $\eta(0,0)=0.0707637$, $T=138.396$, $\tilde f=1.27\times
10^{-26}$. The background radiation was minimized by hand in the right panel by varying $h$ and $\eta(0,0)$.](figs/evol_kdv2){width="\linewidth"}
![\[fig:pov:kdv2\] Same as Figure \[fig:pov:kdv1\], but showing the time-periodic solutions of Figure \[fig:evol:kdv2\] instead of the Stokes waves of Figure \[fig:evol:kdv1\]. The Stokes waves generate new background radiation with each collision while the time-periodic solutions are synchronized with the background waves to avoid generating additional disturbances. ](figs/pov_kdv2){height="2in"}
We minimize $\tilde f$ subject to the constraint $\eta(0,0)=0.0707148$, the third case described in §\[sec:method\] for specifying the amplitude. This causes $\tilde f$ to decrease from $2.3\times 10^{-8}$ to $4.26\times 10^{-27}$ using $M=1200$ grid points and $N=1200$ time-steps (to $t=T$). The results are shown in the left panel of Figures \[fig:evol:kdv2\] and \[fig:pov:kdv2\]. The main difference between the Stokes collision and this nearby time-periodic solution is that the Stokes waves generate additional background ripples each time they collide while the time-periodic solution contains an equilibrium background wave configuration that does not grow in amplitude after the collision. While the background waves in the counter-propagating case (studied in [@water2]) look like small-amplitude standing waves, these background waves travel to the right, but slower than either solitary wave. After computing the $h=0.05$ time-periodic solution, we computed 10 other solutions with nearby values of $h$ and $\eta(0,0)$ to try to decrease the amplitude of the background radiation. The best solution we found (in the sense of small background radiation) is shown in the right panel of Figures \[fig:evol:kdv2\] and \[fig:pov:kdv2\], with $h=0.0503$ and $\eta(0,0)=0.0707637$. The amplitude of the background waves of this solution are comparable to that of the Stokes waves after two collisions.
Our second example is a relative periodic solution in which the initial Stokes waves (the starting guess) are B and C in Figure \[fig:bif:stokes\] instead of A and C. As before, solution C is shifted by $\pi$ when the waves are combined initially, just as in (\[eq:AandC\]). Because the amplitude of the larger wave has been reduced, the difference in wave speeds is smaller, and it takes much longer for the waves to collide. If the waves did not interact, we would have $$\label{eq:cBcC}
c_{B,0} = 0.23246089, \quad c_{C,0} = 0.22808499, \quad
T_0 = \frac{2\pi}{c_{B,0}-c_{A,0}} = 1435.86,$$ where wave B crosses the domain $53.1230$ times in time $T_0$ while wave C crosses the domain $52.1230$ times. The subscript 0 indicates that the waves are assumed not to interact. Since the waves do interact, we have to evolve the solution numerically to obtain useful estimates of $T$ and $\theta$. We arbitrarily rounded $T_0$ to 1436 and made plots of the solution at times $\Delta t = T_0/1200$. We found that $\eta$ is nearly even (up to a spatial phase shift) for the first time at $463\Delta t=554.057$. This was our initial guess for $T/2$. The phase shift required to make $\eta(x+\theta/2,T/2)$ approximately even and $\varphi(x+\theta/2,T/2)$ approximately odd was found by graphically solving $\varphi(x,T/2)=0$. This gives the initial guess $\theta/2=2.54258$. This choice of $T$ and $\theta$ (with $\eta^{B+C}$ and $\varphi^{B+C}$ as initial conditions) yields $f=2.0\times10^{-11}$ and $\tilde f=1.5\times10^{-10}$. We then minimize $f$ holding $E$ and $\theta$ constant, which gives $f=2.1\times10^{-29}$ and $\tilde f=3.0\times10^{-26}$. We note that $\tilde f$ is computed over $[0,T]$, twice the time over which the solution was optimized by minimizing $f$, and provides independent confirmation of the accuracy of the solution and the symmetry arguments of §\[sec:obj:fun\].
The results are plotted in Figure \[fig:evol:kdv3\]. We omit a plot of the initial guess (the collision of Stokes waves) as it is indistinguishable from the minimized solution. In fact, the relative change in the wave profile and velocity potential is about $0.35$ percent, $$\left(\frac{
\|\eta_\text{Stokes} - \eta_\text{periodic}\|^2 +
\|\varphi_\text{Stokes} - \varphi_\text{periodic}\|^2}{
\|\eta_\text{Stokes} - h\|^2 + \|\varphi_\text{Stokes}\|^2}
\right)^{1/2} \le 0.0035,$$ and $T/2$ changes even less, from 554.057 (Stokes) to 554.053 (periodic). By construction, $E$ and $\theta/2$ do not change at all. It was not necessary to evolve the Stokes waves to $T/2$, shift space by $\theta/2$, zero out Fourier modes that violate the symmetry condition (\[eq:init\]), and reset $t=0$ to correspond to this new initial state. Doing so increases the decay rate of the Fourier modes (slope of $\ln|\hat\eta_k|$ vs $k$) by a factor of 1.24 in this example, compared to 3.36 in the previous example, where it is definitely worthwhile.
![\[fig:evol:kdv3\] A relative-periodic solution found using a superposition of the Stokes waves labeled B and C in Figure \[fig:bif:stokes\] as a starting guess. Unlike the previous case, the waves do not fully merge at $t=T/2$. ](figs/evol_kdv3){width="\linewidth"}
The large change from $T_0/2 = 717.93$ to $T/2=554.053$ is due to nonlinear interaction of the waves. There are two main factors contributing to this change in period. The first is that the waves do not fully combine when they collide. Instead, the trailing wave runs into the leading wave, passing on much of its amplitude and speed. The peaks remain separated by a distance of $d=0.52462$ at $t=T/2$, the transition point where the waves have the same amplitude. Thus, the peak separation changes by $\pi-d$ rather than $\pi$ in half a period. The second effect is that the larger wave slows down the smaller wave more than the smaller slows the larger. Recall from Fig. \[fig:AandC\] that each wave induces a negative potential gradient across the other wave that generates a background flow opposing its direction of travel. Quantitatively, when the waves are well separated, we find that the taller and smaller waves travel at speeds $c_B=0.231077=0.994049c_{B,0}$ and $c_C=0.226153=0.991531c_{C,0}$, respectively. The relative speed is then $(c_B-c_C) = 1.12526(c_{B,0}-c_{C,0})$. Thus, $$\label{eq:ineq}
\frac{\pi-d}{c_B-c_C} < \frac{T}{2} <
\frac{\pi-d}{c_{B,0}-c_{C,0}} < \frac{T_0}{2} = \frac{\pi}{c_{B,0}-c_{C,0}},
$$ with numerical values $531.5<554.1<598.0<717.9$. This means that both effects together have overestimated the correction needed to obtain $T$ from $T_0$. This is because the relative speed slows down as the waves approach each other, which is expected since the amplitude of the trailing wave decreases and the amplitude of the leading wave increases in this interaction regime. Indeed, the average speed of the waves is $$\label{eq:average:speed}
\overline{c_B} = \frac{\theta/2 - d/2}{T/2} = 0.993388c_{B,0}, \qquad
\overline{c_C} = \frac{\theta/2 + d/2 - \pi}{T/2} = 0.991737c_{C,0},$$ which are slightly smaller and larger, respectively, than their speeds when well separated. Note that $T/2$ in (\[eq:ineq\]) may be written $T/2=(\pi-d)/(\overline{c_B} - \overline{c_C})$. We used $\theta/2=2.54258+40\pi$ in (\[eq:average:speed\]) to account for the 20 times the waves cross the domain $(0,2\pi)$ in time $T/2$ in addition to the offset shown in Figure \[fig:evol:kdv3\].
Comparison with KdV {#sec:kdv}
===================
In the previous section, we observed two types of overtaking collisions for the water wave: one in which the larger wave completely subsumes the smaller wave for a time, and one where the two waves remain distinct throughout the interaction. Similar behavior has of course been observed for the Korteweg-de Vries equation, which was part of our motivation for looking for such solutions. Lax [@lax:1968] classified overtaking collisions of two KdV solitons as bimodal, mixed, or unimodal. Unimodal and bimodal waves are analogous to the ones we computed above, while mixed mode collisions have the larger wave mostly subsume the smaller wave at the beginning and end of the interaction, but with a two-peaked structure re-emerging midway through the interaction. Lax showed that if $1<c_1/c_2<A=(3+\sqrt{5})/2$, the collision is bimodal; if $c_1/c_2>3$, the collision is unimodal; and if $A<c_1/c_2<3$, the collision is mixed. Here $c_1$ and $c_2$ are the wave speeds of the trailing and leading waves, respectively, at $t=-\infty$. Leveque [@leveque:87] has studied the asymptotic dynamics of the interaction of two solitons of nearly equal amplitude. Zou and Su [@zou:su] performed a computational study of overtaking water wave collisions, compared the results to KdV interactions, and found that the water wave collisions ceased to be elastic at third order. Craig *et. al.* [@craig:guyenne:06] also found that solitary water waves collide inelastically. This does not conflict with our results since we optimize the initial conditions to make the collision elastic. Head on collisions have been studied numerically by Su and Mirie [@su:mirie; @mirie:su], experimentally by Maxworthy [@maxworthy:76], and by a mixture of analysis and computation by Craig *et. al.* [@craig:guyenne:06].
Validation of KdV as a model of water waves has also been studied extensively. A formal derivation may be found in Ablowitz and Segur [@ablowitz:segur]. Rigorous justification has been given by Bona, Colin and Lannes [@bona:lannes], building on earlier work by Craig [@craig:kdv] as well as Schneider and Wayne [@schneider:wayne]. According to [@bona:lannes], some gaps still exist in the theory in the spatially periodic case. Experimental studies of the validity of KdV for describing surface waves have been performed by Zabusky and Galvin [@zabusky:galvin] as well as Hammack and Segur [@hammack:segur:74]. Recently, Ostrovsky and Stepanyants [@ostrovsky] have compared internal solitary waves in laboratory experiments to the predictions of various model equations, including KdV, and give a comprehensive overview of the literature on this subject [@ostrovsky].
Our objective in this section is to determine quantitatively whether the solutions of the water wave equations that we computed in §\[sec:num\] are well-approximated by the KdV equation. Following Ablowitz and Segur [@ablowitz:segur], we introduce a small parameter ${\varepsilon}$ and dimensionless variables $$\hat y = \frac{y}{h}, \qquad
\hat x = \sqrt{{\varepsilon}}\frac{x}{h}, \qquad
\hat t = \sqrt{\frac{{\varepsilon}g}{h}} t, \qquad
\hat \eta = \frac{\eta}{{\varepsilon}h}, \qquad
\hat \phi = \frac{\phi}{\sqrt{{\varepsilon}g h^3}},$$ where $h$ is the fluid depth. We assume the bottom boundary is at $y=-h$ rather than 0 in this derivation, so that $\hat y$ runs from $-1$ to ${\varepsilon}\hat\eta$. The Laplacian becomes $\Delta_{\varepsilon}= h^{-2}\big(
{\varepsilon}\partial_{\hat{x}}^2 + \partial_{\hat{y}}^2 \big)$, which allows for $\hat\phi = \hat\phi_0 + {\varepsilon}\hat\phi_1 + {\varepsilon}^2\hat\phi_2 +
\cdots$ to be computed order by order, with leading terms satisfying $$\hat\phi_{0,\hat y} = 0, \qquad
\hat\phi_1 = -\frac{1}{2}(1+\hat y)^2\hat\phi_{0,\hat x\hat x}, \qquad
\hat\phi_2 = \frac{1}{24}(1+\hat y)^4\hat\phi_{0,\hat x\hat x\hat x\hat x}.$$ Here we used $\Delta\phi=0$ and $\phi_y(x,-h)=0$. Note that $\hat\phi_0$ is independent of $\hat y$, and agrees with the velocity potential $\phi$ on the bottom boundary, up to rescaling: $$\hat\phi_0(\hat x,\hat t) = ({\varepsilon}gh^3)^{-1/2}\phi(x,-h,t).$$ From the equations of motion, $\eta_t = \phi_y - \eta_x\phi_x$ and $\phi_t + \frac{1}{2}\phi_x^2 + \frac{1}{2}\phi_y^2 + g\eta = 0$, one finds that $$\begin{aligned}
\hat\eta_{\hat t} + \hat u_{\hat x} &=
{\varepsilon}\big\{ {\textstyle}\frac{1}{6}\hat u_{\hat x\hat x\hat x} - (\hat\eta\hat u)_{\hat x}
\big\} + O({\varepsilon}^2), \\
\hat u_{\hat t} + \hat\eta_{\hat x} &= {\varepsilon}\big\{ {\textstyle}\frac{1}{2} \hat u_{\hat x\hat x\hat t} - \frac{1}{2}\partial_{\hat x}(\hat u)^2
\big\} + O({\varepsilon}^2),\end{aligned}$$ where $\hat u(\hat x,\hat t) = \partial_{\hat x}\hat\phi_0(\hat x,\hat t)$. Expanding $\hat\eta=\hat\eta_0 + {\varepsilon}\hat\eta_1 +\cdots$, $\hat u=\hat u_0 + {\varepsilon}\hat u_1 +\cdots$, we find that $$\begin{aligned}
\hat\eta_0 = f(\hat x - \hat t; \tau) + g(\hat x + \hat t; \tau), \\
\hat u_0 = f(\hat x - \hat t; \tau) - g(\hat x + \hat t; \tau),
\end{aligned} \qquad
\begin{aligned}
2f_\tau + 3ff_r + (1/3)f_{rrr} &= 0, \\
-2g_\tau + 3gg_l + (1/3)g_{lll} &= 0,
\end{aligned}$$ where we have introduced characteristic coordinates $r=\hat x-\hat t$, $l = \hat x + \hat t$ as well as a slow time scale $\tau={\varepsilon}\hat t$ to eliminate secular growth in the solution with respect to $r$ and $l$ at first order in ${\varepsilon}$; see [@ablowitz:segur] for details. The notational conflict of $g(l,\tau)$ with the acceleration of gravity, $g$, is standard, and will not pose difficulty below.
![\[fig:kdv:cmp1\] Comparison of the solutions of the KdV and Euler equations, initialized identically with the superposition of Stokes waves labeled A and B in Figure \[fig:bif:stokes\]. The final time $T$ is set to $138.399$, as in Fig. \[fig:align:stokes\], when the Euler solution nearly returns to its initial configuration after a single overtaking collision. ](figs/kdv1){width="\linewidth"}
In our case, the waves travel to the right, so we may set $g(l,\tau)=0$ in the formulas above. Returning to dimensional variables, we then have $$\eta(x,t) = h{\varepsilon}f\left(\sqrt{\varepsilon}\left(\frac{x}{h} - \sqrt{\frac{g}{h}} t
\right),\sqrt{\frac{g}{h}}{\varepsilon}^{3/2}t\right),$$ which satisfies $$\label{eq:dim:kdv}
\eta_t + \alpha \eta_x + \frac{3\sqrt{gh}}{2h}\eta\eta_x +
\frac{1}{6}\sqrt{gh}\,h^2\eta_{xxx} = 0,$$ where $\alpha=\sqrt{gh}$. Note that ${\varepsilon}$ drops out. For comparison with the results of §\[sec:num\], we will add $h$ to $\eta$ and set $\alpha=-\frac{1}{2}\sqrt{gh}$ instead. In Figure \[fig:kdv:cmp1\], we compare the solution of (\[eq:dim:kdv\]), with initial condition $\eta(x,0) = \eta_0^{A+B}(x)$, defined similarly to $\eta_0^{A+C}(x)$ in (\[eq:AandC\]), to the solution of the free-surface Euler equations shown in Figs. \[fig:align:stokes\] and \[fig:evol:kdv1\]. Shortly after the waves are set in motion, the KdV solution develops high-frequency oscillations behind the larger peak that travel left and quickly fill up the computational domain with radiation. The solution of the Euler equations remains much smoother. The large peak of the KdV solution also travels $3.4\%$ faster, on average, than the corresponding peak of the Euler solution, so that at $t=138.399$, when the taller Euler wave has traversed the domain 6 times, the taller KdV wave has traversed it $6.206$ times. For our purposes, these discrepancies are much too large for KdV to be a useful model, and we conclude that the first example in §\[sec:num\] is well outside of the KdV regime.
In this comparison, timestepping the KdV equation was done with the 8 stage, 5th order implicit/explicit Runge-Kutta method of Kennedy and Carpenter [@carpenter]. Spatial derivatives were computed spectrally using the 36th order filter of Hou and Li [@hou:li:07]. We found that 2048 spatial grid points and 96000 timesteps was sufficient to reduce the error at $t=138.399$ below $5\times 10^{-6}$ near the larger peak and below $6\times 10^{-7}$ elsewhere, based on comparing the solution to one with 3072 grid points and 192000 timesteps. Our solutions of the Euler equations are much more accurate since there are no second or third spatial derivative terms present to make the equations stiff. Thus, we can use 8th or 15th order explicit timestepping rather than 5th order implicit/explicit timestepping. Monitoring energy conservation and performing mesh refinement studies suggests that we obtain 13–14 digits of accuracy in the solutions of the Euler equations, at which point roundoff error prevents further improvement in double-precision arithmetic.
![\[fig:kdv:cmp2\] Comparison of the solutions of the KdV and Euler equations, both initialized with the superposition of Stokes waves labeled B and C in Figure \[fig:bif:stokes\]. $T=1108.11$ here. ](figs/kdv2){width="\linewidth"}
In Figure \[fig:kdv:cmp2\], we repeat this computation using initial conditions corresponding to the superposition of Stokes waves $\eta_0^{B+C}(x)$, which was used as a starting guess for the second example of §\[sec:num\]. This time the KdV solution does not develop visible high-frequency radiation in the wave troughs, and the solutions of KdV and Euler remain close to each other for much longer. However, the interaction time for a collision also increases, from $T=138.399$ in the first example to $T=1108.11$ here. In Fig. \[fig:kdv:cmp2\], by $t=T/6$, the taller KdV and Euler waves have visibly separated from each other, and by $t=T/2$, when the Euler waves have reached their minimum approach distance, the KdV solution is well ahead of the Euler solution. Thus, while there is good qualitative agreement between the KdV and Euler solutions, they do not agree quantitatively over the time interval of interest. From this point of view, the second example of §\[sec:num\] also lies outside of the KdV regime.
An alternative measure of the agreement between KdV and Euler is to compare the solutions from §\[sec:num\] with nearby relative-periodic solutions of KdV. In other words, we wish to quantify how much the initial conditions and period have to be perturbed to convert a relative-periodic solution of the Euler equations into one for the KdV equations. Since we used a superposition of Stokes waves for the initial guess to find time-periodic and relative-periodic solutions of the Euler equations, we will use a similar superposition (of cnoidal waves) for KdV. The vertical crest-to-trough heights of the three Stokes waves considered in §\[sec:num\] are $$\label{eq:H:ABC}
H_A = 0.028918699, \qquad
H_B = 0.004973240, \qquad
H_C = 0.002683648.$$ Well-known [@kdv:1895; @dingemans] periodic traveling wave solutions of (\[eq:dim:kdv\]) are given by $$\begin{gathered}
\eta(x,t) = h - H + \frac{H}{m}\left(1 - \frac{E(m)}{K(m)}\right) + H{\operatorname}{cn}^2\left(
2K(m)\frac{x-ct}{\lambda}\bigg\vert m\right), \\
\lambda = \sqrt{\frac{16mh^3}{3H}}\,K(m), \qquad
c = \left[1 - \frac{H}{2h} + \frac{H}{mh}\left(1 -
\frac{3E(m)}{2K(m)}\right)\right]\sqrt{gh},\end{gathered}$$ where we added $h$ to $\eta$ to match the change in $\alpha$ from $\sqrt{gh}$ to $-\frac{1}{2}\sqrt{gh}$ in (\[eq:dim:kdv\]). Here $K(m)$ and $E(m)$ are the complete elliptic integrals of the first and second kind, respectively, and ${\operatorname}{cn}(z|m)$ is one of the Jacobi elliptic functions [@dingemans; @gradshteyn]. In our case $\lambda=2\pi$, $g=1$ and $h=0.05$. For each $H$ in (\[eq:H:ABC\]), we solve the $\lambda$ equation for $m$ using Mathematica [@mma], and then evaluate $\eta(x,0)$ on a uniform grid that is fine enough that its Fourier coefficients decay below machine roundoff. The values of $m' = 1-m$ are $$m'_A = 1.81924\times10^{-35}, \qquad
m'_B = 1.98689\times10^{-14}, \qquad
m'_C = 1.79643\times10^{-10}.$$ This approach requires extended precision arithmetic to compute $m$ and evaluate $\eta$, but the running time takes only a few seconds on a typical laptop. A periodized version of the simpler ${\operatorname}{sech}^2$ formula could be used for the first two waves, but decays too slowly for wave $C$ to be spatially periodic to roundoff accuracy. Once these cnoidal waves have been computed, we superpose their initial conditions to form $\eta_0^{A+B}$ and $\eta_0^{B+C}$, just as in §\[sec:num\]. It is well-known that a superposition of $N$ cnoidal waves retain this form when evolved via KdV, with $N$ amplitude and $N$ phase parameters governed by an ODE describing pole dynamics in the complex plane [@kruskal:pole; @airault:kdv; @deconinck:segur]. In the $N=2$ case, the solutions are relative-periodic.
![\[fig:kdv:cmp5\] Comparison of time-periodic solution found in §\[sec:num\] to nearby relative-periodic two phase cnoidal solution of KdV. The periods are $T=138.387$ and $113.079$, respectively. ](figs/kdv5){width=".98\linewidth"}
![\[fig:kdv:cmp4\] Comparison of relative-periodic solution found in §\[sec:num\] to nearby relative-periodic two phase cnoidal solution of KdV. The periods are $T=1108.11$ and $1068.73$, respectively. ](figs/kdv4){width=".98\linewidth"}
Figures \[fig:kdv:cmp5\] and \[fig:kdv:cmp4\] compare the time-periodic and relative-periodic solutions of the Euler equations, computed in §\[sec:num\], to these cnoidal solutions of KdV. Since the periods are different, only the initial conditions are compared. In the larger-amplitude example, shown in Fig. \[fig:kdv:cmp5\], the Euler solution is not as flat in the wave trough as the cnoidal solution due to an additional oscillatory component (the “tuned” radiation). From the difference plot in the right panel, we see that the crest-to-trough amplitude of these higher frequency oscillations is roughly $6\times10^{-4}$, or $2.1\%$ of the wave height $H_A$. The Euler solution is time-periodic with period $T_\text{Euler}=138.387$ while the cnoidal solution is relative-periodic, returning to a spatial phase shift of its initial condition at $T_\text{KdV}=113.079$, which differs from $T_\text{Euler}$ by $18\%$. In the smaller-amplitude example, shown in Fig. \[fig:kdv:cmp4\], both solutions have smooth, flat wave troughs, and it is difficult to distinguish one from the other in the left panel. The crest-to-trough amplitude of the difference in the right panel is roughly $5.5\times10^{-5}$, or $1.1\%$ of $H_B$. The relative change in period is $(T_\text{Euler}-T_\text{KdV})/ T_\text{Euler} = 3.6\%$. While the left panels of Figures \[fig:kdv:cmp5\] and \[fig:kdv:cmp4\] show close agreement between relative-periodic solutions of the Euler and KdV equations at $t=0$, it should be noted that the wave amplitudes of the cnoidal solutions were chosen to minimize the discrepancy in these figures. The change in period by $18\%$ and $3.6\%$, respectively, is perhaps a better measure of agreement.
![\[fig:kdv:cmp7\] Comparison of KdV and Euler solutions, both initialized with a 2-phase cnoidal wave with peaks matching the heights of the Stokes waves labeled A and B (left) or B and C (right) in Fig. \[fig:bif:stokes\]. Here $T=138.387$ (left) and $T=1068.73$ (right). ](figs/kdv7){width=".98\linewidth"}
A final comparison of the two equations is made in Fig. \[fig:kdv:cmp7\], where we evolve the Euler equations with the KdV initial conditions. This requires an initial condition for $\varphi(x)=\phi(x,\eta(x))$, where we have suppressed $t$ in the notation for this discussion since it is held fixed at 0. Based on the derivation presented above, we first solve $\phi_x(x,0) =
\sqrt{g/h}[\eta(x)-h]$ for $\phi$ on the bottom boundary. We then use the approximation $$\varphi(x) \approx \phi(x,0) - \frac{\eta(x)^2}{2}\phi_{xx}(x,0) +
\frac{\eta(x)^4}{24}\phi_{xxx}(x,0)$$ to evaluate $\phi$ on the free surface. In the left panel of Figure \[fig:kdv:cmp7\], the larger wave grows and overturns before $t=T/400$ when evolved under the Euler equations, instead of traveling to the right when evolved via KdV. To handle wave breaking, we switched to an angle-arclength formulation of the free-surface Euler equations [@hls94; @vtxs1]. In the small-amplitude example in the right panel, the Euler solution develops visible radiation and falls slightly behind the KdV solution, although the phases are closer at $T/2$ than the result of evolving the Stokes waves under KdV in Figure \[fig:kdv:cmp2\]. We also tried evaluating $$\phi(x,y)=\sqrt{g/h} \sum_{k=1}^\infty 2k^{-1}\hat\eta_k \sin(kx)\cosh(ky)$$ at $y=\eta(x)$ to obtain the initial condition for $\varphi(x)$, where $\hat\eta_k$ are the Fourier modes of $\eta(x)$ at $t=0$, but the results were worse for the large-amplitude example — the wave breaks more rapidly — and were visually indistinguishable in the small-amplitude example from the results plotted in Fig. \[fig:kdv:cmp7\].
In summary, the large-amplitude time-periodic solution of the Euler equations found in §\[sec:num\] is well outside of the KdV regime by any measure, and the small-amplitude relative-periodic solution is closer, but not close enough to achieve quantitative agreement over the entire time interval of interest.
Conclusion
==========
We have demonstrated that the small amount of background radiation produced when two Stokes waves interact in shallow water can often be tuned to obtain time-periodic and relative-periodic solutions of the free-surface Euler equations. Just as for the Korteweg-de Vries equation, the waves can fully merge when they collide or remain well-separated. However, the comparison is only qualitative as the waves are too large to be well-approximated by KdV theory.
In future work, we will study the stability of these solutions using Floquet theory. Preliminary results suggest that the first example considered above is unstable to harmonic perturbations while the second example is stable. In the stable case, an interesting open question is whether the Stokes waves used as a starting guess for the minimization algorithm, which have the same energy as the relative-periodic solution found, might remain close to it forever, executing almost-periodic oscillations around it. Presumably $\theta$ would need to be varied slightly for this to be true, since $\theta$ is a free parameter that we selected by hand to obtain a small value of $\tilde f$ for the initial guess. Another open question is whether there are analogues for the Euler equations of $N$-phase quasi-periodic solutions of the KdV equation with $N\ge3$. We are confident that the methods of this paper could be used to construct degenerate cases of $N\ge3$ solitary water waves colliding elastically in a time-periodic or relative-periodic fashion, along the lines of what was done for the Benjamin-Ono equation in [@benj2]. Computing more general quasi-periodic dynamics of the form $\eta(x,t)=H(\vec\kappa x + \vec\omega t + \vec\alpha)$, $\varphi(x,t)=\Phi(\vec\kappa x + \vec\omega t + \vec\alpha)$ with $H,\Phi\in C(\mathbb{T}^N)$ and $\vec\kappa$, $\vec\omega$, $\vec\alpha\in\mathbb{R}^N$ seems possible in principle using a more sophisticated shooting method to determine $H$, $\Phi$ and $\vec\omega$. Existence of such solutions for the Euler equations would show that non-integrable equations can also support recurrent elastic collisions even if they cannot be represented as $N$-phase superpositions of elliptic functions.
[10]{}
Mark J. Ablowitz and Harvey Segur, *Solitons and the inverse scattering transform*, SIAM, Philadelphia, 1981.
H. Airault, H. P. McKean, and J. Moser, *Rational and elliptic solutions of the [Korteweg-de Vries]{} equation and a related many-body problem*, Comm. Pure Appl. Math. **30** (1977), 95–148.
D. M. Ambrose and J. Wilkening, *Global paths of time-periodic solutions of the [Benjamin]{}–[Ono]{} equation connecting pairs of traveling waves*, Comm. App. Math. and Comp. Sci. **4** (2009), no. 1, 177–215.
[to3em]{}, *Computation of symmetric, time-periodic solutions of the vortex sheet with surface tension*, Proc. Nat. Acad. Sci. **107** (2010), no. 8, 3361–3366.
[to3em]{}, *Computation of time-periodic solutions of the [Benjamin]{}–[Ono]{} equation*, J. Nonlinear Sci. **20** (2010), no. 3, 277–308.
[to3em]{}, *Dependence of time-periodic votex sheets with surface tension on mean vortex sheet strength*, Procedia IUTAM **11** (2014), 15–22.
G. R. Baker, D. I. Meiron, and S. A. Orszag, *Generalized vortex methods for free-surface flow problems*, J. Fluid Mech. **123** (1982), 477–501.
G. R. Baker and C. Xie, *Singularities in the complex physical plane for deep water waves*, J. Fluid Mech. **685** (2011), 83–116.
J. L. Bona, T. Colin, and D. Lannes, *Long wave approximations for water waves*, Arch. Rational Mech. Anal. **178** (2005), 373–410.
R. K.-C. Chan and R. Street, *A computer study of finite amplitude water waves*, J. Comput. Phys. **6** (1970), 68–94.
G. A. Chandler and I. G. Graham, *The computation of water waves modelled by [Nekrasov’s]{} equation*, SIAM J. Numer. Anal. **30** (1993), no. 4, 1041–1065.
B. Chen and P. G. Saffman, *Numerical evidence for the existence of new types of gravity waves of permanent form on deep water*, Stud. Appl. Math. **62** (1980), 1–21.
M. J. Cooker, P. D. Weidman, and D. S. Bale, *Reflection of a high-amplitude solitary wave at a vertical wall*, J. Fluid Mech. **342** (1997), 141–158.
W. Craig, *An existence theory for water waves and the [Boussinesq]{} and [Korteweg-de Vries]{} scaling limits*, Comm. Partial Diff. Equations **10** (1985), 787–1003.
W. Craig, P. Guyenne, J. Hammack, D. Henderson, and C. Sulem, *Solitary water wave interactions*, Phys. Fluids **18** (2006), 057106.
W. Craig and C. Sulem, *Numerical simulation of gravity waves*, J. Comput. Phys. **108** (1993), 73–83.
A. D. D. Craik, *The origins of water wave theory*, Ann. Rev. Fluid Mech. **36** (2004), 1–28.
[to3em]{}, *George gabriel stokes on water wave theory*, Ann. Rev. Fluid Mech. **37** (2005), 23–42.
B. Deconinck and H. Segur, *Pole dynamics for elliptic solutions of the [KdV]{} equation*, Mathematical Physics, Analysis and Geometry **3** (2000), 49–74.
Maarten W. Dingemans, *Water wave propagation over uneven bottoms, part 2. non-linear wave propagation*, World Scientific, Singapore, 1997.
S. Govindjee, T. Potter, and J. Wilkening, *Cyclic steady states of treaded rolling bodies*, Int. J. Numer. Meth. Engng. (2014), (accepted).
I. S. Gradshteyn and I. M. Ryzhik, *Table of integrals, series and products*, 7th ed., Academic Press, Amsterdam, 2007.
E. Hairer, S. P. Norsett, and G. Wanner, *Solving ordinary differential equations [I]{}: Nonstiff problems, 2nd edition*, Springer, Berlin, 2000.
J. L. Hammack and H. Segur, *The [Korteweg-de Vries]{} equation and water waves. part 2. comparison with experiments*, J. Fluid Mech. **65** (1974), 289–314.
T. Y. Hou and R. Li, *Computing nearly singular solutions using pseudo-spectral methods*, J. Comput. Phys. **226** (2007), 379–397.
T. Y. Hou, J. S. Lowengrub, and M. J. Shelley, *Removing the stiffness from interfacial flows with surface tension*, J. Comput. Phys. **114** (1994), 312–338.
R. S. Johnson, *A modern introduction to the mathematical theory of water waves*, Cambridge University Press, Cambridge, UK, 1997.
C. A. Kennedy and M. H. Carpenter, *Additive [Runge]{}-[Kutta]{} schemes for convection-diffusion-reaction equations*, Appl. Numer. Math. **44** (2003), no. 1–2, 139–181.
D. J. Korteweg and G. de Vries, *On the change of form of long waves advancing in a rectangular canal, and on a new type of long stationary waves*, Philosophical Magazine **39** (1895), 422–443.
R. Krasny, *Desingularization of periodic vortex sheet roll-up*, J. Comput. Phys. **65** (1986), 292–313.
M. D. Kruskal, *The [Korteweg-de Vries]{} equation and related evolution equations*, Nonlinear wave motion, Lectures in Appl. Math., vol. 15, American Mathematical Society, Providence, 1974, pp. 61–83.
P. D. Lax, *Integrals of nonlinear equations of evolution and solitary waves*, Comm. Pure Appl. Math. **21** (1968), 467–490.
R. J. LeVeque, *On the interaction of nearly equal solitons in the kdv equation*, SIAM J. Appl. Math. **47** (1987), 254–262.
M. S. Longuet-Higgins and E. D. Cokelet, *The deformation of steep surface waves on water. [I]{}. a numerical method of computation*, Proc. Royal Soc. A **350** (1976), 1–26.
T. Maxworthy, *Experiments on collisions between solitary waves*, J. Fluid Mech. **76** (1976), 177–185.
G N Mercer and A J Roberts, *[Standing waves in deep water: Their stability and extreme form]{}*, Phys. Fluids A **4** (1992), no. 2, 259–269.
P. A. Milewski, J.-M. Vanden-Broeck, and Z. Wang, *Dynamics of steep two-dimensional gravity–capillary solitary waves*, J. Fluid Mech. **664** (2010), 466–477.
R. M. Mirie and C. H. Su, *Collisions between two solitary waves, [Part II]{}*, J. Fluid Mech. **115** (1982), 475–492.
J. Nocedal and S. J. Wright, *Numerical optimization*, Springer, New York, 1999.
L. A. Ostrovsky and Y. A. Stepanyants, *Internal solitons in laboratory experiments: Comparison with theoretical models*, Chaos **15** (2005), 037111:1–28.
P. Plotnikov and J. Toland, *Nash-moser theory for standing water waves*, Arch. Rat. Mech. Anal. **159** (2001), 1–83.
Chris H. Rycroft and Jon Wilkening, *Computation of three-dimensional standing water waves*, J. Comput. Phys. **255** (2013), 612–638.
G. Schneider and C. E. Wayne, *The long-wave limit for the water wave problem. i. the case of zero surface tension*, Comm. Pure Appl. Math. **53** (2000), 1475–1535.
C. H. Su and R. M. Mirie, *On head-on collisions between two solitary waves*, J. Fluid Mech. **98** (1980), 509–525.
J.-M. Vanden-Broeck, *Elevation solitary waves with surface tension*, Phys. Fluids A **3** (1991), 1989–1993.
G. B. Whitham, *Linear and nonlinear waves*, Wiley, New York, 1974.
J. Wilkening, *Breakdown of self-similarity at the crests of large amplitude standing water waves*, Phys. Rev. Lett **107** (2011), 184501.
[to3em]{}, *Traveling-standing water waves*, (2014), (in preparation).
J. Wilkening and J. Yu, *Overdetermined shooting methods for computing standing water waves with spectral accuracy*, Comput. Sci. Disc. **5** (2012), 014017:1–38.
M. O. Williams, J. Wilkening, E. Shlizerman, and J. N. Kutz, *Continuation of periodic solutions in the waveguide array mode-locked laser*, Physica D **240** (2011), no. 22, 1791–1804.
, *Mathematica, version 8.0*, Champaign, IL, 2010.
N. J. Zabusky and C. J. Galvin, *Shallow-water waves, the [Korteweg-de Vries]{} equation and solitons*, J. Fluid Mech. **47** (1970), 811–824.
Q. Zou and C. H. Su, *Overtaking collisions between two solitary waves*, Phys. Fluids **29** (1986), 2113–2123.
[^1]: This research was supported in part by the Director, Office of Science, Computational and Technology Research, U.S. Department of Energy under Contract No. DE-AC02-05CH11231, and by the National Science Foundation through grant DMS-0955078.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A new formulation of electromagnetism based on linear differential commutator brackets is developed. Maxwell equations are derived, using these commutator brackets, from the vector potential $\vec{A}$, the scalar potential $\varphi$ and the Lorentz gauge connecting them. With the same formalism, the continuity equation is written in terms of these new differential commutator brackets.'
author:
- '*and*'
date: 'Received: 12 September 2009 / Accepted: '
---
[example.eps]{} gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore
Introduction
============
Maxwell equations are first order differential equations in space and time. They are compatible with Lorentz transformation which guarantees its applicability to any inertial frame. A symmetric space-time formulation of any theory will generally guarantees the universality of the theory. With this motivation, we adopt a differential commutator bracket involving first order space and time derivative operators to formulate the Maxwell equations and quantum mechanics. This is in addition to our recent quaternionic formulation of physical laws, where we have shown that many physical equations are found to emerge from a unified view of physical variables \[1\]. In such a formulation, we have found that Maxwell equations emerges from a single equation. Maxwell equations were originally written in terms of quaternions. They were initially written in twenty equations \[a\]. However, later on Maxwell equations are then written in terms of vector in the way that we are familiar today. In our present formulation, Maxwell equations are described by a set of two wave equations representing the evolution of the electric and magnetic fields. This is instead of having four equations. We aim in this paper to write down (derive) the physical equations by vanishing differential commutator brackets. We know that second order partial derivatives commute for space-space variables. We don’t assume here this property is a priori for space and time. To guarantee this, we eliminate the time derivative of a quantity that is acted by a space ($\nabla$) derivative followed by a time derivative, and vice-versa. In expanding the differential commutator bracket, we don’t commute time and space derivative, but rather eliminate the time derivative by the space derivative, and vice versa. This differential commutator bracket may enlighten us to quantize these physical quantities. By employing the differential commutator brackets of the vector $\vec{A}$ and scalar potential $\varphi$, we have derived Maxwell equations without invoking any a priori physical law. Maxwell arrives at his theory of electromagnetism by combing the Gauss, Faraday and Ampere laws. For mathematical consistency, he modified Ampere’s law. He then came with the known Maxwell equations.
Relativistic prelude
====================
From Lorentz transformations one obtain, $$\label{1}
x'=\gamma(x-vt)\,,\qquad y'=y\,,\qquad z'=z \,,\qquad t'=\gamma(t-\frac{v}{c^2}\,x) \,.$$ We see that the commutator bracket $$\label{1}
\left[\triangle t\,, \triangle x\right]=\left[ \triangle t'\,, \triangle x'\right]\,.$$ where we have taken into account in the order of multiplication of the space and time differences, ($\triangle x\,, \triangle t$). This shows that the commutator is Lorentz invariant. This is a new invariant quantity in relativity. We, however, already knew that the square interval is Lorentz invariant, i.e., $(\triangle S)^2 =(\triangle S')^2$ \[2\]. It follows from Eq.(1) that the differential commutator bracket $\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]=0\,$ is Lorentz invariant too, i.e., $\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]=\left[\frac{\partial}{\partial t'}\,, \vec{\nabla}'\right]$. We know that the spatial second order derivatives of a function, $f=f(x,y)$, is commutative, i.e., $\frac{\partial^2f}{\partial x\partial y}=\frac{\partial^2f}{\partial y\partial x}$. We wonder if the commutation of space and time derivatives are equally valid for all physical quantities. Motivated by this hypothesis, we propose the following differential commutator brackets to formulate the physical laws. In particular, we apply these differential commutator brackets, in this work to derive the continuity equation, Maxwell equations.
Differential commutators algebra
================================
Define the three linear differential commutator brackets as follows: $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]=0\,,\qquad \left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]=0\,,
\qquad \left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]=0\,.$$ Equation (3) is correct since partial derivatives commute, i.e., $\frac{\partial^2}{\partial t\partial x}\varphi=\frac{\partial^2}{\partial x\partial t}\varphi$. For a scalar $\psi$ and a vector $\vec{G}$, one defines the three brackets as follows:[^1] $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]\psi=\frac{\partial}{\partial t}\left(\vec{\nabla}\psi\right)-\vec{\nabla}\left(\frac{\partial \psi}{\partial t}\right)\,,$$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{G}=\frac{\partial}{\partial t}\left(\vec{\nabla}\cdot\vec{G}\right)-\vec{\nabla}\cdot\left(\frac{\partial \vec{G}}{\partial t}\right)\,,$$ and $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{G}=\frac{\partial}{\partial t}\left(\vec{\nabla}\times\vec{G}\right)-\vec{\nabla}\times\left(\frac{\partial \vec{G}}{\partial t}\right)\,.$$ It follows that $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\psi\vec{G})= \psi\{\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{G}\}+\{ \left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]\psi\}\cdot\vec{G}\,,$$
$$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right](\psi\vec{G})= \psi\{\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{G}\}+\{\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]\psi\}\times\vec{G}\,,$$
$$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\vec{G}\times\vec{F})= \vec{F}\cdot\{\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{G}\}-\vec{G}\cdot \{\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{F}\}\,,$$
for any vector $\vec{F}$. The differential commutator brackets above satisfy the distribution rule $$\label{1}
\left[\hat{A}\hat{B}\,\,,\hat{C}\right]= \hat{A} \left[\hat{B}\,\,,\hat{C}\right]+ \left[\hat{A}\,\,,\hat{C}\right]\hat{B}\,,$$ where $\hat{A}\,, \hat{B}\,,\hat{C}$ are $\vec{\nabla}\,, \frac{\partial}{\partial t}$. It is evident that the differential commutator brackets identities follow the same ordinary vector identities. We call the three differential commutator brackets in Eq.(3) the grad-commutator bracket, the dot-commutator bracket and the cross-commutator bracket respectively. The prime idea here is to replace the time derivative of a quantity by the space derivative $\vec{\nabla}$ of another quantity, and vice-versa, so that the time derivative of a quantity is followed by a time derivative with which it commutes. We assume here that space and time derivatives don’t commute. With this minimal assumption, we have shown here that all physical laws are determined by vanishing differential commutator bracket.
The continuity equation
=======================
Using quaternionic algebra \[3\], we have recently found that generalized continuity equations can be written as $$\vec{\nabla}\cdot \vec{J}+\frac{\partial \rho}{\partial
t}=0\,,$$ $$\vec{\nabla}(\rho\,c^2)+\frac{\partial
\vec{J}}{\partial t}=0\,,$$ and $$\vec{\nabla}\times
\vec{J}=0\,.$$ Now consider the dot-commutator of $\rho\vec{J}$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\rho\vec{J})=\frac{\partial}{\partial t}\left(\vec{\nabla}\cdot (\rho\vec{J})\right)-\vec{\nabla}\cdot\left(\frac{\partial (\rho\vec{J})}{\partial t}\right)=0\,.$$ Using Eqs.(11) - (13), one obtains $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\rho\vec{J})=c^2\rho\left(\frac{1}{c^2}\frac{\partial^2\rho}{\partial t^2}-\nabla^2\rho\right)+\left(\frac{1}{c^2}\frac{\partial \vec{J}}{\partial t^2}-\nabla^2\vec{J}\right)\cdot\vec{J}=0\,.$$ For arbitrary $\rho$ and $\vec{J}$, Eq.(15) yields the two wave equations $$\label{1}
\frac{1}{c^2}\frac{\partial^2\rho}{\partial t^2}-\nabla^2\rho=0\,,$$ and $$\label{1}\frac{1}{c^2}\frac{\partial \vec{J}}{\partial t^2}-\nabla^2\vec{J}=0\,.$$ Equations (16) and (17) show that the charge and current density satisfy a wave equation travelling at speed of light in vacuum. It is remarkable to know that these two equations are already obtained in \[3\]. Hence, the current-charge density wave equations are equivalent to $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\rho\vec{J})=0\,.$$
Maxwell’s equations
===================
Maxwell’s equations are first order differential equations in space and time of the electromagnetic field, viz., $$\label{2}
\vec{\nabla}\cdot \vec{E}=\frac{\rho}{\varepsilon_{0}}\,,$$ $$\label{2}
\vec{\nabla}\times \vec{B}=\mu_{0}\vec{J}+\frac{1}{c^2}\frac{\partial \vec{E}}{\partial t},$$ $$\vec{\nabla}\times\vec{E}=-\frac{\partial \vec{B}}{\partial t}\,,$$ $$\label{2}
\vec{\nabla}\cdot \vec{B}=0\,.$$ These equations show that charge ($\rho$) and current ($\vec{J}$) densities are the sources of the electromagnetic field. Differentiating Eqs.(20) and using Eq.(21), one obtains $$\frac{1}{c^2}\frac{\partial^2\vec{E}}{\partial
t^2}-\nabla^2\vec{E}=-\mu_0\left(\vec{\nabla}(\rho\,c^2) +\frac{\partial
\vec{J}}{\partial t}\right)\,.$$ Similarly, differentiating Eq.(21) and using Eq.(20), one obtains $$\frac{1}{c^2}\frac{\partial^2\vec{B}}{\partial
t^2}-\nabla^2\vec{B}=\mu_0\left(\vec{\nabla}\times \vec{J}\right)\,.$$ These two equations state that the electromagnetic field propagates with speed of light in two cases:\
(i) charge and current free medium (vacuum), i.e., $\rho=0, \vec{J}=0$, or\
(ii) if the two equations $$\vec{\nabla}(\rho\,c^2) +\frac{\partial
\vec{J}}{\partial t}=0\,,$$ and $$\left(\vec{\nabla}\times \vec{J}\right)=0\,,$$ besides the familiar continuity equation in Eq.(11) $$\vec{\nabla}\cdot\vec{J}+\frac{\partial \rho}{\partial t}=0\,,$$ are satisfied. Equation (23) and (24) resemble Einstein’s general relativity equation where space-times geometry is induced by the distribution of matter present. We see here that the electromagnetic field is produced by any charge and current densities distribution (in space and time). Now define the electromagnetic vector $\vec{F}$ as $$\vec{F}=\vec{B}-\frac{i}{c}\vec{E}\,$$ Adding Eqs.(25) and Eqs.(26) according to Eq.(28), one obtains $$\frac{1}{c^2}\frac{\partial^2}{\partial
t^2}(\vec{B}-\frac{i}{c}\vec{E})-\nabla^2(\vec{B}-\frac{i}{c}\vec{E})=\mu_0\left[\frac{i}{c}\left(\vec{\nabla}(\rho\,c^2) +\frac{\partial
\vec{J}}{\partial t}\right)+\vec{\nabla}\times \vec{J}\right]\,.$$ Applying Eqs.(25), (26) (see \[3\]) in Eq.(29) yields $$\frac{1}{c^2}\frac{\partial^2\vec{F}}{\partial
t^2}-\nabla^2\vec{F}=0\,\,,\qquad \Box^2\vec{F}=0\,.$$ This is a wave equation propagating with speed of light in vacuum ($c$). Hence, Maxwell wave equations can be written as a pure single wave equation of an electromagnetic sourceless complex vector field $\vec{F}$. We call Eqs.(25) - (27) the generalized continuity equations. We have recently obtained these generalized continuity equations by adopting quaternionic formalism for fluid mechanics \[3\]. It is challenging to check wether any real fluid satisfies these equations or not. We have recently shown that Schrodinger, Dirac and Klein - Gordon and diffusion equations are compatible with these generalized continuity equations \[3\]. Using Eqs.(19) and (20), the electric field dot-commutator bracket yields $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{E}=\frac{\partial}{\partial t}\left(\vec{\nabla}\cdot\vec{E}\right)-\vec{\nabla}\cdot\left(\frac{\partial \vec{E}}{\partial t}\right)=\frac{\partial\rho}{\partial t}+\vec{\nabla}\cdot\vec{J}=0\,.$$ This is the familiar continuity equation. Hence, the continuity equation in the commutator bracket form can be written as $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{E}=0\,.$$ Similar, using Eqs.(21) and (22), the magnetic field dot-commutator bracket yields $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{B}=\frac{\partial}{\partial t}\left(\vec{\nabla}\cdot\vec{B}\right)-\vec{\nabla}\cdot\left(\frac{\partial \vec{B}}{\partial t}\right)=0\,.$$ The electric field cross-commutator bracket gives $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{E}=\frac{\partial}{\partial t}\left(\vec{\nabla}\times\vec{E}\right)-\vec{\nabla}\times\left(\frac{\partial \vec{E}}{\partial t}\right)=0\,.$$ Using Eqs.(20) and (21), this yields $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{E}= \frac{1}{c^2}\frac{\partial^2\vec{B}}{\partial t^2}-\nabla^2\vec{B}- \mu_0\left(\vec{\nabla}\times\vec{J}\right)=0\,.$$ This equation is nothing but Eq.(24) above. Similarly, the magnetic field cross-commutator bracket gives $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{B}=\frac{\partial}{\partial t}\left(\vec{\nabla}\times\vec{B}\right)-\vec{\nabla}\times\left(\frac{\partial \vec{B}}{\partial t}\right)=0\ .$$ Using Eqs.(20) and (21) this yields, $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{B}= \frac{1}{c^2}\frac{\partial^2\vec{E}}{\partial t^2}-\nabla^2\vec{E}+ \mu_0\left(\vec{\nabla}(\rho\, c^2)+\frac{\partial \vec{J}}{\partial t}\right)=0\,.$$ This equation is nothing but Eq.(23) above. Hence, Eqs.(35) and (37), i.e., $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{E}= 0\,,\qquad
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{B}= 0\,.$$ represent the combined Maxwell equations. In terms of the vector $\vec{F}$ defined in Eq.(33), the wave equation in Eq.(35) can be written as $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{F}= 0\,,$$ which is also evident from Eq.(28).
Derivation of Maxwell equations from the vector and scalar potentials, $\vec{A}, \varphi$
==========================================================================================
The electric and magnetic fields are defined by the vector $\vec{A}$ and the scalar potential $\varphi$ as follows $$\vec{E}=-\vec{\nabla}\varphi-\frac{\partial \vec{A}}{\partial t}\,\,,\qquad \vec{B}=\vec{\nabla}\times\vec{A}\,.$$ These are related by the Lorentz gauge as $$\vec{\nabla}\cdot\vec{A}+\frac{1}{c^2}\frac{\partial \varphi}{\partial t}=0\,.$$ Comparing this equation with Eq.(11) reveals that the continuity equation is nothing but a gauge condition. This means that a new current density $\vec{J}'$ can be found so that the equation is still intact. We have recently explored such a possibility and showed that it is true \[3\]. With this motivation the physicality of the gauge $\vec{A}$ exhibited by Aharonov–Bohm effect is tantamount to that of the current density $\vec{J}$\[5\]. The grad-commutator bracket of the scalar potential $\varphi$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]\varphi=\frac{\partial}{\partial t}\left(\vec{\nabla}\varphi\right)-\vec{\nabla}\left(\frac{\partial \varphi}{\partial t}\right)= 0\,.$$ Using Eqs.(40) and (41), one obtains $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\right]\varphi=\frac{1}{c^2}\frac{\partial^2\vec{A}}{\partial t^2}-\nabla^2\vec{A}-\mu_0\vec{J}=0\,.$$ This yields the wave equation of the vector field $\vec{A}$ as $$\label{1}
\frac{1}{c^2}\frac{\partial^2\vec{A}}{\partial t^2}-\nabla^2\vec{A}=\mu_0\vec{J}\,.$$ Similarly, the dot-commutator bracket of the vector $\vec{A}$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{A}=\frac{\partial}{\partial t}\left(\vec{\nabla}\cdot \vec{A}\right)-\vec{\nabla}\cdot\left(\frac{\partial \vec{A}}{\partial t}\right)=0\,.$$ Using Eqs.(40) and (41), one obtains $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right]\vec{A}=\frac{1}{c^2}\frac{\partial^2\varphi}{\partial t^2}-\nabla^2\varphi-\frac{\rho}{\varepsilon_0}=0\,.$$ This yields the wave equation of $\varphi$ $$\label{1}
\frac{1}{c^2}\frac{\partial^2\varphi}{\partial t^2}-\nabla^2\varphi=\frac{\rho}{\varepsilon_0}\,.$$ The cross-commutator bracket of the scalar potential $\varphi$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{A}=\frac{\partial}{\partial t}\left(\vec{\nabla}\times\vec{A}\right)-\vec{\nabla}\times\left(\frac{\partial \vec{A}}{\partial t}\right)= 0\,.$$ Using Eqs.(40), one finds $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right]\vec{A}=\frac{\partial \vec{B}}{\partial t}+\vec{\nabla}\times\vec{E}= 0\,.$$ This yields the Faraday’s equation, $$\label{1}
\vec{\nabla}\times\vec{E}= -\frac{\partial \vec{B}}{\partial t}\,.$$ It is interesting to arrive at this result by using the definition in Eq.(40) only. Now consider the dot-commutator bracket of $\varphi\vec{A}$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\varphi\vec{A}) =\frac{\partial}{\partial t}\left(\vec{\nabla}\cdot(\varphi\vec{A})\right)-\vec{\nabla}\cdot\left(\frac{\partial }{\partial t}(\varphi\vec{A})\right)= 0\,.$$ Using Eqs.(40), (41) and the vector identities $$\label{1}
\vec{\nabla}\cdot(\varphi\vec{G})=(\vec{\nabla}\varphi)\cdot\vec{G}+\varphi(\vec{\nabla}\cdot\vec{G})\,,\qquad
\vec{\nabla}\times(\vec{\nabla}\times\vec{G})=\vec{\nabla}(\vec{\nabla}\cdot\vec{G})-\nabla^2\vec{G}\,,$$ Eq.(51) yields $$\label{1}
\varphi\left(\vec{\nabla}\cdot\vec{E}-\frac{\rho}{\varepsilon_0}\right)-c^2\left(\vec{\nabla}\times\vec{B}-\frac{1}{c^2}\frac{\partial \vec{E}}{\partial t}-\mu_0\vec{J}\right)\cdot\vec{A}=0\,.$$ For arbitrary $\varphi$ and $\vec{A}$, Eq.(53) yields the two equations $$\label{1}
\vec{\nabla}\cdot\vec{E}=\frac{\rho}{\varepsilon_0}\,,$$ and $$\label{1}
\vec{\nabla}\times\vec{B}=\mu_0\vec{J}+\frac{1}{c^2}\frac{\partial \vec{E}}{\partial t}\,.$$ Equations (54) and (55) are the Gauss and Ampere equations.
Similarly, the cross-commutator bracket of $\varphi\vec{A}$ $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right](\varphi\vec{A}) =\frac{\partial}{\partial t}\left(\vec{\nabla}\times(\varphi\vec{A})\right)-\vec{\nabla}\times\left(\frac{\partial }{\partial t}(\varphi\vec{A})\right)= 0\,.$$ Using Eq.(40), (41) and the vector identity $$\label{1}
\vec{\nabla}\times(\varphi\vec{G})=(\vec{\nabla}\varphi)\times\vec{G}+\varphi(\vec{\nabla}\times\vec{G})\,,$$ Eq.(56) yields $$\label{1}
\varphi\left(\vec{\nabla}\times\vec{E}+\frac{\partial \vec{B}}{\partial t}\right)-c^2\left(\vec{\nabla}\times\vec{B}-\frac{1}{c^2}\frac{\partial \vec{E}}{\partial t}-\mu_0\vec{J}\right)\times\vec{A}=0\,.$$ For arbitrary $\varphi$ and $\vec{A}$, Eq.(58) yields the two equations $$\label{1}
\vec{\nabla}\times\vec{E}=-\frac{\partial \vec{B}}{\partial t}\,,$$ and $$\label{1}
\vec{\nabla}\times\vec{B}=\mu_0\vec{J}+\frac{1}{c^2}\frac{\partial \vec{E}}{\partial t}\,.$$ Once again, Eqs.(59) and (60) are the Faraday and Ampere equations, respectively. Hence, the four Maxwell equations are completed. To sum up, Maxwell equations are the commutator brackets $$\label{1}
\left[\frac{\partial}{\partial t}\,, \vec{\nabla}\cdot\right](\varphi\vec{A})= 0\,,\qquad \left[\frac{\partial}{\partial t}\,, \vec{\nabla}\times\right](\varphi\vec{A})=0\,.$$
Energy conservation equation
============================
In electromagnetism, the energy conservation equation for electromagnetic field is written as $$\label{1}
\frac{\partial u}{\partial t}+\vec{\nabla}\cdot\vec{S}=-\vec{J}\cdot\vec{E}\,,$$ where $$\label{1}
u=\frac{1}{2}\,\varepsilon_0E^2+\frac{1}{2\mu_0}B^2\,\,,\qquad \vec{S}=\frac{\vec{E}\times\vec{B}}{\mu_0}\,.$$ The energy conservation equation of the electromagnetic field can be easily obtain using the following vector identity $$\label{1}
\vec{\nabla}\cdot(\vec{F}\times\vec{G})=\vec{G}\cdot(\vec{\nabla}\times\vec{F})-\vec{F}\cdot(\vec{\nabla}\times\vec{G})\,.$$ Let now $\vec{E}=\vec{F}\,\,,\,\,\vec{G}=\vec{B}$, so that Eq.(64) becomes $$\label{1}
\vec{\nabla}\cdot(\vec{E}\times\vec{B})=\vec{B}\cdot(\vec{\nabla}\times\vec{E})-\vec{E}\cdot(\vec{\nabla}\times\vec{B})\,.$$ Employing Eqs.(20), (21) and (63), Eq.(65) yields $$\label{1}
\frac{\partial u}{\partial t}+\vec{\nabla}\cdot\vec{S}=-\vec{J}\cdot\vec{E}\,,$$ which is the familiar energy conservation equation of the electromagnetic field \[5\].
Concluding Remarks
==================
By introducing three vanishing linear differential commutator brackets for scalar and vector fields, $\varphi$ and $\vec{A}$ and the Lorentz gauge connecting them, we have derived the Maxwell’s equations and the continuity equation without resort to any other physical equation. Using different vector identities, we have found that no any independent equation can be generated from the three differential commutators brackets.
Acknowledgements {#acknowledgements .unnumbered}
================
Equations (14) and (15) are in the form of coupled wave equations known as inhomogeneous Helmholtz equations. We see that the current density J enters into these equations in a relatively complicated way, and for this reason these equations and are not readily soluble in general. This work is supported by the university of Khartoum research fund. We are grateful for this support.
References {#references .unnumbered}
==========
$[1]$ Arbab, A. I., and Satti, Z., *Progress in Physics*, **2**, 8 (2009).\
$[2]$ Rindler, W., *Introduction to Special Relativity*, Oxford University Press, USA (1991).\
$[3]$ Arbab, A. I., and Widatallah, H. M., *The generalized continuity equations*, unpublished (2009).\
$[4]$ Jackson, J. D., *Classical Electrodynamics*, Wiley, New York, 2nd edition, (1975).\
$[5]$ Aharonov, Y., and Bohm, D., *Significance of electromagnetic potentials in quantum theory*, *Phys. Rev.* **115**, 485 (1959).\
Appendix {#appendix .unnumbered}
========
$$\left[\frac{\partial}{\partial t}\,\,,\vec{\nabla}\times\right](\vec{F}\times\vec{G})=\vec{F}\times\left(\frac{\partial}{\partial t}(\vec{\nabla}\times\vec{G})-\vec{\nabla}\times\frac{\partial\vec{G}}{\partial t}\right)+\left(\vec{\nabla}\times\frac{\partial\vec{F}}{\partial t}-\frac{\partial}{\partial t}(\vec{\nabla}\times\vec{F})\right)\times\vec{G}\qquad (A1)$$ $$\left[\frac{\partial}{\partial t}\,\,,\vec{\nabla}\right](\vec{F}\cdot\vec{G})=\vec{F}\left(\frac{\partial}{\partial t}(\vec{\nabla}\cdot\vec{G})-\vec{\nabla}\cdot\frac{\partial\vec{G}}{\partial t}\right)+\left(\vec{\nabla}\cdot\frac{\partial\vec{F}}{\partial t}-\frac{\partial}{\partial t}(\vec{\nabla}\cdot\vec{F})\right)\vec{G}\qquad\qquad\qquad\qquad (A2)$$
[^1]: See the Appendix for other identities.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Using Girsanov transformations we construct from sticky reflected Brownian motion on $[0,\infty)$ a conservative diffusion on $E:=[0,\infty)^n$, $n \in \mathbb{N}$, and prove by the probabilistic results of [@CK08] that its transition semigroup possesses the strong Feller property for a specified general class of drift functions. By identifying the Dirichlet form of the constructed process, we characterize it as sticky reflected distorted Brownian motion. In particular, the relations of the underlying analytic Dirichlet form methods to the probabilistic methods of random time changes and Girsanov transformations are presented. Moreover, we prove uniqueness of weak solutions to the corresponding stochastic differential equation and apply our results to the dynamical wetting model.\
[^1]\
[^2]
author:
- 'Martin Grothaus [^3] [^4] '
- 'Robert Voßhall [^5] [^6]'
title: |
Strong Feller properties and uniqueness of\
sticky reflected distorted Brownian motion
---
Introduction
============
In [@FGV13] the authors constructed via Dirichlet form techniques a reflected distorted Brownian motion in $E:=[0,\infty)^n$, $n\in\mathbb{N}$, with sticky boundary behavior which solves the system of stochastic differential equations $$\begin{aligned}
\label{sde!}
d\mathbf{X}_{t}^i=\mathbbm{1}_{(0,\infty)}\big(\mathbf{X}^i_t\big)\,\Big(\sqrt{2}\,dB^i_t+\partial_i\ln \varrho \big(\mathbf{X}_t\big)\,dt\Big)+\frac{1}{\beta}\,\mathbbm{1}_{\{0\}}\big(\mathbf{X}^i_t\big)\,dt,\quad i\in I,\end{aligned}$$ or equivalently $$\begin{gathered}
d\mathbf{X}_{t}^i=\mathbbm{1}_{(0,\infty)}\big(\mathbf{X}^i_t\big)\,\Big(\sqrt{2}\,dB^i_t+\partial_i\ln \varrho \big(\mathbf{X}_t\big)\,dt\Big)+d\ell_t^{0,i},\\
\text{with}\quad \ell_t^{0,i}:=\frac{1}{\beta}\int_0^t\mathbbm{1}_{\{0\}}\big(\mathbf{X}_{s}^i\big)\,ds,\quad i\in I,\end{gathered}$$ weakly for quasi every starting point with respect to the underlying Dirichlet form. Here $I:=\{1,\ldots,n\}$, $\beta$ is a real and positive constant and $(B^i_t)_{t\ge 0}$ are one dimensional independent standard Brownian motions, $i\in I$. $\varrho$ is a continuously differentiable density on $E$ such that for all $B \subset I$, $\varrho$ is almost everywhere positive on $E_+(B)$ with respect to the Lebesgue measure and for all $\varnothing\not=B\subset I$, $\sqrt{\varrho\big|_{E_{+}(B)}}$ is in the Sobolev space of weakly differentiable functions on $E_{+}(B)$, square integrable together with its derivative, where $E_{+}(B):=\{ x \in E \big|~x_i >0 \text{ for all } i \in B, ~x_i=0 \text{ for all } i \in I \setminus B \}$. $\varrho$ continuously differentiable on $E$ implies that the drift part $\big(\partial_i\ln \varrho \big)_{i\in I}$ is continuous on $\{ \varrho > 0 \}$. Moreover, $\ell_t^{0,i}$ is the central local time of the solution to (\[sde!\]), i.e., it holds almost surely $$\begin{aligned}
\ell^{0,i}_t=\frac{1}{\beta}\int_0^t\mathbbm{1}_{\{0\}}\big(\mathbf{X}_{s}^i\big)\,ds
&= \lim_{\varepsilon \downarrow 0} \frac{1}{2 \varepsilon} \int_0^t \mathbbm{1}_{(-\varepsilon,\varepsilon)} \big(\mathbf{X}_{s}^i\big)\, d\langle \mathbf{X}^i \rangle_{s}.\end{aligned}$$ A solution to the associated martingale problem is even provided under the weaker assumptions that $\varrho$ is almost everywhere positive, integrable on each set $E_+(B)$ with respect to the Lebesgue measure and that the respective Hamza condition is fulfilled.\
This kind of stochastic differential equation is strongly related to the sticky Brownian motion on the half-line $[0,\infty)$ (which is occasionally also called Brownian motion with delayed reflection or slowly reflecting Brownian motion). In [@EP12] the authors study Brownian motion on $[0,\infty)$ with sticky boundary behavior and provide existence and uniqueness of solutions to the SDE system $$\begin{aligned}
\label{equEP12}
\left\{
\begin{array}{l}
dX_t=\frac{1}{2}d\ell_t^{0+}\big(X\big)+\mathbbm{1}_{(0,\infty)}\big(X_t\big)\,dB_t\\
\mathbbm{1}_{\{0\}}\,dt=\frac{1}{\mu}\,d\ell_t^{0+}\big(X\big),
\end{array}
\right.\end{aligned}$$ for reflecting Brownian motion $X$ in $[0,\infty)$ sticky at $0$, where $X:=\big(X_t\big)_{t\ge 0}$ starts at $x\in [0,\infty)$, $\mu\in (0,\infty)$ is a given constant, $\ell^{0+}\big(X\big)$ is the right local time of $X$ at $0$ and $B:=\big(B_t\big)_{t\ge 0}$ is the standard Brownian motion. In particular, H.-J. Engelbert and G. Peskir show that the system (\[equEP12\]) has a jointly unique weak solution and moreover, they prove that the system (\[equEP12\]) has no strong solution, thus verifying Skorokhod’s conjecture of the non-existence of a strong solution in this case. For an outline of the historical evolution in the study of sticky Brownian motion we refer to the references given in [@EP12] and also to [@KPS10].\
In view of the results provided in [@EP12], the construction of a weak solution as given in [@FGV13] is the only reasonable one. However, the construction via Dirichlet form techniques has the well-known disadvantage that the constructed process solves the underlying stochastic differential equation only for quasi-every starting point with respect to the Dirichlet form. Hence, in the present paper we construct a transition semigroup by Girsanov transformations and investigate its properties in order to strengthen the results of [@FGV13]. In this way, we obtain a diffusion with strong Feller transition function which solves (\[sde!\]) for [*every*]{} starting point in the state space $E$ and furthermore, we also show an ergodicity theorem for [*every*]{} starting point in the state space $E$ under the assumptions on the density given in Condition \[conditions\]. Moreover, we establish connections between the analytic Dirichlet form construction and classical probabilistic methods. Using these relations, we additionally prove uniqueness of weak solutions to (\[sde!\]).\
In the theory of Dirichlet forms it is a common approach to use results of the regularity theory of elliptic partial differential equations in order to deduce that the associated resolvent and semigroup admit a certain regularity and thereby, it is possible to construct a pointwise solution to the underlying martingale problem or stochastic differential equation for an explicitly known set of starting points under very weak assumptions on the density $\varrho$. For example, this has recently been realized in case of distorted Brownian motion on $\mathbb{R}^d$, $d\in \mathbb{N}$, in [@AKR03], in case of absorbing distorted Brownian motion on $\Omega \subset \mathbb{R}^d$, $d\in \mathbb{N}$, in [@BGS13], in the case of reflecting Brownian motion on Lipschitz domains in [@FT96] and in case of reflecting distorted Brownian motion on $\Omega \subset \mathbb{R}^d$, $d\in \mathbb{N}$, under some smoothness condition on the boundary $\partial \Omega$ in [@FG07] and [@BG14]. However, in the present setting which involves not only the Lebesgue measure but also multiple measures on the boundary of the state space $E$ due to the product structure of the problem, the elliptic regularity theory is not yet investigated and from our present point of view the required results are out of reach. For this reason, we choose the probabilistic approach of random time changes and Girsanov transformations in order to obtain a strong Feller transition semigroup which seems to be a new approach in this area.\
Our results apply to the so-called wetting model (also refered to as the Ginzburg-Landau $\nabla\phi$ interface model with entropic repulsion and pinning). More precisely, in a finite volume $\Lambda\subset \mathbb{Z}^d$, $d\in\mathbb{N}$, the scalar field $\boldsymbol{\phi}_t:=\big(\boldsymbol{\phi}_t(x)\big)_{x\in\Lambda}$, $t\ge 0$, is described by the stochastic differential equations $$\begin{gathered}
\label{sde}
d\boldsymbol{\phi}_t(x)=-\mathbbm{1}_{(0,\infty)}\big(\boldsymbol{\phi}_t(x)\big)\sum_{\stackunder{\scriptscriptstyle{|x-y|=1}}{\scriptscriptstyle{y\in\Lambda}}}V'\big(\boldsymbol{\phi}_t(x)-\boldsymbol{\phi}_t(y)\big)\,dt\\
+\mathbbm{1}_{(0,\infty)}\big(\boldsymbol{\phi}_t(x)\big)\sqrt{2}dB_t(x)+d\ell_{t}^{\scriptscriptstyle{0}}(x),\quad x\in\Lambda,\end{gathered}$$ subject to the conditions: $$\begin{aligned}
&\boldsymbol{\phi}_t(x)\ge 0,\quad \ell_{t}^{\scriptscriptstyle{0}}(x)\mbox{ is non-decreasing with respect to }t,\quad \ell^{\scriptscriptstyle{0}}_{0}(x)=0,\\
&\int_0^\infty \boldsymbol{\phi}_t(x)\,d\ell_{t}^{\scriptscriptstyle{0}}(x)=0,\\
&\beta \ell_{t}^{\scriptscriptstyle{0}}(x)=\int_0^t\mathbbm{1}_{\{0\}}\big(\boldsymbol{\phi}_s(x)\big)\,ds\quad\mbox{for fixed }\beta> 0,\nonumber\\\end{aligned}$$ where $\ell_{t}^{\scriptscriptstyle{0}}(x)$ denotes the *local time* of $\boldsymbol{\phi}_t(x)\mbox{ at }0$. Here $|\cdot|$ denotes the norm induced by the euclidean scalar product on $\mathbb{R}^d$, $V\in C^2(\mathbb{R})$ is a symmetric, strictly convex potential and $\big\{(B_t(x))_{t\ge 0}\,|\,x\in\Lambda\big\}$ are independent standard Brownian motions. In dimension $d=2$ this model describes the wetting of a solid surface by a fluid. More details on interface models are e.g. presented in [@Ga02], [@Fu05].\
In [@Fu05 Sect. 15.1] J.D. Deuschel and T. Funaki investigated (\[sde\]) and gave reference to classical solution techniques as developed e.g. in [@WaIk89]. The methods provided therein require more restrictive assumptions on the drift part as in our situation (e.g. the drift is assumed to be bounded and Lipschitz continuous), moreover, do not apply directly (the geometry and the behavior on the boundary differs). First steps in the direction of applying [@WaIk89] are discussed in [@Fu05] by J.-D. Deuschel and T. Funaki.\
As far as we know the only reference that applies to the system of stochastic differential equations (\[sde\]) is [@Gra88]. By means of a suitable choice of the coefficients the system of equations given by [@Gra88 (II.1)] coincides with (\[sde\]), but amongst others the drift part is also assumed to be Lipschitz continuous and boundend. For this reason, it is not possible to apply the results of [@Gra88] to the setting invenstigated by J.-D. Deuschel and T. Funaki, since the potential $V$ naturally causes an unbounded drift (see also Example \[example\]).\
Our paper is organized as follows. In Section \[secmain\] we state the required conditions on the density as well as our main results. In Section \[DFtrans\] we recall some facts about sticky Brownian motion and present the connections of the Dirichlet form constructed in [@FGV13] to classical methods from probability theory. In particular, we establish relations to random time changes and Girsanov transformations. In Section \[Feller\] a Feller transition semigroup is constructed under the conditions given in Section \[secmain\]. This semigroup is used to construct a pointwise solution to (\[sde!\]) and the corresponding Dirichlet form is identified. Moreover, in Section \[appl\] the setting is applied to the dynamical wetting model. Finally, we prove uniqueness of weak solutions to (\[sde!\]) in Section \[uniqueness\].
Main results {#secmain}
============
In the following we denote by $dx_i$ the one dimensional Lebesgue measure and by $\delta_0^i$ the Dirac measure in $0$, where $i=1,\dots,n$ gives reference to the component of $x=(x_1,\dots, x_n) \in E=[0,\infty)^n$. Define the product measure $d\mu_n:= \prod_{i=1}^n (dx_i + \beta \delta_0^i)$ on $(E,\mathcal{B}(E))$. We denote by $d_{\text{euc}}$ the Euclidean metric.\
First, we like to note that the proofs of the results in [@FGV13] are still valid under the following weaker assumptions:
\[thmimpro\] All results of [@FGV13] hold true under the assumption that $\varrho$ fulfills
1. $\varrho$ is $\mu_n$-a.e. positive on $E$ such that $\varrho \in L^1(E;\mu_n)$,
2. $\varrho \in C(E)$,
3. $\sqrt{\varrho_{|E_+(B)}} \in H^{1,2}_{\text{loc}}(E_+(B))$ for every $\emptyset \neq B \subset I$,
4. $\text{cap}(\{\varrho=0\})=0$ (with respect to the form $(\mathcal{E}^{\varrho},D(\mathcal{E}^{\varrho}))$ defined below).
We state the following proposition in order to be able to give afterwards suitable conditions on the density $\varrho$:
\[propindep\] There exists a diffusion process $\mathbb{M}=(\Omega,\mathcal{F},(\mathcal{F}_t)_{t \geq 0}, (X_t)_{t \geq 0}, (\mathbb{P}_x)_{x \in E})$ (called $n$ independent sticky Brownian motions on $[0,\infty)$) solving the SDE $$\begin{aligned}
dX^i_t= \mathbbm{1}_{(0,\infty)}(X^i_t) \sqrt{2} dB^i_t + \frac{1}{\beta}~ \mathbbm{1}_{\{0\}}(X^i_t)dt, \quad i=1,\dots,n,\end{aligned}$$ where $(B_t)_{t \geq 0}$ is an $n$-dimensional standard Brownian motion, and the transition semigroup $(p_t^{\beta,n})_{t >0}$ of $\mathbb{M}$ has the doubly Feller property, i.e. it is a Feller transition semigroup which admits additionally the strong Feller property (see Definition \[defDF\]). Moreover, the Dirichlet form associated to $n$ independent sticky Brownian motions on $[0,\infty)$ is given by the conservative, strongly local, strongly regular Dirichlet form $(\mathcal{E}^n,D(\mathcal{E}^n))$, i.e., the closure on $L^2(E;\mu_n)$ of the bilinear form $$\mathcal{E}^n(f,g)= \int_E \sum_{i=1}^n \mathbbm{1}_{\{ x_i \neq 0 \}} ~ \partial_i f ~\partial_i g ~ d\mu_n \ \ \ \text{ for } f,g \in C_c^1(E).$$
\[conditions\] $\varrho=\phi^2$ is strictly positive, fulfills the conditions (i)-(iii) of Theorem \[thmimpro\] and $$\begin{aligned}
\label{condlnphi} \nabla \ln \phi= \frac{\nabla \phi}{\phi} \in L^{\infty}_{\text{loc}}(E;\mu_n).\end{aligned}$$ Moreover, for every $t >0$ and every compact set $D \subset E$ holds $$\begin{aligned}
\lim_{k \rightarrow \infty} \sup_{x \in D} ~\mathbb{E}_x(\mathbbm{1}_{\{ \tau_k \leq t \}}~Z_t) =0, \label{condZ}\end{aligned}$$ where $(Z_t)_{t \geq 0}$ is given by $$Z_t=\exp\big(\sqrt{2} \sum_{i=1}^n \int_0^t \partial_i \ln \phi(X_s) \mathbbm{1}_{(0,\infty)}(X_s^i) dB_s^i - \sum_{i=1}^n \int_0^t (\partial_i \ln \phi(X_s))^2 \mathbbm{1}_{(0,\infty)}(X_s^i) ds\big)$$ and $\tau_k:= \inf \{ t>0|~ X_t \notin [0,k)^n \}$ with $(X_t)_{t \geq 0}$ and $(B_t)_{t \geq 0}$ as stated in Proposition \[propindep\].
1. All proofs of the following results are still valid if $\varrho$ is not necessarily strictly positive, but Condition \[conditions\] additionally requires (iv) of Theorem \[thmimpro\], $$\begin{aligned}
\nabla \ln \phi= \frac{\nabla \phi}{\phi} \in L^{\infty}_{\text{loc}}(E \backslash \{ \varrho=0\};\mu_n),\end{aligned}$$ $D$ is an arbitrary compact subset of $E \backslash \{ \varrho =0\}$ and $\tau_k$ is defined by $$\tau_k:= \inf \{ t>0|~ X_t \notin [0,k)^n \backslash B_{\frac{1}{k}}(\{ \varrho=0\}) \},$$ where $B_{\frac{1}{k}}(\{ \varrho=0\}):=\{ x \in E|~\inf_{y \in \{\varrho=0\}}~ d_{\text{euc}}(x,y) \leq \frac{1}{k} \}$. In this case, a strong Feller process on the state space $E_1:= E \backslash \{ \varrho=0\}$ can be constructed and the corresponding Dirichlet form is defined analogously but on the space $L^2(E_1;\varrho \mu)$. The additional condition guarantees that the constructed process (using a Girsanov transformation by $(Z_t)_{t \geq 0}$) never hits the set $\{ \varrho=0\}$.
2. (\[condlnphi\]) is equivalent to $\nabla \phi \in L^{\infty}_{\text{loc}}(E;\mu_n)$, since $\phi$ is assumed to be strictly positive and continuous.
3. (\[condZ\]) holds for example if $\sup_{x \in D} \mathbb{E}_x(Z_t^p) < \infty$ for some $p>1$ (see Remark \[remZ\]).
Under the above assumptions on $\varrho$ it holds:
\[thmmain1\] There exits a conservative diffusion process $\mathbb{M}^{\varrho}=(\Omega,\mathcal{F},(\mathcal{F}_t)_{t \geq 0}, (X_t)_{t \geq 0}, (\mathbb{P}^{\varrho}_x)_{x \in E})$ on $E$ with strong Feller transition function $(p_t)_{t \geq 0}$, i.e., $p_t(\mathcal{B}_b(E)) \subset C_b(E)$, such that the associated Dirichlet form is given by the closure of the symmetric bilinear form $(\mathcal{E}^{\varrho},\mathcal{D})$ on $L^2(E;\varrho \mu_n)$, where $$\begin{aligned}
\mathcal{E}^{\varrho}(f,g)&:= \sum_{\emptyset \neq B \subset \{1,\dots,n\}} \mathcal{E}_B(f,g) \\
&= \int_E \sum_{i=1}^n \mathbbm{1}_{\{ x_i \neq 0 \}} ~ \partial_i f ~\partial_i g ~ \varrho d\mu_n \ \ \ \text{ for } f,g \in \mathcal{D}:=C_c^1(E) \end{aligned}$$ with $$\mathcal{E}_B(f,g):= \int_{E} \sum_{i \in B} \partial_i f ~ \partial_i g~\varrho d\lambda_B^{n,\beta},$$ where $d\lambda_B^{n,\beta}:=\beta^{n-\#B} \prod_{j \in B} dx_j \prod_{j \in B^c} \delta_0^j$. In particular, $(p_t)_{t \geq 0}$ fulfills the absolute continuity condition [@FOT94 (4.2.9)], i.e., the transition probabilities $p_t(x,\cdot)$, $x \in E$, $t>0$, given by $p_t(x,A):=\mathbb{P}_x(X_t \in A)$, $A \in \mathcal{B}(E)$, are absolutely continuous with respect to $\varrho \mu_n$.
\[thmmain2\] Let $\mathbb{M}^{\varrho}$ be the diffusion process of Theorem \[thmmain1\]. It holds for each $i=1,\dots,n$ $$\begin{aligned}
\label{main} X_t^i=X_0^i + \sqrt{2} \int_0^t \mathbbm{1}_{(0,\infty)}(X_s^i) dB_s^i + \int_0^t \mathbbm{1}_{(0,\infty)}(X_s^i)~ \partial_i \ln \varrho(X_s) ds + \frac{1}{\beta} \int_0^t \mathbbm{1}_{\{0\}}(X_s^i) ds\end{aligned}$$ $\mathbb{P}^{\varrho}_x$-a.s. for every $x \in E$, where $(B_t^i)_{t \geq 0}$, $i=1,\dots,n$, are independent standard Brownian motions. Moreover, it holds $$\begin{aligned}
\label{eqnergo} \lim_{t \rightarrow \infty} \frac{1}{t} \int_0^t F(X_s) ds= \frac{\int_E F \varrho d\mu_n}{\int_E \varrho d\mu_n} \end{aligned}$$ $\mathbb{P}^{\varrho}_x$-a.s. for every $x \in E$ and $F \in L^1(E; \varrho \mu_n)$.
Let $\Gamma \subset \partial E$ such that $\int_{\Gamma} \varrho d\mu_n >0$. Then it follows by (\[eqnergo\]) that $$\lim_{t \rightarrow \infty} \frac{1}{t} \int_0^t \mathbbm{1}_{\Gamma}(X_s) ds= \frac{\int_{\Gamma} \varrho d\mu_n}{\int_E \varrho d\mu_n} >0$$ $\mathbb{P}^{\varrho}_x$-a.s. for every $x \in E$. This confirms the sticky behavior of the process on the boundary.
The solution to (\[main\]) is unique in law.
Let $\varrho:E \rightarrow (0,\infty)$, $\varrho=\exp(-2H)$, be defined by a potential $V$ with nearest neighbor pair interaction, i.e., $H$ is given by $$\begin{aligned}
\label{hamilt} H(x_1,\cdots,x_n)= \frac{1}{4} \sum_{\stackunder{|i-j|=1}{i,j \in \{0,\dots,n+1\}}} V(x_i-x_j),\end{aligned}$$ where $x_0:=x_{n+1}:=0$ and $V:\mathbb{R} \rightarrow [-b,\infty)$, $b \in [0, \infty)$, fulfills the conditions of [@Fu05 (2.2)]:
1. $V \in C^2(\mathbb{R})$,
2. $V$ is symmetric, i.e., $V(r)=V(-r)$ for all $r \in \mathbb{R}$,
3. $V$ is strictly convex, i.e., $c_{-} \leq V^{\prime \prime}(r) \leq c_{+}$ for all $r \in \mathbb{R}$ and some constants $c_{-},c_{+} >0$.
Denote by $\phi:=\sqrt{\varrho}=\exp(-H)$ the square root of $\varrho$.\
Define $\mathbb{V}^{\prime}(i,x)$ for $i=1,\dots,n$ and $x \in E$ by $$\mathbb{V}^{\prime}(i,x):= \sum_{\stackunder{|i-j|=1}{j \in \{0,\dots,n+1\}}} V^{\prime}(x_i-x_j).$$
In this case, Condition \[conditions\] is fulfilled and the stated results hold accordingly with the drift function given by $\partial_i \ln \varrho=- \mathbb{V}^{\prime}(i,\cdot)$, $i=1,\dots,n$.
Sticky Brownian motion and Dirichlet form transformations {#DFtrans}
=========================================================
Sticky Brownian motion on the halfline {#secsticky}
--------------------------------------
Define the Dirichlet form $(\hat{\mathcal{E}},D(\hat{\mathcal{E}}))$ as the closure of $$\hat{\mathcal{E}}(f,g):= \int_{[0,\infty)} f^{\prime}(x) g^{\prime}(x)dx, \ \ f,g \in C_c^1([0,\infty)),$$ on $L^2([0,\infty);dx)$. It is well-known that reflecting Brownian motion is associated to $(\hat{\mathcal{E}},D(\hat{\mathcal{E}}))$ and $D(\hat{\mathcal{E}})=H^{1,2}((0,\infty))$ is the Sobolev space of order one.\
Let $(\tilde{B}_t)_{t \geq 0}$ be a standard Brownian motion defined on a probability space $(\Omega, \mathcal{F}, \mathbb{P})$. Then $\hat{X}_t:=|x + \sqrt{2} \tilde{B}_t|$, $t \geq 0$, yields reflecting Brownian motion on $[0,\infty)$ starting at $x \in [0,\infty)$ and by Tanaka’s formula $$\begin{aligned}
\label{refltanaka} \hat{X}_t=x + \sqrt{2} \hat{B}_t + L_t^{0+}, \ \ t \geq 0, \end{aligned}$$ where $\hat{B}_t:=\int_0^t \text{sgn}(x+ \sqrt{2}\tilde{B}_s) d\tilde{B}_s$, $t \geq 0$, is a standard Brownian motion and $(L_t^{0+})_{t \geq 0}$ is the right local time in $0$, i.e., $$L_t^{0+}= \lim_{\epsilon \rightarrow 0} \int_0^t \mathbbm{1}_{[0,\epsilon)}(\hat{X}_s) ds$$ in probability. Here, we differ from classical notation by the factor $\sqrt{2}$ (see also Remark \[remsqrt2\]). The Dirichlet form associated to $(\hat{X}_t)_{t\geq 0}$ is $(\hat{\mathcal{E}},D(\hat{\mathcal{E}}))$ and $(L_t^{0+})_{t \geq 0}$ is an additive functional which is in Revuz correspondance with the Dirac measure $\delta_0$ in $0$. Consider the additive functional $A_t:= t + \beta L_t^{0+}$, $t \geq 0$, for some real constant $\beta >0$. Note that $A_0=0$ and $A_t \rightarrow \infty$ a.s. as $t \rightarrow \infty$. Then sticky Brownian motion on $[0, \infty)$ is usually constructed by a random time change using the inverse $\tau(t)$ of $A_t$. More precisely, $X_t:= \hat{X}_{\tau(t)}$ (starting in $x$) solves the stochastic differential equation $$\begin{aligned}
\label{1dstickysde} dX_t= \mathbbm{1}_{(0,\infty)}(X_t) \sqrt{2} dB_t + \frac{1}{\beta}~ \mathbbm{1}_{\{0\}}(X_t)dt,\end{aligned}$$ where $(B_t)_{t \geq 0}$ is a standard Brownian motion. For details on Feller’s Brownian motions and in particular, sticky Brownian motion and its transition semigroup, see e.g. [@EP12], [@KPS10], [@GS72] or [@Kni81].\
In [@FOT94 Chapter 6] and [@ChFu11 Chapter 5] is presented how a random time change by an additive functional affects the underlying Dirichlet form. Let $\mu$ denote the Revuz measure corresponding to $(A_t)_{t \geq 0}$. Clearly, $d\mu=dx + \beta \delta_0$. In particular, $\mu$ has full support $[0,\infty)$. Thus, the Dirichlet form $(\mathcal{E},D(\mathcal{E}))$ on $L^2([0,\infty);\mu)$ associated to $(X_t)_{t \geq 0}$ has the representation $$\mathcal{E}(f,g)=\hat{\mathcal{E}}(f,g) \ \ f,g \in D(\mathcal{E})=D(\hat{\mathcal{E}}) \cap L^2([0,\infty);\mu).$$ In particular, $D(\mathcal{E})=H^{1,2}((0,\infty)) \cap L^2([0,\infty);\mu) = H^{1,2}((0,\infty))$ by Sobolev embedding. Moreover, $C_c^1([0,\infty))$ is dense in $D(\mathcal{E})$ by [@ChFu11 Theorem 5.2.8(i)] and thus, it is a special standard core of $(\mathcal{E},D(\mathcal{E}))$. Hence, the closure of $$\begin{aligned}
\label{1dform}
\mathcal{E}(f,g)= \int_{[0,\infty)} f^{\prime}(x) g^{\prime}(x)dx= \int_{[0,\infty)} \mathbbm{1}_{(0,\infty)}(x)~f^{\prime}(x) g^{\prime}(x)d\mu, \ \ f,g \in C_c^1([0,\infty)),\end{aligned}$$ on $L^2([0,\infty);\mu)$ is the Dirichlet form associated to $(X_t)_{t \geq 0}$.
\[remsqrt2\] Note that our notion for the solution to the equations (\[refltanaka\]) and (\[1dstickysde\]) as reflecting Brownian motion and sticky reflecting Brownian motion on $[0,\infty)$ respectively differs by the factor $\sqrt{2}$ from classical literature in view of the underlying SDE (\[sde!\]). If $(Y^{\gamma}_t)_{t \geq 0}$ solves $$dY^{\gamma}_t = \mathbbm{1}_{(0,\infty)}(Y^{\gamma}_t) dB_t + \frac{1}{\gamma} \mathbbm{1}_{\{0\}}(Y_t^{\gamma}) dt ~\text{ for } \gamma >0,$$ we obtain the solution to (\[1dstickysde\]) by setting $X_t:=\sqrt{2}~ Y^{\sqrt{2}\beta}_t$. This identity is useful in order to derive the resolvent density and transition density for the solution to (\[1dstickysde\]).
Let $F$ be a locally compact separable metric space and denote by $C_0(F):=\{ f \in C(F)|~ \forall \epsilon >0~ \exists K \subset F \text{ compact }: |f(x)| < \epsilon ~ \forall x \in F \backslash K \}$ the space of continuous functions on $F$ vanishing at infinity. We can specify the resolvent and transition semigroup of sticky Brownian motion on $[0,\infty)$. [@KPS10 Corollary 3.10, Corollary 3.11] state the following (see also [@Kni81 Section 6.1]):
\[thmdensity\] The transition function of sticky Brownian motion on $[0,\infty)$ yields a Feller semigroup on $C_0([0,\infty))$, i.e., $p_t(C_0([0,\infty))) \subset C_0([0,\infty))$ and $\lim_{t \downarrow 0} \Vert p_t f -f \Vert_{\infty} =0$ for each $f \in C_0([0,\infty))$. For $\lambda >0$, $x,y \in [0,\infty)$, the resolvent kernel $r_{\lambda}^{\beta}(x,dy)$ of the Brownian motion with sticky origin (i.e., the solution to (\[1dstickysde\])) is given by $$\begin{aligned}
r_{\lambda}^{\beta}(x,dy)= \frac{r_{\lambda}^D(x,\frac{y}{\sqrt{2}})}{\sqrt{2}} dy + \frac{1}{2 (\sqrt{\lambda}+ \beta \lambda )} \big(2 e^{-\sqrt{2 \lambda}(x+\frac{y}{\sqrt{2}})}dy + \sqrt{2} \beta ~e^{-\sqrt{2 \lambda} x} \delta_0(dy)\big),\end{aligned}$$ where $r_{\lambda}^D(x,y)= \frac{1}{\sqrt{2\lambda}} (e^{-\sqrt{2\lambda}|x-y|}-e^{-\sqrt{2\lambda}(x+y)})$ is the resolvent density of Brownian motion with Dirichlet boundary conditions.
Furthermore, by the inverse Laplace transform it follows that, for $t >0$, the transition kernel $p^{\beta}(t,x,dy)$ of the Brownian motion with sticky origin is given by $$\begin{aligned}
\label{semigroup}
p^{\beta}(t,x,dy)=\frac{p^D(t,x,\frac{y}{\sqrt{2}})}{\sqrt{2}} dy + 2 g_{0,\sqrt{2} \beta}(t,x+\frac{y}{\sqrt{2}})dy + \sqrt{2} \beta~ g_{0,\sqrt{2} \beta}(t,x)~ \delta_0(dy), \end{aligned}$$ where $p^D(t,x,y)=p(t,x,y)-p(t,x,-y)$ is the transition density for Brownian motion with Dirichlet boundary conditions, $p(t,x,y)= \frac{1}{\sqrt{2\pi t}} e^{-\frac{(x-y)^2}{2t}}$ and $$g_{0,\gamma}(t,x)= \frac{1}{\gamma} \exp(\frac{2x}{\gamma}+\frac{2t}{{\gamma}^2})~ \textnormal{erfc}(\frac{x}{\sqrt{2t}} + \frac{\sqrt{2 t}}{\gamma}), \ \ \text{for } \gamma >0,~ t >0,~ x \geq 0,$$ with the complementary errorfunction $\textnormal{erfc}(x)=\frac{2}{\sqrt{\pi}} \int_x^{\infty} e^{-z^2} dz$, $x \in \mathbb{R}$.
Note that (\[semigroup\]) implies that $p^{\beta}(t,x,\cdot)$ is absolutely continuous with respect to the measure $d\mu=dx + \beta \delta_0$ for each $x \in [0,\infty)$, $t>0$. Therefore, the so-called [*absolute continuity condition*]{} [@FOT94 (4.2.9)] is fulfilled. In the following we see that the transition semigroup possesses even stronger properties.
Thus, with $p^{\beta}(t,x,dy)$ as above and $p^{\beta}_t$, $t >0$, the transition semigroup of sticky Brownian motion starting in $x \in [0,\infty)$, it holds $$\mathbb{E}_x(f(X_t))=p^{\beta}_tf(x)= \int_{[0,\infty)} f(y) ~p^{\beta}(t,x,dy)$$ for each $f \in C_0([0,\infty))$. Furthermore, the resolvent $r_{\lambda}^{\beta}$ is given by $$\mathbb{E}_x \big( \int_0^{\infty} e^{-\lambda s} f(X_s) ds \big) = \int_0^{\infty} e^{-\lambda s} p_s^{\beta}f(x) ds= r_{\lambda}^{\beta} f(x)=\int_{[0,\infty)} f(y)~ r^{\beta}_{\lambda}(x,dy).$$ The proof of Theorem \[thmdensity\] is based on the so-called first passage time formular (see [@Kni81 (6.4)]).\
Let $\lambda >0$ and define $A^{\beta}:=\lambda - (r_{\lambda}^{\beta})^{-1}$ on $\mathcal{D}:=r_{\lambda}^{\beta} (C_0([0,\infty)))$ (which is independent of $\lambda$). By [@Kni81 Theorem 6.2, Theorem 6.4] is holds that $$\begin{aligned}
\label{genC} A^{\beta} f= f^{\prime \prime}, \ \ \ f \in \mathcal{D}=\{ f \in C_0([0,\infty)) \cap C^2([0,\infty)|~ f^{\prime \prime} \in C_0([0,\infty)) \text{ and } \beta f^{\prime \prime}(0)=f^{\prime}(0) \}. \end{aligned}$$ The condition $\beta f^{\prime \prime}(0)=f^{\prime}(0)$ for $f \in C^2([0,\infty))$ is called Wentzell boundary condition.
\[defDF\] Let $F$ be a locally compact separable metric space. A transition semigroup $p_t$, $t >0$, of an $F$-valued Markov process is said to have the [*Feller property*]{} if $p_t (C_0(F)) \subset C_0(F)$ and $\lim_{t \downarrow 0} \Vert p_t f - f \Vert_{\infty}=0$ for each $f \in C_0(F)$. Furthermore, it is called [*strong Feller*]{} if $p_t(\mathcal{B}_b(F)) \subset C_b(F)$ for each $t >0$. If the transition semigroup has both Feller and strong Feller property, we say that it possesses the [*doubly Feller property*]{}.
We can also deduce the following:
\[1ddoublyfeller\] The transition semigroup $(p_t^{\beta})_{t >0}$ of sticky Brownian motion on $[0,\infty)$ has the doubly Feller property.
In consideration of Theorem \[thmdensity\] it rests to show that $p_t(\mathcal{B}_b([0,\infty))) \subset C_b([0,\infty))$. Let $f \in \mathcal{B}_b([0,\infty))$ and $t >0$. It is well-known that $$\frac{e^{-x^2}}{x + \sqrt{x^2 +2}} < \text{erfc}(x) \leq \frac{e^{-x^2}}{x + \sqrt{x^2 + \frac{4}{\pi}}}$$ for each $x \geq 0$ (see [@AS64 7.1.13]). Let $x \in [0, \infty)$ and $(x_n)_{n \in \mathbb{N}}$ a sequence in $[0,\infty)$ such that $x_n \rightarrow x$ as $n \rightarrow \infty$. Then $G_n(y):=f(y) g_{0,\sqrt{2} \beta}(t,x_n +\frac{y}{\sqrt{2}})$ converges for each fixed $y \in [0,\infty)$ to $G(y):=f(y) g_{0,\sqrt{2} \beta}(t,x +\frac{y}{\sqrt{2}})$ as $n \rightarrow \infty$ by continuity of $g_{0,\sqrt{2} \beta}$ in the second variable. Moreover, for each $y \in [0,\infty)$ holds $$\begin{aligned}
|G_n(y)| &\leq \Vert f \Vert_{\infty} K_1 \exp(\frac{\sqrt{2} x_n +y)}{\beta}) \text{erfc}(\frac{x_n}{\sqrt{2t}}+\frac{y}{2\sqrt{t}}) \\
&\leq \Vert f \Vert_{\infty} K_2 \exp(\frac{y}{\beta}) \text{erfc}(\frac{y}{2 \sqrt{t}}) \\
&\leq \Vert f \Vert_{\infty} K_3 \exp(\frac{y}{\beta}) \exp(-\frac{y^2}{4t})=:H(y)\end{aligned}$$ for suitable constants $K_1$, $K_2$ and $K_3$. Note that the function $H$ is integrable with respect to the Lebesgue measure on $[0,\infty)$. Thus, dominated convergence yields $$\int_{[0,\infty)} G_n(y) dy \rightarrow \int_{[0,\infty)} G(y) dy$$ and by this, we can conclude that $p^{\beta}_tf$ is continuous and bounded.
Denote by $(T_t^{\beta})_{t \geq 0}$ the $L^2([0,\infty);\mu)$-semigroup of $(\mathcal{E},D(\mathcal{E}))$ defined in (\[1dform\]). Then, by the previous considerations, for all $f \in \mathcal{B}_b([0,\infty)) \cap L^2([0,\infty);\mu)$ it holds that $p_t^{\beta} f$ is a $\mu$-version of $T_t^{\beta} f$. Note also that the $L^2([0,\infty);\mu)$-generator $(L,D(L))$ is given by $$Lf(x)= \mathbbm{1}_{(0,\infty)}(x) f^{\prime \prime}(x) + \mathbbm{1}_{\{0\}}(x) \frac{1}{\beta} f^{\prime}(x) \ \ \ \text{ for } f \in D(L)= H^{2,2}((0,\infty)),$$ where $H^{2,2}((0,\infty))$ denotes the Sobolev space of order two. This can be shown using integration by parts, the fact that $D(\mathcal{E})=H^{1,2}((0,\infty))$ and the definition of the space $H^{2,2}((0,\infty))$. For $f \in C_c^2([0,\infty)) \subset D(L)$ such that the Wentzell boundary condition $\beta f^{\prime \prime}(0)=f^{\prime}(0)$ is fulfilled, it holds $Lf=f^{\prime \prime}$ similar to the generator of the $C_0([0,\infty))$-semigroup given in (\[genC\]). However, in the $L^2$-setting the boundary behavior is rather described by the measure $\mu$ instead of the domain of the generator.
Next we will constuct the Dirichlet form corresponding to $n$ independent sticky Brownian motions on $[0,\infty)$, $n \in \mathbb{N}$. In [@BH91 Chapter V, Section 2.1] it is shown how to construct finite tensor products of Dirichlet spaces. Moreover, the corresponding semigroup of the product Dirichlet form has an explicit representation. In our setting this construction yields the semigroup of an $n$-dimensional process on $E=[0,\infty)^n$, $n \in \mathbb{N}$, such that the components are independent sticky Brownian motions on $[0,\infty)$. In particular, this approach justifies the choice of the Dirichlet form structure used in [@FGV13].\
Let $(\mathcal{E}_i,D(\mathcal{E}_i))$, $i=1,\dots,n$, be $n$ copies of the Dirichlet form in (\[1dform\]). Note that each such form is defined on the space $L^2([0,\infty);\mu)$. In accordance with [@BH91 Definition 2.1.1] we define the product Dirichlet form $(\mathcal{E}^n,D(\mathcal{E}^n))$ on $L^2([0,\infty)^n; \mu_n)$ with $d\mu_n=\prod_{i=1}^n (dx_i + \beta \delta_0^i)$ by
$$\begin{aligned}
\label{ndform}
\mathcal{E}^n(f,g):=\sum_{i=1}^n \int_{[0,\infty)^{n-1}} \mathcal{E}_i(f(x_1,\dots, x_{i-1},\cdot,x_{i+1},\dots,x_n),g(x_1,\dots ,x_{i-1},\cdot ,x_{i+1},\dots ,x_n)) \prod_{j \neq i} (dx_j + \beta \delta_0^j)\end{aligned}$$
for $f,g \in D(\mathcal{E}^n)$, where $$\begin{aligned}
D(\mathcal{E}^n):=\{ &f \in L^2([0,\infty)^n;\mu_n)\big|~\text{for each } i=1,\dots,n \text{ and for } \prod_{j \neq i}(dx_j+\beta \delta^j_0)-{a.e. } \\
&(x_1,\dots ,x_{i-1},x_{i+1},\dots ,x_n) \in [0,\infty)^{n-1}: f(x_1,\dots, x_{i-1},\cdot,x_{i+1},\dots,x_n) \in D(\mathcal{E}_i) \}\end{aligned}$$ First, we proof the following:
\[lemdense\] $C_c^1([0,\infty)^n)$ is dense in $D(\mathcal{E}^n)$.
Note that $C_c^1([0,\infty)^n) \subset D(\mathcal{E}^n)$ by definition of $D(\mathcal{E}^n)$.\
W.l.o.g. let $n=2$. By [@BH91 Proposition 2.1.3 b)] $D(\mathcal{E}_1) \otimes D(\mathcal{E}_2)$ is dense in $D(\mathcal{E}^2)$. We show that $C_c^1([0,\infty)) \otimes C_c^1([0,\infty)) \subset C_c^1([0,\infty)^2)$ is dense in $D(\mathcal{E}_1) \otimes D(\mathcal{E}_2)$. Then the assertion follows by a diagonal sequence argument. So let $h \in D(\mathcal{E}_1) \otimes D(\mathcal{E}_2)$ such that $h(x_1,x_2)=f(x_1)g(x_2)$ for $\mu_2$-a.e. $(x_1,x_2) \in [0,\infty)^2$, $f \in D(\mathcal{E}_1)$ and $g \in D(\mathcal{E}_2)$. Choose sequences $(f_k)_{k \in \mathbb{N}}$ and $(g_k)_{k \in \mathbb{N}}$ in $C_c^1([0,\infty))$ such that $f_k \rightarrow f$ in $D(\mathcal{E}_1)$ and $g_k \rightarrow g$ in $D(\mathcal{E}_2)$ as $k \rightarrow \infty$ and define, for $k \in \mathbb{N}$, $h_k \in C_c^1([0,\infty)) \otimes C_c^1([0,\infty))$ by $h_k(x_1,x_2):=f_k(x_1)g_k(x_2)$, $x_1,x_2 \in [0,\infty)$. Then it follows immediately by assumption and the product structure that $h_k \rightarrow h$ as $k \rightarrow \infty$ in $L^2([0,\infty)^2; d\mu)$. Moreover, for $k,l \in \mathbb{N}$ $$\begin{aligned}
\mathcal{E}^2(h_k-h_l)&=\int_{[0,\infty)} \mathcal{E}_1((h_k-h_l)(\cdot,x_2)) (dx_2 + \beta \delta_0^2) + \int_{[0,\infty)} \mathcal{E}_2((h_k-h_l)(x_1, \cdot)) (dx_1 + \beta \delta_0^1)\\
\leq &~\mathcal{E}_1(f_k-f_l)~ \Vert g_k \Vert_{L^2([0,\infty);dx+\beta \delta_0)} + \mathcal{E}_1(f_l)~ \Vert g_k - g_l \Vert_{L^2([0,\infty);dx+\beta \delta_0)} \\
&+ \mathcal{E}_2(g_k-g_l)~ \Vert f_k \Vert_{L^2([0,\infty);dx+\beta \delta_0)} + \mathcal{E}_1(g_l)~ \Vert f_k - f_l \Vert_{L^2([0,\infty);dx+\beta \delta_0)}.\end{aligned}$$ Hence, $\mathcal{E}^2(h_k-h_l) \rightarrow 0$ as $k,l \rightarrow \infty$ and thus, $h_k \rightarrow h$ as $k \rightarrow \infty$ in $D(\mathcal{E}^2)$.
Let $f,g \in C_c^1([0,\infty)^n)$. Then for each $i=1,\dots,n$ and fixed $(x_1,\dots ,x_{i-1},x_{i+1},\dots ,x_n) \in [0,\infty)^{n-1}$ we have $$\begin{aligned}
\mathcal{E}_i&(f(x_1,\dots, x_{i-1},\cdot,x_{i+1},\dots,x_n),g(x_1,\dots ,x_{i-1},\cdot ,x_{i+1},\dots ,x_n)) \\
&= \int_{[0,\infty)} \partial_i f(x_1,\dots ,x_n)~ \partial_i g(x_1,\dots ,x_n) ~dx_i.\end{aligned}$$ Set $\{ j \neq i \}:=\{1,\dots,i,i+1,\dots,n \}$. If $A$ is a subset of some set $I$, we denote by $A^c$ the set $I \backslash A$. Due to the identity $$\prod_{j \neq i} (dx_j + \beta \delta_0^j)= \sum_{A \subset \{j \neq i\}} \beta^{\#A^c} \prod_{j \in A} dx_j \prod_{j \in A^c} \delta_0^j$$ we get by rearranging the terms that $$\mathcal{E}^n(f,g)= \sum_{\emptyset \neq B \subset \{1,\dots,n\}} \mathcal{E}_B(f,g)$$ with $$\mathcal{E}_B(f,g):= \int_{[0,\infty)^n} \sum_{i \in B} \partial_i f ~ \partial_i g~d\lambda_B^{n,s},$$ where $d\lambda_B^{n,\beta}:=\beta^{n-\#B} \prod_{j \in B} dx_j \prod_{j \in B^c} \delta_0^j$. In other words, $(\mathcal{E}^n,D(\mathcal{E}^n))$ defined in (\[ndform\]) coincides with the form defined in [@FGV13 (2.3)] disregarding that in our present setting the density function $\varrho$ is identically one. Moreover, (\[ndform\]) can also be rewritten in the form $$\mathcal{E}^n(f,g)= \int_E \sum_{i=1}^n \mathbbm{1}_{\{ x_i \neq 0 \}} ~ \partial_i f ~\partial_i g ~ d\mu_n \ \ \ \text{ for } f,g \in C_c^1(E).$$ From the present point of view $(\mathcal{E}^n,D(\mathcal{E}^n))$, defined as in (\[ndform\]), is the sum of $n$ subforms and each such form for $i=1,\dots,n$ describes the dynamics of the process on $[0,\infty)^n$ for all configurations where the $i$-th component is not pinned to zero. In contrast, the forms $\mathcal{E}_B$, $\emptyset \neq B \subset \{1,\dots,n\}$ describe the dynamics of the process for all configurations where exactly the components specified by $B$ are non-zero.\
By a minor generalization of the results in [@FGV13] we get the following lemma:
The Dirichlet form $(\mathcal{E}^n,D(\mathcal{E}^n))$ on $L^2([0,\infty)^n;\mu_n)$, $n \in \mathbb{N}$, is conservative, strongly local, strongly regular and symmetric.
Let $x=(x_1,\dots,x_n),y=(y_1,\dots,y_n) \in [0,\infty)^n$, $n \in \mathbb{N}$. Then the transition kernel $p_t^{\beta,n}(x,dy)$ of $n$ independent sticky Brownian motions on $[0,\infty)$ is given by $$p_t^{\beta,n}(x,dy)=\prod_{i=1}^n p_t^{\beta}(x_i,dy_i).$$ Thus, for $f \in C_0([0,\infty)^n)$ we have $$p_t^{\beta,n}f(x)=\int_{[0,\infty)^n} f(y_1,\dots,y_n) \prod_{i=1}^n p_t^{\beta}(x_i,dy_i).$$ By Theorem \[thmdensity\] we have an explicit representation of $p_t^{\beta,n}(x,dy)$ and by the same arguments as in Proposition \[1ddoublyfeller\] the doubly Feller porperty holds also for $p_t^{\beta,n}$:
The transition semigroup $(p_t^{\beta,n})_{t >0}$ of $n$ independent sticky Brownian motions on $[0,\infty)$ has the doubly Feller property.
Let $(T_t^i)_{t \geq 0}$ be the $L^2([0,\infty);\mu)$-semigroup of the forms $(\mathcal{E}_i,D(\mathcal{E}_i))$, $i=1,\dots,n$. Set for $f \in L^2([0,\infty)^n;\mu_n)$, $i=1,\dots,n$, and $\mu_n$-a.e. $(x_1,\dots,x_n) \in [0,\infty)^n$ $$\hat{T}_t^{\beta,i} f(x_1,\dots,x_n):= T_t^i f(x_1,\dots,x_{i-1},\cdot,x_{i+1},\dots,x_n)(x_i).$$ and $$T_t^{\beta,n} f= \hat{T}_t^{\beta,1} \cdots \hat{T}_t^{\beta,n} f.$$ By [@BH91 Proposition 2.1.3 a)] $(T_t^{\beta,n})_{t \geq 0}$ is the $L^2([0,\infty)^n;\mu_n)$-semigroup associated to the form $(\mathcal{E}^n,D(\mathcal{E}^n))$ defined in (\[ndform\]) and the order of the $\hat{T}_t ^{\beta,i}$, $i=1,\dots,n$, is arbitrary.\
Let $f \in \mathcal{B}_b([0,\infty)^n) \cap L^2([0,\infty)^n;\mu_n)$. Then we have for $\mu_n$-a.e. $x=(x_1,\dots,x_n) \in [0,\infty)^n$ $$\begin{aligned}
\label{iter1}
\hat{T}_t^{\beta,n} f(x_1,\dots,x_n)=T_t^n f(x_1,\dots,x_{n-1},\cdot)(x_n) &=p_t^{\beta} f(x_1,\dots,x_{n-1},\cdot)(x_n) \\
&=\int_{[0,\infty)} f(x_1,\dots,x_{n-1},y_n) p^{\beta}_t(x_n,dy_n) \notag\end{aligned}$$ and similarly $$\begin{aligned}
\label{iter2}
\hat{T}_t^{\beta,n-1} \hat{T}_t^{\beta,n} f(x_1,\dots,x_n)= \int_{[0,\infty)} \int_{[0,\infty)} f(x_1,\dots,x_{n-2},y_{n-1},y_n) p^{\beta}_t(x_n,dy_n) p^{\beta}_t(x_{n-1},dy_{n-1}).\end{aligned}$$
Proceeding successively as in (\[iter1\]) and (\[iter2\]), together with the preceding considerations, proves Proposition \[propindep\].
Girsanov transformations {#secgirsanov}
------------------------
We summerize some results on Girsanov transformations of a Markov process and the associated Dirichlet form. The statements can be found in [@Ebe96] and [@FOT94 Chapter 6]. In some cases we do not state the results in full generality, since for our purposes it is sufficient to simplify the assumptions.\
Let $\mathbb{M}=(\Omega,\mathcal{F},(\mathcal{F}_t)_{t \geq 0}, (X_t)_{t \geq 0}, (\mathbb{P}_x)_{x \in F})$ be a $\mu$-symmetric strong Markov process with state space $F \subset \mathbb{R}^n$, $n \in \mathbb{N}$, continuous sample paths and infinite lifetime, where $\mu$ is a positive Radon measure on $(F,\mathcal{B}(F))$ with full support. We suppose that the process is canonical, i.e., $\Omega=C([0,\infty),F)$ and $X_t(\omega)=\omega(t)$ for $\omega \in \Omega$ and $t \geq 0$. Moreover, assume that its Dirichlet form $(\mathcal{E},D(\mathcal{E}))$ on $L^2(F;\mu)$ is strongly regular, strongly local, conservative and that it possesses a square field operator $\Gamma$. We denote its generator by $(L,D(L))$. Suppose that $\mathcal{D}:=C_c^1(F)$ a dense subspace of $D(\mathcal{E})$, $D(L) \cap \mathcal{D}$ is dense in $D(\mathcal{E})$ and for every $f \in \mathcal{D}$ holds $f,\Gamma(f) \in L^{\infty}(F;\mu)$. Denote by $(p_t)_{ t>0}$ the transition semigroup of $\mathbb{M}$, i.e., for $f \in \mathcal{B}_b(F)$ holds $$p_tf(x):= \mathbb{E}_x(f(X_t)),$$ and we suppose that the transition density $p_t(x,\cdot)$, $x \in F$, $t >0$, possesses the absolute continuity condition [@FOT94 (4.2.9)].\
A function $f$ is said to be in $D(\mathcal{E})_{\text{loc}}$ if for any relatively compact open set $G$ there exists a function $g \in D(\mathcal{E})$ such that $f=g$ $\mu$-a.e. on $G$. Fix some $\phi \in D(\mathcal{E})_{\text{loc}} \cap C(F)$ such that $\phi >0$ $\mu$-a.e.. Define $\varrho:=\phi^2$ and the symmetric bilinear form $(\mathcal{E}^{\varrho},\mathcal{D}^{\varrho})$ on $L^2(F; \varrho \mu)$ by $$\begin{aligned}
&\mathcal{D}^{\varrho}:=\{ f \in D(\mathcal{E})|~\int_F (\Gamma(f)+f^2) \varrho d\mu < \infty \}, \\
&\mathcal{E}^{\varrho}(f,g):= \int_F \Gamma(f,g)~ \varrho d\mu. \notag\end{aligned}$$ In particular, $\mathcal{D}^{\varrho}=D(\mathcal{E})$ if $\varrho$ is bounded.\
Under the above assumptions the conditions (D1)-(D3) of [@Ebe96] are fulfilled using the strong regularity of $(\mathcal{E},D(\mathcal{E}))$ and moreover, $\varrho$ is locally bounded. Thus, by [@Ebe96 Theorem 1.1, Corollary 1.3] we can conclude the following:
The symmetric bilinear form $(\mathcal{E}^{\varrho},D(\mathcal{E}))$ is densely defined and closable on $L^2(F;\varrho \mu)$ and its closure $(\mathcal{E}^{\varrho}, D(\mathcal{E}^{\varrho}))$ is a strongly local Dirichlet form. Moreover, $(\mathcal{E}^{\varrho}, D(\mathcal{E}^{\varrho}))=(\mathcal{E}^{\varrho},\overline{\mathcal{D}}))$, i.e., $\mathcal{D}$ is a dense subset of $D(\mathcal{E}^{\varrho})$.
Due to [@FOT94 Theorem 5.5.1] it is possible to give a Fukushima decomposition of the process $\mathbb{M}$ of the form $$\begin{aligned}
\label{fukudec} \ln \phi (X_t) - \ln \phi(X_0)= M_t^{[\ln \phi]} + N_t^{[\ln \phi]} \ \ \mathbb{P}_x-\text{a.s. for each } x \in F, \end{aligned}$$ where $M_t^{[\ln \phi]}$ is a martingale additive functional and $N_t^{[\ln \phi]}$ is a continuous additive functional. The function $\ln \phi$ is possibly unbounded. In this case, the decomposition (\[fukudec\]) requires some localization argument (see e.g. [@FOT94 (6.3.19)]). Define the positive multiplicative functional $(Z_t)_{t \geq 0}$ by $$\begin{aligned}
\label{MF} Z_t=\exp( M_t^{[\ln \phi]} - \frac{1}{2} \langle M^{[\ln \phi]} \rangle_t). \end{aligned}$$ Furthermore, let $(\tilde{p}_t)_{t > 0}$ be defined by $$\tilde{p}_t f(x):= \mathbb{E}_x(Z_t f(X_t))$$ for $f \in \mathcal{B}_b(F)$.\
By [@FOT94 Section 6.3] $(\tilde{p}_t)_{t > 0}$ is a transition function and there exists a corresponding $\varrho \mu$-symmetric right process $\mathbb{M}^{\varrho}=(\Omega,(X_t)_{t \geq 0},(\mathbb{P}^{\varrho}_x)_{x \in F})$. Moreover, the Dirichlet form of $\mathbb{M}^{\varrho}$ is given by $(\mathcal{E}^{\varrho},D(\mathcal{E}^{\varrho}))$. We say that the process $\mathbb{M}^{\varrho}$ and the transition semigrpup $(\tilde{p}_t)_{t >0}$ are the Girsanov transformation of $\mathbb{M}$ and $(\tilde{p}_t)_{t >0}$ respectively by the multiplicative functional $(Z_t)_{t \geq 0}$.
Construction of the strong Feller transition semigroup {#Feller}
======================================================
In [@Chu85] criteria are given under which the doubly Feller property is preserved under the transformation by a multiplicative functional $(Z_t)_{t \geq 0}$. This concept is extended in [@CK08]. It is shown that the conditions on $(Z_t)_{t \geq 0}$ can be weakened. Moreover, the setting is applied to Feynman-Kac and Girsanov transformations. In particular, precise conditions on the Revuz measure of the underlying additive functionals are given. We quote a result of [@CK08] concerning the preservation of the doubly feller property under Girsanov transformations. Since we deal with strong Markov processes with [*continuous sample paths*]{}, we restrict the results to this setting instead of stating them in full generality.\
Let $\mathbb{M}=(\Omega,\mathcal{F},(\mathcal{F}_t)_{t \geq 0}, (X_t)_{t \geq 0}, (\mathbb{P}_x)_{x \in F})$ be again a $\mu$-symmetric strong Markov process with state space $F \subset \mathbb{R}^n$, $n \in \mathbb{N}$, continuous sample paths and infinite lifetime, where $\mu$ is a positive Radon measure on $(F,\mathcal{B}(F))$ with full support. As before, denote by $(p_t)_{ t>0}$ the transition semigroup of $\mathbb{M}$. Assume that $(p_t)_{ t>0}$ possesses the doubly Feller property.\
Let $r_{\lambda}(x,y)$, $\lambda >0$, $x,y \in F$, be the resolvent kernel of $\mathbb{M}$, i.e., the resolvent $(r_{\lambda})_{\lambda>0}$ of $\mathbb{M}$ is given by $$r_{\lambda}f(x)=\int_F f(y) r_{\lambda}(x,y) d\mu(y)$$ for $f \in \mathcal{B}_b(F)$, $\lambda >0$ and $x \in F$. For a Borel measure $\nu$ on $\mathcal{B}(F)$ we define the $\lambda$-*potential of* $\nu$ by $R_{\lambda}\nu(x):= \int_F r_{\lambda}(x,y) d\nu(y)$, $\lambda >0$.\
Let $B$ be a non-empty open subset of $F$ and denote by $B_{\Delta_B}:=B \cup \{\Delta_B\}$ the one-point compactification of $B$. Define $(X_t^B)_{ t \geq 0}$ by $$\begin{aligned}
X_t^B:=
\left\{
\begin{array}{l}
X_t \ \ \text{ if } t < \tau_B \\
\Delta_B \ \text{ if } t \geq \tau_B
\end{array}
\right.\end{aligned}$$ where $\tau_B :=\inf \{ t>0 |~ X_t \notin B \}$. The transition semigroup of $(X_t^B)_{ t \geq 0}$ is given by $$p_t^B(x,A)=\mathbb{P}_x(X_t \in A,~ t < \tau_B)$$ and $$p_t^B(x,\{\Delta_B\}):=1-p_t^B(x,B), \ \ p_t^B(\Delta_B, \{ \Delta_B \}):=1,$$ for $x \in B$, $A \in \mathcal{B}(B)$. A function $f \in \mathcal{B}_b(F)$ is extended to $\Delta_B$ by setting $f(\Delta_B)=0$. For functions of this form, the transition semigroup of $(X_t^B)_{ t \geq 0}$ reads $$p_t^B f(x)=\mathbb{E}_x (f(X_t) \mathbbm{1}_{\{t < \tau_B\}}).$$ The set $B$ is called *regular* if for each $x \in F \backslash B$, we have $\mathbb{P}_x(\tau_B=0)=1$.\
Let $(M_t)_{t \geq 0}$ be a continuous locally square integrable martingale additive functional and denote by $\mu_{\langle M \rangle}$ the Revuz measure of $(\langle M \rangle_t)_{t \geq 0}$. Furthermore, the transition semigroup $(\tilde{p}_t^B)_{t \geq 0}$ is given by $$\tilde{p}_t^Bf(x):=\mathbb{E}_x(Z_t f(X_t) \mathbbm{1}_{\{t < \tau_B\}}),$$ where $Z_t:=\exp (M_t-\frac{1}{2} \langle M \rangle_t )$, $t \geq 0$ and corresponds to the process obtained from $\mathbb{M}^{\varrho}$ (see Section \[secgirsanov\]) killed when leaving $B$. In the special case $B=F$ this definition reduces to the transition semigroup of $\mathbb{M}^{\varrho}$.
A Borel measure $\nu$ on $\mathcal{B}(F)$ is said to be of
1. [*Kato class*]{} if $\lim_{\lambda \rightarrow \infty} \sup_{x \in F} R_{\lambda} \nu(x)=0$,
2. [*extended Kato class*]{} if $\lim_{\lambda \rightarrow \infty} \sup_{x \in F} R_{\lambda} \nu(x) < 1$,
3. [*local Kato class*]{} if $\mathbbm{1}_K \nu$ is of Kato class for every compact set $K \subset F$.
\[thmGT\] Assume that $\frac{1}{2} \mu_{\langle M \rangle}$ is a positive Radon measure of local and extended Kato class and let $B$ be a regular open subset of $F$. Then $(\tilde{p}_t^B)_{t \geq 0}$ has the doubly Feller property. Moreover, $(Z_t)_{t \geq 0}$ is a martingale and $$\begin{aligned}
&\lim_{t \rightarrow 0} \sup_{x \in D} ~\mathbb{E}_x( |Z_t-1| \mathbbm{1}_{\{t < \tau_D\}}) =0 \text{ for any relatively compact open set } D \subset B, \\
& \sup_{0 \leq s \leq t}~ \sup_{x \in B} ~\mathbb{E}_x ( Z_s^p \mathbbm{1}_{\{s < \tau_B \}} ) < \infty \text{ for some } p>1 \text{ and each } t>0.\end{aligned}$$
See [@CK08 Theorem 3.3].
Consider again the $n$ independent sticky Brownian motions on $[0,\infty)$ discussed in Section \[secsticky\] with transition function $(p_t^{\beta,n})_{t >0}$ and Dirichlet form $(\mathcal{E}^n,D(\mathcal{E}^n))$ on $L^2(E;\mu_n)$. In the following, we introduce a density function $\varrho=\phi^2$. Under suitable conditions on $\phi$ it is possible to perform a Girsanov transformation such that the transition semigroup of the transformed process $\mathbb{M}^{\varrho}$ still possesses the strong Feller property (or even the doubly Feller property). By the preceding section the transformed Dirichlet form is of the form considered in [@FGV13]. In this way, we are able to strengthen the results in [@FGV13].
For functions $\phi$ such that the conditions of Theorem \[thmGT\] are fulfilled for $(Z_t)_{t \geq 0}$ as in (\[MF\]) and $B=E$, we immediately get that the transition function has the doubly Feller property and the process $\mathbb{M}^{\varrho}$ solves (\[sde!\]) for every starting point in $E$. Unfortunately, we are also interested in densities $\varrho$ such that the corresponding Revuz measure is not of extended Kato class. Such potentials are of particular interest for the application to the so-called wetting model in the theory of stochastic interface models. For this reason, we construct a strong Feller transition semigroup for a larger class of densities using Theorem \[thmGT\] and an approximation argument. A direct application fails, since the Kato condition on $\mu_{\langle M \rangle}$ ensures that the drift caused by the Girsanov transformation does not “explode”. However, this criterion does only take into account the variation of the drift, but not its direction, which is of particular importance in our setting.
\[example\] Let $n=1$ and $\phi(x):=\exp(-\frac{1}{2} x^2)$. In this case, $(\ln \phi)^{\prime}(x)=-x$. Hence, we expect that the process $\mathbb{M}^{\phi}$ has the representation $$dX_t= \sqrt{2}~ \mathbbm{1}_{(0,\infty)}(X_t) dB_t - 2 X_t~ \mathbbm{1}_{(0,\infty)}(X_t) dt + \frac{1}{\beta} \mathbbm{1}_{\{0\}}(X_t) dt.$$ Note that the additional drift term is always non-positive, since $X_t \in [0,\infty)$ for all $t >0$ and thus, it attracts the process to $0$. However, the logarithmic derivative of $\phi$ is unbounded and the energy measure is even not of extended Kato class. Indeed, $$\begin{aligned}
R_{\lambda} \mu_{\langle \ln \phi \rangle}(x) &= \int_{[0,\infty)} r^{\beta}_{\lambda}(x,y) d\mu_{\langle \ln \phi \rangle}(y) \\
&=2 \int_{[0,\infty)} \big( \frac{1}{\sqrt{2\lambda}} (e^{-\sqrt{2\lambda}|x-y|}-e^{-\sqrt{2\lambda}(x+y)}) + \frac{1}{\sqrt{2 \lambda}+\beta \lambda} 2 e^{-\sqrt{2 \lambda}(x+y)} \big)~ y^2 dy\end{aligned}$$ is unbounded in $x$ for each fixed $\lambda >0$, since $$\int_{[0,\infty)} \frac{1}{\sqrt{2\lambda}} e^{-\sqrt{2\lambda}|x-y|}~ y^2 dy = \frac{1}{\lambda} x^2 - \frac{1}{2 \lambda^2} e^{-\sqrt{2 \lambda} x} + \frac{1}{\lambda^2} \rightarrow \infty \ \text{ as } x \rightarrow \infty,$$ whereas the remaining terms converge to $0$ as $x \rightarrow \infty$. Thus, it is not possible to apply Theorem \[thmGT\] to this specific choice of $\phi$.
Assume that $\phi$ is given such that Condition \[conditions\] is fulfilled. Then $\phi \in D(\mathcal{E}^n)_{\text{loc}}$ and the energy measure $\mu_{\langle \ln \phi \rangle}=\mu_{\langle M^{[\ln \phi]} \rangle}$ is given by $$\begin{aligned}
d\mu_{\langle \ln \phi \rangle}(x)= 2 \sum_{i=1}^n (\partial_i \ln \phi(x))^2~ dx_i \prod_{j \neq i} (dx_j + \beta \delta_0^j) =2 \sum_{i=1}^n \mathbbm{1}_{(0,\infty)}(x_i)~(\partial_i \ln \phi(x))^2~ d\mu_n(x)\end{aligned}$$ and thus, by Revuz correspondence we see that $$\begin{aligned}
\label{variationAF}
\langle M^{[\ln \phi]} \rangle_t = 2 \sum_{i=1}^n \int_0^t (\partial_i \ln \phi(X_s))^2~ \mathbbm{1}_{(0,\infty)}(X_s^i) ds.\end{aligned}$$ By this we can deduce that $(M^{[\ln \phi]}_t)_{t \geq 0}$ has the representation $$\begin{aligned}
\label{AF}
M^{[\ln \phi]}_t= \sqrt{2}~ \sum_{i=1}^n \int_0^t \partial_i \ln \phi(X_s)~ \mathbbm{1}_{(0,\infty)}(X_s^i) dB_s^i.\end{aligned}$$
\[exbounded\] Let ${\nabla}\ln \phi$ additionally be essentially bounded w.r.t. $\mu_n$. Then $\frac{1}{2} \mu_{\langle \ln \phi \rangle}=\frac{1}{2} \mu_{\langle M^{[\ln \phi]} \rangle}$ is of local and extended Kato class.
Let $k \in \mathbb{N}$ and $K:=[0,k)^n$ as well as $\tau_k:=\inf \{ t >0 |~ X_t \notin K \}$. Let $\phi_k$ be given such that $\phi_k=\phi$ on $K$, Condition \[conditions\] is fulfilled for $\phi_k$ and $\nabla \ln \phi_k \in L^{\infty}(E;\mu_n)$. We define the exponential functional $(Z^k_t)_{t \geq 0}$ by $$Z^k_t := \exp(M^{[\ln \phi_k]}_t - \frac{1}{2} \langle M^{[\ln \phi_k]} \rangle_t ).$$ Note that we are in fact only interested in the restriction of $\phi$ to the set $K$, since the function is used to define a Girsanov transformation of $(p_t^{\beta,n})_{t >0}$ which is killed when leaving $K$. Nevertheless, in order to give meaning to $Z_t^k$ for $t \geq \tau_k$, we extend $\phi_k$ to $E$.
\[strongfeller\] Let $\varrho=\phi^2$ be given as in Condition \[conditions\] and $Z_t=\exp(M_t^{[\ln \phi]}- \frac{1}{2} \langle M^{[\ln \phi]} \rangle_t)$, $t \geq 0$. Then the transition function $(p_t)_{t \geq 0}$ defined by $p_tf(x)=\mathbb{E}_x(Z_tf(X_t))$ for $f \in \mathcal{B}_b(E)$ and $x \in E$ which corresponds to the strong Markov process $\mathbb{M}^{\varrho}$ has the strong Feller property.
Let $k>0$ and $K=[0,k)^n$. $K$ is regular, i.e., $\mathbb{P}_x(\tau_K =0)=1$ for each $x \in E \backslash K$. We define the transition function $(p^k_t)_{t \geq 0}$ similar as $(p_t)_{t \geq 0}$ by $p_t^k f(x):=\mathbb{E}_x(Z_t^k f(X_t) \mathbbm{1}_{\{t <\tau_K\}})$. By the assumptions on $\phi_k$, Example \[exbounded\] and Theorem \[thmGT\], $(p_t^k)_{t \geq 0}$ has the doubly Feller property for each $k>0$. Let $f \in \mathcal{B}_b(E)$ and choose a constant $C(f) < \infty$ such that $|f(x)| \leq C(f)$ for all $x \in E$. Clearly, $p_tf \in \mathcal{B}_b(E)$. Hence, it suffices to show that $p_tf$ is continuous. We have for $x \in D:=[0,d]^n$, $d>0$, $$\begin{aligned}
| p_tf(x) - p_t^kf(x)| &=| \mathbb{E}_x(Z_t f(X_t)) -\mathbb{E}_x(Z_t^k f(X_t) \mathbbm{1}_{\{t < \tau_K\}}) | \\
&=|\mathbb{E}_x(Z_t f(X_t) \mathbbm{1}_{\{ t \geq \tau_K\}})| \\
&\leq C(f)~ |\mathbb{E}_x(Z_t \mathbbm{1}_{\{ t \geq \tau_K\}})| \\
&\leq C(f)~ \sup_{x \in D} |\mathbb{E}_x(Z_t \mathbbm{1}_{\{ t \geq \tau_K\}})|
\rightarrow 0 \quad \text{as } k \rightarrow \infty.\end{aligned}$$ uniformly on $D$ by (\[condZ\]). Hence, $p_t f$ is continuous on $D$ for each $d >0$ and so $p_t f \in C_b(E)$.
\[remZ\] Let $D \subset E$ be compact. Then $\lim_{k \rightarrow \infty} \sup_{x \in D} ~\mathbb{E}_x(\mathbbm{1}_{\{ \tau_k \leq t \}}~Z_t) =0$ holds for example if there exists some $p>1$ such that $\sup_{x \in D} \mathbb{E}_x(Z_t^p) < \infty$. Indeed, let $1<q<\infty$ such that $\frac{1}{p}+\frac{1}{q}=1$. Then $$\sup_{x \in D} \mathbb{E}_x(\mathbbm{1}_{\{\tau_k \leq t\}} Z_t) \leq \sup_{x \in D} \mathbb{E}_x(Z_t^p)^{\frac{1}{p}}~ \sup_{x \in D} \big(\mathbb{P}_x(\tau_k \leq t)\big)^{\frac{1}{q}}.$$ Define $C_t:= \max_{i=1,\dots,n} \max_{0 \leq s \leq t} X_s^i$ for $t \geq 0$. Then for $x \in D$ and $k >d$ $$\mathbb{P}_x( \tau_k \leq t) \leq \mathbb{P}_0( C_t \geq k-d ) \leq n~ \sqrt{\frac{t}{2 \pi}} \frac{4}{k-d} \exp(-\frac{(k-d)^2}{2t}) =: C(k) \rightarrow 0 \ \text{ as } k \rightarrow \infty$$ due to [@KS98 p.96,(8.3)’], since $C_t \leq \max_{i=1,\dots,n} \max_{0 \leq s \leq t} |B_s^i|$ almost surely with respect to $\mathbb{P}_0$.
By Section \[secgirsanov\] there exists a strong Markov process $\mathbb{M}^{\varrho}$ with transition semigroup $(p_t)_{t \geq 0}$ and the Dirichlet form associated to $\mathbb{M}^{\varrho}$ is given by the closure of $(\mathcal{E}^{\varrho},\mathcal{D})$ on $L^2(E;\varrho \mu_n)$. Note that in this case $\mathcal{D} \cap D(L) \supset C_c^2(E)$ and $C_c^2(E)$ is also dense in $D(\mathcal{E}^n)$. Indeed, Lemma \[lemdense\] is based on the fact that $C_c^1([0,\infty))$ is dense for the one dimensional form which also holds for $C_c^2([0,\infty))$ (and even $C_c^{\infty}([0,\infty))$) by [@ChFu11 Theorem 5.2.8(i)]. The strong Feller property is shown in Theorem \[strongfeller\] and the last statement holds by [@FOT94 Exercise 4.2.1].
The statement follows by the results proven in [@FGV13 Corollary 4.18, Theorem 5.6] considering that the absolute continuity condition [@FOT94 (4.2.9)] is fulfilled.
Application to the dynamical wetting model {#appl}
==========================================
Densities corresponding to potential energies
---------------------------------------------
In the following, let $\phi \in C^2(E)$ be strictly positive such that $\phi \in L^2(E;\mu_n)$. Set $H:=- \ln \phi$ (thus, $\phi=\exp(-H)$) and assume additionally that there exist real constants $K_1 \geq 0,~K_2$ and $K_3$ such that
1. $H(x) \geq -K_1$ for all $x \in E$,
2. $\partial_i H(x) \leq K_2$ for all $x \in \{ x_i=0\}:=\{ x \in E|~x_i=0\}$, $i=1,\dots,n$,
3. $\partial_i^2 H(x) \leq K_3$ for all $x \in E$, $i=1,\dots,n$.
If we can verify (\[condZ\]), Condition \[conditions\] is fulfilled and thus, the results of Theorem \[thmmain1\] and Theorem \[thmmain2\] hold accordingly.\
Using (\[AF\]), (\[variationAF\]) and Ito’s formula we see that $$\begin{aligned}
M_t^{[\ln \phi]} - \frac{1}{2} \langle M^{[\ln \phi]} \rangle_t &= \sqrt{2} \sum_{i=1}^n \int_0^t \partial_i \ln \phi (X_s) \mathbbm{1}_{(0,\infty)} (X_s^i) dB_s^i -\sum_{i=1}^n \int_0^t \big(\partial_i \ln \phi(X_s) \big)^2 \mathbbm{1}_{(0,\infty)}(X_s^i) ds \notag \\
&=H(X_0)-H(X_t)+ \frac{1}{\beta} \sum_{i=1}^n \int_0^t \partial_i H(X_s) \mathbbm{1}_{\{0\}}(X_s^i) ds \label{AFdecomp} \\
&\ \ + \sum_{i=1}^n \int_0^t \big( \partial_i^2 H(X_s) - \partial_i H(X_s)^2 \big) \mathbbm{1}_{(0,\infty)}(X_s^i) ds \notag \\
&\leq H(x) + K_1 + \frac{n}{\beta} K_2 t + n K_3 t \notag\end{aligned}$$ $\mathbb{P}_x$-a.s. for each $x \in E$.\
Let $p>1$ be arbitrary, $D \subset E$ compact. Then it holds $$\sup_{x \in D}~ \mathbb{E}_x(Z_t^p) \leq \exp\big(p~( \sup_{x \in D} H(x) + K_1 + \frac{n}{\beta} K_2 t + n K_3 t)\big) < \infty \quad \text{for every } t>0.$$ Thus, in view of Remark \[remZ\], (\[condZ\]) holds true.
Densities corresponding to potential energies given by pair potentials
----------------------------------------------------------------------
Assume that $H$ is given by a potential with nearest neighbor pair interaction, i.e., $H$ is defined as in (\[hamilt\]). In particular, $\kappa:=\int_{\mathbb{R}} \exp(-V(r)) dr < \infty$, $V$ is convex, $V^{\prime}(0)=0$ and $V^{\prime}$ is non-decreasing. Then, we have $H(x) \geq -\frac{n}{2} b$, $$\partial_i H (x)= \frac{1}{2} \sum_{\stackunder{|i-j|=1}{j \in \{0,\dots,n+1\}}} V^{\prime}(x_i-x_j) \ \ \big(= \frac{1}{2} V^{\prime}(x_i-x_{i-1}) + \frac{1}{2} V^{\prime}(x_i-x_{i+1}) \big)$$ for $i=1,\dots,n$ and moreover, $$\partial^2_i H (x)=\frac{1}{2} \sum_{\stackunder{|i-j|=1}{j \in \{0,\dots,n+1\}}} V^{\prime \prime}(x_i-x_j) \ \ \big(= \frac{1}{2} V^{\prime \prime}(x_i-x_{i-1}) + \frac{1}{2} V^{\prime \prime}(x_i-x_{i+1}) \big).$$ Since $\partial_i H (x)=\frac{1}{2} (V^{\prime}(-x_{i-1}) + V^{\prime}(-x_{i+1})) \leq 0$ if $x_i=0$, we get $\partial_i H(x) \leq 0$ for all $x \in \{ x_i=0\}$ and furthermore, $\partial_i^2 H(x) \leq c_+$ , $i=1,\dots,n$. Thus, (i)-(iii) above are fulfilled with $K_1:=\frac{n}{2}b$, $K_2:=0$ and $K_3:=c_+$. Note that $\phi \in L^2(E;\mu_n)$.
Uniqueness of weak solutions {#uniqueness}
============================
Let $\varrho=\phi^2$ be given as in Condition \[conditions\]. Then the solution to (\[main\]) is unique in law.
By [@GS72 §24, Theorem 1, Corollary 1] the one dimensional sticky Brownian motion on $[0,\infty)$ is unique in law. Thus, the same holds true for $n$ independent sticky Brownian motions for each $n \in \mathbb{N}$. Finally, we can conclude that the solution to (\[main\]) is unique in law due to [@WaIk89 Chapter IV, Theorem 4.2], since its law is constructed by a Girsanov transformation.
Acknowledgment {#acknowledgment .unnumbered}
--------------
We thank Torben Fattler for helpful comments and discussions. R. Vo[ß]{}hall gratefully acknowledges financial support in the form of a fellowship of the German state Rhineland-Palatine.
[BDG01]{}
S. Albeverio, Y. Kondratiev and M. R[ö]{}ckner. Strong Feller properties for distorted Brownian motion and applications to finite particle systems with singular interactions. In [*Finite and Infinite Dimensional Analysis in Honor of Lenorad Gross*]{}, volume 317 of [*Contemp. Math.*]{}. Amer. Math. Soc., Providence, RI, 2003.
S. Albeverio and M. R[ö]{}ckner. Dirichlet form methods for uniqueness of martingale problems and applications. In [*Stochastic analysis ([I]{}thaca, [NY]{}, 1993)*]{}, volume 57 of [*Proc. Sympos. Pure Math.*]{}, pages 513–528. Amer. Math. Soc., Providence, RI, 1995.
M. Abramowitz and I.A. Stegun. . , Dover Publications, Inc. , New York, ninth Dover printing, tenth GPO printing, 1964.
B. Baur and M. Grothaus. Construction and strong Feller property of distorted elliptic diffusion with reflecting boundary. , 40(4):391–425, 2014.
B. Baur, M. Grothaus and P. Stilgenbauer. Construction of $\mathcal{L}^p$-strong Feller Processes via Dirichlet Forms and Applications to Elliptic Diffusions. , 38(4):1233–1258, 2013.
N. Bouleau and F. Hirsch. . Walter de Gruyter & Co., Berlin, 1991.
Z. Chen and M. Fukushima. , volume 35 of [*London Mathematical Society Monographs*]{}. Princeton University Press, Princeton and Oxford, 2012.
Z. Chen and K. Kuwae. On doubly Feller property. , 46:909-930, 2009.
K.L. Chung. Doubly-Feller Process with Multiplicative Functional. In [*Seminar on Stochastic Processes, 1985*]{}, volume 12 of [*Progress in Probability and Statistics*]{}, pages 63–78. Birkh[ä]{}user, Boston, 1986.
A. Eberle. Girsanov-type transformations of local Dirichlet forms: an analytic approach. , 33:497–531, 1996.
H. -J. Engelbert and G. Peskir. Stochastic differential equations for sticky Brownian motion. , 86(6):993–1021.
T. Fattler and M. Grothaus. Strong Feller properties for distorted Brownian motion with reflecting boundary condition and an application to continuous $N$-particle systems with singular interactions. , 246:217–241, 2007.
T. Fattler, M. Grothaus, and R. Voßhall. Construction and analysis of a sticky reflected distorted Brownian motion. To appear in [*Annales de l’Institut Henri Poincar[é]{}*]{}, 2014.
M. Fukushima, Y. Oshima and M. Takeda. , volume 19 of [*de Gruyter Studies in Mathematics*]{}. Walter de Gruyter & Co., Berlin, 2011.
T. Funaki. Stochastic interface models. In [*Lectures on probability theory and statistics*]{}, volume 1869 of [*Lecture Notes in Math.*]{}, pages 103–274. Springer, Berlin, 2005.
M. Fukushima and M. Tomisaki. Construction and decomposition of reflecting diffusions on Lipschitz domains with Hölder cusps. , 106:521–557, 1996.
G. Giacomin. Limit theorems for random interface models of [G]{}inzburg-[L]{}andau [$\nabla\phi$]{} type. In [*Stochastic partial differential equations and applications ([T]{}rento, 2002)*]{}, volume 227 of [*Lecture Notes in Pure and Appl. Math.*]{}, pages 235–253. Dekker, New York, 2002.
C. Graham. The martingale problem with sticky reflection conditions, and a system of particles interacting at the boundary. , 24(1):45–72, 1988.
I.I. Gikhman and A.V. Skorochod. , Band 72 of [*Ergebnisse der Mathematik und ihrer Grenzgebiete*]{}. Springer-Verlag, Berlin, 1972.
N. Ikeda and Sh. Watanabe. , volume 24 of [*North-Holland Mathematical Library*]{}. North-Holland Publishing Co., Amsterdam, second edition, 1989.
O. Kallenberg. . Springer-Verlag, New York, 2001.
F.B. Knight. , volume 18 of [*Mathematical surveys*]{}. American Mathematical Society, Providence, Rhode Island, 1981.
V. Kostrykin, J. Potthoff and R. Schrader. Brownian motion on metric graphs: feller Brownian motions on intervals revisited. , 2010.
I. Karatzas and S.E. Shreve. , volume 113 of [*Graduate Texts in Mathematics*]{}. Springer-Verlag, New York, 1998.
Z.M. Ma and M. R[ö]{}ckner. . Universitext. Springer-Verlag, Berlin, 1992.
[^1]: **Mathematics Subject Classification 2010**. *60K35, 60J50, 60J55, 60J35 , 82C41.*
[^2]: **Keywords**: *Sticky reflected distorted Brownian motion, strong Feller properties, Skorokhod decomposition, Wentzell boundary condition, interface models.*
[^3]: University of Kaiserslautern, P.O.Box 3049, 67653 Kaiserslautern, Germany.
[^4]:
[^5]:
[^6]:
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Magnetorotational instability (MRI) is one of the fundamental processes in astrophysics, driving angular momentum transport and mass accretion in a wide variety of cosmic objects. Despite much theoretical/numerical and experimental efforts over the last decades, its saturation mechanism and amplitude, which sets the angular momentum transport rate, remains not well understood, especially in the limit of high resistivity, or small magnetic Prandtl numbers typical to interiors (dead zones) of protoplanetary disks, liquid cores of planets and liquid metals in laboratory. Using direct numerical simulations, in this paper we investigate the nonlinear development and saturation properties of the helical magnetorotational instability (HMRI) – a relative of the standard MRI – in a magnetized Taylor-Couette flow at very low magnetic Prandtl number (correspondingly at low magnetic Reynolds number) relevant to liquid metals. For simplicity, the ratio of azimuthal field to axial field is kept fixed. From the linear theory of HMRI, it is known that the Elsasser number, or interaction parameter determines its growth rate and plays a special role in the dynamics. We show that this parameter is also important in the nonlinear problem. By increasing its value, a sudden transition from weakly nonlinear, where the system is slightly above the linear stability threshold, to strongly nonlinear, or turbulent regime occurs. We calculate the azimuthal and axial energy spectra corresponding to these two regimes and show that they differ qualitatively. Remarkably, the nonlinear state remains in all cases nearly axisymmetric suggesting that HMRI turbulence is quasi two-dimensional in nature. Although the contribution of non-axisymmetric modes increases moderately with the Elsasser number, their total energy remains much smaller than that of the axisymmetric ones.'
address: |
$^1$Helmholtz-Zentrum Dresden-Rossendorf, Dresden 01328, Germany\
$^2$Department of Physics, Faculty of Exact and Natural Sciences, Tbilisi State University, Tbilisi 0179, Georgia\
$^3$Abastumani Astrophysical Observatory, Ilia State University, Tbilisi 0162, Georgia\
$^4$University of Bremen, Center of Applied Space Technology and Microgravity (ZARM), Bremen 28359, Germany
author:
- 'G. Mamatsashvili$^{1,2,3}$, F. Stefani$^{1}$, A. Guseva$^{4}$ and M. Avila$^{4}$'
bibliography:
- 'biblio.bib'
title: 'Quasi-two-dimensional nonlinear evolution of helical magnetorotational instability in a magnetized Taylor-Couette flow'
---
Introduction
============
The magnetorotational instability (MRI, [@Velikhov59; @Balbus_Hawley98]) is one of the most important processes in magnetized differentially rotating conducting fluids, which largely determines their dynamics and evolution. It is a powerful linear instability arising as a result of the combined effect of weak magnetic field and radially decreasing angular velocity. The MRI is believed to operate in a vast variety of cosmic objects, ranging from astrophysical disks and stars to liquid-metal cores of planets. While discovered as early as 1959 [@Velikhov59], the astrophysical significance of MRI was first recognized only three decades later by Balbus & Hawley [@Balbus_Hawley91], who demonstrated that weak magnetic fields can destabilize Keplerian accretion disks around such diverse objects as supermassive black holes, black holes in X-ray binaries and young stellar objects, which would otherwise be linearly stable according to Rayleigh’s criterion. The linear growth of MRI, which taps into the free energy of differential rotation, eventually breaks down into magnetohydrodynamic (MHD) turbulence [@Hawley_etal95; @Fromang13]. This turbulence transports angular momentum outward and, as a consequence, matter inward in the disk, yielding mass accretion rates onto the central object close to observationally inferred values. More recently, MRI has also been discussed for explaining angular momentum transport in the Sun, massive stars and neutron stars [@Kagan_Wheeler14; @Wheeler_etal15; @Masada_etal15; @Spada_etal16] and also in the context of the geodynamo [@Petitdemange_etal08; @Petitdemange_etal13].
The MRI was originally discovered theoretically in a classical Taylor-Couette (TC) flow of a conducting fluid between two concentric rotating cylinders threaded by an external vertical (along the common axis of cylinders) magnetic field [@Velikhov59; @Chandrasekhar60] – a setup which is also best suited and most frequently used nowadays for experimental investigations. The conducting medium filling the gap between cylinders is usually a liquid metal (sodium, gallium) characterized by an extremely small ratio of viscosity to Ohmic resistivity, or magnetic Prandtl number, $Pm=\nu/\eta \sim 10^{-6}-10^{-5}$. By suitably adjusting the rotation rates of outer and inner cylinders, the TC flow profile can be made very close to the Keplerian one [@Ji_etal06; @Schartman_etal12; @Edlund_Ji15; @Lopez_Avila17], offering a unique possibility to study the disk MRI problem in laboratory as well, which has been up to now mostly carried out both via analytical means and numerical simulations.
First experimental efforts to study MRI in the laboratory were made at the University of Maryland [@Sisan_etal04], at Princeton University [@Nornberg_etal10; @Roach_etal12], and at Helmholtz-Zentrum Dresden-Rossendorf (HZDR) [@Stefani_etal06; @Stefani_etal09; @Seilmayer_etal14]. The liquid sodium spherical Couette experiment in Maryland had produced coherent velocity/magnetic field fluctuations showing up in a parameter region reminiscent of MRI [@Sisan_etal04]. The liquid GaInSn TC experiments in Princeton was designed to investigate the standard version of MRI (SMRI) with only an axial magnetic field being imposed. In this case, the azimuthal magnetic field (which is an essential participant in the MRI process) must be produced from the axial field by induction effects, which are proportional to the magnetic Reynolds number $Rm$ of the flow. $Rm$, in turn, is proportional to the hydrodynamic (HD) Reynolds number $Re$ according to $Rm = Pm\cdot Re$. Therefore, in order to achieve $Rm \gtrsim 1$ necessary for SMRI to operate (see e.g., [@Sano_Miyama99; @Ji_etal01]), taking into account the very small values of $Pm$, $Re \sim 10^5-10^6$ is needed. This makes SMRI experiments with TC flows extremely challenging, although evidence for slow magneto-Coriolis waves [@Nornberg_etal10] and for a free-Shercliff layer instability [@Roach_etal12] have already been obtained. Within the DRESDYN project at HZDR [@Stefani_etal12], it is planned to set-up a large liquid sodium TC experiment, which will allow to broaden the parameter range in such a way as to make SMRI reachable.
On the theoretical side, SMRI in TC flow has been extensively studied both in the linear and nonlinear regimes, with more focus on the low-$Pm$ regime as it is typical in experiments. Still, the typical Reynolds numbers $Re\sim 10^3$ used in these analyses are orders of magnitude smaller than experimental values $Re\gtrsim
10^6$, which are extremely demanding from the computational point of view. The linear analysis identified critical magnetic Reynolds numbers for the onset of the instability and discussed possibilities for its experimental detection (e.g., [@Ji_etal01; @Ruediger_Zhang01; @Goodman_Ji02; @Kirillov_Stefani10]). In particular, it was shown that the linear growth rate is determined by magnetic Reynolds number and Lundquist number (i.e., ratio of the magnetic diffusion time to the Alfvén crossing time), which should be both larger than unity for SMRI to operate. The saturation properties of SMRI following its exponential growth was investigated both in the weakly nonlinear regime, i.e., near the instability threshold, [@Umurhan_etal07a; @Umurhan_etal07b; @Clark_Oishi16a; @Clark_Oishi16b] and in the fully nonlinear regime [@Knobloch_Julien05; @Liu_etal06a; @Liu08; @Ebrahimi_etal09; @Gissinger_etal12; @Gellert_etal12; @Wei_etal16]. It was demonstrated that SMRI saturates on the resistive time scale by modifying (reducing) the background velocity shear responsible for it and strengthening the background field. The dependence of the saturation level on the imposed field as well as on the magnetic Reynolds number, Prandtl number and Lundquist number was explored in greater detail.
While SMRI has been explored quite extensively both in TC flows and astrophysical disk context since the 1990s, its relative – helical magnetorotational instability (HMRI) – has become a subject of active theoretical and experimental research only in the last decade. These studies were initiated by Hollerbach & R[ü]{}diger [@Hollerbach_Ruediger05], who realized that adding an azimuthal magnetic field to the axial field can destabilize highly resistive flows at much smaller field strengths and at several orders of magnitude smaller Reynolds numbers, reducing the critical value from $Re\sim 10^6$ needed for SMRI with purely axial field to $Re\sim
10^3$. Because this instability takes place in the presence of helical magnetic field, composed of azimuthal and axial components, it was termed HMRI. Like the SMRI, the HMRI also draws free energy from the background shear flow and transports angular momentum outward. Subsequently, HMRI has been widely studied theoretically by means of linear modal stability analysis [@Liu_etal06; @Priede_etal07; @Priede_Gerbeth09; @Kirillov_Stefani10; @Priede11; @Kirillov_Stefani13; @Kirillov_etal14; @Ruediger_etal06]. It was shown that HMRI represents a destabilized inertial wave [@Liu_etal06] and its growth rate is governed by the Reynolds number and the Hartmann number, $Ha$, characterizing the magnetic field strength. As a result, HMRI can persist even at extremely small magnetic Prandtl numbers typical to liquid metals, in contrast to SMRI, which is generally suppressed under this condition. Another difference with SMRI is that HMRI is restricted to rotational profiles with comparably steep negative shear or extremely steep positive shear, smaller than so-called the lower Liu limit or larger than the upper Liu limit, respectively [@Liu_etal06; @Priede11]. Although the transition between HMRI and SMRI is monotonic when decreasing the ratio of azimuthal to axial field [@Hollerbach_Ruediger05], the mathematics of this connection turned out to be quite subtle, including the formation of a spectral exceptional point where the original inertial mode coalesces with the slow magnetocoriolis mode [@Kirillov_Stefani10]. Moreover, the recent nonmodal analysis revealed a fundamental connection between the modal growth of HMRI and the nonmodal dynamics in the corresponding HD problem [@Mamatsashvili_Stefani16]. The relevance of HMRI for astrophysical Keplerian disks has also been considered: it seems to be a promising candidate to replace SMRI in weakly ionized parts of disks with small $Pm$, for example in the “dead zones” of protoplanetary disks or the outer parts of accretion disks around black holes. However, this issue is not yet fully resolved and under debate, because under standard conditions Keplerian shear alone might not be steep enough for HMRI to work [@Liu_etal06] unless other physical factors, such as specific electrical boundary conditions [@Ruediger_Hollerbach07; @Ruediger_Schultz08] or additional axial currents within the fluid are involved [@Kirillov_Stefani13].
The remarkable feature of HMRI to survive at very small magnetic Prandtl numbers and to set in at moderate Reynolds numbers makes it an ideal playground for experimental studies with liquid metals. Indeed, shortly after its theoretical discovery, a series of specially designed TC experiments [@Stefani_etal06; @Stefani_etal09] provided the first experimental evidence of HMRI at the liquid metal facility PROMISE and reproduced the main results of the linear theory, such as the stability threshold, the wavenumber and frequency of the HMRI-wave. In the experiments, the saturation amplitude as well as the propagation speed of the HMRI wave were measured in detail as a function of the system parameters ($Re$, $Ha$, ratio of rotation frequencies of outer and inner cylinders, etc.) and some differences with the numerical results were pointed out (see e.g., figure 10 in [@Stefani_etal09]).
This prompted further theoretical studies of HMRI in TC flows both with and without endcaps, focusing more on its nonlinear development and saturation. However, even today this is still a less explored area than the linear HMRI. First axisymmetric numerical simulations explored the dependence of the established HMRI-wave amplitude and propagation speed on the Reynolds number and the rotation ratio of the inner and outer cylinders as well as the role of endcaps [@Szklarski_Ruediger06; @Liu_etal07; @Szklarski07]. However, the parameter range adopted in these studies was much narrower than that in the above experiments, being near the linear instability threshold of HMRI, when its dynamics is weakly nonlinear (see also [@Clark_Oishi16b]). More recent axisymmetric simulations [@Child_Hollerbach16; @Hollerbach_etal17] looked into the intrinsic nonlinear dynamics of HMRI in an axially infinite TC domain, avoiding complications because of the endcaps [@Avila12]. Different types of nonlinear regimes were analyzed and the applicability of the generalized quasi-linear approximation was tested by tracing the dynamics of large-scale Fourier modes. Despite these efforts, which undoubtedly contributed to a better understanding of the nonlinear dynamics of HMRI, detailed physics of its saturation and sustenance is still missing, especially when comparison with experiments is concerned. Evidently, it is this nonlinear saturated state which is observed in experiments, while the initial transient phase is much harder to identify due to the limited sensitivity of the applied ultrasonic flow measurement techniques.
In some respect, more progress has been made recently on the nonlinear dynamics of a “sibling” of HMRI – the azimuthal magnetorotational instability (AMRI). This non-axisymmetric instability, which emerges in the presence of an imposed purely azimuthal fields in TC liquid metal flows, had been first identified after HMRI [@Seilmayer_etal14; @Hollerbach_etal10] (see also [@Kirillov_etal14] and references therein). Using numerical simulations, Guseva et al. [@Guseva_etal15; @Guseva_etal16; @Guseva_etal17] probed much broader ranges of Reynolds and Hartmann numbers than those done for HMRI so far and identified different regimes of nonlinear saturation, from supercritical Hopf bifurcation near the linear instability threshold up to a catastrophic transition to spatio-temporal defects, which are mediated by a subcritical subharmonic Hopf bifurcation, and ultimately to turbulence. The scaling of the angular momentum transport of AMRI in the nonlinear regime with respect to Reynolds, Hartmann and Prandtl numbers was also explored in these papers.
Motivated by the experimental results of [@Stefani_etal09] and by the recent progress on understanding the nonlinear dynamics of AMRI [@Guseva_etal15], in this paper we investigate the evolution of HMRI, from its linear exponential growth to nonlinear saturation in liquid metal TC flows at small magnetic Prandtl numbers using numerical simulations. We consider infinite cylinders with periodic boundary conditions in the axial direction in order to avoid complications because of the endcaps (Ekman pumping) and concentrate on the intrinsic dynamics of HMRI driven by the combination of an imposed helical magnetic field and differential rotation of the flow. Distinctly from the above-mentioned previous studies on nonlinear HMRI, we do not restrict ourselves to axisymmetric perturbations only and allow non-axisymmetric modes to naturally develop during dynamical evolution, so that we can quantify the degree of non-axisymmetry of the saturated state of HMRI depending on the system parameters. It is well known that HMRI is axisymmetric in the linear regime, i.e. the most unstable modes do not have any azimuthal variation [@Hollerbach_Ruediger05; @Priede_etal07; @Priede11; @Child_Hollerbach16], and it is therefore important to know if this feature is retained in the nonlinear regime, which in turn can shed light on the saturation mechanism. We show that different regimes of the saturation are realized in the flow and analyze the characteristics of these states, such as angular momentum transport and energy spectra. This study, aiming at understanding the basic nonlinear dynamics of HMRI – underlying mechanism and properties of its saturation – is intended to be guiding and preparatory for the upcoming liquid sodium TC experiments in the DRESDYN facility at HZDR, which will combine and enhance the previous experiments on HMRI [@Stefani_etal09], AMRI [@Seilmayer_etal14] and on the current-driven kink-type Tayler instability [@Seilmayer_etal12].
The paper is organized as follows. The basic equations are introduced in section 2. Direct numerical simulations of the nonlinear saturation and evolution of HMRI as well as the spectral characteristics of the nonlinear state are presented in section 3. Summary and conclusions are given in section 4.
Main equations
==============
The basic equations of non-ideal MHD governing the motion of an incompressible conductive fluid with constant kinematic viscosity $\nu$ and Ohmic resistivity $\eta$ are $$\frac{\partial {\bf u}}{\partial t}+({\bf u}\cdot \nabla) {\bf
u}=-\frac{1}{\rho}\nabla p+\frac{1}{\mu_0\rho}(\nabla \times {\bf
B})\times {\bf B} + \nu\nabla^2 {\bf u},$$ $$\frac{\partial {\bf B}}{\partial t}=\nabla\times \left( {\bf
u}\times {\bf B}\right)+\eta\nabla^2{\bf B},$$ $$\nabla\cdot {\bf u}=0,~~~\nabla\cdot {\bf B}=0,$$ where $\rho$ is the constant density, $p$ is the thermal pressure, ${\bf u}$ is the velocity, ${\bf B}$ is the magnetic field and $\mu_0$ is the magnetic permeability of vacuum.
The equilibrium state is an axisymmetric cylindrical magnetized TC flow between two coaxial cylinders with the inner radius $R_i$ and the outer radius $R_o$, rotating with the angular velocities $\Omega_i$ and $\Omega_o$, respectively. The flow is threaded by an externally imposed helical magnetic field ${\bf
B}_0=(0,B_{0\phi}(r),B_{0z})$ consisting of constant axial, $B_{0z}$, and radially varying current-free azimuthal, $B_{0\phi}(r)=\beta B_{0z}R_i/r$, components in cylindrical coordinates $(r,\phi,z)$, where $\beta$ is the dimensionless parameter characterizing the helicity of the field. We ignore the effects of endcaps in order to gain insight into the basic/intrinsic nonlinear evolution of HMRI. The ratio of the radii $a\equiv
R_i/R_o=0.5$ and the height of the cylinders, $L_z=10d$, where $d=R_o-R_i=R_i$ is the gap width, are chosen as in PROMISE [@Stefani_etal09]. An unperturbed flow between the cylinders, as follows from Eqs. (1)-(3), is a standard vertically uniform and axisymmetric TC flow, ${\bf U}_0=(0,r\Omega(r),0)$, with the angular velocity given by $$\Omega(r)=\frac{\Omega_oR_o^2-\Omega_iR_i^2}{R_o^2-R_i^2}+\frac{\Omega_i-\Omega_o}{R_o^2-R_i^2}\frac{R_i^2R_o^2}{r^2},$$ pressure $p_0(r)$ and constant density $\rho_0$, satisfying the hydrostatic balance equation $\rho_0r\Omega^2=dp_0/dr$. The rotation ratio of the cylinders is fixed to $\mu=\Omega_o/\Omega_i=0.27$, slightly larger than the Rayleigh line $\mu_c=0.25$ at $R_i/R_o=0.5$, ensuring that purely HD instabilities are excluded, so that the flow can become unstable solely due to the presence of the magnetic field. In the following, we introduce the nondimensional variables by using the gap width $d$ as the unit of length, $\Omega_i^{-1}$ as the unit of time, $\Omega_id$ as the unit of velocity and $\rho_0\Omega_i^2d^2$ as the unit of pressure and energy density. (Since in the present case $d=R_i$, this velocity scale is in fact the rotational velocity of the inner cylinder, $\Omega_iR_i$.) The magnetic field is normalized by the imposed axial field $B_{0z}$. The Reynolds number is defined in terms of the rotation rate of the inner cylinder $$Re=\frac{\Omega_id^2}{\nu}.$$ Its value is chosen to be $Re = 6000$ throughout the paper, which is within the range used in the related experiments [@Stefani_etal09]. In this paper, we consider a highly resistive fluid with very small magnetic Prandtl number, $Pm=\nu/\eta=1.4\times 10^{-6}$, typical to the liquid metal alloy GaInSn, so that the corresponding magnetic Reynolds number of the flow is also small, $Rm=Re\cdot Pm=8.4\times 10^{-3}$.
The strength of the imposed axial field is measured by the Hartmann number, $$Ha=\frac{B_{0z}d}{\sqrt{\mu_0\rho_0\nu\eta}},$$ whereas the strength of the azimuthal field relative to the vertical one is measured by the helicity $\beta$ parameter introduced above. Everywhere below this parameter will be fixed to $\beta=4.53$, as adopted in some of the experiments [@Stefani_etal09]. Previous linear analysis showed that HMRI is effective for relatively strong azimuthal fields, $\beta \gtrsim 1$ (e.g., [@Hollerbach_Ruediger05; @Liu_etal06; @Kirillov_Stefani10; @Kirillov_etal14]).
Consider perturbations of the velocity, pressure and magnetic field about the equilibrium, ${\bf u}'={\bf u}-{\bf U}_0$, $p'=p-p_0$, ${\bf b}'={\bf B}-{\bf B}_0$. In the limit $Rm \ll 1$, which applies here, the magnetic field perturbation induced by the perturbed flow is much smaller than the imposed field and scales with $Rm$ [@Zikanov_Thess98; @Willis_Barenghi02; @Liu_etal06], so we change it to ${\bf b}'\rightarrow Rm\cdot {\bf b}'$, such that the new ${\bf
b}'$ is comparable to the perturbed velocity ${\bf u}$. Substituting this into Eqs. (1)-(3), taking into account that the imposed magnetic field is current-free, and using the above normalization, we arrive at the following equations governing perturbations with arbitrary amplitude in nondimensional form to first order in $Rm$ (primes henceforth will be omitted): $$\frac{\partial {\bf u}}{\partial t}+({\bf U}_0\cdot \nabla) {\bf
u}=-\nabla p-({\bf u}\cdot \nabla) {\bf U}_0+ \frac{Ha^2}{Re}(\nabla
\times {\bf b})\times {\bf B}_0 + \frac{1}{Re}\nabla^2 {\bf u}-({\bf
u}\cdot \nabla) {\bf u},$$ $$Rm\left[\frac{\partial {\bf b}}{\partial t}-\nabla\times \left({\bf
U}_0\times {\bf b}\right)-\nabla\times \left({\bf u}\times {\bf
b}\right)\right]=\nabla\times \left({\bf u}\times {\bf
B}_0\right)+\nabla^2{\bf b},$$ $$\nabla\cdot {\bf u}=0,~~~\nabla\cdot {\bf b}=0,$$ where the terms proportional to $Rm$ on the left hand side of induction equation (6), including the time-derivative, are usually neglected in the limit of small $Rm\rightarrow 0$ (inductionless, or quasi-static approximation, see e.g., [@Willis_Barenghi02; @Szklarski_Ruediger06; @Child_Hollerbach16]). In that case, the advective derivative, or the inertial term involving the perturbed velocity, $({\bf u}\cdot \nabla) {\bf u}$, is the only nonlinearity that remains in these equations. Here we preferred to keep these small terms for completeness. In the chosen nondimensional units, the rotational frequency (4) of the background velocity becomes $$\Omega(r)=\frac{1}{1-a^2}\left(\mu-a^2+\frac{a^2}{(1-a)^2}\frac{1-\mu}{r^2}\right)$$ and the imposed field is ${\bf B}_0=(0,\beta/r,1)$. Since $\eta,
\mu, \beta$ and $Re$ are fixed, only $Ha$ is a free parameter, entering Eqs. (5)-(7) through the Elsasser number, or the interaction parameter: $$\Lambda\equiv \frac{Ha^2}{Re}=\frac{B_{0z}^2}{\mu_0\rho_0\eta
\Omega_i},$$ which is composed of the imposed magnetic field, rotation and resistivity and is independent of viscosity. The interaction parameter has been shown to play a special role in the linear dynamics of the HMRI, as it determines its modal growth rate, the corresponding critical wavenumber and other characteristics [@Priede11; @Kirillov_etal14; @Mamatsashvili_Stefani16]. Specifically, in the inviscid case ($Re\rightarrow \infty$), the HMRI can exist in a certain range of $\Lambda$ and disappears at sufficiently large or small values of this parameter. In the first limit, the magnetic field is too strong and stabilizes the flow, whereas in the second limit it is too weak to support the instability. At finite $Re$, this implies that for a fixed $Ha$, HMRI operates in a range of $Re$ and for a fixed $Re$, in a range of $Ha$ [@Priede_etal07; @Priede_Gerbeth09; @Ruediger_etal06], which was also confirmed experimentally [@Stefani_etal07; @Stefani_etal09]. Since $\Lambda$ mainly determines the linear dynamics of HMRI, it is natural to explore how the nonlinear dynamics and saturation properties of HMRI depend on this parameter, which is our main goal. This is the main reason why we decided to follow the dynamics as a function of $\Lambda$. We anticipate that, like in the linear theory, this parameter will be decisive in the nonlinear outcome of the HMRI. Previous related nonlinear studies [@Szklarski_Ruediger06; @Szklarski07] focused mostly on the variation of the saturated state’s characteristics with the Reynolds number at fixed Hartmann number. Besides, for these parameters, HMRI was in the weakly nonlinear regime. In this connection, we mention that the interaction parameter also plays an important role in the dynamics of forced MHD turbulence with mean field in the small-$Pm$ regime [@Zikanov_Thess98].
Numerical method
----------------
We solve the basic Eqs. (1)-(3) using the pseudo-spectral code from [@Guseva_etal15]. It is based on the Fourier expansion in the axial $z$- and azimuthal $\phi$-directions and the finite-difference method in the radial direction. The nonlinear terms are calculated using the pseudospectral method and are de-aliased using the 3/2-rule. Integration in time is done with a second-order scheme based on the implicit Crank-Nicolson method. Although the inductionless approximation holds in the present case of small $Pm$, the code updates the induction Eq. (2) in time in a general manner without invoking this approximation. Further details of the code and its validation tests on HD and MHD problems in TC flows can be found in [@Guseva_etal15].
The cylindrical flow domain in nondimensional units spans the radial range $(R_i,R_o)=(1,2)$ divided into $N_r=100$ grid points, azimuthal range $(0, 2\pi)$ and axial range $(0,L_z)=(0,10)$ with, respectively, $N_{\phi}=80$ and $N_z=400$ spatial Fourier modes. The radial grid points are placed at collocation points of Chebyshev polynomials (despite the code is based on finite differences in $r$) that ensures higher resolution near the cylinder surfaces. Thus, although the HMRI is thought to be predominantly axisymmetric at least in the linear regime, we still allow for non-axisymmetric modes to naturally develop in the nonlinear regime by encompassing azimuthal wavenumbers, $m$, in the range $(-N_{\phi}/2,
N_{\phi}/2)$. A resolution test we performed (not shown here) showed that the adopted resolution is well sufficient for the present problem, capturing a whole range of wavenumbers from dynamically important intermediate ones, corresponding to the most unstable HMRI modes, down to wavenumbers dominated by viscous dissipation, as indicated by the energy spectra calculations in section 3.3 below. The radial boundary conditions at the cylinders are standard no-slip for the velocity perturbation, ${\bf u}|_{r=R_i,R_o}=0$, and insulating for the magnetic field perturbation, meaning that current does not penetrate into cylinders, that is, $\nabla \times {\bf
b}=0$ at $r<R_i$ and $r>R_o$. These radial boundary conditions are implemented in the code (see details in [@Willis_Barenghi02; @Willis_Barenghi02a; @Guseva_etal15]), while periodic conditions are imposed for all variables in the axial $z$-direction, with a period $L_z$, and in the azimuthal $\phi$-direction. In this first effort to gain a deeper insight into the process of the nonlinear saturation of HMRI, we vary the $\Lambda$ number at fixed $Re$ and other parameters of the flow (which is equivalent to varying $Ha$).
Energy and torque
-----------------
The perturbation energy density, composed of kinetic and magnetic parts, $e=({\bf u}^2+Ha^2Pm\cdot {\bf b}^2)/2$ (here ${\bf b}$ is the normalized magnetic field perturbation as defined in Eqs. 5-7), and torques at the cylinders, governing the evolution of angular momentum $l=ru_{\phi}$ (in nondimensional units), are among the important characteristics/diagnostics of the flow. We see that at very small $Pm$, the magnetic energy is much smaller than the kinetic one. The evolution equations for these quantities, taking into account the boundary conditions, are readily derived from Eqs. (5)-(7). As a result, we get for the total energy $E=\int edV$, to leading order in $Rm$ [@Priede_etal07; @Willis_Barenghi02a] $$\frac{dE}{dt}=-\int u_ru_{\phi}r\frac{d\Omega}{dr}dV+D,$$ where $D$ is the dissipation function $$D=-\frac{1}{Re}\int [(\nabla\times {\bf u})^2+Ha^2(\nabla \times
{\bf b})^2]dV.$$ The first term in Eq. (8) is the Reynolds stress $u_ru_{\phi}$ multiplied by the shear, $rd\Omega/dr$, and describes energy exchange between the basic flow and perturbations, which is therefore due to the shear and is not affected by the magnetic field. The dissipation function $D$ is always negative definite and consists of two parts. The first part describes the usual viscous dissipation of the kinetic energy, while the second part describes the Joule losses and in general is comparable to the former. Since the magnetic energy is negligible in this case, the Joule losses act as an additional effective dissipation of special kind for the kinetic energy through the kinetic-magnetic exchange term proportional to $\Lambda$ in Eq. (5) [@Zikanov_Thess98]. Note that a net contribution from nonlinear terms has canceled out in Eq. (8) after averaging over the domain due to the boundary conditions. Thus, only Reynolds stress, when positive, can act as a source of perturbation energy, extracting it from the background flow due to the shear. The Maxwell stress, $-b_rb_{\phi}$, does not play a role because of the smallness of the induced field in the low-$Rm$ regime. The nonlinear term, not directly tapping into the flow energy and therefore not changing the total perturbation energy, acts only to redistribute energy among different wavenumbers. In the quasi-steady saturated state, energy injection by the Reynolds stress balances dissipation losses. Since $d\Omega/dr<0$, this implies that the stress should be positive for the long-term sustenance of this state.
For the total angular momentum, $L=\int ldV$, we get $$\frac{dL}{dt}=G_i-G_o,$$ where $$G_{i,o}=-\frac{R_{i,o}^3}{Re}\int_0^{2\pi}
\int_0^{L_z}\frac{d}{dr}\left(\frac{u_{\phi}}{r}\right)|_{r=R_{i,o}}d\phi
dz$$ are the total viscous torques exerted, respectively, by the inner and outer cylinders on the flow. These torques also characterize angular momentum transport by the perturbations. Other – advection (i.e., Reynolds stress) and magnetic – torques vanish after volume integration due to the boundary conditions and do not contribute to the total angular momentum evolution. In the equilibrium (laminar) state, the torques at both cylinders have the same absolute value, $G_{lam}=-(2\pi L_z/Re) R_i^3d\Omega/dr|_{r=R_i}=-(2\pi L_z/Re)
R_o^3d\Omega/dr|_{r=R_o}=4\pi L_z(1-\mu)a^2/Re(1-a^2)(1-a)^2>0$, implying that the inner cylinder tries to increase the angular momentum of the basic flow, while the outer cylinder to decrease. These two torques are in balance, resulting in the stationary rotational profile (4). Below we measure the torque in the perturbed state against the laminar torque, i.e., consider the ratio $G_{i,o}/G_{lam}$ and redefine $G_{i,o}/G_{lam} \rightarrow
G_{i,o}$. Its advantage is that it does not depend on the normalization of the velocity.
Results
=======
We move to the analysis of the HMRI, from its linear growth to nonlinear saturation. To have an idea about the regimes of the onset and operation of HMRI in the given TC flow configuration, in figure 1 we show the growth rate of the most unstable mode (which is axisymmetric) as a function of $\Lambda$ and $Re$, obtained by solving the linear stability problem with the same parameters and boundary conditions. From this figure we see that HMRI operates in a certain interval of $\Lambda$ provided the Reynolds number is larger than a critical value, $Re>Re_c=1244$. The instability interval with respect to the interaction parameter broadens with the increase of $Re$, but converges in the inviscid limit (not shown in this figure). It is also seen in figure 1 that at sufficiently high $Re\gtrsim 3000$, the maximum growth of HMRI is mainly determined by $\Lambda$ and attained at its nearly constant value. For the chosen $Re=6000$, the instability first appears in the flow at the critical value $\Lambda_c=0.022$, reaches a maximum at $\Lambda=0.15$ and then decreases again, disappearing at $\Lambda=1.38$. Having outlined the main linear features of HMRI, we can now move to nonlinear evolution.
We initialize the simulations by imposing random noise perturbations of the velocity and magnetic field on the equilibrium flow. The subsequent evolution of the total energy $E$ and torques $G_i,G_o$ are shown in figure 2 at different interaction parameters, which are all above the critical value $\Lambda_c=0.022$ when HMRI first emerges. (We also checked that, as expected, the runs at $\Lambda<\Lambda_c$ eventually decay, although a further study with a variable initial amplitude of perturbations is required to see whether there is a possibility of subcritical transition in this case.) As we will see below, the dynamics in these cases differ qualitatively not only in terms of the temporal evolution, but also in its spatial appearance and spectral characteristics. After an initial transient phase, the most unstable mode eventually emerges and grows. The exponential growth phase lasts until the velocity amplitude becomes large enough for the nonlinear term in Eq. (5) (which is the dominant nonlinearity at $Rm\ll 1$) to come into play and halt this growth. The duration of this phase depends on $\Lambda$: if it is modestly larger than $\Lambda_c$, the saturation takes a fraction of viscous time $O(Re)$, whereas if $\Lambda$ is only slightly above $\Lambda_c$, saturation can take much longer, from several to tens of viscous times. This divergence of the saturation time at the bifurcation point (critical slowing down) is well known from dynamical systems (e.g., [@Kuznetsov]). As a result, the flow settles down into a statistically steady state where total energy and torques oscillate around well-defined mean values. The time variations become more irregular and the saturated energy and torque increase with increasing the interaction parameter. This behavior of the saturation level with $\Lambda$ will be investigated below in more detail. Because the volume-averaged angular momentum is also quasi-stationary in time, both torques have nearly the same absolute values, although the inner torque is more oscillatory at larger $\Lambda$ due to the appearance of smaller scale turbulent eddies near the inner cylinder (see below). This implies that the HMRI, like the SMRI, transports angular momentum outwards, which is expected, since it also derives energy from shear.
Saturation mechanism
--------------------
Let us see what is a main mechanism underlying the saturation of HMRI. During its initial exponential growth, the main nonlinear term (advective derivative in Eq. 5) also gradually gains strength and transfers the energy from the most unstable wavenumbers to larger and smaller ones. In the saturated state, the linear energy extraction rate due to the Reynolds stress associated with the unstable wavenumbers matches the nonlinear transfer rate from these wavenumbers to larger ones, where energy is ultimately dissipated due to viscosity. Together with $\Lambda$, the linear growth rate of HMRI is determined by the radial shear of the azimuthal velocity characterized by the Rossby number, $Ro=(r/2\Omega)d\Omega/dr$ (e.g., [@Liu_etal06; @Kirillov_etal14]), which is a function of radius in a TC flow (for a Keplerian flow $Ro = -0.75$ everywhere). In the exponential growth phase, a slowly increasing feedback of the nonlinear term on the large-scale azimuthal velocity (which is initially a TC profile given by Eq. 4) rearranges the radial distribution of shear ($Ro$), such as to reduce it by absolute value in the bulk of the flow. This is clearly illustrated in figure 3, which shows the radial profile of $Ro$, calculated with the total azimuthal velocity (i.e., initial TC flow plus the azimuthal velocity perturbation averaged over azimuthal $\phi-$ and axial $z$-directions) in the saturated state. It considerably deviates from the initial profile corresponding to the TC flow and results in the reduction of the linear growth rate of the HMRI and hence of the energy extraction rate from the mean flow. This is confirmed by figure 4, which shows a parallel evolution of energies in two simulations one starting with the saturated/modified mean azimuthal velocity profile for the run shown in figure 3 at $\Lambda=0.05$ and the other starting with the original TC profile (4) for the same other parameters of the flow, following only early exponential growth phase in the linear regime. Comparing the growth rates, derived from the inclination of ${\rm ln}(E)$ versus time, we clearly see that the saturated profile indeed results in a reduced exponential growth rate for the perturbation energy in the initial linear regime, which for the chosen value of $\Lambda$ is smaller than that in the presence of originally imposed TC flow by a factor of 3.77. Saturation sets in when the energy gain by the linear instability, being reduced by the nonlinearity, eventually becomes equal to the nonlinear transfer of the energy. The saturation levels in these two cases also differ, because, as we checked, the corresponding established nonlinear states have different structures – the number of rolls in the axial direction. Thus, there is no unique flow attractor, which would otherwise result in the same saturation amplitudes for these two profiles. Obviously, the deviation of the shear profile from the initial one due to the nonlinear feedback is larger, the higher is $\Lambda$, i.e., the more unstable is the flow. So, the saturation, where the energy gain by the linear HMRI is in balance with the nonlinear transfers, occurs through the modification (reduction) of the shear of the mean rotational velocity, which in turn defines the growth rate of HMRI. The modified shear profile in the quasi-steady state then stays practically unchanged during the evolution. This type of saturation mechanism is in fact similar to that for SMRI in TC flow discussed in [@Knobloch_Julien05; @Umurhan_etal07b; @Ebrahimi_etal09; @Clark_Oishi16a; @Clark_Oishi16b]. As for the initially imposed mean magnetic field ${\bf B}_0$, it basically does not change in the saturated state, because, as mentioned above, magnetic field perturbations are usually orders of magnitude smaller the imposed field in the low-$Rm$ regime.
{width="33.00000%"} {width="33.00000%"} {width="33.00000%"}
Properties of the saturated state
---------------------------------
Figure 5 shows the time-averaged values of the energy, $\langle
E\rangle$ (a), the inner torque $\langle G_{i}\rangle$ (b) and the dissipation rate $\langle D\rangle$ (c) in the quasi-steady saturated state as a function of $\Lambda$. Since this state is on average constant in time, the saturated mean values of both torques are essentially the same, $\langle G_{i}\rangle=\langle
G_{o}\rangle$, at each $\Lambda$, as it should be (see also [@Guseva_etal17]). This figure allows us to clearly see different regimes of the nonlinear saturation of the HMRI in TC flow and transitions between them. When the interaction parameter increases, the instability first appears at $\Lambda=\Lambda_c=0.022$ via classical supercritical Hopf bifurcation [@Knobloch96]. For $\Lambda>\Lambda_c$, but still near the instability threshold, the perturbations are weakly nonlinear and the saturated energy is proportional to $\propto
\Lambda-\Lambda_c$, as expected for Hopf bifurcation. The saturated torque and dissipation function also exhibit a similar dependence on $\Lambda$ as the energy. As mentioned above, in this case of $|\Lambda-\Lambda_c| << \Lambda_c$ the growth rate is small and the saturation time is long, several tens of viscous time. A typical spatial structure of the velocity in this regime is shown in figure 6(a), which consists of a well-organized chain of regular Taylor vortices uniformly filling the domain and resembles the corresponding most unstable mode from which it originated. As evident from the isosurface plot of the axial velocity, this weakly nonlinear state is axisymmetric. Five pairs of vortices fit in the domain, implying that the dominant most unstable azimuthal and axial wavenumbers are $(m,k_z)=(0,2\pi n_z/L_z)$ with $n_z=\pm 5$, which will be confirmed by spectra calculations presented in the next section. This picture is consistent with the pattern of the unstable modes of the HMRI observed in previous linear stability studies [@Hollerbach_Ruediger05] and nonlinear simulations in the similar weakly nonlinear regime [@Szklarski_Ruediger06]. This regime holds until about $\Lambda_{tr}=0.0232$, where an abrupt transition to less organized and irregular (turbulent) regime takes place, which is marked by a sharp inflection point on the curves of the saturated values in figure 5. Beyond this point both $\langle
E\rangle$ and $\langle G_{i}\rangle$ exhibit a different behavior with $\Lambda$ than the linear one in the weakly nonlinear regime. When $\Lambda$ is just above $\Lambda_{tr}$, they start with a slower increase, followed by a steeper monotonic increase starting from $\Lambda=0.025$. The transition to the strongly nonlinear regime is reflected in the clear change of temporal behavior of the total energy and torques – oscillations emerge in their evolution, which become more and more irregular and of higher amplitude with increasing the effect of nonlinearity as $\Lambda$ grows. This is illustrated in figure 7, which shows the time evolution of $G_i$ before (at $\Lambda=0.023$) and after (at $\Lambda=0.026$) the transition value $\Lambda_{tr}$ and also in figure 2 at higher $\Lambda$. The emerging oscillations in the later case are due to a wider spectrum of higher frequency modes (inertial waves, see below) excited in the stronger nonlinear regime. However, already after $\Lambda \simeq 0.08$, the saturated energy and torques do not seem to increase anymore and the character of the time-variation are more or less similar. In the next section, we will see that not only the time evolution properties, but also the power spectra of the saturated state of HMRI drastically and abruptly change when going from the weakly nonlinear to the strongly nonlinear regime.
{width="50.00000%"} {width="50.00000%"} {width="50.00000%"} {width="50.00000%"}
Figures 6(b)-6(d) show the spatial structures of the radial $u_r$ and axial $u_z$ velocities in the saturated state of HMRI at different $\Lambda$ after the transition point, which differ qualitatively from those in the weakly nonlinear regime in figure 6(a). Now the eddies of different shapes and sizes have appeared in the domain, being distributed non-uniformly along the axial direction. With increasing $\Lambda$, more and more small-scale vortices emerge mostly near the inner cylinder, while larger ones remain near the outer cylinder (figures 6(c) and 6(d)). As a result, the flow takes the form of fully developed turbulence. This is most clearly seen in the maps of the axial velocity. These eddies change shape in a random manner on the turnover (dynamical) time, which is the shorter the smaller is the eddy. As a result, the torque at the inner cylinder, $G_i$, displays faster chaotic oscillations due to the small-scale eddies than the torque at the outer cylinder,$G_o$, due to larger scale eddies, as seen in figures 2(b) and 2(c). Note also how the axisymmetry of the nonlinear state changes with $\Lambda$ (see also figure 9 below). After the transition point $\Lambda_{tr}$, before the emergence of smaller-scale eddies at higher $\Lambda \gtrsim 0.1$, the nonlinear state preserves near-axisymmetry, as evidenced by the isosurface plots of the axial velocity, for example, in the case of $\Lambda=0.05$ in figure 6(b). So, we observe a type of self-sustained quasi two-dimensional (2D, i.e., dependent only on $r$ and $z$ coordinates) MHD turbulence driven/fed by HMRI, analogous to that previously found in hydrodynamically unstable TC flow with much stronger and purely azimuthal background magnetic field [@Zikanov_etal14]. We calculate its spectral characteristics in the next section. However, with further increasing $\Lambda$, starting from about $\Lambda=0.07$, the nonlinear state slightly deviates from axisymmetry, as shown in figures 6(c) and 6(d), respectively, for $\Lambda=0.1$ and $0.2$ (see also figure 9 below). One can discern in these plots that the non-axisymmetry is associated mostly with smaller-scale eddies near the inner cylinder, while the larger scale ones near the outer cylinder stay mostly axisymmetric. Noticing that the saturated mean azimuthal velocity profile in figure 3 develops a Rayleigh-unstable region, $Ro < -1$, near the inner cylinder at this larger $\Lambda$, we may conjecture that the emergence of small-scale eddies (spatial defects) in the same radial interval is due to a HD instability of the nonlinearly modified flow profile. The spectral analysis presented below will give a more quantitative characterization of the degree of the non-axisymmetry in this case. In this regard, we would like to caution that studying the nonlinear dynamics of HMRI at any values of the Elsasser number (i.e., for any strength of the imposed magnetic field) with purely axisymmetric simulations might lead to the incomplete picture of the saturation and nonlinear dynamics of HMRI, overlooking the role of these modes. An advantage of the present analysis, including non-axisymmetric modes, is that it enables us to establish at which magnetic field strength (i.e., interaction parameter) the nonlinear saturated state of HMRI deviates from axisymmetric configuration (see below).
The transition from weakly nonlinear to strongly nonlinear and ultimately turbulent regime for HMRI is actually analogous to the nonlinear transition found for AMRI with increasing the Hartmann number of the imposed purely azimuthal field at constant $Re$ by Guseva et al. [@Guseva_etal15]. (Since Reynolds number is fixed, increasing the interaction parameter corresponds to the increase of Hartmann number.) As in our case, in the weakly nonlinear regime, a regular pattern of vortices emerge in the saturated state via Hopf bifurcation, except that for HMRI it is a traveling wave, as will be evident from spectral analysis below, while for AMRI it is a standing wave. With further increase of Hartmann number, so-called defects appear. The spatial structure at this stage is similar to that shown in figure 6(b) with nonuniformly distributed irregular vortices, not far from the transition point. Subsequently, a fully developed turbulent state arises from these defects at even higher Hartmann numbers – a stage which can be identified in our case with that depicted in figures 6(c) and 6(d). It is also expected that a similar hysteresis phenomenon mediated by an edge state dividing weakly nonlinear and strongly nonlinear (turbulent) states, as found by Guseva et al. for AMRI, also exists for HMRI. However, we did not study this possibility here: the system is always followed starting with the same random initial conditions when changing the interaction parameter. In this case, after the exponential growth, the flow always ends up in the chaotic state if $\Lambda$ is larger than the transition value. Identifying edge states and the hysteresis in the transition process for HMRI will be a subject of future study. Such edge states are known to play an important role in the turbulence transition in HD shear flows (see [@Avila_etal13] and references therein).
Spectral characteristics
------------------------
Further insight into the nature of the quasi-steady states of the HMRI at different interaction parameters, which we have described so far in physical space, can be gained by spectral analysis. It is well known from linear theory that in the case of the background helical magnetic field, $z$-reflection symmetry is broken [@Knobloch96; @Hollerbach_Ruediger05; @Kirillov_etal14] and, consequently, in the given right-handed configuration of the axial and azimuthal fields ($\Omega B_{0\phi}B_{0z}>0$), HMRI is characterized by inertial waves that travel downwards (opposite the $z$-axis) [@Liu_etal06; @Kirillov_etal14]. As a result, the subsequent nonlinear state is composed of these downward propagating waves. In a real experiment, both upward and downward propagating waves are present because of the reflection from the endcaps, but the one that is unstable tends to be dominant [@Stefani_etal09]. Below we analyse the axial and azimuthal spectra of these waves. To this end, we decompose each velocity component in the azimuthal and axial directions (in which they are periodic), $$\bar{u}_i(r,m,k_z)=\frac{1}{2\pi L_z}\int_0^{2\pi} \int_0^{L_z}
u_i(r,\phi,z){\rm e}^{-{\rm i}m\phi-{\rm i}k_zz}d\phi dz,$$ where the index $i=r,\phi,z$ and $m=0,1,2,...$ and $k_z=2\pi
n_z/L_z$ with $n_z=0,\pm 1, \pm 2,...$, are the azimuthal and axial wavenumbers, respectively. We also define radially integrated spectral energy density, $${\cal E}(m,k_z)=\pi L_z\int_{R_i}^{R_0}
(|\bar{u}_r|^2+|\bar{u}_{\phi}|^2+|\bar{u}_z|^2)rdr,$$ as well as the azimuthal, $\hat{\cal E}_m$, and axial, $\hat{\cal
E}$, spectral energy densities via $$\hat{\cal E}_m=\sum_{k_z}{\cal E}(m,k_z), ~~~\hat{\cal
E}=\sum_{m}{\cal E}(m,k_z),$$ so that their sum over wavenumbers is equal to the total energy, $E=\sum_{m,k_z} {\cal E}(m,k_z)=\sum_{m}\hat{\cal
E}_m=\sum_{k_z}\hat{\cal E}$. These energy spectra are among the main characteristics/diagnostics in the nonlinear (turbulent) regime in flows. In particular, the azimuthal spectrum characterizes the total contribution of different $m$ in the dynamics and hence can serve as a measure of the non-axisymmetry, while the axial $k_z$ spectra provide information on the energy-injection due to axisymmetric HMRI modes (see e.g., [@Zikanov_etal14]). We use these spectra below to characterize the dynamics in the quasi-steady state. This state, however, exhibits, as we have seen above, irregular oscillations, making the spectra noisy. To avoid this, we also average $\hat{\cal E}_m$ and $\hat{\cal E}$ in time over the whole duration of the quasi-steady state in the simulations.
Figure 8 shows the time-averaged axial energy spectrum $\hat{\cal
E}$ in the saturated state at different $\Lambda$. It is seen that qualitatively different spectra correspond to weakly and strongly nonlinear saturation regimes, and the rearrangement from the former to the latter type, like the saturated values of energy and torque (figure 5), occurs suddenly near the transition point $\Lambda_{tr}=0.0232$. At small $\Lambda\gtrsim \Lambda_c$, the linearly most unstable axial wavenumber of HMRI, $k_z=2\pi
n_z/L_z=3.14$ with the mode number $n_z=5$, is dominant also in the saturated state and carries most of the power. In physical space, this corresponds to five pair of vortices, as shown in figure 6(a). The modes with wavenumbers, which are multiples of the most unstable axial wavenumber, are excited due to weak nonlinear self-interaction of the dominant mode in the saturation process. These subharmonics (separate spikes in figure 8) have orders of magnitude smaller power compared to the dominant one, whereas at all other $k_z$ the spectral power is essentially zero. Remarkably, the energy spectrum changes qualitatively when the interaction parameter approaches a transition value $\Lambda_{tr}$, where the effect of nonlinearity is more appreciable. As a result, power in other wavenumbers, lying between the multiple wavenumbers, abruptly increases, i.e., the spectrum is quickly filled and becomes continuous, although the multiple harmonics of the most unstable wavenumber still have larger power than other ones. But already at $\Lambda=0.025$, slightly larger than the transition point, the spectrum takes a smoother shape, continuously well populated at all wavenumbers, as it is characteristic of a turbulent state. With further increasing $\Lambda$, this spectrum moves upwards and converges starting from about $\Lambda=0.05$. Now, the maximum comes again at the wavenumber of the most unstable HMRI mode, $k_z\sim 2-3$, which mainly determines the energy injection from the basic flow into turbulent fluctuations. As it is seen in figure 8, at intermediate wavenumbers in the range $5\lesssim k_z\lesssim 20$, this converged spectrum displays the power-law dependence close to $k_z^{-3}$ typical for 2D HD turbulence [@Kraichnan67], as opposed to the Kolmogorov spectrum, $k^{-5/3}$, for three-dimensional turbulence or to the $k^{-2}$ spectrum for rotating HD turbulence [@Mueller07]. Then, at $k_z\gtrsim 20$ the spectrum decays faster because of viscous dissipation. The appearance of the quasi-2D turbulence in physical space, corresponding to these spectra, we have already seen in figure 6. The energy spectrum with a similar power-law dependence were also reported for 2D MHD turbulence in TC flow with a large purely azimuthal magnetic field in the highly resistive (inductionless) limit, $Rm \ll 1$, in the above-mentioned work [@Zikanov_etal14]. In that case, however, the turbulence is due to HD instability and forced to be 2D by the imposed azimuthal field. By contrast, in the present case, where the TC flow is hydrodynamically stable, the observed turbulence is of magnetic origin, triggered and energetically supplied by HMRI, and hence quasi-2D from the outset. This implies that the turbulence we observe here cannot be described within standard Kolmogorov phenomenology, which anyway is inapplicable in the case of high resistivity, because the Joule dissipation in this regime is anisotropic and acts at all scales in the flow (see e.g., [@Zikanov_Thess98]). In this situation, it is not possible generally to define an inertial range in a classical sense, where only nonlinear term $({\bf u}\cdot \nabla){\bf u}$ in Eq. (5) operates and transfers energy towards large wavenumbers. The presence of the power-law interval in the energy spectrum different from Kolmogorov one that has been found here and in [@Zikanov_etal14] is a confirmation of that. The similarity of the spectra and the associated power-law indices, despite different driving mechanisms of the turbulence in our case and in [@Zikanov_etal14], indicates that these spectral features could be generic to quasi-2D MHD turbulence, which can occur in a magnetized resistive TC flow.
{width="50.00000%"} {width="50.00000%"}
As we have found above, the saturated state remains nearly axisymmetric, despite non-axisymmetric modes emerging with increasing $\Lambda$. To quantify the role of the latter modes, in figure 9 we plot the ratio of the sum of azimuthal energies $\hat{\cal E}_m$ of all the non-axisymmetric modes with $m\geq 1$ to the energy of the axisymmetric modes with $m=0$, measured by the parameter $A=\sum_{m\geq 1}\hat{\cal E}_m/\hat{\cal E}_0$, as a function of $\Lambda$ (a) as well as the typical azimuthal energy spectrum in the strongly nonlinear state (b). From this figure, it is seen that non-axisymmetric modes are essentially absent at small $\Lambda$; they start to emerge from about $\Lambda=0.07$, after which the total energy of all non-axisymmetric modes relative to that of axisymmetric ones increases with $\Lambda$. However, the former still remains much smaller than the latter, particularly in the range $0.07\leq \Lambda \leq 0.2$, where the saturation level reaches a maximum (figure 5). This is also confirmed by the azimuthal spectrum of the energy, $\hat{\cal E}_m$, in figure 9(b). Since non-axisymmetric modes are present only at $\Lambda\gtrsim
0.07$, we show the azimuthal spectrum at typical values $\Lambda=0.1$ and 0.2, corresponding to the strongly nonlinear (turbulent) regime. It reaches a maximum at $m=0$ and rapidly decreases with $m$. The energy of the first non-axisymmetric modes with $m=1$ is already by more than an order of magnitude smaller than that of the axisymmetric ones, $\hat{\cal E}_1/\hat{\cal
E}_0$=0.007 and 0.04, respectively, at $\Lambda=0.1$ and 0.2, and the energy of higher non-axisymmetric modes is even lower.
Summary and conclusion
======================
In this paper, we investigated the development of HMRI in an infinite/periodic TC flow domain at very small magnetic Prandtl numbers by following its evolution from the linear growth phase to nonlinear saturation using direct numerical simulations. To focus on the basic dynamics of HMRI arising from the combined effect of differential rotation and helical magnetic field, we simplified the analysis by ignoring the effects of endcaps that would induce Ekman pumping. We analyzed the dynamics with respect to the interaction parameter, or Elsasser number, $\Lambda$, which is known to be a central parameter determining the linear evolution of HMRI. As distinct from previous nonlinear analyzes of HMRI, which focused only on axisymmetric modes, an advantage of our study is that by allowing for non-axisymmetric modes in the simulations, it enabled us to establish for which interaction parameters the role of the latter modes becomes important. We confirmed that in the nonlinear regime, just as SMRI, HMRI transports angular momentum outward. We demonstrated that different regimes of nonlinear saturation are realized when the interaction parameter changes. At smaller values of this parameter, just above the stability threshold, the HMRI saturates in the weakly nonlinear regime and the corresponding spatial structure consists of well-organized regular vortices, which are axisymmetric. In this case, the most unstable mode of instability dominates, which is also supported by our spectral analysis. However, with increasing the interaction parameter, at a certain value an abrupt transition to strongly nonlinear (turbulent) state takes place, marked by different, irregularly oscillating behavior of the energy and stress. The saturated values of the energy and stresses as a function of $\Lambda$ exhibit a sharp inflection point, corresponding to this transition, and then increase with until about $\Lambda=0.1$ and then slowly decrease (figure 5), since the effectiveness of the HMRI itself is decreasing at higher magnetic fields. The spatial structure of the nonlinear state is predominantly axisymmetric and represents a type of self-sustained quasi-2D (in $(r,z)$-plane) MHD turbulence due to HMRI. Generally, classical 2D MHD turbulence is decaying and requires external forcing for long-term maintenance [@Biskamp03]. So, here we demonstrated the existence of self-sustained quasi-2D MHD turbulence in a TC flow at high resistivity, or small magnetic Prandtl numbers caused and supplied by HMRI at the expense of the flow energy. We also calculated the energy spectrum as a function of axial wavenumber and found that it increases with $\Lambda$, but converges from about $\Lambda=0.05$, and at intermediate axial wavenumbers $k_z$ exhibits a power-low dependence close to $k_z^{-3}$, as typical for 2D HD turbulence [@Kraichnan67]. It would be interesting to probe in future studies whether this quasi-2D turbulent state persists as the Reynolds number increases or a transition to three-dimensional turbulence occurs.
In both weakly and strongly nonlinear regimes, the saturation mechanism appears to be general, consisting in the modification, or reduction of the radial shear profile of the mean azimuthal velocity in the bulk of the flow, which, in turn, results in the reduction of the exponential growth rate of HMRI to match energy transfer rate due to the nonlinear (advection) term. Such a nonlinear saturation mechanism was already discussed for SMRI in a TC flow [@Knobloch_Julien05; @Umurhan_etal07a; @Ebrahimi_etal09; @Clark_Oishi16a; @Clark_Oishi16b].
The present analysis, although simplified by excluding the effects of endcaps, is a first step towards understanding the experimental manifestation of HMRI, where it is already in the saturated nonlinear regime. The insight gained from this study will be a stepping stone for future numerical studies incorporating endcaps and conducing boundaries, where the situation is complicated by Ekman circulations, penetrating from the endcaps in the bulk of the flow, and for subsequent comparison with planned experiments on HMRI within the DRESDYN project at HZDR. Nevertheless, as demonstrated in [@Guseva_etal15] on the example of AMRI, the results obtained with periodic boundary conditions along the axial direction are still useful to interpret the experiments. In the future, we plan to do analogous studies of HMRI in a TC flow with endcaps and compare with the present results. To date, such realistic three-dimensional simulations of HMRI, which are necessary for understanding the related experimental results, have not been undertaken yet because of high computational cost. Although previous early simulations explored the nonlinear development of HMRI in TC flow with endcaps, they considered a narrow range of parameters and/or were axisymmetric by design [@Szklarski_Ruediger06; @Liu_etal07; @Szklarski07], not capturing different regimes of the nonlinear saturation.
Another interesting venue of research extending the present analysis is to explore the nonlinear dynamics of HMRI and associated angular momentum transport for quasi-Keplerian rotation relevant to protoplanetary disks, whose dense and cold interiors are too resistive for SMRI to operate. The nonlinear development of HMRI has not been explored also for positive shear profiles relevant to the near-equator strip of the solar tachocline. These topics are of a current research interest and have been studied so far only in the linear regime [@Kirillov_Stefani13; @Kirillov_etal14; @Stefani_Kirillov15].
This work was supported by the Alexander von Humboldt Foundation and by the German Helmholtz Association in frame of the Helmholtz Alliance LIMTECH. Computations were done on the high-performance Linux cluster Hydra at the Helmholtz-Zentrum Dresden-Rossendorf. We would like to thank R. Hollerbach for sharing and assisting with his code and valuable discussions on HMRI at the initial stage of this work. We also thank O. Zikanov for useful discussions on low-$Rm$ MHD turbulence.
References {#references .unnumbered}
==========
| {
"pile_set_name": "ArXiv"
} |
---
author:
- |
Vincent Bruneau$^1$, Adrien Doradoux$^{1,2}$, Pierre Fabrie$^2$\
$^1$ Université de Bordeaux, IMB, CNRS UMR5251, 351 cours de la libération, 33405 Talence - France\
$^2$ Bordeaux INP, Institut de Mathématiques de Bordeaux, CNRS UMR5251, ENSEIRB-MATMECA, Talence - France
bibliography:
- 'bibli.bib'
title: 'Convergence of a Vector Penalty Projection Scheme for the Navier-Stokes Equations with moving body'
---
**Résumé**
Dans ce travail nous analysons un schéma de projection vectorielle (voir [@AN08]) pour traiter le déplacement d’un corps solide dans un fluide visqueux incompressible dans le cas où l’interaction du fluide sur le solide est négligeable. La présence de l’obstacle dans le domaine solide est modélisée par une méthode de pénalisation.\
Nous montrons la stabilité du schéma et la convergence des variables vitesse-pression vers une limite quand le paramètre $\e$ qui assure une faible divergence et le pas de temps $\delta t$ tendent vers $0$ avec une contrainte de proportionalité $\e = \lambda \delta t$.\
Finalement nous montrons que le problème converge au sens faible vers une solution des équations de Navier-Stokes avec une condition aux limites de non glissement sur la frontière immergée quand le paramètre de pénalisation $\eta$ tend vers $0$.
Introduction {#introduction .unnumbered}
============
Main result
===========
Mathematical recalls
====================
Stability analysis {#sec:stability}
==================
Convergence analysis when $\e$ and $\delta t$ tend to 0 {#sec:convergence_epsilon}
=======================================================
Convergence towards the Navier-Stokes Equations {#sec:convergence_eta}
===============================================
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Spectral unmixing is a crucial processing step when analyzing hyperspectral data. In such analysis, most of the work in the literature relies on the widely acknowledged linear mixing model to describe the observed pixels. Unfortunately, this model has been shown to be of limited interest for specific scenes, in particular when acquired over vegetated areas. Consequently, in the past few years, several nonlinear mixing models have been introduced to take nonlinear effects into account while performing spectral unmixing. These models have been proposed empirically, however without any thorough validation. In this paper, the authors take advantage of two sets of real and physical-based simulated data to validate the accuracy of various nonlinear models in vegetated areas. These physics-based models, and their corresponding unmixing algorithms, are evaluated with respect to their ability of fitting the measured spectra and of providing an accurate estimation of the abundance coefficients, considered as the spatial distribution of the materials in each pixel.'
author:
- |
Nicolas Dobigeon$^{(1)}$, Laurent Tits$^{(2)}$, Ben Somers$^{(3)}$,\
Yoann Altmann$^{(1)}$ and Pol Coppin$^{(2)}$\
$^{(1)}$ University of Toulouse, IRIT/INP-ENSEEIHT/TéSA, Toulouse, France\
`{Nicolas.Dobigeon,Yoann.Altmann}@enseeiht.fr`\
$^{(2)}$ Department of Biosystems, Katholieke Universiteit Leuven, 3001 Leuven, Belgium\
`{Laurent.Tits,Pol.Coppin}@biw.kuleuven.be`\
$^{(3)}$ Division Forest, Nature and Landscape, Katholieke Universiteit Leuven, 3001 Leuven, Belgium\
`[email protected]` [^1]
bibliography:
- 'D:/Dropbox/strings\_all\_ref.bib'
- 'D:/Dropbox/all\_ref.bib'
title: A comparison of nonlinear mixing models for vegetated areas using simulated and real hyperspectral data
---
Introduction {#sec:introduction}
============
Spectral unmixing (SU) of hyperspectral images consists of extracting the spectral responses $\Vmat{1},\ldots,\Vmat{\nbmat}$ of the $\nbmat$ macroscopic materials (or *endmembers*) present in the imaged scene and, for each pixel $\Vpix{\nopix}$ of the image ($\nopix=1,\ldots,\nbpix$), estimating the corresponding proportions $\abond{1}{\nopix},\ldots,\abond{\nbmat}{\nopix}$ (or *abundances*) that represent the spatial distributions of these materials over the area of interest [@Bioucas2012jstars]. The first automated unmixing techniques have been proposed in the early 1990’s [@Keshava2002]. When no prior knowledge is available regarding the studied scene, SU can be usually decomposed into two successive steps. First, the endmembers are extracted from the image and, subsequently, the proportions of the materials are estimated in a so-called *inversion* step. A vast majority of the endmember extraction algorithms (EEA) and inversion techniques exploit some geometrical concepts that are intrinsically related to an assumption of a linear mixing process to explain the observed pixels. In other words, under this linear mixing model (LMM), each observed pixel of a given image is assumed to result from the linear combination of the $\nbmat$ endmember spectra $$\label{eq:LMM}
\Vpix{\nopix}^{(\textrm{LMM})} = \sum_{\nomat=1}^{\nbmat} \abond{\nomat}{\nopix}
\Vmat{\nomat} + \Vnoise{\nopix} = \MATmat\Vabond{\nopix} +
\Vnoise{\nopix}$$ where $\Vabond{\nopix}=\left[\abond{1}{\nopix},\dots,\abond{\nbmat}{\nopix}\right]^T$ denotes the proportions of the $\nbmat$ materials in the $\nopix$th pixel, $\MATmat=\left[\Vmat{1},\ldots,\Vmat{\nbmat}\right]$ is the endmember matrix and $\Vnoise{\nopix}$ stands for an additive residual term accounting for the measurement noise and modeling error. Since the mixing coefficients $\abond{1}{\nopix},\ldots,\abond{\nbmat}{\nopix}$ are expected to represent the actual spatial distribution of the materials in the $\nopix$th pixel, they are commonly subject to the following positivity and sum-to-one (or additivity) constraints $$\label{eq:constraints_LMM}
\left\{
\begin{array}{ll}
\abond{\nomat}{\nopix}\geq 0, & \forall \nomat,\ \forall \nopix \\
\sum_{\nomat=1}^{\nbmat} \abond{\nomat}{\nopix}=1, & \forall \nopix . \\
\end{array}
\right.$$ This LMM has received a considerable attention in the image processing and remote sensing literature since it represents an acceptable first-order approximation of the physical processes involved in most of the scenes of interest [@Keshava2002]. Consequently, it has motivated a lot of research works that aim at developing efficient endmember extraction algorithms (EEA), designed to recover pure spectral signatures in the image, and inversion techniques to estimate the abundance coefficients for a given (estimated or *a priori* known) set of endmembers. Comprehensive overviews of these EEA and inversion methods can be found in [@Keshava2002; @Somers2011; @Bioucas2012jstars]. Specifically, two main approaches have been advocated to solve the inversion step, that can be formulated as a constrained optimization problem solved by fully constrained least square algorithms [@Heinz2001; @Theys2009; @Heylen2011tgrs] or as a statistical estimation problem solved within a Bayesian framework [@Dobigeon2008; @Eches2010ip; @Eches2011icassp].
However, for specific applications, LMM has demonstrated some difficulties to accurately describe real mixtures [@Dobigeon2014]. Notably, intimate mixtures of minerals are characterized by spatial scales typically smaller than the path length followed by the photons, which violates one fundamental assumption for considering a linear model. Analyzing such mixtures, e.g., composed of minerals, requires to resort to complex physical models coming from the radiative transfer theory. Various approximating models have been proposed in the spectroscopic literature, such as the popular Hapke’s model [@Hapke1981]. More recently, this model or related alternatives have been exploited in the hyperspectral literature to derive unmixing algorithms dedicated to remotely sensed images [@Guilfoyle2001; @Nascimento2010]. Broadwater *et al.* derived various kernel-based unmixing techniques that implicitly relied on the Hapke model [@Broadwater2009whispers; @Broadwater2010whispers; @Broadwater2011whispers]. In [@Close2012spie; @Close2012spieb; @Heylen2014], the authors combined linear and intimate mixing processes in single models to improve flexibility.
Conversely, scenes acquired over vegetated areas are also known to be subjected to more complex interactions that can not be properly taken into account by a simple LMM [@Borel1994; @Ray1996; @Zhang1998; @Chen2006; @Fan2009; @Somers2009; @Tits2012igarss; @Somers2014]. Indeed, for these specific scenarios addressed in this paper, differences in elevation between the transparent $3$D vegetation canopies and the relatively flat soil surfaces submit photons to multipath and scattering effects. Similar interaction effects have been also encountered when analyzing urban scenes [@Huard2011whispers; @Fontanilles2011; @Meganem2014]. Therefore, various attempts have been conducted to overcome the intrinsic limitations of the LMM. A large family of nonlinear models that have been proposed to analyze vegetated areas can be described as $$\label{eq:NLMM}
\Vpix{\nopix} = \MATmat\Vabond{\nopix} +
\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)
+ \Vnoise{\nopix}.$$ In , the observed pixel is composed of a linear contribution similar to the LMM and an additive nonlinear term $\boldsymbol{\mu}\left(\cdot\right)$ that may depend on the endmember matrix $\MATmat$, the abundance coefficients in $\Vabond{\nopix}$ and additional nonlinearity coefficients $\Vnabond{\nopix}$ introduced to adjust the amount of nonlinearity in the pixel. This class of models includes the bilinear models [@Altmann2011whispers], the quadratic-linear model [@Meganem2014], the post-nonlinear model [@Altmann2012ip] and the bilinear-bilinear model [@Eches2014grsl] (the most commonly used will be fully described in Section \[sec:models\]).
However, to our knowledge, most of these models have been derived following physical or intuitive considerations, without any careful and thorough analysis of their ability to properly describe real mixtures while performing spectral unmixing. In this article, we propose to fill this gap by evaluating the relevance of various nonlinear models when used for spectral unmixing of images acquired over vegetated areas. Specifically, requirements for ensuring the quality of a model in this specific applicative context are threefold: *(i)* the model should not depend on external parameters related to the studied scene (e.g., leaf index area, geometry or illumination incidence) since this prior knowledge is generally not available, *(ii)* this model should be still sufficiently flexible to fit the real observations in various external conditions, despite the ignorance of these unknown external parameters, *(iii)* it should be able to account for the relative spatial distribution of the materials in the pixel, with the prime objective to estimate the abundance coefficients. In particular, mainly because of the two first requirements enounced above, advanced nonlinear models proposed in the remote sensing literature (e.g., [@Borel1994]) will not be considered in this study since they need a detailed prior knowledge regarding the analyzed scene.
To meet this challenge, we take advantage of an interesting set of simulated and in-situ collected hyperspectral data. First, we use a detailed virtual orchard and forest model constructed in a physically based ray-tracing environment using detailed sub-models for the description of tree geometry, leaf and soil bidirectional reflectance and diffuse illumination [@Stuckens2009; @VanderZande2010; @Tits2012]. The model has been thoroughly validated with field observations [@Stuckens2009], and more recently we could provide, based on a comparison with in-situ data, strong evidence that our ray tracing model realistically describes the spectral scattering and thus nonlinearity observed in vegetated areas [@Somers2014]. Second, we use data from an in-situ experiment in a commercial citrus orchard. This experiment comprised in-situ measured mixed pixel reflectance spectra, pixel specific endmember spectra and subpixel cover fraction distributions. This unique dataset of in-situ measured mixed pixel reflectance spectra has previously been used to study nonlinearity in fruit orchards [@Somers2009].
The paper is organized as follows. Section \[sec:models\] introduces the main nonlinear models that have been proposed in the literature to describe mixtures encountered in vegetated areas. The ray-tracer based simulated data and the in-situ measurements used to validate these models are described in Section \[sec:data\]. The experiment results obtained by using the previously introduced nonlinear models on the two sets of data are reported in Section \[sec:results\]. A comprehensive discussion on these results is conducted in Section \[sec:discussion\]. Section \[sec:conclusion\] concludes the paper.
Nonlinear mixing models {#sec:models}
=======================
Bilinear models {#subsec:bilinear}
---------------
To take into account the scattering effects the photons are subjected to before reaching the sensor, a wide class of nonlinear models are derived by defining the nonlinear component $\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)$ in as a sum of bilinear terms [@Altmann2011whispers] $$\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)
\triangleq \sum_{i=1}^{R-1}\sum_{j=i+1}^{R} \nabond{i,j}{\nopix} \Vmat{i}\odot \Vmat{j}$$ where the operator $\odot$ stands for a termwise product $$\label{eq:term_BLMM}
\Vmat{i}\odot \Vmat{j} \triangleq \left(
\begin{array}{c}
\mat{1}{i}\mat{1}{j} \\
\vdots \\
\mat{\nbband}{i}\mat{\nbband}{j} \\
\end{array}
\right).$$ The set of nonlinearity coefficients $\left\{\nabond{i,j}{\nopix}\right\}_{i,j}$ allows the amount of nonlinearity in the $p$th pixel to be adjusted between each pair of materials $\Vmat{i}$ and $\Vmat{j}$. Most of the various bilinear models of the literature mainly differ by the definition of these coefficients $\nabond{i,j}{\nopix}$ and the associated constraints they are subject to. The most common models, that will be evaluated in Section \[sec:results\], are recalled below.
In [@Somers2009] and [@Nascimento2009spie], the authors propose to include the nonlinearity coefficients $\left\{\nabond{i,j}{\nopix}\right\}_{i,j}$ within the set of constraints defined by the LMM, leading to $$\label{eq:FM}
\Vpix{\nopix}^{(\textrm{NM})} \triangleq \sum_{\nomat=1}^{\nbmat} \abond{\nomat}{\nopix}
\Vmat{\nomat} + \sum_{i=1}^{R-1}\sum_{j=i+1}^{R} \nabond{i,j}{\nopix} \Vmat{i}\odot \Vmat{j} + \Vnoise{\nopix}$$ with $$\label{eq:constraints_NLMM}
\left\{
\begin{array}{l}
\abond{\nomat}{\nopix}\geq 0, \quad \forall \nomat,\ \forall \nopix \\
\nabond{i,j}{\nopix}\geq 0, \quad \forall \nomat,\ \forall i\neq j \\
\sum_{r=1}^{\nbmat} \abond{\nomat}{\nopix}+
\sum_{i=1}^{\nbmat-1}\sum_{j=i+1}^{\nbmat} \nabond{i,j}{\nopix}
= 1, \ \ \forall \nopix.
\end{array}
\right.$$
Note that this model, denoted NM for Nascimento’s model in this article, reduces to the LMM when $\nabond{i,j}{\nopix}= 0$, $\forall
i\neq j$. This is an interesting property since the LMM is known to be an admissible first approximation of the actually involved physical processes[^2]. However, in a more general case (i.e., $\nabond{i,j}{\nopix}\neq0$), the abundance coefficients $\left\{\abond{r}{\nopix}\right\}_{r=1}^R$ are not subject to the sum-to-one constraints defined in .
In [@Fan2009], Fan *et al.* have defined the nonlinearity coefficients $\nabond{i,j}{\nopix}$ as the product of the abundances, $\nabond{i,j}{\nopix}\triangleq \abond{i}{\nopix}
\abond{j}{\nopix}$, under the LMM-based constraints in , leading to the so-called Fan’s Model (FM) $$\label{eq:FM}
\Vpix{\nopix}^{(\textrm{FM})} \triangleq \sum_{\nomat=1}^{\nbmat} \abond{\nomat}{\nopix}
\Vmat{\nomat} + \sum_{i=1}^{R-1}\sum_{j=i+1}^{R} \abond{i}{\nopix} \abond{j}{\nopix} \Vmat{i}\odot \Vmat{j} +
\Vnoise{\nopix}.$$ The main motivation for relating the amount of nonlinear interactions (governed by $\nabond{i,j}{\nopix}$) to the amount of linear contribution (governed by $\abond{i}{\nopix}$ and $\abond{j}{\nopix}$) is straightforward: the more a given material is present in the pixel, the more nonlinear interactions may occur. In particular, if a component $\Vmat{i}$ is absent in the $p$th pixel, then $\abond{i}{\nopix} = 0$ and consequently $\nabond{i,j}{\nopix}=0$, which means that there are no interactions between the material $\Vmat{i}$ and any other materials $\Vmat{j}$ ($j\neq i$). Note however that this bilinear model does not extend the LMM.
To cope with this latter limitation, the generalized bilinear model (GBM) [@Halimi2011] weights the products of abundances $\abond{i}{\nopix} \abond{j}{\nopix}$ by additional free parameters $\gamma_{i,j,p}\in (0,1)$ that tune the amount of nonlinear interactions, leading to $\nabond{i,j}{\nopix} \triangleq
\gamma_{i,j,p}\abond{i}{\nopix} \abond{j}{\nopix} $ and $$\label{eq:GBM}
\Vpix{\nopix}^{(\textrm{GBM})} \triangleq \sum_{\nomat=1}^{\nbmat} \abond{\nomat}{\nopix}
\Vmat{\nomat} + \sum_{i=1}^{R-1}\sum_{j=i+1}^{R} \gamma_{i,j,p}\abond{i}{\nopix} \abond{j}{\nopix} \Vmat{i}\odot \Vmat{j} +
\Vnoise{\nopix}.$$ The GBM has the nice properties of i) generalizing the LMM by enforcing $\gamma_{i,j,p}=0$ ($\forall i,j$), similarly to NM but contrary to FM and ii) having the amount of nonlinear interactions to be proportional to the material abundances, similarly to FM but contrary to NM.
Post-nonlinear mixing model
---------------------------
Inspired by pioneered works in blind source separation [@Taleb1999], Altmann *et al.* have introduced in [@Altmann2012ip] a nonlinear model that relies on a $2$nd-order polynomial expansion of the nonlinearity, $$\label{eq:term_PPNM}
\begin{split}
\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)
&\triangleq b_p \left(\MATmat\Vabond{p}\right) \odot
\left(\MATmat\Vabond{p}\right)\\
&= b_p \sum_{i=1}^R\sum_{j=1}^R \abond{i}{\nopix} \abond{j}{\nopix} \Vmat{i}\odot \Vmat{j}
\end{split}$$ leading to the following polynomial post-nonlinear mixing model (PPNM) $$\Vpix{\nopix}^{(\textrm{PPNM})} = \MATmat\Vabond{p} + b_p \sum_{i=1}^R\sum_{j=1}^R \abond{i}{\nopix} \abond{j}{\nopix} \Vmat{i}\odot \Vmat{j}+ \Vnoise{p}.$$ The PPNM has demonstrated a noticeable flexibility to model various nonlinearities not only for unmixing purposes [@Altmann2012ip] but also to detect nonlinear mixtures in the observed image [@Altmann2013ip]. This model has also the great advantage of having the amount of nonlinearity to be governed by a unique parameter $b_p$ in each pixel, contrary to NM or GBM. Eq. also shows PPNM includes bilinear terms $\Vmat{i}\odot \Vmat{j}$ ($j \neq i$) similar to those involved in the NM, FM and GBM, and also quadratic terms $\Vmat{i}\odot \Vmat{i}$, which may account for interactions between similar materials.
Unmixing algorithms {#subsec:algorithm}
-------------------
To evaluate the accuracy of the mixing models of interest, the pixels of the in-situ and simulated data are unmixed with respect to each model. When analyzing the pixels with the LMM, the nonlinear contribution $\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)$ is set to zero. Based on the prior knowledge of the endmember signatures $\MATmat$, the abundance vector $\Vabond{\nopix}$ associated with each pixel $\Vpix{\nopix}$ is estimating by solving the constrained minimization problem $$\label{eq:criterion_LMM}
\hatVabond{\nopix} =
\operatornamewithlimits{argmin}_{\Vabond{\nopix}}
\left\|\Vpix{\nopix} - \MATmat\Vabond{\nopix}\right\|_2^2 \quad
\text{s.t.} \quad \eqref{eq:constraints_LMM}.$$
In this work, to solve this problem, the fully constrained least square (FCLS) algorithm [@Heinz2001] is used.
Moreover, when analyzing the pixels with nonlinear mixing models, the abundance vector $\Vabond{\nopix}$ and the nonlinearity parameter vector $\Vnabond{\nopix}$ associated with each pixel $\Vpix{\nopix}$ are estimated by solving the following constrained optimization problem $$\label{eq:criterion_NLMM}
\left(\hatVabond{\nopix}, \hatVnabond{\nopix}\right) =
\operatornamewithlimits{argmin}_{\Vabond{\nopix},\Vnabond{\nopix}}
\left\|\Vpix{\nopix} - \MATmat\Vabond{\nopix} -
\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)\right\|_2^2.$$ Depending on the considered model, the set of constraints imposed to the abundance vector $\Vabond{\nopix}$ and the possible nonlinear coefficient vector $\Vnabond{\nopix}$ may differ. For the FM, GBM and PPNM, the abundance vector $\Vabond{\nopix}$ should satisfy the LMM-based constraints , while for the NM, this constraint is applied to the joint vector $\left[\Vabond{\nopix},\Vnabond{\nopix}\right]$. Similarly, the nonlinear coefficient vector $\Vnabond{\nopix}$ for the GBM and PPNM should satisfy constraints that depend on the considered model and the nonlinearity component $\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)$ in or depends also on the considered nonlinear model.
For the experimental results reported in Section \[sec:results\], the FCLS algorithm is used to solve the NM-based unmixing problem since NM can be interpreted as a linear mixture of an extended set of endmembers, as shown in [@Nascimento2009spie]. The FM parameters are estimated with the algorithm detailed in [@Fan2009], based on a first-order Taylor series expansion of the nonlinearity $\boldsymbol{\mu}\left(\MATmat,\Vabond{\nopix},\Vnabond{\nopix}\right)$. Finally, the gradient descent and the subgradient descend algorithms developed in [@Halimi2011igarss] and [@Altmann2012ip] are used to solve the GBM- and PPNM-based unmixing problems, respectively. Interested readers are invited to refer to these works for detailed information regarding the optimization schemes.
Data description {#sec:data}
================
The mixing models and corresponding unmixing algorithms detailed in the previous sections are compared using simulated and real hyperspectral images. It is worth noting that, for both kinds of datasets, actual pure component spectral signatures (i.e., endmember spectra) and quantitative spatial distributions of these components (i.e., abundances) are available as ground truth in each pixel of the considered images. These datasets[^3] are described in this section.
Simulated dataset
-----------------
Two types of synthetic hyperspectral image data were generated from a ray tracing experiment. First, synthetic but realistic fully calibrated virtual scenes, namely citrus orchards and a forest, have been designed using methods developed in [@Stuckens2009] and [@VanderZande2008], respectively, which will be explained in more detail in the following paragraphs. Then, corresponding hyperspectral images have been simulated using an extended version of the physically based ray tracer (PBRT) [@Pharr2004]. In PBRT, a scene is defined using submodels to describe the various components of the scene: illumination sources, sensor platform, material optical properties, integrator and geometry descriptions. For the different generated images, the illumination has been modelled to closely agree with the average circadian illumination from April until September, corresponding to a midlatitude northern hemisphere growing season. The illumination has been composed of a combination of direct and diffuse light calculated from $350$ to $2500$nm with a $10$nm interval. The citrus trees and weeds of the orchard scenes (see paragraph \[sec:orchard\]) and the trees of the forest scene (see paragraph \[sec:forest\]) have been constructed as triangular meshes by implementing the algorithm introduced in [@Weber1995]. Their material properties have been described by a bidirectional scattering distribution function (BSDF) model [@Stuckens2009].
### Orchard scenes {#sec:orchard}
The fully calibrated virtual citrus orchard developed in [@Stuckens2009] has been used to create two different orchard scenes: (*i*) an orchard consisting of citrus trees and a soil background, leading to two-endmember mixtures and (*ii*) an orchard consisting of citrus trees, a soil background and weed patches, leading to three-endmember mixtures. Each orchard scene consists of $20\times20$ pixels, with a pixel size of $2\textrm{m}\times2\textrm{m}$. The exact per-pixel abundances are known for the three components, as well as the reference spectral signatures. More precisely, for the soil endmember, the pure spectral signature consisted of the fully sunlit soil uncontaminated by the surrounding trees. For the tree endmember, the soil background of the orchard was replaced by a perfectly absorbing background, to minimize the influence of the background on the tree signature. A $5$cm resolution image of $4$m by $4$m was rendered above a canopy with one row. Only the pixels containing a tree fraction greater than $0.95$ were retained and averaged to provide the pure tree signature. As such, the tree spectral signatures are an integration of all components of a tree, including sunlit and shaded leafs, branches and stems. For the weed endmember spectral signature, a similar approach to the tree endmember was used, replacing the soil background with a perfectly absorbing background, and removing all trees from the orchard. A $4$m by $4$m image with $1$cm resolution was rendered over a weed patch, selecting only those pixels with a weed fraction greater than $0.95$. Finally, these pixels were averaged to provide the spectral signature of the weeds. The resulting endmember spectra are depicted in Fig. \[fig:orchard\_endmembers\].
![Two- and three-endmember orchard synthetic dataset. Endmember spectra: soil (black), weed (red) and tree (green).[]{data-label="fig:orchard_endmembers"}](endmembers_orchard.pdf){width="0.75\columnwidth"}
The orchards have been constructed with a row spacing of $4.5$m, tree spacing of $2$m, row azimuth of $7.3^{\circ}$ and an average tree height of $3$m. This composition is consistent with the reference orchard, located in Wellington, South Africa ($33.58^{\circ}$S, $18.93^{\circ}$E), used to calibrate the virtual orchard [@Stuckens2009]. Spectral input data for citrus leaves and stems, soils and weeds have been measured using a full-range ($350$-$2500$nm) analytic spectral devices (ASD) Fieldspec JR spectroradiometer with a $25^{\circ}$ foreoptic. The weed spectrum has been chosen as of the Lolium sp. species. A Haplic Arenosol [@FAO1998] typical for commercial citrus orchards in the Western Cape Province in South Africa has been used in the simulations [@Somers2010a]. An example of a high resolution image of $20\textrm{m} \times20\textrm{m}$ of the two-endmember orchard is depicted in Fig \[fig:orchardExample\] (a), while the three-endmember orchard is shown in Fig \[fig:orchardExample\] (b). For a detailed description of the design, modalities and application of the virtual orchard, the reader is invited to consult [@Stuckens2009].
[0.49]{} ![High resolution images of the two orchards with (a) two endmembers, i.e., tree and soil, and (b) three endmembers, i.e., tree, soil and weeds. []{data-label="fig:orchardExample"}](Orchard_noweeds "fig:"){width="\columnwidth"}
[0.49]{} ![High resolution images of the two orchards with (a) two endmembers, i.e., tree and soil, and (b) three endmembers, i.e., tree, soil and weeds. []{data-label="fig:orchardExample"}](Orchard_weeds "fig:"){width="\columnwidth"}
### Forest scene {#sec:forest}
The virtual forest consisted of a soil background planted with trees selected from the species-specific tree pools. More precisely, to simulate the forest scene, $3$D tree geometry descriptions were available for beech (*Fagus sylvatica* L.) and poplar (*Populus nigra* L. var. *“italica”* Muench). Each tree was characterized by a specific structure based on its age (i.e., $20$ years old). All leaves were assigned a species-specific reflectance and transmittance spectrum extracted from the leaf optical properties experiment (LOPEX) dataset [@Hosgood1995]. Examples of the soil, beech and pop endmember signatures are depicted in Fig \[fig:forest\_endmembers\].
![Forest synthetic dataset. Example of the generated endmember spectra: soil (black), beech (red) and pop (green).[]{data-label="fig:forest_endmembers"}](endmembers_forest_one_instance.pdf){width="0.75\columnwidth"}
To achieve a nearly $100\%$ canopy cover, the average tree spacing has been set to $5$m for the beech trees and $1$m for the poplars. A series of six forest scenes has been rendered providing a gradual transition from a forest scene completely dominated by one species to a scene dominated by the other species. More precisely, $20\%$ of the beech trees have been randomly replaced by poplar trees in the subsequent scene. Each forest scene consisted of $15\times15$ pixels, with a pixel size of $30 \textrm{m} \times 30 \textrm{m}$. In Fig \[fig:forestExample\], a detail is shown of a $30$m pixel, for the forest consisting of 60$\%$ beech trees and 40$\%$ poplars. Note here that the spatial resolution of the forest scene is significantly larger than the resolution of the orchard scene detailed in paragraph \[sec:orchard\]. These choices allow different plant production systems to be covered, with various species combinations, sets of endmembers and spatial resolution scales.
![High resolution detail of a $30$m pixel of the forest with 60$\%$ beech trees and 40$\%$ poplars.[]{data-label="fig:forestExample"}](forest_detail.\figextension){width="0.75\columnwidth"}
In-situ measurement {#sec:in situ}
-------------------
In addition, an experiment was conducted in the same orchard used for the calibration of the virtual orchard described in paragraph \[sec:orchard\]. Significant weed cover, dominantly *Lolium sp. L.* ($\approx30\%$ of the inter-row spacing, concentrated in dense patches) was present. Throughout the orchard, in-situ measured reflectance spectra of $60$ mixed ground plots were collected, i.e., $25$ mixtures of tree and soil, $25$ mixtures of tree and weed, and $25$ mixtures of tree, soil and weed. Reflectance measurements were performed in August using a spectroradiometer with a $25^{\circ}$ fore-optic, covering the $350-2500$nm spectral domain (Analytic Spectral Devices, Boulder, USA). The measurements were taken from nadir at a height of $4$m. For each measured mixed pixel, the plot-specific pure endmember spectra and ground cover fraction distributions were determined. Specifically, to mitigate the impact of nonlinear mixing from endmember variability, plot-specific endmembers were acquired by measuring a number of pure spectra in each plot, as illustrated in Fig. \[fig:setup\]. One set of soil, weed and tree endmember spectra is depicted in Fig. \[fig:insitu\_endmembers\].
![Experimental set-up to determine plot-specific soil and tree endmember signatures for each plot (from [@Somers2009]). The areas T1, T2 and T3 (S1, S2, and S3, resp.) identify the sub-plots selected for the measurements of pure tree (soil, resp.) spectra. These measurements are averaged to provide the plot-specific tree (soil, resp.) endmember signature.[]{data-label="fig:setup"}](fig_setup_new_small.\figextension){width="0.75\columnwidth"}
![Two- and three-endmember in-situ measurements. Example of the measured endmember spectra: soil (black), weed (red) and tree (green).[]{data-label="fig:insitu_endmembers"}](endmembers_insitu_one_instance.pdf){width="0.75\columnwidth"}
Information on the ground cover composition of each of the measured mixed pixels was extracted from digital photographs (SONY DSC-P8/3.2 megapixel cyber shot camera, positioned in nadir). A more detailed description on the experimental setup, depicted in Fig. \[fig:setup\], can be found in [@Somers2009].
Experimental results
====================
The relevance of the mixing models under test, namely LMM, FM, NM, GBM and PPNM, and associated unmixing algorithms, is evaluated with respect to i) their ability of accurately describing the physical processes yielding the considered mixtures and ii) their ability of providing meaningful estimations of the abundance coefficients, to properly account for the spatial distribution of the materials over each observed pixel. More precisely, let $\hatVabond{\nopix}$ and $\hatVnabond{\nopix}$ denote the abundance and nonlinearity coefficient vectors estimated by the algorithms introduced in paragraph \[subsec:algorithm\]. First, the average square reconstruction error (RE) is measured as $$\label{eq:RE}
\mathrm{RE} = \frac{1}{\nbband\nbpix} \sum_{\nopix=1}^{\nbpix} \left\|\Vpix{\nopix}-\hatVpix{\nopix}\right\|^2$$ where $\left\|\cdot\right\|$ stands for the usual Euclidean norm ($\left\|\bfx\right\| = \sqrt{\bfx^T\bfx}$). In the right-hand side of , $\Vpix{\nopix}$ ($\nopix=1,\ldots,\nbpix$) are the observed pixels whereas $\hatVpix{\nopix}$ are the corresponding estimates given by $$\hatVpix{\nopix} = \MATmat\hatVabond{\nopix} + \boldsymbol{\mu}\left(\MATmat,\hatVabond{\nopix},\hatVnabond{\nopix}\right)$$ where $\boldsymbol{\mu}\left(\cdot\right)$ is equal to $0$ for the LMM or stands for the additional nonlinear contribution for the nonlinear models (see Section \[sec:models\]).
Since the actual endmember spectra and abundance coefficients (that satisfy the constraints in ) are perfectly known for each pixel of the considered scenes, these REs can also be computed from pixels reconstructed following the LMM and FM with the actual values of the abundances. These two “oracle” models are denoted o-LMM and o-FM in what follows. In particular, the RE associated with the o-LMM provides interesting information regarding the actual level of nonlinearities in the considered pixels. Note also that such oracle performance can not be computed for the other nonlinear models, since NM is based on a different abundance definition (e.g., they do not follow the constraints ) and GBM and PPNM require the prior knowledge of additional (unknown) parameters.
Moreover, to visualize the reconstruction error as a function of the wavelength, a signed error, defined as the mean reconstruction difference in the $\ell$th band, is also computed as $$\mathrm{RD}_{\ell} = \frac{1}{\nbpix} \sum_{\nopix=1}^{\nbpix}
\left(\pix{\ell}{\nopix}-\hatpix{\ell}{\nopix}\right).$$ Finally, to measure the accuracy of the abundance estimation, the mean square errors (MSE) between the actual abundance vectors $\Vabond{\nopix}$ and the corresponding estimated $\hatVabond{\nopix}$ ($\nopix=1,\dots,\nbpix$) are computed as follows $$\mathrm{MSE} = \frac{1}{\nbmat\nbpix} \sum_{\nopix=1}^{\nbpix} \left\|\Vabond{\nopix}-\hatVabond{\nopix}\right\|^2.$$
\[sec:results\]
Simulated dataset
-----------------
### Virtual orchard {#subsubsec:results_orchard}
The unmixing results for the simulated orchard scenes are shown in Table \[tab:synth\_orchard\_RMSE\] in terms of MSE and RE. From these results, for both two- and three-endmembers, one can conclude that NM and LMM perform similarly in term of RE, while PPNM and FM provide the best results and, in particular, significantly better than LMM. It is interesting to note that, for the $2$-endmember mixtures, GBM does not provide smaller RE than LMM, as expected. Indeed, as highlighted in paragraph \[subsec:bilinear\], GBM reduces to LMM if $\gamma_{i,j,p}=0,\ \forall i,j$, which is supposed to confer to GBM more flexibility than LMM. This might indicate that the unmixing algorithm associated with GBM has not properly converged for this dataset. This point is discussed in more details in Section \[sec:discussion\]. Regarding the abundance MSE, NM and LMM provide similar errors for two-endmember mixtures and all nonlinear models perform better than LMM for three-endmember mixtures.
$2$ endm. $3$ endm.
---- ----------- --------------------------------------------------------- --------------------------------------------------------
\* LMM $7.70 $ $5.81 $
o-LMM $15.0 $ $10.40 $
FM ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{1.24}}} $ ${\mathbf{0.91}} $
o-FM $10.20 $ $7.66 $
NM $7.70 $ $5.81 $
GBM $10.13 $ $0.94 $
PPNM ${\mathbf{1.28}} $ ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.91}}} $
\* LMM ${\mathbf{0.96}} $ $3.17 $
FM ${1.13} $ ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{2.27}}} $
NM ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.92}}} $ ${\mathbf{2.44}} $
GBM $1.47 $ ${2.45} $
PPNM ${1.22} $ ${2.62} $
: Two- and three-endmember orchard synthetic dataset. Abundance MSE ($\times 10^{-2}$) and RE ($\times 10^{-4}$) for various linear/nonlinear mixing models.\[tab:synth\_orchard\_RMSE\]
In Fig. \[fig:synth\_orchard\_wavelength\], the RDs are depicted as functions of wavelength, for the different linear and nonlinear mixing models. From this figure, it appears that the nonlinearities occurring in spectral bands ranging from $1400$nm to $2500$nm are of high intensity (see the plot associated with the oracle-LMM, in black dashed line) but are rather well described by the various nonlinear models.
![Two- and three-endmember orchard synthetic dataset. Reconstruction difference $\textrm{RD}_{\ell}$ as a function of wavelength for various linear/nonlinear mixing models: LMM (black), oracle-LMM (black, dashed line), FM (blue), oracle-FM (blue, dashed line), NM (magenta), GBM (red) and PPNM (green).[]{data-label="fig:synth_orchard_wavelength"}](results_simulated_orchard_error_wavelength_new.pdf){width="0.75\columnwidth"}
### Virtual forest
For the simulated forest scenes, the unmixing results are reported in Table \[tab:synth\_forest\_RMSE\]. These results are computed for four scene compositions, with increasing proportions from $20\%$ to $80\%$ of beech trees with respect to poplars (see paragraph \[sec:forest\]). The first three images provided a sequence of images with increasing nonlinearity, as shown by the RE obtained with the oracle-LMM, ranging from $2.11$ to $5.36$ ($\times
10^{-4}$). The fourth image, composed of $80\%$ of poplars and $20\%$ of beech trees, seems to be subject to nonlinearities of lower intensity, since the oracle-LMM RE is $3.15 \times 10^{-4}$.
As with the previous dataset, NM together with PPNM provides the best model fit for all images, i.e. with lowest RE, and the best abundance estimates in terms of MSE. The abundance estimation performance of the different models is also decreasing with increasing nonlinear mixing effects in the images, even though the RE remained almost constant for NM and PPNM. FM performed poorly and LMM and GBM lead to similar results.
---- ------- -------------------------------------------------------- -------------------------------------------------------- -------------------------------------------------------- --------------------------------------------------------
$0.2$ $0.4$ $0.6$ $0.8$
$0.8$ $0.6$ $0.4$ $0.2$
\* LMM $ 0.92 $ $ 1.78 $ $ 1.84 $ $ 0.88 $
o-LMM $ 2.11 $ $ 4.37 $ $ 5.36 $ $ 3.15 $
FM $ 1.37 $ $ 3.33 $ $ 4.56 $ $ 3.39 $
o-FM $ 7.01 $ $ 20.54 $ $ 33.03 $ $ 24.56 $
NM $ {\mathbf{0.23}} $ $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.11}}} $ $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.10}}} $ $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.12}}} $
GBM $ 0.92 $ $ 1.78 $ $ 1.84 $ $ 0.88 $
PPNM $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.13}}} $ $ {\mathbf{0.15}} $ $ {\mathbf{0.14}} $ $ {\mathbf{0.12}} $
\* LMM $ 0.73 $ $ 2.44 $ $ 4.98 $ $ 3.18 $
FM $ 1.43 $ $ 5.16 $ $ 11.65 $ $ 14.95 $
NM $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.25}}} $ $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.58}}} $ $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.66}}} $ $ {\mathbf{0.45}} $
GBM $ 0.72 $ $ 2.45 $ $ 5.01 $ $ 3.22 $
PPNM $ {\mathbf{0.40}} $ $ {\mathbf{0.80}} $ $ {\mathbf{0.93}} $ $ {\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{0.62}}} $
---- ------- -------------------------------------------------------- -------------------------------------------------------- -------------------------------------------------------- --------------------------------------------------------
: Three-endmember forest synthetic dataset. Abundance MSE ($\times 10^{-2}$) and RE ($\times 10^{-4}$) for various linear/nonlinear mixing models. \[tab:synth\_forest\_RMSE\]
Fig. \[fig:synth\_forest\_wavelength\] shows the RDs as functions of wavelength. From the RD associated with the oracle-LMM, it clearly appears that the nonlinearity effects mostly occur in the spectral range $700\textrm{nm}-1400\textrm{nm}$, especially for the $20-80\%$ and $80-20\%$ scenes. All nonlinear mixing models provide good model fits, except the FM, as already shown by the REs reported in Table \[tab:synth\_forest\_RMSE\].
![Three-endmember forest synthetic dataset. Reconstruction difference $\textrm{RD}_{\ell}$ as a function of wavelength for various linear/nonlinear mixing models: LMM (black), oracle-LMM (black, dashed line), FM (blue), oracle-FM (blue, dashed line), NM (magenta), GBM (red) and PPNM (green).[]{data-label="fig:synth_forest_wavelength"}](results_simulated_error_wavelength_new.pdf){width="0.75\columnwidth"}
In-situ measurements
--------------------
Three types of in-situ measured mixed pixels were available to test the different mixing models, i.e., tree-weed, tree-soil and tree-soil-weed mixtures (see paragraph \[sec:in situ\]). In Table \[tab:RMSE\_real\], the reconstruction error of the mixed signal and the accuracy of the estimated abundances are depicted. From the RE associated with the oracle-LMM, it appears that most nonlinearities occur in the tree-soil mixtures. Once again, PPNM is the mixing model that reconstructs the mixed signatures the best, while FM performed worse than the LMM. For the abundance accuracy, MSE results are less homogeneous than those obtained with the various simulated datasets. Depending on the type of the mixture, GBM or PPNM are the best unmixing model, while FM gives the lowest abundance estimation accuracies.
---- ----------- ---------------------------------------------------- ---------------------------------------------------- ----------------------------------------------------
tree-weed tree-soil tree-soil-weed
\* LMM $16.4$ $27.1$ $6.80$
o-LMM $33.9$ $50.0$ $37.4$
FM $17.7$ $16.4$ $10.9$
o-FM $26.0$ $40.7$ $53.0$
NM $16.3$ $26.8$ $2.13$
GBM $15.9$ $15.2$ $6.71$
PPNM ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{3.07}}}$ ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{1.82}}}$ ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{1.21}}}$
\* LMM ${\mathbf{12.5}}$ $2.78$ $6.42$
FM $13.5$ $2.88$ $8.15$
NM ${12.6}$ ${\mathbf{2.71}}$ ${\mathbf{5.80}}$
GBM ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{12.2}}}$ ${2.86}$ $6.39$
PPNM $13.0$ ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{2.57}}}$ ${\textcolor[rgb]{0.00,0.00,1.00}{\mathbf{4.83}}}$
---- ----------- ---------------------------------------------------- ---------------------------------------------------- ----------------------------------------------------
: Two- and three-endmember in-situ measurements. Abundance MSE ($\times 10^{-2}$) and RE ($\times 10^{-4}$) for various linear/nonlinear mixing models.\[tab:RMSE\_real\]
The RDs obtained on the in-situ measurements are depicted in Fig. \[fig:in\_situ\_error\_wavelength\]. Similarly to the previous analyzed dataset, most of the nonlinear effects seem to occur in the $700\textrm{nm}-1400\textrm{nm}$ spectral range, while being very small in the visible range. From these plots, most of the mixing model appear not sufficiently accurate to capture the nonlinearities in the observed mixtures, except the PPNM.
![Two- and three-endmember in-situ measurements. Reconstruction difference $\textrm{RD}_{\ell}$ as a function of wavelength for various linear/nonlinear mixing models: LMM (black), oracle LMM (black, dashed line), FM (blue), oracle FM (blue, dashed line), NM (magenta), GBM (red) and PPNM (green).[]{data-label="fig:in_situ_error_wavelength"}](results_in_situ_error_wavelength.pdf){width="0.75\columnwidth"}
Discussion {#sec:discussion}
==========
The various datasets used during the experiments enable the assessment of the performance of different unmixing models, and the evaluation of the relevance of using nonlinear mixing models to properly describe mixtures observed in vegetated areas. As the exact per-pixel endmembers are known, the effects of endmember spectral variability can be strongly reduced. Consequently, the simulated or measured mixed pixels can be fully characterized by the abundances, and the influence of the nonlinear mixing effects on the unmixing accuracy could be evaluated. To qualitatively and quantitatively evaluate the mixing models and corresponding unmixing algorithms, general trends emerge from the results presented in Section \[sec:results\]. These findings are reported in what follows.
Quantifying the amount of nonlinearity with o-LMM
-------------------------------------------------
Since the endmember signatures as well as the abundance coefficients are perfectly known for each pixel of the considered scenes, the modeling error (i.e., the RE) obtained with the oracle-LMM could be considered as the mis-modeling introduced by nonlinear mixing effects. For all three data sets, a significant RE can be observed with the oracle-LMM, demonstrating the presence of nonlinear mixing effects, as already shown in [@Borel1994; @Somers2009; @Tits2012igarss], for example. In particular, the results reported in Table \[tab:RMSE\_real\] show that the in situ-measurements are submitted to highly nonlinear effects. Conversely, from Table \[tab:synth\_forest\_RMSE\], the forest synthetic dataset seems to be less subjected to these nonlinear effects. Overall, from the results reported in the previous section, the mixed pixel signatures seem to be better represented by nonlinear mixing models, and specifically PPNM and NM. However, all nonlinear mixing models can not be advocated to better describe mixed pixels than LMM, such as the GBM and NM for the simulated orchard data (see Table \[tab:synth\_orchard\_RMSE\]), and the FM for the simulated forest data (see Table \[tab:synth\_forest\_RMSE\]) and the in-situ orchard data (see Table \[tab:RMSE\_real\]). This shows that these nonlinear mixing models do not necessarily better represent the mixed signatures.
On the use of reconstruction error to assess a mixing model
-----------------------------------------------------------
It is also important to note that a better modeling of the mixed pixels does not necessarily result in a better estimation of the abundances. For instance, PPNM, which has been shown to be the most accurate to model nonlinearly mixed spectral signatures, sometimes lead to less accuracy with respect to the abundance estimation when compared to LMM, in particular for the three-endmember mixtures in the simulated orchard data (see Table \[tab:synth\_orchard\_RMSE\]) and for the tree-weed mixtures in the in-situ data (see Table \[tab:RMSE\_real\]). In the results of the simulated forest, the same trend can be observed: in spite of increasing nonlinear mixing effects, the REs remain almost constant for both the PPNM and the NM, while the accuracy of the estimated abundances decreases (see Table \[tab:synth\_forest\_RMSE\]). As a consequence, the model fitting error, widely used in the remote sensing literature to monitor the performance of the unmixing algorithm, can not be used as the unique figure-of-merit to evaluate the relevance of a given mixing model.
Mis-modeling with respect to wavelength
---------------------------------------
All nonlinear mixing models considered in Section \[sec:models\] and used in the experiments reported in Section \[sec:results\] implicitly assume the same amount of nonlinearity for each wavelength of the spectral domain. Indeed, they are basically defined by cross-products between the endmember spectra, without introducing any weighting functions that would depend on the spectral bands. However, from the RDs depicted in Fig.’s \[fig:synth\_orchard\_wavelength\], \[fig:synth\_forest\_wavelength\] and \[fig:in\_situ\_error\_wavelength\], it clearly appears that the mis-modeling is drastically subjected to the influence of the wavelength. This corroborates the results of Somers *et al.* who also noticed similar behavior for the bilinear mixing model [@Somers2014]. Most of the nonlinear models under test lead to reconstructed mixtures with the same admissible accuracy as the LMM in the visible range ($400\textrm{nm}-700\textrm{nm}$). Conversely, a clear degradation of the modeling performance can be observed in the $700\textrm{nm}-1400\textrm{nm}$ spectral range for most linear and nonlinear models, except for the PPNM. In particular, the RDs associated with the oracle-LMM demonstrate the important level of nonlinearity in the near-infrared region. This finding has been widely observed in the literature [@Huete1985; @Goel1988; @Jacquemoud2000].
Dealing with the unmixing algorithm intrinsic limitations
---------------------------------------------------------
For both LMM and FM models, oracle measures of performance have been computed since these models are fully described by the a priori known abundance coefficients, explicitly considered as the spatial distributions of the materials over the imaged pixels. However, for the other nonlinear mixing models, unmixing algorithms need to be used to infer all the parameters involved in the model specification (e.g., abundances and nonlinearity parameters). Unfortunately, the optimization problems to be solved, formulated in and , to recover the abundance coefficients are not totally straightforward, mainly due to the constraints and/or the nonlinearity. As a consequence, the reliability of the obtained results, in terms of RE and abundance MSE, should be carefully analyzed, indeed mitigated. More precisely, part of the REs may consist of approximation errors induced by the unmixing algorithms themselves, in particular when these iterative algorithms converge toward a stationary point which is not the global minimizer of the objective function. Consequently, the abundance estimates may be biased since subjected to these approximation errors. As a manifest example, one can consider the fitting performance of the GBM. By definition, this model generalizes both LMM and FM and, thus, should provide at least similar RE to the lowest RE among those obtained with LMM and FM. However, this is not the case for the orchard synthetic dataset, as already highlight in paragraph \[subsubsec:results\_orchard\] (see Table \[tab:synth\_orchard\_RMSE\]). This is an archetypal instance of the limitations of the GBM-based unmixing algorithm.
Conclusion {#sec:conclusion}
==========
This paper attempted to make a first step toward a full quantitative assessment of linear and nonlinear mixing models to properly described mixtures observed in hyperspectral images acquired over vegetated areas. The conducted work exploited two kinds of hyperspectral data, whose main advantages lies in the availability of ground truth, that consists of the actual material signatures (endmember spectra) and their corresponding spatial repartitions in the pixels (abundance coefficients). The first set of hyperspectral data consisted of physically-based simulated images, while the second set of hyperspectral data came from real in-situ measurements. Various linear and nonlinear mixing models were used to analyze these data. They were evaluated in terms of spectral mis-modeling (i.e., reconstruction error) and abundance estimation accuracy. From the obtained results, it clearly appeared that the polynomial post-nonlinear mixing model undeniably provided, by far, the best reconstruction of the mixed pixels. It also persistently led to admissible abundance estimates, regardless of the considered scene. More generally, depending of the analyzed mixtures, the Nascimento model, the Fan model or the polynomial post-nonlinear model provided the most interesting results with respect to the abundance estimates. However, it was worth noting that the results presented in this work needed to be mitigated by the intrinsic limitations of the resorted unmixing algorithms, that could induce estimate biases. Finally, it is important to admit that the results reported in this work are only valid for $2$- and $3$-endmember mixtures. Generalizing or extending these findings to more complex scenes would require further investigation.
[^1]: Part of this work has been funded by the Hypanema ANR Project n$^\circ$ANR-12-BS03-003 and the Research Foundation Flanders (FWO) project G.0677.13N.
[^2]: It is widely admitted that the pixel spectrum measured by the sensor can be accurately described by the LMM when *(i)* the photons are not subjected to multipath effects and *(ii)* the materials are arranged side-by-side in the scene (as a checkerboard structure) [@Keshava2002].
[^3]: They will be available online at <http://www.biw.kuleuven.be/m3-biores/geomatics/data/>.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We fit the energy distribution of the IceCube starting events by a model which involves four parameters in the neutrino spectrum, namely three normalizations $n_e,n_\mu,n_\tau$ and a common power-law index $\gamma$, with a fixed background simulated by IceCube. It is found that the best fit index is $\gamma = 2.7$ with $\chi^2_{\rm min} =
32.3/24\,{\rm dof}$. As for the two parameter model involving a democratic normalization and an index, the best fit is at $\gamma = 2.8$ with $\chi^2_{\rm min}
= 33.9/26\,{\rm dof}$. The flavored model and the democratic model do not have much difference in the quality of the (energy-spectrum) fit. The standard $1:1:1$ composition is not disfavored by the current data.
---
=19.5pt
[MISC-2014-08]{}
[**The spectrum and flavor composition of the astrophysical neutrinos in IceCube** ]{}
Atsushi Watanabe
\
[(December, 2014)]{}
Introduction {#intro}
============
IceCube has recently made a great success in the observation of the high-energy neutrinos of extraterrestrial origin. In their three years of data, 37 events have been found in $30$ TeV–$2$ PeV energy range [@IC1; @IC2; @IC3]. They have concluded that the hypothesis of the atmospheric neutrino origin is rejected at 5.7$\sigma$, heralding a new era of high-energy astronomy. The analysis with a lowered threshold down to $1$ TeV also shows a significant contribution from the astrophysical component [@IC4]. Neutrino sky which will be seen by the existing and the future neutrino telescopes will provide indispensable information to understand the origin of cosmic rays, physics of Gamma Ray Bursts, GZK processes, etc. Since the first announcement of the two PeV cascades, many authors have speculated about the sources of the observed high-energy neutrinos [@astro].
While the high-energy neutrinos are unique astronomical messengers, they may also play an interesting role in particle physics. Neutrino decay [@decay], pseudo-Dirac neutrinos [@pDirac], and Lorentz/CPT violation [@CPT] have been discussed for long time as new physics testable by high-energy neutrinos. More recently, the isolated nature of the events around $1$ PeV [@IC3] have triggered a variety of intriguing ideas such as decay of long-lived particles [@PeVDM], exotic mediators for neutrino absorbtions [@abs], new physics in the detection processes [@detection]. Obviously more data is needed to deal with such diverse hypotheses and speculations.
Flavor ratios for the three types of neutrinos are one of the key information for making further progress in these subjects [@FR1]. One of the benchmark ratios for the source fluxes is $\Phi^0_e:\Phi^0_\mu:\Phi^0_\tau = 1:2:0$. Lepton mixing changes this ratio to $\simeq 1:1:1$ at Earth [@FR2]. Depending on astrophysical processes at the sources or new physics involved in the production, propagation and detection, the democratic composition at Earth may be significantly changed [@FR1; @FR2].
In this paper, we study the flavor composition of high-energy neutrinos by using the three years data of IceCube [@IC3]. Making the normalizations of power-law fluxes be flavor dependent, we fit the data and report the best fit and the intervals for the normalizations.
This issue was first addressed in Ref. [@Palomares-Ruiz], where they found that the $1:1:1$ composition at Earth with $E_\nu^{-2}$ spectrum is disfavored at $92\%$ CL with the best fit composition $1:0:0$. They analyzed the total number of the shower and the track events which are integrated over the deposited energies. A goal of this paper is to study the impact of the energy distributions on the determination of the flavor ratios. We model the astrophysical neutrino fluxes for each flavor $\Phi_\alpha\,(\alpha =
e,\mu,\tau)$ as $\Phi_\alpha \,=\, n_\alpha E_\nu^{-\gamma}$, where $n_\alpha$ is the (flavor dependent) normalization, $E_\nu$ is the neutrino energy, and $\gamma$ is the spectral index. The model parameters to be determined are $n_\alpha$ and $\gamma$. By seeking the global minimum of a $\chi^2$ function (see Section 3) with respect to these four parameters, we study the interplay between the flavor ratios and the spectral index. Our emphasis is, however, not on the numbers themselves given by the analysis, but on the qualitative differences between the flavored model and the usual democratic model, which may be highlighted by taking account of the energy distribution.
This paper is organized as follows. In Section \[events\], the calculations for the number of events by the astrophysical neutrinos are demonstrated. In Section \[flavor\], we discuss the method of the statistical analysis and show the results. Section \[summary\] is for conclusions.
Number of events {#events}
================
Astrophysical neutrino events
-----------------------------
Following Ref. [@IC2; @IC3], we focus on the neutrino events whose vertices are contained in the detector volume (so-called “starting events”). The neutrinos leave their signals via neutrino–nucleon ($\nu N$) scattering. There are two main topologies of the neutrino events; the showers and the tracks. The electron neutrinos $\nu_e$ trigger the shower events by the charged current (CC) and the neutral current (NC) interactions. The muon neutrinos $\nu_\mu$ produce both tracks and showers by the CC and NC interactions, respectively. The tau neutrinos $\nu_\tau$ with the energies less than $\sim 1$ PeV produce showers by CC and NC, whereas $\nu_\tau$ with energies greater than $1$ PeV may produce distinct events called double-bang and lolipop [@tau]. In this paper, we assume $\nu_\tau$ triggers only showers since we focus on the neutrino events whose energies are less than a few PeV[^1].
Let us first focus on the down-going events where the attenuation by Earth is irrelevant. The number of the shower events by the CC interactions of $\nu_e N$ and $\nu_\tau N$ are given by $$\begin{aligned}
\nu^{\rm sh}_{\rm CC} \,=\,
2\pi T N_A \!\int\!\! dE_\nu \,\, V_{\rm CC}^{e,\tau}\,\,
\sigma_{\rm CC}\,\,\Phi_{e,\tau},
\label{rate1}\end{aligned}$$ where $T=988$ days of exposure time, $N_A = 6.022\times 10^{23}\,{\rm g^{-1}}$, $E_\nu$ is the neutrino energy, $V_{\rm CC}^{e}$ and $V_{\rm CC}^{\tau}$ are the effective masses of the detector [@IC2], $\sigma_{CC}$ is the $\nu N$ total cross section for the CC interactions [@cross], $\Phi_{e}$ and $\Phi_{\tau}$ stand for the $\nu_e$ and $\nu_\tau$ fluxes, respectively. The factor $2\pi$ accounts for the integration over Southern sky under the assumption that the neutrino fluxes are isotropic. In the CC channel of $\nu_e N$ and $\nu_\tau N$, almost all neutrino energy is converted to the electromagnetic deposited energy ($E_{\rm em}$). In what follows, we assume $E_\nu = E_{\rm em}$ for these CC processes.
The number of the shower events by the NC interactions of $\nu_\alpha N$ ($\alpha =
e,\mu,\tau$) are given by $$\begin{aligned}
\nu^{\rm sh}_{\rm NC} =
2\pi TN_A \!\int_{E_0/\langle y \rangle}^{E_1/\langle y \rangle}\!\! dE_\nu \,\,
V_{\rm NC} \,\,
\sigma_{\rm NC}\,\,\Phi_{\alpha},
\label{rate2}\end{aligned}$$ where $V_{\rm NC}$ is the effective mass for the NC processes [@IC2], $\sigma_{\rm NC}$ is the $\nu N$ total cross section for the NC interactions [@cross], and $\langle y \rangle$ is the mean inelasticity, which is the mean energy fraction carried by the kicked quark in the final state [@cross]. The formula Eq. (\[rate2\]) shows the number of events for the shower energy between $E_0$ and $E_1$.
Finally, the number of the track events by the CC interactions of $\nu_\mu N$ is given by $$\begin{aligned}
\nu^{\rm tr} =
2\pi T N_A \!\int_{E_0/\langle y \rangle}^{E_1/\langle y \rangle}\!\!
dE_\nu \,\, V_{\rm CC}^{\mu}\,\,
\sigma_{\rm CC}\,\,\Phi_{\mu},
\label{rate3}\end{aligned}$$ where $V_{\rm CC}^{\mu}$ is the effective mass for the $\nu_\mu$ CC process [@IC2]. The out-going muons produced inside the instrumental volume usually escape from the volume, such that the showers at the starting vertices dominantly contribute to the deposited energies. In this work, we assume the deposited energies are equal to the starting shower energies, and use Eq. (\[rate3\]) for the track events whose deposited energies between $E_0$ and $E_1$.
For the up-going events (the events induced by the neutrinos coming from Northern sky), the events are calculated by Eq. (\[rate1\]), (\[rate2\]) and (\[rate3\]) with the replacement $(2\pi) \to (2\pi)S(E_\nu)$, where $S(E_\nu)$ is the shadow factor [@cross] varying from zero to unity, which accounts for the attenuation of the neutrinos by Earth. The calculations for the antineutrino are done by replacing the cross-sections which are slightly different from the ordinary ones [@cross].
Astrophysical neutrino fluxes
-----------------------------
In this work, we consider isotropic diffuse fluxes for the astrophysical neutrinos. In order to make the model be sensitive to the neutrino flavors in a simple way, let the normalization of the astrophysical neutrino flux for each flavor be independent, while assuming the spectra follow a common power law; $$\begin{aligned}
\Phi_\alpha \,=\, n_\alpha E_\nu^{-\gamma}, \quad\quad(\alpha = e,\mu,\tau).
\label{flux1}\end{aligned}$$ Fig. \[fig1\] shows typical examples of the deposited energy distributions of the events. The solid line shows the summation of the astrophysical neutrino events calculated by Eqs.(\[rate1\])–(\[flux1\]) and the background events (shown by the long-dashed line) simulated by the IceCube collaboration [@IC3]. The IceCube estimation of the total events is also shown by the short-dashed line for comparison. The black dots are the observed data. In accordance with Ref. [@IC3], the flux for each flavor is set as $n_\alpha = 0.95\times 10^{-8}\,{\rm GeV\, cm^{-1}\, s^{-1}\, sr^{-1}}$ for $\alpha = e,\mu,\tau$ with $\gamma = 2.0$, and the neutrino/antineutrino ($\nu/\bar{\nu}$) fraction is taken to be unity. It is seen from Fig. \[fig1\] that the estimation by Eqs.(\[rate1\])–(\[flux1\]) agrees well with the IceCube analysis, up to the large discrepancy at the bin for $E_{\rm em} = 10^{6.6}-10^{6.8}\,{\rm GeV}$. The shower and the track fraction of the astrophysical neutrino events are $82$% and $18$%, respectively. These numbers are also agree with Ref. [@IC3].
The large discrepancy around $E_{\rm em} = 10^{6.7}\,{\rm GeV}$ is due to the Glashow resonance [@GR1], which is the resonant production of the $W^-$ boson in $\overline{\nu}_e e$ scattering at $E_\nu = 6.3\,{\rm PeV}$. This effect is not included in Fig. \[fig1\]. The significance of this resonance strongly depends on the $\nu/\bar{\nu}$ fraction [@GR2], which would be a nuisance to the current purpose. Since no events larger than $\sim 2\,{\rm PeV}$ have been observed, we first avoid the uncertainty from the $\nu/\bar{\nu}$ fraction by assuming that the power-law fluxes have a cutoff at $E_\nu = 3.0\,{\rm PeV}$. In this case, the $\nu/\bar{\nu}$ fraction does not make much difference on the result of the following flavor analysis. The effect of the Glashow resonance on the fluxes without cutoff is discussed later (see Table \[tab2\] and the related text). In what follows, we set the $\nu/\bar{\nu}$ ratio to be unity as a typical example. Such a ratio is realized if the neutrinos are produced on source by the proton-proton scattering.
Flavor compositions {#flavor}
===================
We assume that the shower and track events are Poisson distributed around mean values $\mu^{\rm sh}$ and $\mu^{\rm tr}$. The observed data is fitted by minimizing the logarithm of the likelihood ratio of the current model to the saturated model [@PDG] $$\begin{aligned}
\chi^2 \,=\, \chi^2_{\rm shower} \,+\, \chi^2_{\rm track},
\label{chisq}\end{aligned}$$ $$\begin{aligned}
\chi^2_{\rm shower} \,=\, 2 \sum_{i=1}^{14}
\left(\, \mu^{\rm sh}_i - N^{\rm sh}_i + N^{\rm sh}_i\ln\frac{N^{\rm sh}_i}
{\mu^{\rm sh}_i}\,
\right),\end{aligned}$$ where $i$ labels the energy bins (see in Fig. \[fig1\]), $N^{\rm sh}$ is the observed shower events [@IC3]. The mean of the shower events $\mu^{\rm sh}$ is given by $$\begin{aligned}
\mu^{\rm sh} \,=\, \nu^{\rm sh} + b^{\rm sh},\end{aligned}$$ where $\nu^{\rm sh}$ stands for $\nu^{\rm sh}_{\rm CC} + \nu^{\rm sh}_{\rm NC}$ summed over the up and down-going, the neutrino and antineutrino components. $b^{\rm sh}$ is the background shower events. The symbols with the subscript $i$ stand for the values for the $i$-th bin. The function $\chi^2_{\rm track}$ is defined in the same manner as $\chi^2_{\rm shower}$.
For the background estimations of $b^{\rm sh}$ and $b^{\rm tr}$, we use the numbers in Ref. [@IC3]; the binned expected numbers for “atmospheric neutrino $(\pi/K)$” and “muon flux” shown in Fig. 2 of Ref. [@IC3]. In order to breakdown the atmospheric neutrino events into the showers and the tracks, we assume that the atmospheric neutrino events are solely induced by $\nu_\mu$ and its CC and NC reactions are identified as the tracks and the showers, respectively. This estimates that the track events account for $76\%$ of the atmospheric neutrino events in each energy bin[^2].
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
![Best fit and intervals of $n_e, n_\mu, n_\tau$ in the case of $E_\nu^{-2}$ spectrum ($\gamma = 2.0$). The three panels show the regions in the three dimensional $(n_e,n_\mu,n_\tau)$ space projected to the two dimensional planes. The symbol $\star$ stands for the best fit, and the inner (outer) region filled dark (light) is the 68% (95%) region.[]{data-label="abc"}](a.eps "fig:")
![Best fit and intervals of $n_e, n_\mu, n_\tau$ in the case of $E_\nu^{-2}$ spectrum ($\gamma = 2.0$). The three panels show the regions in the three dimensional $(n_e,n_\mu,n_\tau)$ space projected to the two dimensional planes. The symbol $\star$ stands for the best fit, and the inner (outer) region filled dark (light) is the 68% (95%) region.[]{data-label="abc"}](b.eps "fig:") ![Best fit and intervals of $n_e, n_\mu, n_\tau$ in the case of $E_\nu^{-2}$ spectrum ($\gamma = 2.0$). The three panels show the regions in the three dimensional $(n_e,n_\mu,n_\tau)$ space projected to the two dimensional planes. The symbol $\star$ stands for the best fit, and the inner (outer) region filled dark (light) is the 68% (95%) region.[]{data-label="abc"}](c.eps "fig:")
------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
For a fixed value of the spectral index $\gamma$, the best fit of $n_e$, $n_\mu$, $n_\tau$ is given by the minimum of Eq. (\[chisq\]). In addition to the best fit, we report the regions which satisfy $\chi^2< \chi^2_{\rm min} + 3.53\,(7.82)$ as approximate 68% (95%) confidence regions [@PDG].
Fig. \[abc\] shows the best fit and the intervals in the case of $E_\nu^{-2}$ spectrum ($\gamma = 2.0$). The three panels show the projections of the regions in the three dimensional $(n_e,n_\mu,n_\tau)$ space. The symbol $\star$ stands for the best fit, and the inner (outer) region filled in dark (light) colors is the 68% (95%) region. The best fit is $n_e =3.0\times 10^{-8}$, $n_\mu =3.9\times 10^{-9}$, $n_\tau = 0$ in the unit of ${\rm GeV \,cm^{-1}\, s^{-1}\, sr^{-1}}$ where $\chi^2_{\rm min} = 42.7/25\,{\rm dof}$. Starting from the minimum, the $\chi^2$ function is well increasing along the $n_e$ axis, whereas it sharply stands up only toward the increasing direction along the $n_\mu$ and $n_\tau$ axes. Although the best fit of $n_\mu$ is not zero, the increasing of $\chi^2$ is moderate along the decreasing $n_\mu$ direction.
The standard $\Phi_e : \Phi_\mu : \Phi_\tau = 1:1:1$ hypothesis is represented by the $n_e = n_\mu = n_\tau$ trajectory in the $(n_e, n_\mu, n_\tau)$ space. It is seen from Fig. \[abc\] that $1:1:1$ is lying on outside of the $68\%$ region. The minimum of $\chi^2$ along the $n_e = n_\mu = n_\tau$ trajectory is $\left. \chi^2_{\rm min}\right|_{n_e = n_\mu = n_\tau} = 46.9/27\,{\rm dof}$, which means the $n_e = n_\mu = n_\tau$ trajectory is tangent to the $76\%$ surface in the $(n_e, n_\mu, n_\tau)$ space. If we change the background assumption by replacing the track fraction $76\%$ with $69\%$($50\%$), $\left. \chi^2_{\rm min}\right|_{n_e = n_\mu = n_\tau}$ goes down from $46.9$ to $44.5$($39.0$).
![Minimum of $\chi^2$ (Eq. (\[chisq\])) for each value of the spectral index $\gamma$. The left panel shows the case where $n_e$, $n_\mu$ and $n_\tau$ are independent, while the right panel shows the case where the condition $n_e = n_\mu = n_\tau$ is imposed. The horizontal lines show the minimum $+$ $1$, $4$, and $9$, as the references for the $1$,$2$ and $3\sigma$ ranges of $\gamma$.[]{data-label="chigam"}](chisqgam.eps "fig:") ![Minimum of $\chi^2$ (Eq. (\[chisq\])) for each value of the spectral index $\gamma$. The left panel shows the case where $n_e$, $n_\mu$ and $n_\tau$ are independent, while the right panel shows the case where the condition $n_e = n_\mu = n_\tau$ is imposed. The horizontal lines show the minimum $+$ $1$, $4$, and $9$, as the references for the $1$,$2$ and $3\sigma$ ranges of $\gamma$.[]{data-label="chigam"}](chisqgam2.eps "fig:")
The minimum of the flavored model with $E_\nu^{-2}$ spectrum is $\chi^2_{\rm min} =
42.7/ 25\,{\rm dof}$, which means that this fit must be also poor. Better fits are obtained with the larger values of $\gamma$. Fig. \[chigam\] shows the minimum of $\chi^2$ (Eq. (\[chisq\])) for each value of the spectral index $\gamma$. The left panel shows the case where $n_e$, $n_\mu$ and $n_\tau$ are independent, while the right panel shows the case where the condition $n_e = n_\mu = n_\tau$ is imposed. The left panel tells us that the global minimum is away from $\gamma = 2.0$. The minimum is achieved at $\gamma = 2.7$, where $\chi^2_{\rm min} = 32.3/ 24\,{\rm dof}$, which is more acceptable than $\gamma = 2.0$. In the right panel of Fig. \[chigam\], $\chi^2_{\rm min} = 33.9/ 26\,{\rm dof}$ at $\gamma = 2.8$. When we omit the events below $\sim 60\,{\rm TeV}$ and perform a fit without the lower three bins, the best fit index becomes $\gamma = 2.3$ for the flavored model and $\gamma = 2.4$ for the democratic model, in agreement with Ref. [@IC3].
Fig. \[abc2\] shows the regions for the normalization constants in the case of $\gamma = 2.7$. In the plots, the normalization parameters are taken as $$\begin{aligned}
E_\nu^2 \Phi_\alpha = n_\alpha \left( \frac{E_\nu}{10^5\,{\rm GeV}} \right)^{-0.7},
\quad\quad (\alpha = e,\mu,\tau).\end{aligned}$$ The best fit is $n_e = 4.9\times 10^{-8}$, $n_\mu = 5.8\times 10^{-9}$, $n_\tau =0$ in the unit of ${\rm GeV \,cm^{-1}\, s^{-1}\, sr^{-1}}$. Compared with $\gamma = 2.0$ (Fig. \[abc\]), wider ranges are allowed for $\gamma = 2.7$. The $n_e = n_\mu = n_\tau$ trajectory is tangent to the $38\%$ surface, which means the $1:1:1$ ratio is consistent with the current observation.
----- ----------------- -------------- -------------- --------------
[68%]{} [95%]{} [68%]{} [95%]{}
$T$ $0$-$0.47$ $0$-$0.63$ $0$-$0.53$ $0$-$0.70$
$R$ $0.62$-$\infty$ $0$-$\infty$ $0$-$\infty$ $0$-$\infty$
----- ----------------- -------------- -------------- --------------
: Crude intervals for the flux ratios $T = \Phi_\mu/(\Phi_e + \Phi_\mu + \Phi_\tau)$ and $R = \Phi_e/\Phi_\tau$. In each column for $\gamma=2.0$ and $\gamma = 2.7$, the left (right) item shows the interval corresponding to the 68% (95%) region presented in Fig. \[abc\] and Fig. \[abc2\].[]{data-label="tab1"}
Note in passing that we are able to put the intervals on the flux ratios frequently quoted in literature. The two ratios $T \equiv \Phi_\mu/(\Phi_e + \Phi_\mu + \Phi_\tau)$ and $R \equiv \Phi_e/\Phi_\tau$ are often discussed [@FR2]. As a crude estimate of the confidence intervals, we show in Table. \[tab1\] the ranges of the functions $T$ and $R$ under the domain of the 68%(95%) regions of $(n_e, n_\mu, n_\tau)$ (the space defined by $\chi^2 \leq \chi^2_{\rm min} + 3.53\,(7.82)$). Notice that this does not take into account the cancellation of the uncertainties, so that the actual intervals may be narrower than shown here.
------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------
![ The same plots as in Fig. \[abc\], but for $E_\nu^{-2.7}$ spectrum ($\gamma = 2.7$). []{data-label="abc2"}](a2.eps "fig:")
![ The same plots as in Fig. \[abc\], but for $E_\nu^{-2.7}$ spectrum ($\gamma = 2.7$). []{data-label="abc2"}](b2.eps "fig:") ![ The same plots as in Fig. \[abc\], but for $E_\nu^{-2.7}$ spectrum ($\gamma = 2.7$). []{data-label="abc2"}](c2.eps "fig:")
------------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------
Finally, we comment on the effects of the Glashow resonance and the misidentification (mis-ID) of the track events. It is pointed out that 30% of the track events could be misidentified as showers [@ICf], and this effect has strong impacts on the determination of the flavor composition [@Palomares-Ruiz2]. In fact, we find that both of these effects have moderate impacts on the quality of the fit, but they significantly change the best fit of the flavor ratio and the exclusion level of $1:1:1$ in each model.
The results are summarized in Table \[tab2\]. As is expected, the inclusion of mis-ID allows lager fractions of $\nu_\mu$ and reduces the exclusion level of $1:1:1$. In the flavored model with the four parameters $(\gamma, n_\alpha)$ being floated, the best fit ratio $1:0.1:0$ is changed to $1:0.2:0$, and the exclusion limit 38% goes down to 12%. The best fit value of $\gamma$ is not changed by the mis-ID effects.
On the other hand, the effect of the Glashow resonance shifts the best fit of $\gamma$ to a slightly softer value, mitigating the conflict between the null observation and the enhanced event rate at the resonance bin. A striking change is the pusing-up of the $\nu_\tau$ component at the best fit. There are two reasons for this increasing of $\nu_\tau$. The first reason is the difference of the detection efficiencies of $\nu_e$ and $\nu_\tau$ at the lower energies. According to Ref. [@IC2], the effective volume of $\nu_e$ is as twice as large as $\nu_\tau$ around $40$-$100\,{\rm TeV}$. As the spectrum gets soft, the events less than $\sim 100\,{\rm TeV}$ get too large, so that $\nu_\tau$ is preferred for its lower detection rate than $\nu_e$. In fact, in the search of the best fit ratio, $\nu_\tau$ becomes dominant over $\nu_e$ for $\gamma \gtrsim 2.8$ in any flavored model. The second reason is that $\nu_\tau$ can account for the shower events while keeping the resonance event suppressed. This effect may slightly push down the value of $\gamma$ at which $\nu_\tau$ overcomes $\nu_e$. Since the inclusion of the Glashow resonance favors softer spectra, the best fit of the $\nu_\tau$ fraction accordingly increases due to the first reason mentioned above.
[cccc]{}Model & Best fit & $\chi^2_{\rm min}$ & Exclusion level of 1:1:1\
$(\gamma, n_\alpha)$-free (4P)&
----------------
$\gamma = 2.7$
$1:\,0.1\,:0$
----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 32.3/24 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 38%\
mis-ID &
----------------
$\gamma = 2.7$
$1:\,0.2\,:0$
----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 32.2/24 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 12%\
GR &
-----------------
$\gamma = 2.9$
$1:\,0.1\,:0.7$
-----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 32.9/24 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 27%\
mis-ID+GR &
-----------------
$\gamma = 2.8$
$1:\,0.4\,:0.7$
-----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 32.8/24 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 10%\
$\gamma=2.0,\, n_\alpha$-free (3P)&
--
--
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
---------------
$1:\,0.1\,:0$
---------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 42.7/25 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 76%\
mis-ID &
--
--
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
---------------
$1:\,0.4\,:0$
---------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 42.2/25 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 47%\
GR &
--
--
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
-----------------
$1:\,0.2\,:0.8$
-----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 54.1/25 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& 32%\
$(\gamma, n)$-free (2P)&
--
--
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
----------------
$\gamma = 2.8$
----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 33.9/26 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& -\
mis-ID &
--
--
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
----------------
$\gamma = 2.8$
----------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
&
------------------------------------------
$\chi^2_{\rm min} = 32.8/26 \,{\rm dof}$
------------------------------------------
: Summary of the best fit, $\chi^2_{\rm min}$, and the Feldman-Counsins exclusion level [@FC] of $1:1:1$. “$(\gamma, n_\alpha)$-free (4P)” stands for the model where the four parameters $\gamma, n_e, n_\mu, n_\tau$ are floated. “$\gamma=2.0, n_\alpha$-free (3P)” is the model where three parameters $n_e,n_\mu,n_\tau$ are free with the fixed index $\gamma = 2.0$. The third model “$(\gamma,n)$-free (2P)” is the case where $\gamma$ and the normalization $n=n_e=n_\mu=n_\tau$ are varied. The sub-items “mis-ID”, “GR”, and “mis-ID+GR” show the options that include the effect of the 30% misidentification of the tracks as showers, the Glashow Resonance without the energy cutoff, and the combination of both, respectively. In the column of “Best fit”, the ratios shows $1:n_\mu/n_e: n_\tau/n_e$ at the best fit values of the normalizations. []{data-label="tab2"}
& -\
Conclusions {#summary}
===========
The current data of the IceCube’s starting events seemingly shows a paucity of the muon events. Above $30\,{\rm TeV}$, just eight tracks have been observed against the background of $8.4\pm 4.2$ cosmic ray muon events and $6.6^{+5.9}_{-1.6}$ atmospheric neutrino events. If this tendency would hold, it suggests that the standard $1:1:1$ scenarios should be revised, and may even indicate the existence of some new physics.
In this work, we have studied the flavor composition of the astrophysical neutrinos observed in IceCube, especially focusing on the impact of the spectral index $\gamma$. Our point is not to give a precise estimation for the best fit and the intervals of the relevant parameters, but to illustrate important qualitative features in the flavor and the spectrum analysis of the astrophysical neutrinos. For this purpose, we consider the model with the three-flavor normalizations $n_e$,$n_\mu$,$n_\tau$ and a common index $\gamma$ kept independent (the flavored model), and compare it to the usual model with a common normalization and an index (the democratic model).
It is found that the global minimum of the flavored model is at $\gamma = 2.7$ with $\chi^2_{\rm min} = 32.3/24\,{\rm dof}$. As for the democratic model, the best fit is at $\gamma = 2.8$ with $\chi^2_{\rm min} = 33.9/26\,{\rm dof}$. The democratic model and the flavored model do not have much difference in the quality of the (energy-spectrum) fit. The standard $1:1:1$ composition is consistent with the current data.
However, the flavor compositon may affect the interval determination of $\gamma$. The left panel of Fig. \[chigam\] shows that the $\chi^2$ does not quickly stand up as $\gamma$ increases, indicating that the determination of $\gamma$ might be more challenging for the flavored model than for the democratic case. The current background model does not leave much room for the track contributions from the astrophysical neutrinos at lower energies. Thus the $1:1:1$ case gets trouble at the lower energy bins as $\gamma$ becomes large, whereas the flavored model can avoid the conflict by taking the configuration where the muon component is suppressed. The inference of the spectral index may become a nontrivial task once the flavor degress of freedom are switched on.
Acknowledgments {#acknowledgments .unnumbered}
---------------
I thank Werner Rodejohann for his contribution in the early stage of this work. I also thank Thomas Schwetz and Hiroaki Sugiyama for useful discussions on statistics.
[99]{} M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Phys. Rev. Lett. [**111**]{} (2013) 021103 \[arXiv:1304.5356 \[astro-ph.HE\]\]. M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Science [**342**]{} (2013) 1242856 \[arXiv:1311.5238 \[astro-ph.HE\]\]. M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], Phys. Rev. Lett. [**113**]{} (2014) 101101 \[arXiv:1405.5303 \[astro-ph.HE\]\]. M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], arXiv:1410.1749 \[astro-ph.HE\]. I. Cholis and D. Hooper, JCAP [**1306**]{} (2013) 030 \[arXiv:1211.1974 \[astro-ph.HE\]\]; O. E. Kalashev, A. Kusenko and W. Essey, Phys. Rev. Lett. [**111**]{} (2013) 4, 041103 \[arXiv:1303.0300 \[astro-ph.HE\]\]; D. B. Fox, K. Kashiyama and P. Me’szaro’s, Astrophys. J. [**774**]{} (2013) 74 \[arXiv:1305.6606 \[astro-ph.HE\]\]; F. W. Stecker, Phys. Rev. D [**88**]{} (2013) 4, 047301 \[arXiv:1305.7404 \[astro-ph.HE\]\]; K. Murase and K. Ioka, Phys. Rev. Lett. [**111**]{} (2013) 12, 121102 \[arXiv:1306.2274 \[astro-ph.HE\]\]; R. Laha, J. F. Beacom, B. Dasgupta, S. Horiuchi and K. Murase, Phys. Rev. D [**88**]{} (2013) 043009 \[arXiv:1306.2309 \[astro-ph.HE\]\]; H. Gao, B. Zhang, X. F. Wu and Z. G. Dai, Phys. Rev. D [**88**]{} (2013) 043010 \[arXiv:1306.3006 \[astro-ph.HE\]\]; K. Murase, M. Ahlers and B. C. Lacki, Phys. Rev. D [**88**]{} (2013) 12, 121301 \[arXiv:1306.3417 \[astro-ph.HE\]\]; L. A. Anchordoqui, H. Goldberg, M. H. Lynch, A. V. Olinto, T. C. Paul and T. J. Weiler, Phys. Rev. D [**89**]{} (2014) 8, 083003 \[arXiv:1306.5021 \[astro-ph.HE\]\]; S. Razzaque, Phys. Rev. D [**88**]{} (2013) 10, 103003 \[arXiv:1307.7596 \[astro-ph.HE\]\]; M. Ahlers and K. Murase, Phys. Rev. D [**90**]{}, no. 2, 023010 (2014) \[arXiv:1309.4077 \[astro-ph.HE\]\]; N. Fraija, Mon. Not. Roy. Astron. Soc. [**437**]{} (2014) 2187 \[arXiv:1310.7061 \[astro-ph.HE\]\]; A. M. Taylor, S. Gabici and F. Aharonian, Phys. Rev. D [**89**]{} (2014) 10, 103003 \[arXiv:1403.3206 \[astro-ph.HE\]\]; M. Ahlers and F. Halzen, Phys. Rev. D [**90**]{} (2014) 043005 \[arXiv:1406.2160 \[astro-ph.HE\]\]; J. Becker Tjus, B. Eichmann, F. Halzen, A. Kheirandish and S. M. Saba, Phys. Rev. D [**89**]{} (2014) 12, 123005 \[arXiv:1406.0506 \[astro-ph.HE\]\]; A. Bhattacharya, R. Enberg, M. H. Reno and I. Sarcevic, arXiv:1407.2985 \[astro-ph.HE\]; W. Winter, Phys. Rev. D [**90**]{} (2014) 10, 103003 \[arXiv:1407.7536 \[astro-ph.HE\]\]; C. Y. Chen, P. S. B. Dev and A. Soni, arXiv:1411.5658 \[hep-ph\].
J. F. Beacom, N. F. Bell, D. Hooper, S. Pakvasa and T. J. Weiler, Phys. Rev. Lett. [**90**]{} (2003) 181301 \[hep-ph/0211305\]. S. T. Petcov, Phys. Lett. B [**110**]{} (1982) 245; M. Kobayashi, C. S. Lim and M. M. Nojiri, Phys. Rev. Lett. [**67**]{} (1991) 1685. V. A. Kostelecky and M. Mewes, Phys. Rev. D [**69**]{} (2004) 016005 \[hep-ph/0309025\]; G. Barenboim and C. Quigg, Phys. Rev. D [**67**]{} (2003) 073024 \[hep-ph/0301220\].
B. Feldstein, A. Kusenko, S. Matsumoto and T. T. Yanagida, Phys. Rev. D [**88**]{} (2013) 1, 015004 \[arXiv:1303.7320 \[hep-ph\]\]; A. Esmaili and P. D. Serpico, JCAP [**1311**]{} (2013) 054 \[arXiv:1308.1105 \[hep-ph\]\]; T. Higaki, R. Kitano and R. Sato, JHEP [**1407**]{} (2014) 044 \[arXiv:1405.0013 \[hep-ph\]\]; A. Bhattacharya, R. Gandhi and A. Gupta, arXiv:1407.3280 \[hep-ph\]; Y. Ema, R. Jinno and T. Moroi, JHEP [**1410**]{} (2014) 150 \[arXiv:1408.1745 \[hep-ph\]\]; A. Esmaili, S. K. Kang and P. D. Serpico, arXiv:1410.5979 \[hep-ph\]; C. S. Fong, H. Minakata, B. Panes and R. Z. Funchal, arXiv:1411.5318 \[hep-ph\]; E. Dudas, Y. Mambrini and K. Olive, arXiv:1412.3459 \[hep-ph\]. K. Ioka and K. Murase, PTEP [**2014**]{} (2014) 6, 061E01 \[arXiv:1404.2279 \[astro-ph.HE\]\]; M. Ibe and K. Kaneta, Phys. Rev. D [**90**]{} (2014) 5, 053011 \[arXiv:1407.2848 \[hep-ph\]\]; K. Blum, A. Hook and K. Murase, arXiv:1408.3799 \[hep-ph\]; T. Araki, F. Kaneko, Y. Konishi, T. Ota, J. Sato and T. Shimomura, arXiv:1409.4180 \[hep-ph\]. A. N. Akay, O. Cakir, Y. O. Gunaydin, U. Kaya, M. Sahin and S. Sultansoy, arXiv:1409.5896 \[hep-ph\]; J. I. Illana, M. Masip and D. Meloni, arXiv:1410.3208 \[hep-ph\]; J. F. Beacom, N. F. Bell, D. Hooper, S. Pakvasa and T. J. Weiler, Phys. Rev. D [**68**]{} (2003) 093005 \[Erratum-ibid. D [**72**]{} (2005) 019901\] \[hep-ph/0307025\]. W. Rodejohann, JCAP [**0701**]{} (2007) 029 \[hep-ph/0612047\]; S. Pakvasa, W. Rodejohann and T. J. Weiler, JHEP [**0802**]{} (2008) 005 \[arXiv:0711.4517 \[hep-ph\]\]; S. Choubey and W. Rodejohann, Phys. Rev. D [**80**]{} (2009) 113006 \[arXiv:0909.1219 \[hep-ph\]\]; D. Meloni and T. Ohlsson, Phys. Rev. D [**86**]{} (2012) 067701 \[arXiv:1206.6886 \[hep-ph\]\]; F. Vissani, G. Pagliaroli and F. L. Villante, JCAP [**1309**]{} (2013) 017 \[arXiv:1306.0211 \[astro-ph.HE\]\]; X. J. Xu, H. J. He and W. Rodejohann, JCAP [**12**]{} (2014) 039 \[arXiv:1407.3736 \[hep-ph\]\]. O. Mena, S. Palomares-Ruiz and A. C. Vincent, Phys. Rev. Lett. [**113**]{} (2014) 9, 091103 \[arXiv:1404.0017 \[astro-ph.HE\]\]; S. Palomares-Ruiz, O. Mena and A. C. Vincent, arXiv:1411.2998 \[astro-ph.HE\]. J. G. Learned and S. Pakvasa, Astropart. Phys. [**3**]{} (1995) 267 \[hep-ph/9405296, hep-ph/9408296\]. R. Gandhi, C. Quigg, M. H. Reno and I. Sarcevic, Astropart. Phys. [**5**]{} (1996) 81 \[hep-ph/9512364\]; R. Gandhi, C. Quigg, M. H. Reno and I. Sarcevic, Phys. Rev. D [**58**]{} (1998) 093009 \[hep-ph/9807264\].
S. L. Glashow, Phys. Rev. [**118**]{} (1960) 316. L. A. Anchordoqui, H. Goldberg, F. Halzen and T. J. Weiler, Phys. Lett. B [**621**]{} (2005) 18 \[hep-ph/0410003\]; A. Bhattacharya, R. Gandhi, W. Rodejohann and A. Watanabe, JCAP [**1110**]{} (2011) 017 \[arXiv:1108.3163 \[astro-ph.HE\]\]; V. Barger, L. Fu, J. G. Learned, D. Marfatia, S. Pakvasa and T. J. Weiler, Phys. Rev. D [**90**]{} (2014) 121301 \[arXiv:1407.3255 \[astro-ph.HE\]\].
K. A. Olive [*et al.*]{} \[Particle Data Group Collaboration\], Chin. Phys. C [**38**]{} (2014) 090001. M. G. Aartsen [*et al.*]{} \[IceCube Collaboration\], arXiv:1502.03376 \[astro-ph.HE\]. S. Palomares-Ruiz, A. C. Vincent and O. Mena, arXiv:1502.02649 \[astro-ph.HE\]. G. J. Feldman and R. D. Cousins, Phys. Rev. D [**57**]{} (1998) 3873 \[physics/9711021 \[physics.data-an\]\].
[^1]: The taus produced from $\nu_\tau$-CC decay to muons in $17.4$% branching ratio, and such events are classified as tracks. The inclusion of this track events slightly changes the following results on the flavor composition. However, the best fit values of the spectral index $\gamma$ are not changed.
[^2]: A more realistic number given in Ref. [@IC3] is $69$%. If we use this number in the following analysis, the best fit values of $\gamma$ and $n_\alpha$ are accordingly changed by a few percent.
| {
"pile_set_name": "ArXiv"
} |
Frequent polarization reversals, or spin-flips, of a stored polarized beam in a high energy scattering asymmetry experiments may greatly reduce systematic errors of spin asymmetry measurements. A spin-flipping technique is being developed by using rf magnets running at a frequency close to the spin precession frequency, thereby creating spin-depolarizing resonances; the spin can then be flipped by ramping the rf magnet’s frequency through the resonance. We studied, at the Indiana University Cyclotron Facility Cooler Ring, properties of such rf depolarizing resonances in the presence of a nearly-full Siberian snake and their possible application for spin-flipping. By using an rf-solenoid magnet, we reached a 98.7$\pm$1$\%$ efficiency of spin-flipping. However, an rf-dipole magnet is more practical at high energies; hence, studies of spin-flipping by an rf-dipole are underway at IUCF. I would like to thank Professor Alan Krisch for his valuable advice, enouragement and support. I also thank my dissertation committee members their helpful comments and suggestions to this thesis. I thank my fellow collaborators from Michigan, Indiana, Protvino and Brookhaven for many hours they devoted to this experiment. I would especially like to thank Rick Phelps for teaching me a lot about how the experiment works.
I would like to thank the IUCF technical staff for their hard work on successful experiment operation. In particular, I want to thank the operators headed by Gary East with help from Terry Sloan for giving us beam when we needed it, the cryogenics experts Kevin Komisarcik and John Vanderwerp for keeping our snake cool, and the polarizaed ion source group headed by Vladimir Derenchuk for making polarized protons out of hydrogen gas.
Finally, I would like to thank my parents, my big brother and my little sister for their love and support, and my wife Svetlana for always being there when I needed her.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Thanks to the high detection quality of the INDRA array, signatures related to the dynamics (spinodal decomposition) and thermodynamics (negative microcanonical heat capacity) of a liquid-gas phase transition have been simultaneously studied in multifragmentation events in the Fermi energy domain. The correlation between both types of signals strongly supports the existence of a first order phase transition for hot nuclei.'
address:
- 'Institut de Physique Nucléaire, IN2P3-CNRS, F-91406 Orsay Cedex, France'
- 'LPC Caen, IN2P3-CNRS/ENSICAEN et Université, F-14050 Caen cedex, France'
- 'GANIL, DSM-CEA/IN2P3-CNRS, B.P. 5027, F-14021 Caen Cedex, France'
- 'Institut de Physique Nucléaire, IN2P3-CNRS et Université, Villeurbanne, France'
- ' National Institute for Physics and Nuclear Engineering, Bucharest-Măgurele, Romania'
- 'Dipartimento di Scienze Fisiche e Sezione INFN, Universitá di Napoli, Napoly, Italy'
author:
- 'B. Borderie, R. Bougault, P. Désesquelles, E. Galichet, B. Guiot [^1], Ph. Lautesse, N. Le Neindre, J. Marie, M. Pârlog, M. Pichon, M. F. Rivet, E. Rosato, G. Tăbăcaru [^2], M. Vigilante and J. P. Wieleczko'
title: 'Liquid-gas phase transition in hot nuclei studied with INDRA'
---
Introduction
============
The main goal of multifragmentation studies concerns its relation with subcritical and/or critical phenomena. Thus it is fully connected to the nature of the phase transition of nuclear matter which is expected of the liquid-gas type due to the specific form of the nucleon-nucleon interaction. Despite the large number of experimental results, fundamental questions could not be addressed for a long time. We are facing an exciting but also very complex subject where only very sophisticated detector arrays like INDRA [@I3-Pou95; @I5-Pou96] can produce the high quality experiments needed [@MDA00; @Hau00; @Elli00; @Xu00; @I46-Bor02; @MDA02; @Radu02; @Kle02; @Ell02]. From collisions between nuclei we hope to reveal a phase transition for those finite objects and derive the related equation of state of nuclear matter. In parallel to the experimental effort a big theoretical effort started a few years ago to characterize and propose signatures of phase transitions in finite systems [@Gro01; @Cho02; @Gul03]. Phase transitions should be reconsidered from a more general point of view, out of the thermodynamical limit. Thus, experiments on nuclear multifragmentation are also used as test-bench for the elaboration of the statistical physics of finite systems.
The paper describes some major recent results obtained with the INDRA array. It is organized as follows. After some information concerning the description of collisions and associated simulations, the dynamics of phase transition in hot nuclei and the related proposed signature will be presented. Then specific phase transition signals related to the thermodynamics of finite systems will be discussed in terms of robustness. Finally, from systematic studies, the coherence between signals will be demonstrated.
Description of collisions and multifragmentation
================================================
Semi-classical simulations of nucleus-nucleus collisions based on the nuclear Botzmann equation (VUU,LV,BUU,BNV) [@Kru85; @Gre87; @Ber88; @Bon94], were very successful in reproducing a variety of experimental dynamical observables. However as they follow the time evolution of the one body density, considering only the average effect of collisions between particles (two-body collisions), they ignore fluctuations about the main trajectory of the system (deterministic description), which becomes a severe drawback if one wants to describe processes involving instabilities, bifurcations or chaos. Such approaches are appropriate during the early stages of nuclear collisions, when the system is hot and compressed, but they become inadequate when expansion and cooling have brought the system in the instability (spinodal) zone of the phase diagram. In such a scenario, it is essential to include the fluctuations. This is done in Quantum Molecular Dynamics methods and in the Stochastic Mean Field (SMF) approach. We shall now focus on SMF approach, a natural evolution of deterministic semi-classical simulations, which was compared to INDRA data.
Brownian one-body dynamics (BOB)
--------------------------------
In BOB simulations, fluctuations in one-body dynamics are introduced by employing a Brownian force in the kinetic equations. The starting point is the so-called Boltzmann-Langevin equation (BLE): $$\frac{\partial f}{\partial t} = \{h[f],f\} + \overline{I}[f] + \delta I[f]$$ which was introduced for heavy-ion collisions in references [@Ayi88; @Ran90]. $f$ is the one-body phase space density. The first term on the r.h.s. produces the collisionless propagation of $f$ in the self-consistent one-body field described by the effective Hamiltonian. The second term (second source of evolution called collision term) represents the average effect of the residual Pauli-suppressed two-body collisions; this is the term included in LV,BUU and BNV simulations. The third term is the Langevin term which accounts for the fluctuating part of the two-body collisions. Exact numerical solutions of the BLE are very difficult to obtain and have only be calculated for schematic cases in one or two dimensions [@Cho91]. Therefore various approximate treatments of the BLE have been developed. The basic idea of BOB [@Cho94] is to replace the fluctuating term by $$\delta\tilde{I} [f] = - \delta \bi{F} [f] . \frac{\partial f}{\partial
\bi{p}}$$ where $\delta \bi{F} (\bi{r},t)$ is the associated Brownian force ($<\delta \bi{F}>=0$). Since the resulting Brownian one-body dynamics mimics the BL evolution, the stochastic force is assumed to be local in space and time. The strength of the force is adjusted to reproduce the growth of the most unstable modes for infinite nuclear matter in the unstable spinodal region (see next section).
Comparison with data
--------------------
The comparison was made for two very heavy fused systems produced in Xe+Sn and Gd+U central collisions which undergo multifragmentation with the same available excitation energy ( $\sim$ 7 MeV per nucleon). Stochastic mean-field simulations were performed for head-on collisions, thus neglecting shape effects. Ingredients of simulations are detailed in [@I29-Fra01]. After BOB simulation a second step, starting from the phase space configuration of the primary fragments as given by BOB, followed the fragment statistical deexcitation while preserving space-time correlations. Finally the events were filtered to account for the experimental device. These complete simulations very well reproduce the observed multiplicity and charge distributions of fragments [@I29-Fra01]. Particularly the independence of the charge distribution against the mass of the system experimentally observed is recovered [@I12-Riv98]. More detailed comparisons also show a good agreement [@I29-Fra01; @TabBol00]. Kinetic properties of fragments are rather well accounted for especially for the Gd+U system. For Xe+Sn, the calculated energies fall $\sim$ 20% below the measured values which remains satisfactory if one remembers that there were no adjustable parameters in the simulation.
What have we learnt ?
---------------------
From the good agreement between dynamical simulations and data a complete scenario is proposed [@I29-Fra01]. After a gentle compression ($\sim$1.25$\rho_{0}$), maximum at 40 fm/c, systems expand and enter the unstable (negative compressibility) spinodal region at around 80 fm/c. Slightly later ($\sim$ 100 fm/c) thermal equilibrium is achieved ($<T>$=4 MeV) at low density ($\sim$0.4$\rho_{0}$) and systems have lost $\sim$5% of their initial masses by emitting preequilibrium particles. The radial velocities at the surface ($\sim$0.1c) reveal the expansion of the systems and density fluctuations (spinodal instabilities) have time to develop leading to the formation of fragments. Around 250 fm/c after the beginning of collisions the fragments do not interact any more, they are in thermal equilibrium and still bear an average excitation energy of $\sim$3 MeV per nucleon: it is the freeze-out time with the associated freeze-out volume, it corresponds to the end of these BOB simulations.
Dynamics of the phase transition and related signal
===================================================
Thermodynamics describes phase transitions in terms of “static conditions”. Information on existence and coexistence of phases are derived depending on thermodynamical parameters (temperature, pressure,…). How to pass from a phase to another ? What is the time needed ? To answer these questions, dynamics of phase transitions must be studied. A phase transition is a dynamical phenomenon with its proper kinetics. The aim of this section is to discuss the dynamics of the phase transition which can be involved in hot unstable nuclei. Two mechanisms are candidates. (i) Nucleation which is associated with a transport of matter; diffusion is the key mechanism which controls the kinetics; we are in presence of heterogeneous fluctuations. (ii) Spinodal decomposition which occurs when a system spontaneously develops local fluctuations of concentration or density which grow exponentially with time; characteristic time is related to the most unstable modes. In contrast with nucleation, we are here dealing with homogeneous fluctuations. Given the agreement discussed in the previous section, we will now focus on spinodal decomposition.
Spinodal decomposition: from nuclear matter to nuclei
-----------------------------------------------------
In the spinodal region some modes are amplified because of the instability. Their associated characteristic times are almost identical, around 30- 50 fm/c, depending on density ($\rho_{0}/2$-$\rho_{0}/8)$ and temperature (0-9MeV) [@Colo97; @Idi94]. Within this scenario a breakup into nearly equal-sized “primitive” fragments should be favoured in relation to the wawelengths of the most unstable modes [@Ayi95]. However this simple picture is predicted to be strongly blurred by several effects: beating of different modes, coalescence of nascent fragments, secondary decay of excited fragments and, above all, finite size effects [@Jac96]. Therefore only a weak proportion of multifragmentation events with nearly equal-sized fragments is expected.
The related fossil signal: enhanced production of nearly equal-sized fragments
------------------------------------------------------------------------------
A few years ago a new method called higher order charge correlations [@Mor96], which has been recently improved [@Des02], was proposed to enlighten any extra production of events with specific fragment partitions. The high sensitivity of the method makes it particularly appropriate to look for small numbers of events as those expected to have kept a memory of spinodal decomposition properties. Thus, such a charge correlation method allows to examine model independent signatures that would indicate a preferred decay into a number of equal-sized fragments in events from experimental data or from simulations. All fragments ($Z \geq 5$) of one event with fragment multiplicity $M = \sum_Z n_Z$, where $n_Z$ is the number of fragments with charge $Z$ in the partition, are taken into account. By means of the normalized first and second order moments of the fragment charge distribution in the event: $${\langle Z \rangle}= \frac{1}{M} \sum_Z n_Z Z \; , \; \; \;
\sigma_Z^2 = \frac{1}{M} \sum_Z n_Z (Z - {\langle Z \rangle})^2
\label{equ1}$$ one may define the higher order charge correlation function: $$\left. 1+R(\sigma_Z, {\langle Z \rangle})=\frac{Y(\sigma_Z, {\langle Z \rangle})}{Y'(\sigma_Z, {\langle Z \rangle})}
\right| _{M}
\label{equ3}$$ Here, the numerator $Y(\sigma_Z, {\langle Z \rangle})$ is the yield of events with given ${\langle Z \rangle}$ and $\sigma_Z$ values and the evaluation of the denominator (uncorrelated yield) is deduced from the exact multinomial formula under total charge conservation constraint (see ref. [@Des02; @I40-Tab02] for details).
We shall now discuss the results obtained for fused systems produced in $^{129}$Xe central collisions on $^{nat}$Sn at four incident energies (32, 39, 45 and 50 AMeV) [@I40-Tab02]. At 32 AMeV the present analysis fully confirms the previous one obtained with the original method [@I31-Bor01] and the extra-percentage of events with nearly equal-sized fragments is maximum at 39 AMeV incident energy. The excitation function is displayed in fig. \[fig8\]. Information on the associated thermal excitation energies (and extra radial collective energy) involved over the incident energy domain studied can be provided by the SMM model [@Bon95] which well describes static and dynamic observables of fragments. Starting from a freeze-out volume fixed at three times the normal volume, the thermal excitation energies of the dilute and homogeneous system, extracted from SMM, vary from 5.0 to 7.0 AMeV and the added radial expansion energy remains low: from 0.5 to 2.2 AMeV [@T16Sal97; @T25NLN99]. A rise and fall of the percentage of “fossil partitions” from spinodal decomposition is measured. Results from BOB simulations ( head-on $^{129}$Xe on $^{119}$Sn collisions at 32 AMeV) are also shown; although all events in the simulation arise from spinodal decomposition, only a very small fraction of the final partitions ($\sim$ 1%) have kept nearly equal-sized fragments. Mainly due to finite size effects the signature of spinodal decomposition can only reveal itself as a weak “fossil” signal.
As a conclusion of that section we can say that, supported by theoretical simulations, we interpret our data as a signature of spinodal instabilities as the origin of multifragmentation of those fused systems in the Fermi energy domain. Spinodal decomposition describes the dynamics of a first order phase transition, the present observations thus support the existence of such a transition for hot finite nuclear matter.
![Abnormal production of events with nearly equal-sized fragments ( $\sigma_{Z} < 3$ corresponds to the upper limit of the islet of peaks - see [@I40-Tab02] for details) as a function of thermal excitation energy (deduced from SMM): full points. The incident and radial energy scales are also indicated. The open point refers to the result from BOB simulations; the average thermal energy is used.[]{data-label="fig8"}](figure1.eps){width="\textwidth"}
Thermodynamics of finite systems and related signals
====================================================
In section 2 we have shown that experimental charge and multiplicity distributions were reproduced with BOB simulations while average fragment kinetic energies were accounted for within 20 %. These same data were also well accounted for with the statistical model SMM [@T16Sal97; @T25NLN99; @LenBol00]. From these two findings we can deduce that the dynamics involved is sufficiently chaotic to finally explore enough of the phase space in order to describe fragment production through a statistical approach. Selected events correspond to “statistical samples” (which have nothing to do with ergodic arguments) and therefore we can go further and use statistical mechanics of finite systems.
Different signals and their robustness
--------------------------------------
From statistical physics of finite systems three direct signals of a first order phase transition are proposed: (i) the caloric curve at constant pressure, which is predicted to present a negative curvature, (ii) the presence of negative microcanonical heat capacity [@Cho02] and (iii) the existence of a bimodality of the event distribution as a function of an order parameter [@Cho01]. They are all related to a curvature anomaly of a thermodynamical potential due to surface effects in the unstable region of coexistence.
Do those signals have equivalent robustness ? Experimentally one does not explore the caloric curves at constant pressure nor at constant volume. In fact many different measured caloric curves can be generated from experiments depending on the path followed on the microcanonical equation of state surface (T versus excitation energy and average volume). Therefore signals from shapes of caloric curves ( plateau-like) appear not as robust as the two others signals. The method proposed to measure the microcanonical heat capacity uses partial energy fluctations. Abnormal large fluctuations (as compared to the canonical expectation) are predicted to be always seen, independently of the path, if microcanonical negative heat capacity is present [@Cho00]. Finally bimodality is predicted to be very robust.
Abnormal kinetic energy fluctuations and negative microcanonical heat capacity
------------------------------------------------------------------------------
The prescription proposed to estimate microcanonical heat capacity is based on the fact that for a given total energy of a system, the average partial energy stored in a part of the system is a good microcanonical thermometer, while the associated fluctuations can be used to construct the heat capacity [@Cho02]. From experiments the most simple decomposition of the total energy $E^{*}$ is in a kinetic part $E_{1}$ and a potential part $E_{2}$ (Coulomb energy + total mass excess). However these quantities have to be determined at freeze-out and consequently it is necessary to trace back this configuration on an event by event basis. This needs, in principle, the knowledge of the freeze-out volume and of all the particles evaporated from primary hot fragments including the undetected neutrons. Consequently some reasonable working hypotheses are used, eventually constrained by specific experimental results [@MDA02]. An example of results obtained from INDRA data is displayed in fig. \[fig:b5\].
![ Microcanonical heat capacity per nucleon as a function of the excitation energy (corrected in average for the radial collective energy) measured in central collisions between Xe and Sn. From [@NLNBorm00][]{data-label="fig:b5"}](figure2.eps){width="\textwidth"}
Observation of correlated signals
=================================
The spinodal region is also thermally unstable. This property has a well known consequence at the thermodynamical limit: the condition of maximal entropy (Maxwell construction) fixes a constant temperature for a given isobar in the coexistence region. But for finite systems the entropy presents a convex dip in that region and a direct consequence is the occurrence of a negative heat capacity. Thus, when spinodal decomposition is observed, one must measure a negative heat capacity. The reverse is not true because spinodal decomposition needs sufficient time to occur (about 100 fm/c, see subsection 3.1). Thus, systems may penetrate the spinodal region and multifragment through nucleation rather than spinodal decomposition.
Therefore both signals (spinodal decomposition and negative microcanonical heat capacity) have been simultaneously studied on different fused systems which undergo multifragmentation [@T25NLN99; @I40-Tab02; @T32Gui02]. Results are summarized in table\[table1\].
system Ni+Au Ni+Ni Xe+Sn Xe+Sn Ni+Au Xe+Sn Xe+Sn
------------------ ------------- ------------- ------------- ------------- ------------- ------------- -------------
Incident Energy
AMeV 32. 32. 32. 39. 52. 45. 50.
Thermal energy
AMeV 5.0$\pm$1.0 5.0$\pm$1.0 5.0$\pm$0.5 6.0$\pm$0.5 6.0$\pm$1.0 6.5$\pm$0.5 7.0$\pm$0.5
Radial energy
AMeV 0. 0.8$\pm$0.5 0.5$\pm$0.2 1.0$\pm$0.3 0. 1.5$\pm$0.4 2.2$\pm$0.4
Spinodal
decomposition no yes? yes yes yes yes no
Negative microc.
heat capacity no yes yes yes yes yes? no
\[table1\]
For the different systems we have also indicated the associated thermal and radial collective energies extracted (see subsection 3.2. and reference [@T32Gui02] for details). We observe a correlation between the two signals, which strongly supports the presence of a first order phase transition for hot nuclei undergoing multifragmentation in the Fermi energy domain. Moreover this correlation reinforces the fact that spinodal decomposition (even if evidenced by a small fossil signal) is indeed the dynamics of the phase transition. Both signals are present when a total (thermal+radial) energy in the range 5.5-7.0 AMeV is measured. Note that the effect of a very gentle compression phase leading to 0.5-1.0 AMeV radial expansion energy plays the same role as a slightly higher thermal energy (Ni+Au system at 52 AMeV). This can be understood in terms of a required threshold for expansion energy; in the latter case this threshold should be reached by thermal expansion only.
Other signals
=============
Several other signals are under studies using INDRA data. The so called Fisher scaling is observed on a large set of central (fused systems) and peripheral collisions [@NLNBorm02]. It was tentatively used in reference [@Ell02] to derive information on the critical point of finite neutral nuclear matter; we notice that, for the fused systems previously discussed, the observed “pseudo” critical point appears inside the coexistence zone (presence of the two correlated signals discussed in the previous section). This last remark is also valid for analysis of data within the universal fluctuations framework (see [@Bot01] - fluctuations of the size of the heaviest fragment). Within this theory, transition from an ordered phase to a disordered phase can be well identified through the largest fluctuations nature provides for an order parameter. Finally one can make the more general comment that for finite nuclei the largest fluctuations measured (kinetic energy and size of the heaviest fragment) are associated to the coexistence region (see also [@Gul03]).
Concerning bimodality, encouraging results have been recently obtained using data from peripheral Au+Au collisions studied with INDRA at GSI [@PicBorm03]. A correlation between bimodality and negative microcanonical heat capacity is observed.
Conclusions and perspectives
============================
The correlation observed between the presented signals completely argues for the existence of a liquid-gas type phase transition in hot nuclei. The nature of the dynamics of the phase transition has been evidenced; spinodal instabilities are thus shown for the first time in a finite system. Many systematic studies including correlations between different signals, independent measurements to characterize systems at freeze-out, are now required for establishing a metrology of the phase diagram and the related equation of state for hot nuclei and nuclear matter. The introduction of the N/Z degree of freedom appears also for the future as a second very exciting challenge.
[10]{} url \#1[`#1`]{}urlprefix\[2\]\[\][[\#2](#2)]{}
J. Pouthas et al., Nucl. Instr. and Meth. in Phys. Res. A 357 (1995) 418.
J. Pouthas et al., Nucl. Instr. and Meth. in Phys. Res. A 369 (1996) 222.
M. D’Agostino et al., Phys. Lett. B 473 (2000) 219.
J. A. Hauger et al. ([EOS collaboration]{}), Phys. Rev. C 62 (2000) 024616.
J. B. Elliott et al. ([EOS collaboration]{}), Phys. Rev. C 62 (2000) 064603.
H. Xu et al., Phys. Rev. Lett. 85 (2000) 716.
B. Borderie, J. Phys. G: Nucl. Part. Phys. 28 (2002) R217 and references therein.
M. D’Agostino et al., Nucl. Phys. A 699 (2002) 795.
A. H. Raduta et al., Phys. Rev. C 65 (2002) 054610.
M. [Kleine Berkenbusch]{} et al., Phys. Rev. Lett. 88 (2002) 022701.
J. B. Elliott et al., Phys. Rev. Lett. 88 (2002) 042701.
D. H. E. Gross, Microcanonical Thermodynamics - Phase Transitions in “small” systems, World Scientific, Singapore, 2001.
P. Chomaz et al., T. Dauxois et al. (eds.) Dynamics and Thermodynamics of systems with long range interactions, Springer-Verlag, Heidelberg, 2002, vol. 602 of Lecture Notes in Physics, 68–129.
F. Gulminelli, LPCC 03-06 (2003) and references therein.
H. Kruse et al., Phys. Rev. C 31 (1985) 1770.
C. Grégoire et al., Nucl. Phys. A 465 (1987) 317.
G. F. Bertsch et al., Phys. Rep. 160 (1988) 189.
A. Bonasera et al., Phys. Rep. 243 (1994) 1.
S. Ayik et al., Phys. Lett. B 212 (1988) 269.
J. Randrup et al., Nucl. Phys. A 514 (1990) 339.
P. Chomaz et al., Phys. Lett. B 254 (1991) 340.
P. Chomaz et al., Phys. Rev. Lett. 73 (1994) 3512.
J. D. Frankland et al. (INDRA collaboration), Nucl. Phys. A 689 (2001) 940.
M. F. Rivet et al. (INDRA collaboration), Phys. Lett. B 430 (1998) 217.
G. Tăbăcaru et al. ([INDRA collaboration]{}), G. Bonsignori et al. (eds.) Proc. of the International Conference on Structure of the Nucleus at the Dawn of the Century, Bologna, Italy, World Scientific, 2001, Vol. 1, Nucleus-Nucleus collisions, 321.
M. Colonna et al., Nucl. Phys. A 613 (1997) 165.
D. Idier et al., Ann. Phys. Fr. 19 (1994) 159.
S. Ayik et al., Phys. Lett. B 353 (1995) 417.
B. Jacquot et al., Phys. Lett. B 383 (1996) 247.
L. G. Moretto et al., Phys. Rev. Lett. 77 (1996) 2634.
P. Désesquelles, Phys. Rev. C 65 (2002) 034604.
G. Tăbăcaru et al., Eur. Phys. J. A 18 (2003) in press, nucl-ex/0212018.
B. Borderie et al. (INDRA collaboration), Phys. Rev. Lett. 86 (2001) 3252.
J. Bondorf et al., Phys. Rep. 257 (1995) 133.
S. Salou, thèse de doctorat, Université de Caen (1997), [GANIL T 97 06]{}.
N. [Le Neindre]{}, thèse de doctorat, Université de Caen (1999), [LPCC T 99 02]{}.
N. [Le Neindre]{} et al. ([INDRA collaboration]{}), G. Bonsignori et al. (eds.) Proc. of the International Conference on Structure of the Nucleus at the Dawn of the Century, Bologna, Italy, World Scientific, 2001, Vol. 1, Nucleus-Nucleus collisions, 221.
P. Chomaz et al., Phys. Rev. E 64 (2001) 046114.
P. Chomaz et al., Phys. Rev. Lett. 85 (2000) 3587.
N. [Le Neindre]{} et al. ([INDRA collaboration]{}), I. Iori et al. (eds.) Proc. XXXVIII Int. Winter Meeting on Nuclear Physics, Bormio, Italy, Ricerca scientifica ed educazione permanente, 2000, 404.
B. Guiot, thèse de doctorat, Université de Caen (2002), [GANIL T 02 04]{}.
N. [Le Neindre]{} et al. ([INDRA collaboration]{}), I. Iori et al. (eds.) Proc. XL Int. Winter Meeting on Nuclear Physics, Bormio, Italy, Ricerca scientifica ed educazione permanente, 2002, 144.
R. Botet et al., Phys. Rev. Lett. 86 (2001) 3514.
M. Pichon et al. ([INDRA and ALADIN collaborations]{}), I. Iori et al. (eds.) Proc. XLI Int. Winter Meeting on Nuclear Physics, Bormio, Italy, Ricerca scientifica ed educazione permanente, 2003, 149.
[^1]: Present address: INFN, Sezione di Bologna and Dipartimento di Fisica, Università di Bologna, Italy
[^2]: Present address: Cyclotron Institute, Texas A&M University, College Station, Texas 77845, USA
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In the present work we investigate the possibility of superluminal information transmission in quantum theory. We give simple and general arguments to prove that the general structure (Hilbert’s space plus instantaneous state reduction) of the theory allows the existence of superluminal communication. We discuss how this relates with existing no-signalling theorems.'
address: 'Dipartimento di Informatica , Universita’ di Milano'
author:
- Giovanni Andrea Fantasia
title: Superluminality in Quantum Theory
---
[^1]
Introduction {#introduction .unnumbered}
============
Since the EPR paradox appeared [@EPR35] the question of which were the sense of non-locality in quantum theory was a very fundamental question to address [@Red]. One main problem is to know if quantum non-locality implies superluminal communication between two separated parts A and B. Several arguments are given to exclude this hypotheses [@Bohm] [@Ghi80] [@Shi]: they form well-known no-signalling theorems. However these theorems have all the common hypotheses that operators associated with the two separated parts commute: \[A,B\]=0. We show possible gedanken experiments which violate this hypotheses: in effect we give two examples that prove that the general structure (Hilbert’s space plus instantaneous state reduction) of quantum theory allows the existence of superluminal communication.
Superluminal communication {#superluminal-communication .unnumbered}
==========================
In this section we propose two different protocols for superluminal communication and we discuss how they circumvent no-signalling theorems. Suppose that a quantum system, say a particle, is in the state $$\label{particle}
\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}+{\left\vertB\right\rangle})$$ where ${\left\vertA\right\rangle}$ and ${\left\vertB\right\rangle}$ represent two long distance separated spatial localization of the particle. We assume that trough all the duration of the protocol dispersion of the particle will be not relevant. So ${\left\vertA\right\rangle}$ represents the particle localized in a finite volume in region A e so does ${\left\vertB\right\rangle}$. If we make a measurement testing if particles is in the state ${\left\vertA\right\rangle}$ this action is represented by the projector ${\left\vertA\right\rangle}{\left\langleA\right\vert}$ where ${\left\langleA\right\vert}$ has the properties $$\label{proj}
{\left\langleA\right\vert}A\rangle=1;{\left\langleA\right\vert}B\rangle=0$$ In this way observing the state (\[particle\]) using ${\left\vertA\right\rangle}{\left\langleA\right\vert}$ will transform the state of the system in ${\left\vertA\right\rangle}$ or in with equal probability. Analogously we can define a projector
$$\label{projB}
{\left\vertB\right\rangle}{\left\langleB\right\vert}$$
\
Let construct a new projector $$\label{newproj}
\frac{1}{\sqrt{2}}{\left\vertA+B\right\rangle}\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}$$
defining $$\nonumber
\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}=\frac{1}{\sqrt{2}}({\left\langleA\right\vert}+{\left\langleB\right\vert})$$ where $\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}$ has the properties $$\begin{aligned}
\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}\frac{1}{\sqrt{2}}(A+B)\rangle=1&\\\frac{1}{\sqrt{2}}{\left\langle
A+B\right\vert}\frac{1}{\sqrt{2}}(A-B)\rangle=0\end{aligned}$$ In this way observing the state ${\left\vertA\right\rangle}$ using projector (\[newproj\]) will transform the state of the system in $\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}+{\left\vertB\right\rangle})$ or in $\frac{1}{\sqrt{2}}({\left\vert(A\right\rangle}-{\left\vertB\right\rangle})$ with equal probability. In fact ${\left\vertA\right\rangle}=\frac{1}{\sqrt{2}}(\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}+{\left\vertB\right\rangle})+\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}-{\left\vertB\right\rangle}))$ where $\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}+{\left\vertB\right\rangle})$ and $\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}-{\left\vertB\right\rangle})$ are the eigen-vectors of projector $\frac{1}{\sqrt{2}}{\left\vertA+B\right\rangle}\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}$ . Making now a projection with ${\left\vertB\right\rangle}{\left\langleB\right\vert}$ and whatever was the state leaved by $\frac{1}{\sqrt{2}}{\left\vertA+B\right\rangle}\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}$ we will find the particle in the state ${\left\vertB\right\rangle}$ and so with probability $\frac{1}{2}$ we teleportate the particle from A to B.\
Using a large collection of particle will permit us to raise the probability that at least one particle reaches B close as we want to $1$.\
Summarizing we have a particle in the localized state ${\left\vertA\right\rangle}$, part A makes a measurement projecting onto $\frac{1}{\sqrt{2}}({\left\vertA\right\rangle}+{\left\vertB\right\rangle})$ then part B makes a second measurement projecting onto ${\left\vertB\right\rangle}$ in a finite amount of time. This protocol leads with probability $\frac{1}{2}$ to teleportate particle from A to B.\
The key feature of this protocol is the use of projector (\[newproj\]). We see that this projector doesn’t commute with the one associated with B (\[projB\]). In fact $$\begin{aligned}
[\frac{1}{\sqrt{2}}{\left\vertA+B\right\rangle}\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}{\left\vertB\right\rangle}{\left\langleB\right\vert},
{\left\vertB\right\rangle}{\left\langleB\right\vert}\frac{1}{\sqrt{2}}{\left\vertA+B\right\rangle}\frac{1}{\sqrt{2}}{\left\langleA+B\right\vert}]=&\\
\frac{1}{2}{\left\vertA+B\right\rangle}{\left\langleB\right\vert}-\frac{1}{2}{\left\vertB\right\rangle}{\left\langleA+B\right\vert}=\frac{1}{2}({\left\vertA\right\rangle}{\left\langleB\right\vert}-{\left\vertB\right\rangle}{\left\langleA\right\vert}\neq
0\end{aligned}$$ A protocol that uses only statistical correlations between spin systems could be find in [@Ghi00].\
There is however a physical objection that may be done to measurements of the kind (\[newproj\]). In effect to do this measurement part A would physically operate on both region of space A and B because this measurement test a non-local properties of the system. We can test if the particle is in A or in B just operating locally in A (or in B) but it is not clear if measurement (\[newproj\]) may be physically achieved by part A operating locally in A (or in B) or even non-locally in A and in B. Even if this operator is formally a well defined self-adjoint operator it is hard to imagine a physically (local or non-local) realization of such a measurement. Anyway this problem doesn’t seems to affect the next example of superluminal communication.\
Suppose now part A has a particle in the state ${\left\vertA\right\rangle}$. So the particle is strictly localized around A and probability that B detect the particle is very near to 0. Part A performs (locally) a precision measurement on the momentum of the particle. This “collapse” the wave function from ${\left\vertA\right\rangle}$ to ${\left\vertA'\right\rangle}$. The new state ${\left\vertA'\right\rangle}$, in momentum domain, is strictly localized around a random value $\lambda$ that is the result of precision momentum measurement. Obviously, as indetermination relation between momentum and position $2\Delta x\Delta p\geq\hbar$ requires, the new state ${\left\vertA'\right\rangle}$ is more spread in space than old state ${\left\vertA\right\rangle}$ so that now probability that part B detects the particle will be increased. So part A could send instantaneous information to B. Again the protocol circumvent the no-signalling theorems because measurement operators of part A and part B doesn’t commute. We do now some formal calculations to validate this protocol.\
Suppose part A and B laying on a line. Part A is located at the origin and B is located at a distance d from A. Part B detects if the particle is located in the region \[d-k,d+k\]. So the starting state ${\left\vertA\right\rangle}$ written in space domain is $$\label{stateA}
{\left\vertA(x)\right\rangle}=\frac{1}{\sqrt[4]{\pi\sigma^2}}\hspace{4pt}e^{-\frac{x^2}{2\sigma^2}}$$ ($\hbar=1$) with the variance small enough to let probability that part B detect the particle near to 0 $$\label{smallprob}
\int^{d+k}_{d-k}\parallel\frac{1}{\sqrt[4]{\pi\sigma^2}}\hspace{4pt}e^{-\frac{x^2}{2\sigma^2}}\parallel^2 dx\approx
0$$ Fourier transform of state ${\left\vertA\right\rangle}$ will give us the wave function written in momentum domain $$\label{momentum1}
F({\left\vertA\right\rangle})={\left\vertA(p)\right\rangle}=\sqrt[4]{\frac{\sigma^2}{\pi}}\hspace{4pt}e^{-\frac{\sigma^2p^2}{2}}$$ After the precision momentum measurement, supposed we got the random value $p=\lambda$, the state ${\left\vertA'\right\rangle}$ written in momentum domain will be $$\label{statexmom}
{\left\vertA'(p)\right\rangle}=\sqrt[4]{\frac{\overline{\sigma}^2}{\pi}}\hspace{4pt}e^{-\frac{\overline{\sigma}^2(p-\lambda)^2}{2}}$$ with $\overline{\sigma}\gg\sigma$. In the space domain state ${\left\vertA'\right\rangle}$ will be $$\label{statexspc}
{\left\vertA'(x)\right\rangle}=F^{-1}(A'(p))=\frac{1}{\sqrt[4]{\pi\overline{\sigma}^2}}\hspace{4pt}e^{-\frac{x^2}{2\overline{\sigma}^2}-\frac{i\lambda
x}{2}}$$ The probability that part B will detect the particle is now $$\label{prob2}
\int^{d+k}_{d-k}\parallel \frac{1}{\sqrt[4]{\pi\overline{\sigma}^2}}\hspace{4pt}e^{-\frac{x^2}{2\overline{\sigma}^2}-\frac{i\lambda
x}{2}}\parallel^2 dx=\int^{d+k}_{d-k}
\frac{1}{\sqrt[2]{\pi\overline{\sigma}^2}}\hspace{4pt}e^{-\frac{x^2}{\overline{\sigma}^2}}dx>0$$ and we see that doesn’t depend from the $\lambda$ measured momentum.
Conclusion {#conclusion .unnumbered}
==========
Even if superluminal velocity in unitary evolutions of quantum system was already known [@Chi] [@Heg] we think that we have here provides some clear examples of how hypotheses of instantaneous wave collapse in quantum theory will permit superluminal communication. This is not in conflict with existing no-signalling theorems because we don’t use commuting operators. Our main conclusion is that quantum theory doesn’t peacefully coexist with special relativity and no-signalling requirements will impose new postulates to quantum theory. In effect it turns out that a sufficient condition for no-signalling [@Peres] is $[A,B]=0$ where A are B are measurement operators associated with part A and B but , as we have seen, we can imagine several situations in which $[A,B]\neq 0$.
[1]{}
N. Rosen A. Einstein, B. Podolsky, Phys. Rev. **47** (1935), 777.
D. Terno A. Peres, quant-ph/0212023.
A.Shimony, Proc. of the Int. Symp. on Foundations of Quantum Mech., eds. S. Kamefuchi (Phys. Soc. of Japan, Tokyo, 1993).
R. Chiao, quant-ph/9811019.
B.J.Hiley D.Bohm, *The undivided universe: An onthological interpretation of quantum theory*, Routledge, London-New York, 1993.
G.C.Ghirardi, Found. Phys. **30** (2000), 9.
T. Weber G.C.Ghirardi, A. Rimini, Lett. Nuovo Cimento **27** (1980), 293.
G.Hegerfeldt, quant-ph/9809030.
M.Redhead, *Incompleteness, nonlocality and realism*, Clarendon press, Oxford, 1987.
[^1]: I would like to thank E.Angeleri, G.Degli Antoni, F.Thaheld, S.Srikanth, E.Recami for useful discussions Thank also STMicroelectronics for financial support
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Investigations of living organisms have led biologists and physicians to introduce fundamental concepts, including Brownian motion, the First Law of Thermodynamics, Poiseuille’s Law of fluid flow, and Fick’s Law of diffusion into physics. Given the prominence of viscous forces within and around cells and the experience of identifying and quantifying such resistive forces, biophysical cell biologists have an unique perspective in discovering the viscous forces that cause moving particles to respond to an applied force in a nonlinear manner. Using my experience as a biophysical cell biologist, I show that in any space consisting of a photon gas with a temperature above absolute zero, Doppler-shifted photons exert a velocity-dependent viscous force on moving charged particles. This viscous force prevents charged particles from exceeding the speed of light. Consequently, light itself prevents charged particles from moving faster than the speed of light. This interpretation provides a testable alternative to the interpretation provided by the Special Theory of Relativity, which contends that particles are prevented from exceeding the speed of light as a result of the relativity of time.'
address: |
Department of Plant Biology,\
Cornell University\
Ithaca, New York, 14853 USA\
author:
- Randy Wayne
date: 'October 9, 2009'
title: 'Charged Particles are Prevented from Going Faster than the Speed of Light by Light Itself: A Biophysical Cell Biologist’s Contribution to Physics [^1]'
---
=by -1
PACS 03.30.+p, 03.65.Pm, 42.25.-p
Introduction
============
*“Ask not what physics can do for biology, ask what biology can do for physics.”*
-Stanislaw Ulam [@1]
Cells and the organelles within them live in a world whose dimensions fall between those claimed by the world of macroscopic theoretical physics and those claimed by the world of microscopic theoretical physics [@2; @3]. In working in this world of neglected dimensions, biophysical cell biologists have used the known laws of physics to make great strides in understanding the physical basis of life [@4; @5; @6; @7; @8; @9; @10; @11; @12; @13; @14; @15; @16; @17; @18; @19; @20; @21; @22; @23; @24; @25; @26]. However, in working in the world of cellular dimensions, biophysical cell biologists also have a unique opportunity to contribute to “new physics” by looking for physical laws that are capable of encompassing theoretical macrophysics and microphysics. This is not such a wild speculation when one considers that much of what we call “physics” comes from the study of living organisms [@27]. The wave theory of light as a description of diffraction came from Thomas Young’s [@28] endeavor to understand vision; the discovery of Brownian motion came from Robert Brown’s [@29; @30] study of pollen and pollination; the First Law of Thermodynamics came from Robert Mayer’s [@31] observation that the venous blood of people living in the warm climate of Java is redder and thus more oxygenated than the venous blood of people living in the colder German climate; the eponymous Law of laminar flow came from Jean Poiseuille’s [@32] work on describing the flow of blood; and the eponymous law of diffusion came from Adolf Fick’s [@33] work on describing transmembrane solute movement in kidneys.
For over four centuries, Newton’s *Philosphiae Naturalis Principia Mathematica* has provided a method for describing the frame of the System of the World, and the three laws of motion described in Book One of the *Principia* have formed the theoretical foundations of terrestrial and celestial mechanics [@34; @35]. According to Newton’s First Law, *“Every body continues in its state of rest, or of uniform motion in a right line, unless it is compelled to change that state by forces impressed on it.”* While this statement is often called the law of inertia, it is better characterized as the assumption of no friction. That is, while the inertia of the body provides a resistance to any change in motion; the body is considered to be *inert* to the medium though which the body moves so that the medium provides no resistance to the movement of the *inert*ial body.
Once one assumes that friction is negligible, Newton’s Second Law follows. According to Newton’s Second Law, *“The change of motion [\[]{.nodecor}*dmv*/*dt*[\]]{.nodecor} is proportional to the motive force impressed; and is made in the direction of the right line in which that force is impressed.”* That is, the time rate of change of momentum of an object is proportional to the motive force applied to the object. Assuming, along with Newton, that the mass of the body is a constant, it is the acceleration of the body and not the velocity that is proportional to the applied motive force. Newton’s Second Law (***F*** = *m*$\frac{d\textit{\textbf{v}}}{d\textit{t}}$), which serves to describe everything from the falling of an apple to the orbit of the Moon, implies that any particle with a constant and invariant mass (*m*) can be accelerated from rest to any velocity (***v***) in time (*t*) by the application of a large enough constant force (***F***) or the application of a small force for a long enough time.
In contrast to the Laws given by Newton in Book One of the *Principia*, are the lesser-known propositions given in Book Two that describe the motion of bodies in resisting mediums. While the Laws in Book One can be considered to be simple and elegant descriptions of Platonic or ideal situations that have become the foundation of theoretical macroscopic physics, the lesser-known propositions in Book Two, which Newton wrote in haste, can be considered to describe complex fudge factors that depend on a number of variables, including the density and tenacity of the medium as well as the size and shape of the inertial body [@36]. These fudge factors, which give the resistance \[viscous force\] of the medium, were added or subtracted by Newton in order to deduce the magnitude of the “absolute force” applied to that body. The effects of these fudge factors on movement were quantified by George Stokes [@37], who provided the foundation for “non-Newtonian” physics. By studying mixtures of beach sand or gelatin, Osborne Reynolds [@38] and Herbert Freundlich [@39], respectively realized that resistive media could have an infinite number of viscosities that depended on the velocity of the particle moving through the medium. Media whose resistance increased with velocity were called dilatant and media whose resistance decreased with velocity were called thixotropic. Realizing that the structure and composition of the highly dynamic and ever changing cytoplasmic space determined the characteristics of the motile processes that take place within it, William Seifriz, Nobur$\widehat{\textnormal{o}}$ Kamiya and their colleagues championed the use of experimental techniques to characterize the nonlinear resistance to movement provided by the living protoplasm itself [@40; @41; @42; @43; @44; @45; @46; @47; @48; @49]. As I will describe below, by using biophysical cell biological thinking, it is possible to integrate the nonlinear properties of the viscous force into Newton’s Second Law to construct a simple, elegant, quantitative and testable physical theory that is more robust than Newton’s Second Law in that it also encompasses microphysical phenomena.
Influenced by the field theory of Maxwell [@50; @51], scientists in the late nineteenth century generally thought that the elementary unit of electricity known as an electron was nonmaterial in nature and resulted from the release of a center of strain in the nonmaterial aether [@52; @53; @54; @55; @56]. Going against the conventional wisdom, J. J. Thomson [@57] resurrected Gustav Fechner’s and Wilhelm Weber’s idea that electrons might be corpuscles of matter rather than a released element of an nonmaterial aether and applied Newton’s Laws to determine the mass of the electrons that made up cathode rays [@58; @59; @60]. While Newton’s Second Law was known to describe the motion of corpuscular bodies from apples to planets, it failed to describe the motion of the corpuscular carrier of electrical charge. In his characterization of the mass-to-charge ratio of the electron, Thomson [@61] found that the amount of force required to accelerate an electron increased nonlinearly as the electron’s velocity asymptotically approached the speed of light (c). This indicated to Thomson that the mass of the electron increased with velocity. Thomson [@62; @63] concluded that the increased mass followed from Maxwell’s equations and was a result of the magnetic field produced in the aether by the moving charge.
Interpreting the mass of an electron did not turn out to be straightforward. Hendrik Lorentz [@64; @65; @66] postulated that the mass of an electron moving through an isotropic space was anisotropic and calculated that the longitudinal mass or the component of the mass of an electron parallel to the direction of motion increased with an increase in velocity by a factor of $\sqrt[3]{1 - v^{2}/c^{2}}$, while the transverse mass or the component of the mass of an electron perpendicular to the direction of motion increased with an increase in velocity by a factor of $\sqrt{1 - v^2/c^2}$. The reciprocal of the latter expression is commonly known as the Lorentz factor ($\gamma$).
Based only on his postulates of the principle of relativity and the constancy of the speed of light, Albert Einstein [@67] independently derived the velocity-dependent longitudinal and transverse components of the mass of the electron in the dynamical part of the paper entitled, *On the electrodynamics of moving bodies* and concluded that the increase in mass might result from the withdrawal of energy from the electrostatic field and thus might only be apparent. He further wrote that *“in comparing different theories of the motion of the electron we must proceed very cautiously.”* Nevertheless, for many physicists, the interpretation that the mass of an electron increased in a velocity-dependent manner was useful in explaining the nonlinear nature of the force-acceleration relation for an electron [@68; @69; @70; @71; @72; @73].
While various mechanical and electromagnetic interpretations of the relationship between impulse (the product of force and time) and velocity were proffered in the first decades of the 20th century [@74; @75], by the 1920s, there was a general agreement that the nonlinear relationship between impulse and velocity was explained in terms of the relativity of space-time [@76; @77; @78; @79]. That is, the duration of time is a relative quantity that depends on the relative velocity of the electron and the experimenter in the laboratory frame.
According to this kinematic space-time interpretation of Einstein’s Special Theory of Relativity, the duration of time (d*t*) varies in a velocity-dependent manner according to the following equation: $$dt_{proper} = \sqrt{(1 - v^2/c^2)}\ \ dt_{improper},$$ where $dt_{proper}$ and $dt_{improper}$ are the proper duration and improper duration of time, respectively. According to the space-time interpretation of the Special Theory of Relativity, the force-acceleration relation is nonlinear because a moving particle experiences a constant force for a shorter duration of time ($dt_{proper}$) than it would if the particle were at rest in the laboratory frame ($dt_{improper}$). That is, as a particle goes faster and faster, it experiences the motive force for a shorter and shorter duration of time.
Since movement is one of the fundamental attributes of life [@4], biophysical cell biologists spend a great deal of time investigating movement, including the movement of mechanochemical proteins such as myosin, dynein and kinesin [@80; @81; @82; @83]; the movement of water and ions through channels in membranes [@84; @85; @86; @87; @88; @89]; the movement of small molecules through plasmodesmata [@90]; the movement of proteins from their site of synthesis to their site of action [@91]; the movement of membrane vesicles, tubules [@92; @93; @94]; and organelles [@95; @96] throughout the viscous cytoplasm [@97; @98]; the movement of chromosomes during mitosis and meiosis [@99; @100; @101]; the movement of cilia and flagella [@102; @103]; and the movement of the polymers of the extracellular matrix that allow the growth of plant cells [@104; @105; @106; @107]. Each one of these cellular movements, as well as others not listed here, involves a motive force that must overcome a non-negligible viscous force. Consequently, the acceleration is not proportional to the applied force. In fact, given the low Reynolds Numbers [@108], which are system-dependent dimensionless numbers that relate the motive force to the viscous force [@109], in a cell, movement ceases immediately after the removal of a motive force, and it is the velocity rather than the acceleration of a body that is proportional to the motive force [@110; @111; @112].
Given the prominence of viscous forces within and around a cell [@113; @114] and the experience of identifying and quantifying such resistive forces, the first question a biophysical cell biologist asks when confronted with the nonlinear relationship between force and the acceleration of an electron is, “Can the nonlinear relationship between force and acceleration observed for particles moving at speeds approaching the speed of light be explained by invoking the presence of a viscous force (without reintroducing the contentious nineteenth century aether)?” Biophysical cell biologists are at an advantage in discovering the viscous forces that cause moving particles to respond to an impulse in a nonlinear manner (Fig. 1).
![The impulse needed to accelerate an electron (*m* = 9.1 x 10$^{-31}$ kg) from rest (***v*** = 0) to a given velocity assuming Newton’s Second Law is valid and impulse = $\int$ ***F** dt* = *m**v*** or that Newton’s Second Law is invalid and impulse = $\int$ ***F** dt*$_{improper}$ $\neq$ *m****v*****.[]{data-label="f1"}](figure1.eps){width="12cm"}
Results and Discussion
======================
When a biophysical cell biologist examines the movement of or in a cell, he or she asks, “What are the physical attributes of the space through which the body moves?” A biophysical cell biologist strives to answer this question in spite of the fact that the structure and composition or organization of the space itself is not static. Likewise, when one studies the movement of a particle through a space, one must ask the same question. Let us consider the movement of an electron through space where the motive force is provided by an electric field. At any temperature above 0 K, the space consists of a radiation field composed of photons. The photons can be considered to have a black body distribution [@115] (Fig. 2).
![The distribution of energy density with respect to wavelength in a photon gas with a temperature of 300 K. The peak will shift to the left and the area under the curve will increase for a photon gas with a temperature $>$ 300 K; and the peak will shift to the right and the area under the curve will diminish for a photon gas with a temperature $<$ 300 K.[]{data-label="f2"}](figure2.eps){width="12cm"}
The electron moving through the black body radiation field experiences the photons that make up the field as being Doppler shifted [@67; @116; @117]. The photons that collide with the front [@118] of the moving electron will be blue shifted (have a higher frequency) and the photons that collide with the back of the moving electron will be red shifted (have a lower frequency) (Fig. 3).
![An electron moving through a photon gas experiences the photons as being Doppler shifted. The photons that strike the front of the electron and slow it down are blue shifted and the photons that strike the back of an electron and speed it up are red shifted.[]{data-label="f3"}](figure3.eps){width="12cm"}
Since the electron is moving at a velocity (***v***) relative to the center of momentum of the radiation field, as a result of the Doppler effect, the electron will experience the radiation field as being anisotropic whereas an observer who is at rest with the radiation field will observe it as being isotropic. Consequently, I will describe the radiation experienced by the moving particle with an original relativistic wave equation that describes the propagation of light waves between inertial frames moving relative to each other at velocity ***v*** and satisfies the requirements set by the Michelson-Morley experiment [@119; @120; @121]. This new relativistic wave equation is given by:
$$\frac{\partial^2\Psi}{\partial t^2} = \textnormal{c}c'\frac{\sqrt{c - v \cos\theta}}{\sqrt{c + v \cos\theta}}\ \frac{\partial^2\Psi}{\partial x^2},\\$$
or $$\frac{\partial^2\Psi}{\partial t^2} = \textnormal{c}c'\frac{\sqrt{1 - (v \cos\theta)/c}}{\sqrt{1 + (v \cos\theta)/c}}\ \frac{\partial^2\Psi}{\partial x^2},$$
where *v* is the absolute value of the velocity and *$\theta$* is the angle between the velocity of a particle and the velocity of a photon (or wave). When the movements of a photon and a particle are parallel, *$\theta$* = 0 radians and cos *$\theta$* = 1; and, when the movements of a photon and a particle are antiparallel, *$\theta$* = $\pi$ radians and cos *$\theta$* = -1 (see appendix 1).
Different aspects of the speed of a photon or a light wave are represented by c and *c’*. The parameter c, which is absolute and independent of the velocity of the source or the observer, gives the speed of the photon or wave through the vacuum and is equal to the square root of the reciprocal of the product of the electric permittivity ($\epsilon_{o}$) and the magnetic permeability ($\mu_{o}$) of the vacuum. It was the idea of the absolute nature of the speed of light that originally attracted Max Planck to Einstein’s Special Theory of Relativity [@122]. However, as a consequence of the Doppler Effect, light also has another characteristic speed *c’*, which is local and depends on the relative velocity of the source and observer. *c’* gives the ratio of the angular frequency ($\omega$) of the source in its inertial frame to the angular wave number (*k*) observed in any inertial frame (*c’* $\equiv$ $\omega_{source}/\textit{k}_{observer}$). At any relative velocity between the source and the observer, c*c’*($\frac{\sqrt{1 - (\textit{v} cos \theta)/c}}{\sqrt{1 + (\textit{v} cos \theta)/c}}) = c^{2}$. After canceling c on both sides, we get a relativistic dispersion relation that must be satisfied when the general plane wave solution of the relativistic wave equation has the form: $\Psi = \Psi_{o} e^{i(k_{observer} r - \omega_{source}\frac{\sqrt{c - v cos\theta}}{\sqrt{c + v cos \theta}}t)}$. The relativistic dispersion relation is: $$\textit{c'}(\frac{\sqrt{1 - (\textit{v} \cos \theta)/c}}{\sqrt{1 + (\textit{v} \cos \theta)/c}}) = c$$ At *v* = 0, the new relativistic wave equation reduces to Maxwell’s wave equation [@50].
Introducing the perspicuous correction factor (c = $\textit{c'}\frac{\sqrt{1 - (\textit{v} \cos \theta)/c}}{\sqrt{1 + (\textit{v} \cos \theta)/c}}$) into Maxwell’s wave equation ensures the invariance of this new relativistic wave equation when describing waves traveling between two inertial frames. A relativistic Doppler equation is obtained naturally from the dispersion relation:
$$k_{observer} = k_{source}\frac{\sqrt{1 - (v \cos\theta)/c}}{\sqrt{1 + (v \cos\theta)/c}} \\$$
$$k_{observer} = k_{source}\frac{(1 - v \cos\theta)/c)}{\sqrt{1 - (v^2 \cos^2\theta)/c^2}}.$$
The relativistic Doppler equation allows one to transform angular wave number instead of length and duration between inertial frames.
According to eq. (6), the angular wave numbers in inertial frames moving relative to each other at velocity ***v*** are related to each other by a Galilean transformation in the numerator and a Lorentz-like transformation in the denominator. The Galilean transformation in the numerator is dominant at low velocities while the Lorentz-like transformation in the denominator is dominant at high velocities.
The experimental observations of Ives and Stillwell [@117] on the effect of velocity on the displacement of the spectral lines of hydrogen confirm the utility and validity of using the new relativistic wave equation. However, whereas the Special Theory of Relativity [@67; @123] predicts the transverse Doppler Effect in an inertial system, the relativistic wave equation given above does not predict any Doppler shift exactly perpendicular to the velocity of an inertial particle. Both the Special Theory of Relativity and the relativistic wave equation presented above predict that averaging the forward and backward longitudinal Doppler-shifted light will give the Lorentz factor also known as the “time dilation” factor as observed by Ives and Stillwell [@117]. Consequently, experiments that average the forward and backward longitudinal Doppler shifts [@117; @124] are consistent with both treatments.
The linear momentum of a photon is given by $\hbar$*k*, where $\hbar$ is Planck’s constant divided by 2$\pi$, and *k*, the angular wave number, is equal to 2$\pi$/$\lambda$ [@125]. As a consequence of the relativistic Doppler effect, upon interacting with a photon, the change in the linear momentum of an electron (d$\hbar$*k*$_{\textit{electron}}$) moving with speed *v* is velocity dependent and is given by:
$$\textnormal{d}\hbar k_{electron} = \hbar k_{source}\frac{\sqrt{c - v \cos\theta}}{\sqrt{c + v \cos\theta}} \\$$
$$\textnormal{d}\hbar k_{electron} = \hbar k_{source} \frac{(1 - (v \cos\theta)/c)}{\sqrt{1 - (v^2 \cos^2\theta)/c^2}}.$$
For convenience, I will split eq. (8) into two equations–one for an electron moving parallel (cos $\theta$ = 1) relative to the photons or waves propagating from the source and one for an electron moving antiparallel (cos $\theta$ = -1) relative to the photons or waves propagating from the source. The momentum of the light experienced by the back of an electron traveling parallel to a photon propagating from the source is:
$$\textnormal{d}\hbar k_{electron} = \hbar k_{source}\frac{(1 - v/c)}{\sqrt{1 - v^2/c^2}}.$$
This equation is reminiscent of the equation that describes the Compton Effect, where $\hbar$*k*$_{\textit{source}}$ [@126] is the momentum of a photon before a collision and d$\hbar$*k*$_{\textit{electron}}$ is the momentum transferred from the photon to the “back” of the electron as it is pushed forward [@127]. The momentum of the light experienced by a particle traveling antiparallel to light propagating from the source is:
$$\textnormal{d}\hbar k_{electron} = \hbar k_{source}\frac{(1 + v/c)}{\sqrt{1 - v^2/c^2}}.$$
This equation is similar to the equation that describes the inverse Compton Effect, where $\hbar$*k*$_{\textit{source}}$ is the momentum of a photon before a collision and d$\hbar$*k*$_{\textit{electron}}$ is the momentum transferred from the photon to the “front” of the electron as it is pushed back [@128].
Let us assume that the moving electron interacts with one photon from the front and one photon from the back. The net momentum experienced by this electron moving through a photon gas is antiparallel to the velocity of the electron and would be:
$$\textnormal{d}\hbar k_{electron} = \hbar k_{source}\frac{(1 - v/c) - (1 + v/c)}{\sqrt{1 - v^2/c^2}} \\$$
$$\textnormal{d}\hbar k_{electron} = \hbar k_{source}\frac{(-\frac{2v}{c})}{\sqrt{1 - v^2/c^2}}.$$
Since the momentum of a blue-shifted photon, striking the front of a moving electron is greater than the momentum of a red-shifted photon striking the back of a moving electron, the momentum of the moving electron decreases. The average decrease of momentum (*p* = *mv* = d$\hbar$*k*$_{\textit{electron}}$) experienced by a moving electron upon colliding with one coaxial photon in a photon gas would be [@129]:
$$\textnormal{d}\hbar k_{electron} = -\frac{(\frac{1}{2})\hbar k_{source}(\frac {2v}{c})}{\sqrt{1 - v^2/c^2}} \\$$
$$\textnormal{d}\hbar k_{electron} = -\frac{\hbar k_{source}(\frac{v}{c})}{\sqrt{1 - v^2/c^2}},$$
where the negative sign indicates that the momentum of the electron moving through the radiation field decreases.
Since the “average photon” in the photon gas can strike the moving electron at any angle from 0 to $\pm \frac{\pi}{2}$ with differing effectiveness, the average transfer of momentum [@130] from the radiation field to the electron is $\hbar$*k*$_{\textit{source}}$$\frac{(\frac{1}{4})(\frac{v}{c})}{\sqrt{1 - v^2/c^2}}$ for a single collision. As a consequence of the omnipresent existence of a photon gas in space at all temperatures greater than absolute zero, contrary to Newton’s First Law, moving charged particles are not inert to the photons in the photon gas through which they move and all bodies composed of charged particles will slow down. In an adiabatic photon gas, this will result in an increase in temperature and an increase in the peak angular wave number of the photons in the photon gas. In an isothermal photon gas, this will result in an expansion of the photon gas. Since a photon gas exists at all temperatures above absolute zero, Newton’s First Law is only valid for charged particles at absolute zero and absolute zero, according to the Third Law of Thermodynamics, is unattainable [@131]. By taking into consideration the thermodynamics of the photon gas occupying the space through which the electrons move, I will recast Newton’s Second Law in an alternative form that applies to charged particles [@132] moving at velocities close to the speed of light.
The velocity-dependent relativistic Doppler-shifted momentum of the photon gas provides the basis for a velocity-dependent viscous force (**F**$_{Dopp}$) that counteracts the applied force (**F**$_{app}$) used to accelerate a particle.
$$\textbf{F}_{app} + \textbf{F}_{Dopp} = m\frac{d\textbf{v}}{dt},$$
where **F**$_{app}$ and **F**$_{Dopp}$ are antiparallel by definition.
The viscous force exerted by the radiation field on a moving particle is a function of the collision rate between the moving particle and the photons in the field. The collision rate ($\frac{\textit{dn}}{\textit{dt}}$) depends on the photon density ($\rho$), the speed of the particle (*v*) and the cross section of the photon ($\sigma$) according to the following equation:
$$\frac{dn}{dt} = \rho v \sigma.$$
The photon density is a function of the absolute temperature. The absolute temperature on Earth and in the cavity of some accelerators, including the LINAC at Stanford University, is close to 300 K, while the absolute temperature of the cosmic microwave background radiation and in the cavity of other accelerators, including the LINAC at Jefferson Laboratory and the Large Hadron Collider at CERN, is 2.73 K.
Assuming a blackbody distribution of energy, the photon density can be calculated from Planck’s black body radiation distribution formula. According to Planck [@115], the energy density per unit wavelength interval (*u*) is given by:
$$u = \frac{8\pi hc}{\lambda^5}\frac{1}{exp[\frac{hc}{\lambda kT}] - 1}.$$
The peak wavelength of a photon gas can be obtained by differentiating eq. (17) with respect to wavelength or by simply using Wien’s distribution law:
$$\lambda_{peak} = 2.89784\ \textnormal{x}\ 10^{-3} \textnormal{mK}/T = w/T,$$
where *w* is called the Wien coefficient and is equal to 2.89784 x 10$^{-3}$ mK. The peak wavelengths ($\lambda_{peak}$) in 300 K and 2.73 K radiation fields are 9.66 x 10$^{-6}$ m and 1.87 x 10$^{-3}$ m, respectively. The energies of photons with these wavelengths are given by Planck’s equation:
$$E = \frac{hc}{\lambda_{peak}},$$
and are 2.06 x 10$^{-20}$ J/photon and 1.06 x 10$^{-22}$ J/photon for the peak photons in a 300 K and 2.73 K radiation field, respectively. The total energy density (*U*) of a radiation field can be determined by using the Stefan-Boltzmann equation or by integrating eq. (17) over wavelengths from zero to infinity [@133].
$$U = \int {u\ d\lambda}, \\$$
$$U = \int {\frac{8\pi hc}{\lambda^5}\frac{1}{(exp[\frac{hc}{\lambda kT}] - 1)}\ d\lambda}, \\$$
$$U = \frac{8\pi k^4T^4}{c^3 h^3} \int {\frac{x^3}{[exp(x) -1]}\ dx}, \\$$
$$U = \frac{8\pi k^4T^4}{c^3h^3}(\frac{\pi^4}{15}), \\$$
$$U = \frac{8\pi^5 k^4}{15c^3h^3}\ (T^4) = 7.57\ \textnormal{x}\ 10^{-16}\ (T^4).$$
The quantity 7.57 x 10$^{-16}$ J m$^{-3}$ K$^{-4}$, known as the radiation constant, is equal to (4$\sigma_{B}$/c), where $\sigma_{B}$ represents the Stefan-Boltzmann constant (5.6704 x 10$^{-8}$ J m$^{-2}$ s$^{-1}$ K$^{-4}$). The total energy densities (*U*) of photon gases with temperatures of 300 K and 2.73 K are 6.13 x 10$^{-6}$ J/m$^{3}$ and 4.02 x 10$^{-14}$ J/m$^{3}$, respectively. The photon densities ($\rho$) in 300 K and 2.73 K radiation fields, which are obtained by dividing eq. (24) by eq. (19), are 2.98 x 10$^{14}$ photons/m$^{3}$ and 3.79 x 10$^{8}$ photons/m$^{3}$, respectively.
Although light is often modeled as an infinite plane wave or a mathematical point, the phenomena of diffraction and interference indicate that a photon has neither an infinite nor a nonexistent width, but something in between [@134; @135]. By analyzing the relationship between the size of silver halide grains and the light-induced darkening of film, Ludwik Silberstein introduced the cross sectional area of Einstein’s light quantum or *light dart* as a useful working hypothesis to quantitatively describe the photographic process [@136; @137; @138]. Silberstein calculated the cross sectional area of a 470 nm photon to be between 8.1 x 10$^{-15}$ m$^{2}$ and 97.3 x 10$^{-15}$ m$^{2}$. I also assume that a photon has a finite width as well as a wavelength and that its geometrical cross section [@139] ($\sigma$) is given by:
$$\sigma = \pi r^2,$$
where *r* is the radius of the photon (see appendix 2). The radius of a photon can be estimated by following the example of Niels Bohr and using a mixture of classical and quantum reasoning. I make use of the fact that all photons, independent of their wavelength, have the same quantized angular momentum (*L* = $\hbar$), and that classically, angular momentum is equal to *m$\omega$$\Gamma$r$^{2}$* or $\omega$*I*. *I* is the moment of inertia and $\Gamma$ is a geometrical factor that relates *I* to *mr$^{2}$* such that $\Gamma$ = *I*/*mr$^{2}$*. I am assuming that $\Gamma$ is equal to unity, which would be correct if the photon consisted of a point mass at the end of a mass-less string of length *r*.
By using *E* = *m*c$^{2}$ = $\hbar\omega$ and assuming that the equivalent mass (*m*) of a photon that interacts with matter is given by $\hbar\omega$/c$^{2}$, then the radius (*r*) of a photon will be equal to $\sqrt{\frac{\hbar}{m\omega}}$ = $\sqrt{\frac{\hbar c^{2}}{\hbar\omega^{2}}}$ = $\sqrt{\frac{c^{2}}{\omega^{2}}}$ = c/$\omega$ = $\frac{1}{k}$ = $\frac{\lambda}{2\pi}$, which is the reciprocal of the angular wave number [@140] and equal to the wavelength of the photon divided by 2$\pi$. Thus, the geometrical cross section, which is related to its angular wave number and wavelength, is given by:
$$\sigma = \pi (\frac{1}{k})^2 = \pi (\frac{\lambda}{2\pi})^2 = \frac{\lambda^2}{4\pi}.$$
According to this reasoning, the cross sections of thermal (300 K) photons and microwave (2.73 K) photons are 7.43 x 10$^{-12}$ m$^{2}$ and 2.78 x 10$^{-7}$ m$^{2}$, respectively. The cross sectional area of 470 nm photons given by eq. (26) is 17.6 x 10$^{-15}$ m$^{2}$, consistent with the values of 470 nm photons given by Silberstein [@136; @137].
According to eq. (16), the collision rate ($\frac{\textit{dn}}{\textit{dt}}$) between a moving charged particle and photons in a photon gas is dependent on the speed of the charged particle. After factoring in the photon densities and the cross section of the peak photons, I find that the collision rates are equal to (2214.14 collisions/m)*v* and (105.36 collisions/m)*v* for 300K and 2.73 K radiation fields, respectively, where *v* is the speed of the charged particle relative to the observer who experiences the photon gas as being isotropic. For a given speed, the collision rate increases with the temperature of the radiation field. This is because the photon density increases with the third power of the temperature while the geometrical cross section decreases with the first power of the temperature. In a 300 K thermal radiation field, at speeds approaching the speed of light, the collision rate will be about 6.64 x 10$^{11}$ s$^{-1}$, while it will be about 3.16 x 10$^{10}$ s$^{-1}$ for a 2.73 K microwave radiation field.
The velocity-dependent viscous force (**F**$_{Dopp}$) exerted on a moving electron by the photons in the photon gas is given by the product of the collision rate and the average velocity-dependent momentum of a photon:
$$\textbf{F}_{Dopp} = -\frac{(\rho \sigma v)(\hbar k_{source})(\frac {1}{4})(\frac {v}{c})}{\sqrt{1 - v^2/c^2}}, \\$$
$$\textbf{F}_{Dopp} = -\frac{(\rho \sigma h/4\lambda_{source})(\frac{v^2}{c})}{\sqrt{1 - v^2/c^2}}.$$
At *v* = 0, there is no net viscous force and the average momenta (h/4*$\lambda_{source}$*) of photons in a thermal radiation field and a microwave radiation field are 1.72 x 10$^{-29}$ kg m/s and 8.86 x 10$^{-32}$ kg m/s, respectively. Since when *v* = 0, the collisions with the photons in the photon gas are random, the electron will exhibit Brownian motion.
We can define the product of *$\rho$* and *$\sigma$* as the linear photon density (*$\rho_{L}$*), replace h in eq. (28) with $\frac{e^{2}}{\epsilon_{o}c\alpha}$, using the definition of the fine structure constant [@138; @139] ($\alpha$), substitute $\mu_{o}$ for 1/$\epsilon_{o}c^{2}$, and replace *$\lambda_{source}$* with *w*/*T* to get,
$$F_{Dopp} = -\frac{[\rho_{L} T e^2\mu_{o}/4w\alpha](v^2)}{\sqrt{1 - v^2/c^2}}.$$
This expression of the viscous force shows explicitly that the viscous force depends on the temperature, the square of the charge of the moving particle, the magnetic permeability of the vacuum, and the fine structure constant, which quantifies the strength of the interaction between a charged particle and the radiation field. The viscous force vanishes, as either the charge of the moving particle or the temperature goes to zero.
The linear photon density ($\rho_{L}$, in m$^{-1}$) is only a function of temperature and can be written as $\frac{\sigma_{B}w^{3} T}{\pi hc^2}$ or $7.375\ $m$^{-1}$ K$^{-1}$ $T$. By combining the constants in eq. 29, the viscous force experienced by a univalent particle can be expressed exclusively in terms of temperature and velocity:
$$\textbf{F}_{Dopp} = -[1.41\ \textnormal{x}\ 10^{-39}\ \textnormal{N}\ \textnormal{s}^{2}\ \textnormal{m}^{-2}\ \textnormal{K}^{-2}]\ T^{2}\frac{(v^2)}{\sqrt{1 - v^2/c^2}}.$$
The magnitude of the friction on a univalent particle caused by the viscous force depends on the velocity of the particle and the square of the temperature. The coefficient of friction of the photon gas is given by $\frac{F_{Dopp}}{v}$ = \[1.41 x 10$^{-39}$ $\textnormal{N}\ \textnormal{s}^{2}\ \textnormal{m}^{-2}\ \textnormal{K}^{-2}]\ T^{2} \frac{(v)}{\sqrt{1 - v^2/c^2}}$. The power dissipated by the photon gas is given by $vF_{Dopp}$. The power dissipated not only depends on the temperature but will increase the temperature and/or the volume of the space composed of the photon gas. Such an effect may have been important in the expansion of the Universe.
The equation of motion that accounts for the temperature and velocity-dependent viscous force experienced by a univalent particle is: $$F_{app} - [1.41\ \textnormal{x}\ 10^{-39}\ \textnormal{N}\ \textnormal{s}^{2}\ \textnormal{m}^{-2}\ \textnormal{K}^{-2}]\ T^{2}\frac{(v^2)}{\sqrt{1 - v^2/c^2}} = m\frac{dv}{dt}$$ which is equivalent to: $$F_{app} + F_{Dopp} = m\frac{dv}{dt}.$$
Any equation that describes motion at velocities close to the speed of light must reduce to Newton’s Second Law for motion at small velocities [@143]. Indeed eq. (31) reduces to Newton’s Second Law when $v <<$ c and $F_{app}$ $>>$ \[1.41 x 10$^{-39}$ $\textnormal{N}\ \textnormal{s}^{2}\ \textnormal{m}^{-2}\ \textnormal{K}^{-2}]\ T^{2}\frac{(v^2)}{\sqrt{1 - v^2/c^2}}$. The integration of eq. (31) with respect to distance yields a quantitative relationship between the apparent mass of a charged particle and energy [@144].
At temperatures greater than 0 K, the photons, which make up the photon gases that occupy all space, act as dilatant, shear-thickened, viscous “non-Newtonian” solutions. While I am not attempting to resurrect the complex and self-contradictory 19th century aether by any means, oddly enough, the rheological concept of dilatancy was first developed by Osborne Reynolds [@38; @145] when he was contemplating the nature of the luminous ether. Quantum electrodynamics handles the viscous force in terms of renormalization (see appendix 3).
Equation (31), which is based on biophysical cell biological analogies [@146] contrasts with Planck’s [@147] relativistic version of Newton’s Second Law, which is:
$$F_{app} = \frac {d}{dt}\frac{mv}{\sqrt{1 - v^2/c^2}}.$$
Depending on the assumptions [@148; @149; @150; @151; @152; @153] used to differentiate eq. (33), the following solutions have been proffered:
$$F_{app}\sqrt{1 - v^2/c^2} = m(\frac {dv}{dt}),\\$$
and
$$F_{app}\sqrt[3]{1 - v^2/c^2} = m(\frac {dv}{dt}).$$
In contrast to eq. (31), where the term that includes the “Lorentz factor” is subtracted from the applied force, in eqs. (34) and (35), the applied force is multiplied by the terms that include the “Lorentz factor.” Equations (31), (34), and (35) all predict that the relation between the impulse needed to accelerate an electron from rest to a given velocity and that velocity will be nonlinear (Fig. 4). However, the opto-mechanical Doppler Effect model based on biophysical cell biological reasoning, further predicts that the applied force necessary to overcome the viscous force in order to accelerate a charged particle from rest to velocity $v$ in time $t$ will be temperature dependent.
![The impulse needed to accelerate an electron from rest to a given velocity at two different temperatures as predicted by the opto-mechanical Doppler model (eq. 31) using the geometrical cross section, two interpretations of the Special Theory of Relativity (eqs. 34 and 35), and the opto-mechanical model using the Thompson scattering cross section (see Appendix 2). Equations (34) and (35) have no temperature dependence.[]{data-label="f4"}](figure4.eps){width="12cm"}
A test of whether the magnitude of the impulse needed to accelerate an electron from rest to a given velocity is influenced by temperature will determine whether the limiting speed of light is a consequence of the viscous force provided by the Doppler-shifted photons that populate every space as predicted by the biophysical cell biologist’s model or due to the relativity of time as predicted by Einstein’s Special Theory of Relativity. The test can be performed by measuring the impulse necessary to accelerate an electron from rest to a given velocity [@154] at different temperatures.
Conclusion
==========
While it is possible that there is more than one mechanism to prevent charged particles from moving faster than the speed of light, the causal, picturesque and testable biophysical cell biologist’s opto-mechanical working hypothesis presented here indicates that light can act as an ultimate speed limit to any charged particle because, at velocities approaching the speed of light, the photon gas that occupies the space through which the particles move is not a “Newtonian” photon gas with a single and trivial viscosity but is a “non-Newtonian”, shear-thickened or dilatant photon gas that becomes infinitely viscous as the velocity of a moving charged particle approaches the speed of light. That is, as a result of the Doppler Effect, light itself, and not the relativity of time, may prevent charged particles from moving faster than the speed of light.
Acknowledgement
===============
I dedicate this paper to my brother Scott Wayne.
Appendix 1
==========
It is possible that Maxwell’s second order wave equation [@50] is not the best starting point for describing relativistic phenomenon since his second order wave equation does not contain a relative velocity term. Consequently I introduced a new form-invariant relativistic wave equation, which is consistant with known phenomena that depend on the relative velocity of the source and observer. Since the derivations of eqs. (31) and (32) critically depend on the introduction of this new relativistic wave equation, here I show that the new relativistic wave equation is consistent with the two fundamental principles upon which Einstein based the Special Theory of Relativity [@67]. I assume that the new relativistic wave equation, which is form invariant and consistent with the principle of relativity, is the equation of motion that describes the properties of light experienced by an electron moving at velocity $v$ relative to the center of momentum of the photon gas, which is at rest relative to the inertial frame of the observer: $$\frac{\partial^2\Psi}{\partial t^2} = \textnormal{c}c'\ \frac{\sqrt{\textnormal{c} - v \cos\theta}}{\sqrt{\textnormal{c} + v \cos\theta}}\ \nabla^2\Psi.$$
I also assume that the following equation is a general plane wave solution to the second order relativistic wave equation given above: $$\Psi = \Psi_{o} e^{i(\textbf{k}_{observer}\cdot\textbf{r}\ -\ \omega_{source}\frac{\sqrt{c - v cos\theta}}{\sqrt{c + v cos \theta}}t)}.$$
By substituting eq. (37) into eq. (36) and taking the spatial and temporal partial derivatives, we can obtain the relativistic dispersion relation: $$\textnormal{c}c' (\frac{\sqrt{\textnormal{c} - v \cos\theta}}{\sqrt{\textnormal{c} + v \cos\theta}}) i^{2}k_{observer}^{2}\Psi = i^{2}\omega_{source}^{2}(\frac{\textnormal{c} - v \cos\theta}{\textnormal{c} + v \cos\theta})\Psi.$$
After canceling like terms, we get: $$\textnormal{c}c'k_{observer}^{2} = \omega_{source}^{2}\frac{\sqrt{\textnormal{c} - v \cos\theta}}{\sqrt{\textnormal{c} + v \cos\theta}}.$$
Since $c'$ $\equiv$ $\frac{\omega_{source}}{k_{observer}}$, the above equation simplifies to: $$\textnormal{c}k_{observer} = \omega_{source}\frac{\sqrt{\textnormal{c} - v \cos\theta}}{\sqrt{\textnormal{c} + v \cos\theta}}.$$
By solving for c, the speed of the wave, we get the relativistic dispersion relation: $$\textnormal{c} = \frac{\omega_{source}}{k_{observer}}\frac{\sqrt{\textnormal{c} - v \cos\theta}}{\sqrt{\textnormal{c} + v \cos\theta}}.$$
The relativistic dispersion relation tells us that while the observed angular wave number of light and consequently its observed momentum varies in a velocity-dependent manner, the speed of light (c) is invariant and independent of the motion of the source or observer, consistent with the principle of the constancy of the speed of light. That is, while the speed of light is independent of the relative velocity between the source and the observer, the relative velocity between the source and the observer results in a “stretching” or “compressing” of the light wave without changing its speed. Since the stretched light waves transfer a smaller quantity of momentum per unit time compared with the compressed light waves, an electron moving through a photon gas with a temperature greater than 0 K experiences a velocity-dependent viscous force.
Appendix 2
==========
The degree of nonlinearity between the applied force and the rate of change in momentum predicted by eqs. (31) and (32) depends quantitatively on the calculated geometrical cross sections of the photons that comprise the photon gas through which the charged particle moves. It is natural to wonder what is the effect of replacing the geometrical cross section of a photon with the Thomson scattering cross section [@155], which is not a geometrical cross section but a measure of the probability that a photon will interact with an electron. The Thomson cross section is used to model the Sunyaev-Zel’dovich effect in which high energy electrons are decelerated by the cosmic microwave background and produce x-rays as a result of inverse Compton scattering [@156]. In contrast to the geometrical cross section, the Thomson scattering cross section is wavelength- and thus temperature-independent. The Thomson scattering cross section ($\sigma_{T}$) is 6.6524586 x 10$^{-29}$ m$^{2}$, which is orders of magnitude smaller than the geometrical cross section of the photon presented in this paper or calculated by Silberstein [@136; @137]. Using the Thomson scattering cross section instead of the geometrical cross section, the formula for the temperature- and velocity-dependent viscous force given by eq. (31) becomes: $$F_{app} - \frac{\sigma_{T} \sigma_{B}}{c^{3}} T^{4}\frac{(v^2)}{\sqrt{1 - v^2/c^2}} = m\frac{dv}{dt}$$ After combining the constants we get: $$F_{app} - [1.40\ \textnormal{x}\ 10^{-61}\ \textnormal{N}\ \textnormal{s}^{2}\ \textnormal{m}^{-2}\ \textnormal{K}^{-4}]\ T^{4}\frac{(v^2)}{\sqrt{1 - v^2/c^2}} = m\frac{dv}{dt}$$
While the temperature dependence is greater in the equation that uses the Thompson scattering cross section than in eq. (31), the coefficient in the equation that uses the Thomson scattering cross section is approximately twenty-two orders of magnitude smaller than the coefficient in eq. (31) that uses the geometrical cross section, and, as shown in Fig. 4, the viscous force that results from the relativistic Doppler effect would only be detectable at velocities infinitessimally close to the speed of light.
Since the acceleration of an electron parallel to the electric field of the incident radiation may also be retarded by the photon gas and as a result of the retardation, dissipate energy in a manner that may affect the scattering cross section, I calculated a scattering cross section using the Larmor formula in a manner following Thomson [@155] and Purcell [@157] except that I substituted $a^{2} = (\frac{eE}{m} + F_{Dopp})^{2}$ for $a^{2} = e^{2}E^{2}/m^{2}$ where $E$ represents the electric field of the incident radiation. The Larmor formula describes the influence of acceleration on the power emitted by a moving charge with mass $m$ and charge $e$. The resulting formula for the scattering cross section ($\sigma_{s}$) is: $$\sigma_{s} = \frac{e^{4}}{6 \pi \epsilon_{o}^{2} m^{2} c^{4}} + \frac{2F_{Dopp}e^{3}}{6 \pi \epsilon_{o}^{2} m^{2} c^{4} E} + \frac{F_{Dopp}^{2}e^{2}}{6 \pi \epsilon_{o}^{2} m^{2} c^{4} E^{2}}$$
The first term is equal to the Thomson scattering cross section, which is both temperature- and velocity-independent. Thus the temperature- and velocity-dependent cross section can be written as: $$\sigma_{s} = \sigma_{T} + \frac{2F_{Dopp}e^{3}}{6 \pi \epsilon_{o}^{2} m^{2} c^{4} E} + \frac{F_{Dopp}^{2}e^{2}}{6 \pi \epsilon_{o}^{2} m^{2} c^{4} E^{2}}$$
When either the temperature or velocity approaches zero, the above scattering cross section reduces to the Thomson scattering cross section. While the temperature- and velocity-dependent cross section given above is greater than $\sigma_{T}$, it is still small enough that, if it were an accurate representation of the cross section of a photon, then the viscous force caused by the relativistic Doppler effect would only be detectable at velocities infinitessimally close to the speed of light.
The derivation of the Thomson scattering cross section assumes that the electric field produced by the electron is symmetric. However, the symmetry of the electric field produced by a moving electron may depend on its velocity. For example, the measured value of the electric field, as well as its divergence, may be greater in front of a moving charge than behind it if the measurements occur over a finite time. I postulate that at a distance $r$, the electric field produced by a moving charge ($q_{m}$) may be given by:
$$\textbf{E} = \frac{1}{4 \pi \epsilon_{o}} \frac {1 - \frac{vcos \theta}{c}}{\sqrt{1-\frac{v^{2} cos^{2} \theta}{c^{2}}}}\frac{q_{m}}{r^{2}}\hat{\textbf{r}}.$$
Such an asymmetrical distribution of the electric field along the direction of motion contrasts with the symmetrical velocity-dependent distribution of the electric field postulated by the Special Theory of Relativity [@158]. By taking into consideration the possibility of a velocity-dependent asymmetry in the electric field of a moving charge it may be possible to derive a more relevant scattering cross section to model the viscous force.
Standard scattering theory based upon interaction cross sections are unable to account for a detectable temperature dependence in the impulse-velocity curve. Consequently, a discovery of a temperature dependence in an experimentally-obtained impulse-velocity curve would support the validity of using the geometrical cross section as well as reify the proposal that light itself prevents charged particles from moving faster than the speed of light. On the other hand, the demonstration of a temperature independence in an experimentally-obtained impulse-velocity curve would support the assumption that the photon is best modeled as a mathematical point with a geometrically undefined interaction cross section and it is the relativity of time that prevents any particle from moving faster than the speed of light.
Appendix 3
==========
I consider the mass of a charged particle moving through space composed of a photon gas to be constant and invariant. However, the effect of Doppler-shifted photons on resisting the movement of an electron moving through a photon gas consisting of real photons can be interpreted in terms of an increase in the effective mass of an electron that occurs as a consequence of the dressing or renormalizing of the electron as it interacts with the virtual photons of the quantum electrodynamical vacuum [@159]. The velocity- and temperature-dependent ratio of the apparent mass that results from the viscous force to the constant mass ($m$) is given by the following equation: $$\frac{m_{apparent}}{m} = \frac{F_{applied}}{F_{applied} + F_{Doppler}}$$ After rearranging, we get the apparent mass or effective mass of an electron: $$m_{apparent} = \frac {m}{1 + \frac{F_{Dopp}}{F_{applied}}}$$ where the second term in the denominator, which is velocity- and temperature-dependent, is analogous to the self-energy of the dressed electron.
[99]{}
H. Frauenfelder, P. G. Wolynes and R. H. Austin, *Rev. Mod. Phys.* **71**, S419 (1999). W. Ostwald, *An Introduction to Theoretical and Applied Colloid Chemistry. The World of Neglected Dimensions*, John Wiley $\&$ Sons, New York, 1922. A. Frey-Wyssling, *Macromolecules in Cell Structure*, Harvard University Press, Cambridge, 1957. T. H. Huxley, in *Lay Sermons, Addresses, and Reviews* , D. Appleton and Co., New York, 1890, p. 120. J. Tyndall, in *Fragments of Science* Vol. II, D. Appleton and Co., New York, 1898, p. 135. A. V. Hill, *Living Machinery*, Harcourt, Brace and Co., New York, 1927. D. Burns, *An Introduction to Biophysics*, Macmillan, New York, 1929). R. Höber, *Physical Chemistry of Cell and Tissues*, Blakiston, Philadelphia, 1945. A. Szent-Györgyi, *Nature of Life*, Academic Press, New York, 1948. A. Szent-Györgyi, *Introduction to a Submolecular Biology*, Academic Press, New York, 1960. A. Frey-Wyssling, ed, *Deformation and Flow in Biological Systems*, North-Holland Publishing Co., Amsterdam, 1952. M. Tazawa, *Protoplasma* **48**, 342 (1957). L. V. Heilbrunn, *The Viscosity of Protoplasm*. Protoplasmatologia, Springer-Verlag, Vienna, 1958. N. Kamiya, *Protoplasmic streaming*. Protoplasmatologia. Bd 8,3a, Springer-Verlag, Vienna, 1959. N. Kamiya, *Annu. Rev. Plant Physiol.* **11**, 324 (1960). N. Kamiya, *Annu. Rev. Plant Physiol.* **32**, 205 (1981). N. Kamiya, *Bot. Mag. Tokyo* **99**, 441 (1986). G. von Békésy, *Experiments in Hearing*, McGraw-Hill, New York, 1960. D. M. Needham, *Machina Carnis. The Biochemistry of Muscular Contraction in its Historical Development*, Cambridge University Press, Cambridge, 1971. J. C. Eccles, *The Understanding of the Brain*, McGraw-Hill, New York, 1973. A. F. Huxley, *Reflections on Muscle*, Princeton University Press, Princeton, 1980. R. K. Clayton, *Photosynth. Res.* **19** , 207 (1988). K. J. Niklas, *Plant Biomechanics. An Engineering Approach to Plant Form and Function*, University of Chicago Press, Chicago, 1992. H. E. Huxley, *Ann. Rev. Physiol.* **58**, 1 (1996). G. Feher, *Annu. Rev. Biophys. Biomol. Struct.* **31**, 1 (2002). P. S. Nobel, *Physicochemical and Environmental Plant Physiology*, Academic Press, San Diego, 2009. R. Wayne, *Plant Cell Biology. From Astronomy to Zoology*, Elsevier Academic Press, Amsterdam, 2009. T. Young, *A Course of Lectures on Natural Philosophy and the Mechanical Arts*, Printed for Joseph Johnson, London, 1807. R. Brown, in *The Miscellaneous Botanical Works of Robert Brown* Vol. I, Robert Hardwicke, London, 1866, p. 463. R. Brown, in *The Miscellaneous Botanical Works of Robert Brown* Vol. I, Robert Hardwicke, London, 1866, p. 479. J. R. Mayer, in W. R. Grove, H. Helmholtz, J. R. Mayer, M. Faraday, J. Liebig, W. B. Carpenter, and E. L. Youmans, *The Correlation and Conservation of Forces. A Series of Expositions*, D. Appleton and Co., New York, 1868, p. 316. J. L. M. Poiseuille, *Experimental Investigations upon the Flow of Liquids in Tubes of Very Small Diameter* Translated by W. H. Herschel. Rheological Memoirs. E. C. Bingham, ed. Vol. 1, Number 1, Lancaster Press, Inc., Easton, 1940. A. Fick, *Phil. Mag.* **10**, 30 (1855). F. Cajori, *Sir Isaac Newton’s Mathematical Principles of Natural Philosophy and his System of the World*, University of California Press, Berkeley, 1946. H. Bondi, in *Let Newton Be! A New Perspective on his Life and Works* eds. J. Fauvel, R. Flood, M. Shortland, and R. Wilson, Oxford University Press, Oxford, 1988, p. 241. R. S. Westfall, *Science* **179**, 751 (1973). G. G. Stokes, in *Mathematical and Physical Papers* Vol. III, Cambridge University Press, Cambridge, 1922, p. 1. O. Reynolds, *Phil. Mag. Ser. 5.* **20**, 469 (1885). H. Freundlich, in *A Symposium on the Structure of Protoplasm* ed. W. Seifriz, Iowa State College Press, Ames, 1942. p. 85. H. Freundlich and W. Seifriz, *Z. phys. Chem.* **104**, 233 (1923). W. Seifriz, *J. Rheology* **1**, 261 (1930). W. Seifriz, *Protoplasm*, McGraw-Hill, New York, 1936. W. Seifriz and J. Plowe, *J. Rheology* **2**, 263 (1931). N. Kamiya and W. Seifriz, *Exp. Cell Res.* **6**, 1 (1954). N. Kamiya, *Protoplasma* **45**, 513 (1956). N. Kamiya, *Annu. Rev. Plant Physiol. Plant Mol. Biol*. **40**, 1 (1989). N. Kamiya and K. Kuroda, Proc. IVth Intern. Congr. Rheology. Part 4. Symp. Biorheol., John Wiley, New York, 1965, p. 157. N. Kamiya and K. Kuroda, *Biorheology* **10**, 179 (1973). M. Tazawa, *Cell Struc. Funct.* **24**, 55 (1999). J. C. Maxwell, *Phil. Trans. Roy. Soc. Lond.* **155**, 459 (1865) . J. C. Maxwell, *A Treatise on Electricity and Magnetism*, Dover, New York, 1954. O. Lodge, *Modern Views of Electricity*, Macmillan, London, 1889. G. J. Stoney, *Phil. Mag. Ser. 5.* **38**, 418 (1894). J. Larmor, *Aether and Matter*, Cambridge University Press, Cambridge, 1900. W. Kaufmann, *The Electrician* November 8, 1901. *48*, 95 (1901). E. Whittaker, *A History of the Theories of Aether and Electricity. The Classical Theories*, Thomas Nelson and Sons, London, 1951. J. J. Thomson, *The Corpuscular Theory of Matter*, Archibald Constable $\&$ Co., London, 1907. A. Bennett, R. Heikes, P. Klemens, A. Maradudin and S. Banigan, *Electrons on the Move*, Walker and Co., New York, 1964. E. A. Davis and I. Falconer, *J. J. Thomson and the Discovery of the Electron*, Taylor $\&$ Francis, London, 1997. I. Falconer, *Physics Education* **32**, 226 (1997). J. J. Thomson, *Phil. Mag. Ser 5.* **44**, 293 (1897). J. J. Thomson, *Phil. Mag. Ser. 5.* **11**, 229 (1881). J. J. Thomson, *Recollections and Reflections*, Macmillan, New York, 1937. H. A. Lorentz, *Proc. Roy. Netherlands Acad. Arts Sci.* **6**, 809 (1904). H. A. Lorentz, *Problems of Modern Physics*, Ginn and Co., Boston, 1927. H. A. Lorentz, *The Theory of Electrons*, Dover, New York, 1952, 2nd ed. A. Einstein, in *The Collected Papers of Albert Einstein*. Vol. 2. English Translation, Princeton University Press, Princeton, 1989, p. 140. M. Jammer, *Concepts of Mass in Classical and Modern Physics*, Dover, New York, 1961. M. Born, *Einstein’s Theory of Relativity*, Dover, New York, 1962. R. P. Feynman, R. B. Leighton and M. Sands, *The Feynman Lectures on Physics* Vol I, Addison-Wesley, Reading, 1963. H. Bondi, *Relativity and Common Sense. A New Approach to Einstein*, Dover, New York, 1964. D. Bohm, *The Special Theory of Relativity*, W. A. Benjamin, New York, 1965. T. R. Sandlin, *Amer. J. Phys.* **59**, 1032 (1991). J. T. Cushing, *Amer. J. Phys.* **49**, 1133 (1981). A. K. Wr$\acute{\textnormal{o}}$blewski, *Acta Physica Polonica B* **37**, 11 (2006). C. G. Adler, *Amer. J. Phys.* **55**, 739 (1989). N. D. Mermin, *It’s about Time. Understanding Einstein’s Relativity*, Princeton University Press, Princeton, 2005. S. Bais, *Very Special Relativity*, Harvard University Press, Cambridge, 2007. L. B. Okun, *Amer. J. Phys.* **77**, 430 (2009). M. P. Sheetz, R. Chasan and J. A. Spudich, *J. Cell Biol.* **99**, 1867 (1984). T. Shimmen, *Bot. Mag. Tokyo* **101**, 533 (1988). K. Svoboda and S. M. Block, *Cell* **77**, 773 (1994). K. Ito, M. Ikebe, T. Kashiyama, T. Mogami, T. Kon and K. Yashimoto, *J. Biol. Chem.* **282**, 19534 (2007). N. Kamiya and M. Tazawa, *Protoplasma* **46**, 394 (1956). J. Dainty and B. Z. Ginzburg, *Biochim. Biophys. Acta* **79**, 102 (1964). M. Tazawa and N. Kamiya, *Ann. Rep. Biol. Works Fac. Sci. Osaka Univ.* **13**, 123 (1965). A. Finkelstein, *Water Movement through Lipid Bilayers, Pores, and Plasma Membranes. Theory and Reality*, John Wiley $\&$ Sons, New York, 1987. J. I. Schroeder, *J. Gen. Physiol.* **92**, 667 (1988). R. Wayne and M. Tazawa, *Protoplasma \[Suppl. **2**\]*, 116 (1988). P. B. Goodwin, *Planta* **157**, 124 (1983). G. Blobel, *Proc. Natl. Acad. Sci. USA* **77**, 1496 (1980). G. E. Palade, *Science* **189**, 347 (1975). J. E. Rothman, *Harvey Lectures* **86**, 65 (1992). R. W. Schekman, *Harvey Lectures* **90**, 41 (1996). W. Haupt, *Ann. Rev. Plant Physiol.* **33**, 204 (1982). S. Takagi, E. Kamitsubo and R. Nagai, *Protoplasma* **168**, 153 (1992). K. Luby Phelps, D. L. Taylor and F. Lanni, *J. Cell Biol.* **102**, 2015 (1986) . E. Kamitsubo, M. Kikuyama and I. Kaneda, *Protoplasma \[Suppl. **1**\]*, 10 (1988). S. Inoué, in *Biophysical Science-A Study Program* eds. J. L. Oncley, F. O. Schmitt, R. C. Williams, M. D. Rosenberg and R. H. Holt, John Wiley $\&$ Sons, Inc., New York, 1959, p. 402. R. B. Nicklas, *J. Cell Biol.* **97**, 542 (1983). P. K. Hepler, *J. Cell Biol.* **100**, 1363 (1985). M. A. Sleigh, *The Biology of Cilia and Flagella*, Macmillan, New York, 1962. R. Kamiya and G. B. Witman, *J. Cell Biol.* **98**, 97 (1984). Y. Masuda, *Bot. Mag. Tokyo* Special Issue **1**, 103 (1978). J.-P. Métraux and L. Taiz, *Plant Physiol.* **61**, 135 (1978). A. Okamoto-Nakazato, *J. Plant Res.* **115**, 309 (2002). T. E. Proseus and J. S. Boyer, *Ann. Bot.* **98**, 93 (2006). E. M. Purcell, *Amer. J. Phys.* **45**, 3 (1977). O. Reynolds, in *Papers on Mechanical and Physical Subjects* Vol. II. (1881-1900), Cambridge University Press, Cambridge, 1901, p. 51. N. Kamiya and K. Kuroda, *Protoplasma* **50**, 144 (1958). M. Tazawa and U. Kishimoto, *Plant Cell Physiol.* **9**, 361 (1968). R. E. Williamson, *J. Cell Sci.* **17**, 655 (1975). S. Vogel, *Life in Moving Fluids. The Physical Biology of Flow*, Princeton University Press, Princeton, 1981. M. W. Denny, *Air and Water. The Biology and Physics of Life’s Media*, Princeton University Press, Princeton, 1993. M. Planck, *Theory of Heat*, Macmillan, New York, 1949. Many biophysical cell biologists are familiar with microscopes based on the Doppler Effect (J. Earnshaw and M. Steer, *Proc. Roy. Micro. Soc.* **14**, 108 (1979). H. E. Ives and G. R. Stillwell, *J. Opt. Soc. Am.* **28**, 215 (1938). My treatment is not based on the common assumption that an electron is a mathematical point. Even the existence of a mathematical point is nothing more than an unproven definition used by Euclid to build his system of geometry; and the success of geometry does not prove that a mathematical point exists in reality (J. L. Synge, *Science, Sense and Nonsense* W. W. Norton $\&$ Co., New York, 1951). After observing a “point” under the microscope, Robert Hooke (R. Hooke, *Micrographia* Printed by Jo. Martyn and Ja. Allestry. London, 1665) found that the “Point of a Needle \[which\] is commonly reckon’d for one…if view’d with a very good Microscope… appears a broad, blunt, and very irregular end; not resembling a Cone, as is imagin’d….” Likewise, while the idea of defining an elementary particle as a mathematical point is a good starting point in science, it too is making use of an unproven definition and better resolution of the particle in the real world may lead to a visualization of its extension. In this paper, I assume that all elementary particles in reality have extension and moving particles have a “front” and a “back”. The assumption that an electron has extension requires the additional assumption that the charge is indivisible. Since the light source, mirrors and detector in the interferometer are all in the same inertial frame, in the Michelson-Morley experiment *v* = 0 and the speed of light given by eq. 4 would be the same in any and all directions. A. A. Michelson and E. W. Morley, *Amer. J. Sci.* **34**, 333 (1887). A. A. Michelson, *Light Waves and their Uses*, University of Chicago Press, Chicago, 1907. M. Planck, *Scientific Autobiography and Other Papers*, Williams $\&$ Norgate, London, 1950. A. Einstein, 1907. in *The Collected Papers of Albert Einstein*. Vol. 2. English Translation, Princeton University Press, Princeton, 1989, p. 232. S. Reinhardt, G. Saathoff, H. Buhr, L. A. Carlson, A. Wolf, D. Schwalm, S. Karpuk, C. Novotny, G. Huber, M. Zimmermann, R. Holzwarth, T. Udem, T. W. Hänsch and G. Gwinner, *Nature Physics* **3**, 861 (2007). A. Einstein, 1917. in *The World of the Atom*. eds. H. A. Boorse and L. Motz, Basic Books, New York, 1966, p. 888. Later in this paper, the momentum of the source will be estimated by the momentum of the photons at the peak of the black body distribution curve. A. H. Compton, *Phys. Rev.* **21**, 483 (1923). E. Feenberg and H. Primakoff, *Phys. Rev.* **73**, 449 (1948). This analysis assumes that the photon is re-emitted isotropically [@125]. If the radiation is reflected or emitted at the same angle that it is absorbed, the viscous force would be twice as large. Assuming isotropy in the center of moment frame of the photon gas, where the linear momentum coming from any direction $\hbar k$($\theta$,$\varphi$) = $\hbar k$(0,0), the total linear momentum coming from all directions is $\hbar k$ = $\int^{2\pi}_{0}$ $\int^{\pi}_{0}$ $\hbar k$($\theta$,$\varphi$) sin$\theta$ d$\theta$ d$\varphi$ = $4\pi\hbar k(0,0)$. The total linear momentum coming from all directions per unit area per unit time is $\hbar k$ $\int^{2\pi}_{0}$ $\int^{\frac{\pi}{2}}_{0}$ cos $\theta$ sin $\theta$ d$\theta$ d$\varphi$ = $\frac{\hbar k}{4}$. W. Nernst, *The New Heat Theorem. Its Foundations in Theory and Experiment*, Dover, New York, 1969. This way of thinking also applies to neutral particles, including neutrons and neutrinos that have a magnetic moment that may form an electrical dipole that can couple to the radiation field. It would not apply to uncharged particles without a magnetic moment. Let *x* = $\frac{hc}{\lambda k \textit{T}}$. $\int$ $\frac{x^{3}}{[exp(\textit{x})-1]}$ *dx* = $\frac{\pi^{4}}{15}$. H. A. Lorentz, *Nature* **113**, 608 (1924). R. Wayne, in, R. Wayne, *Light and Video Microscopy*, Elsevier Academic Press, Amsterdam, 2009, p. 277. L. Silberstein, *Phil. Mag. Ser. 6* **44**, 257 (1922). L. Silberstein and A. P. H. Trivelli, *Phil. Mag. Ser. 6* **44**, 956 (1922). A. P. H. Trivelli and L. Richter, *Phil. Mag. Ser. 6* **44**, 252 (1922). While these geometrical cross sections appear large, using eq. (25), the geometrical cross section calculated for a 10 MeV photon would be 1.23 x 10$^{-27}$ m$^{2}$ or 0.123 barn, within the range of the experimentally determined photon cross sections. The cross section is typically a measure of the probability that any given reaction will occur and the total cross section is a measure of the probability that all possible reactions will occur. The cross sections for individual processes that make up the total cross sections vary by many orders of magnitude and may be less than, equal to or greater than the geometrical cross section. Here I assume that a charged particle in thermal equilibrium with the black body radiation field has a resonance for photons in the radiation field with every possible angular wave number and thus the probability of an electron interacting with the radiation field is unity. Consequently, the effective cross section equals the geometrical cross section. This calculation can also be based on the fact that light radiated from an object provides that object with linear momentum antiparallel to the direction of radiation (G. N. Lewis, *Phil. Mag. Ser. 6.* **16**, 705 (1908)). Semi-classically, the equivalent momentum of the photon of light is equal to *m**v***. Since the photon travels at the speed of light (c), its equivalent momentum is given by *m*c. According to quantum theory, the momentum of the photon is given by $\hbar k$. By equating the classical and quantum descriptions of momentum, the equivalent mass of a photon is given by the absolute value of $\hbar$*k*/c, which is equal to $\hbar\omega$/c$^{2}$. Friedrich Hasenöhl derived the relationship, *E* $\approx m$c$^{2}$, entirely based on classical reasoning making use of Maxwell’s light pressure and equating the Poynting vector to the momentum vector multiplied by c$^{2}$ (P. Lenard, *Great Men of Science. A History of Scientific Progress*, G. Bell and Sons, London, 1933; W. Pauli, Theory of Relativity, Dover, New York, 1958). R. P. Feynman, *Quantum Electrodynamics*, W. A. Benjamin, New York, 1962. R. P. Feynman, *QED. The Strange Theory of Light and Matter*, Princeton University Press, Princeton, 1985. P. Frank, *Einstein. His Life and Times*, Alfred Knopf, New York, 1947. Since the integral of force with respect to distance gives energy, integrating eq. 31 with respect to distance gives the relationship between energy the apparent mass of a charged particle at velocities close to c. O. Reynolds, *Nature* **33**, 429 (1886). J. A. Thomson, *Introduction to Science*, Henry Holt and Company, New York, 1911. M. Planck, *Verh. der Deutschen Physikalische Gesellschaft* **8**, 136 (1906). G. N. Lewis, *Science* **30**, 84 (1909). G. N. Lewis and R. C. Tolman, *Phil. Mag. Ser. 6.* **18**, 510 (1909). R. C. Tolman, *Phil. Mag. Ser. 6.* **21**, 296 (1911a). R. C. Tolman, *Phil. Mag. Ser. 6.* **22**, 458 (1911b). R. C. Tolman, *Phil. Mag. Ser. 6.* **23**, 375 (1912). D. Kleppner and R. J. Kolenkow, *An Introduction to Mechanics*, McGraw-Hill, New York, 1973. W. Bertozzi, *Amer. J. Phys.* **32**, 551 (1964). J. J. Thomson, *The Corpuscular Theory of Matter*, Archibald Constable $\&$ Company, London, 1907. R. A. Sunyaev and Ya. B. Zel’dovich, *Ann. Rev. Astron. Astrophys.* **18**, 537 (1980). E. M. Purcell, *Electricity and Magnetism. Second Edition*, McGraw-Hill, Boston, 1985. W. G. V. Rosser, *Classical Electromagnetism via Relativity*, Plenum Press, New York (1968). R. D. Mattuck, *A Guide to Feynman Diagrams in the Many-Body Problem*, McGraw-Hill, Boston, 1967.
[^1]: Send any remarks to [[email protected]]{}
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
[**Abstract**]{}
We study the action of space-time symmetries on quantum fields in the presence of small departures from locality determined by dynamical gravity. It is shown that, under such relaxation of locality the symmetries of the theory cannot be described within the usual framework of Lie algebras but rather in terms of non-co-commutative Hopf algebras or “quantum groups". Similar “quantizations" of space-time symmetries are expected to emerge in the low-energy limit of certain quantum gravity models and have been used to describe the symmetries of various non-commutative space-times. Our result provides an intuitive characterization of the mechanism that could lead to the emergence of deformed co-products in models of quantum relativistic symmetries.
author:
- Michele Arzano
title: 'Quantum fields, non-locality and quantum group symmetries'
---
Introduction
============
Symmetries play a prominent role in theoretical physics as they allow to establish constraints and make predictions for physical processes without knowing the detailed structure of the system under study. Exact symmetries, however, are rarely realized in Nature. Indeed it seems that our present knowledge of particle physics, up to the energies probed by experiments so far, owes much to a systematic, “controlled", symmetry breaking. Moreover as we gain sensitivity in the probes which test the symmetry principles we assume as fundamental, often such exact symmetries appear only as approximations, at leading order in some physical scale, of more fundamental ones. In some cases the need for such generalizations is suggested by a radical incompatibility between the framework in which the symmetries are described and certain fundamental aspects of the theory at hand. A nice example of this is given by the transition from Galilean to Lorentz/Poincaré relativistic symmetries: the old Galilean framework for the description of symmetries was at odds with the intrinsic Lorentzian nature of Maxwell’s theory of electro-magnetism.\
Nowadays theoretical physics is facing a puzzle which might reflect a similar state of affairs. Local quantum field theory (LQFT), even if extremely successful as an effective field theory in its range of validity, seem to grossly overcount the number of degrees of freedom in a given region of space. In fact “holographic" arguments predict a non-extensive scaling of the number of degrees of freedom for a given region of space determined by the area of the region [@Bousso:2002ju] while the degrees of freedom of local quantum fields scale with the volume. The emergence of non-locality is usually indicated as the cause for such tension. Indeed, according to a common intuition (see [@Giddings:2005id; @ArkaniHamed:2007ky] for recent discussions), locality (or microscopic causality) should be an approximate concept in quantum gravity since once the background metric becomes dynamical and is allowed to fluctuate the notion of space-like separation of two events potentially loses its meaning. Our description of particle physics in terms of local field theory thus relies on the assumption that in an ideal setting even if an intrinsic non-locality is present its negligibly small effects will become important only in the ultraviolet where the effective description is supposed to break down anyway. This expectation, however, turns out to be wrong [@ArkaniHamed:2007ky] when for example such tiny effects are amplified by a very large number of states. In these special cases the knowledge of how our effective theory is modified by non-locality becomes of vital importance.\
In this Letter we argue that there is a qualitative difference between usual LQFT and quantum fields in the presence of an intrinsic non-locality. In fact, while in the former case external space-time symmetries are described by the action of a Lie algebra on the asymptotic free states, in the presence of deviations from locality the characterization of such symmetries requires the use of non-trivial Hopf algebras known as “quantum groups".\
In the next Section we will briefly recall how symmetries are described in the framework LQFT with particular emphasis on the relation between locality and the additive action of symmetry generators on asymptotic states. In Section III we present our main argument, namely that the failure of strict locality requires a description of the symmetries of the theory in terms of non-cocommutative Hopf algebras (“quantum groups"), and we link our considerations to specific models of quantum group symmetries that have been studied in the literature. The last Section contains a summary and outlook.
Symmetries and local quantum fields
===================================
Let us start by recalling the notion of locality (micro-causality) in quantum field theory and its implications for the symmetries of the theory.\
Strictly speaking a field operator in LQFT, $\phi(x)$, is an “operator valued distribution". This means that the corresponding operator acting on the Hilbert space of the theory is obtained by smearing $\phi(x)$ with an appropriate $C^{\infty}$ test function $$\phi(f)=\int \phi(x) f(x)\, dx\, .$$ If the function $f$ vanishes outside a bounded region $\phi(f)$ is a [*localized*]{} operator, if $f$ does not vanish but is fast decreasing with all its derivatives then $\phi(f)$ is a [*quasi-local*]{} operator [@Haag:1992hx]. A localized operator is said to be [*local*]{} iff $$\label{loc2}
[\phi(f),\phi(g)]=0$$ when the supports of the test functions $f$ and $g$ are space-like separated. Now consider the translated operator $$\phi(f;x)=U(x)\phi(f)U^{-1}(x)\, ,$$ due to (\[loc2\]) the commutator $$\label{comm4}
[\phi(f; t,\vec{x}_1),\phi(g; t ,\vec{x}_2)]$$ vanishes for some finite value of $|\vec{x}_1-\vec{x}_2|$ if $\phi(f)$ and $\phi(g)$ are localized operators. On the other hand the commutator (\[comm4\]) for quasi-local operators [*does not vanish but falls off to zero faster than any inverse power of the spatial separation $|\vec{x}_1-\vec{x}_2|$*]{} [@Haag:1992hx].\
The construction of the asymptotic states of a general LQFT relies exclusively on quasi-local operators. Indeed, in the Haag-Ruelle formalism [@Haag:1992hx; @Landau:1970fs], one constructs from appropriately smeared polynomials of the field operators a quasi-local operator $q(f,t)$ which creates a one-particle state $q(f,t)|\,0>=|f>$ with “wavefunction" $<\vec{p}\,|f\,>=f(\vec{p})$ independent of $t$. One can show that $q(f_1,t)...q(f_n,t)|\,0>$ has a strong limits for $t\rightarrow\pm\infty$ leading to the the asymptotic free states $|f_1,...,f_n>_{\mathrm{out}, \mathrm{in}}$ [@Haag:1992hx; @Landau:1970fs]. Under the assumption of [*asymptotic completeness*]{} the collections of $|f_1,...,f_n>_{\mathrm{out}, \mathrm{in}}$ span the entire Hilbert (Fock) space of physical states $\mathcal{F}(\mathcal{H})$ [@Haag:1992hx; @Landau:1970fs]. There will be a unitary operator, the $S$-matrix, such that $|f_1,...,f_n>_\mathrm{out}=S\,|f_1,...,f_n>_\mathrm{in}$. We are interested in the interplay between external (geometrical) symmetries and quantum fields. A key fact is that any symmetry describes certain properties which are preserved by the dynamics and thus [*is fully characterized in terms of its action on the asymptotic, free state, configurations*]{}.\
Let us consider the simple example of a massive real scalar field. A [*symmetry transformation*]{} of the theory is a one-parameter, continous, abelian unitary operator $U(\tau)$ in the space of physical states which commutes with the $S$-matrix and transforms one-particle states into themselves. The symmetry transformation is said to possess a [*generator*]{} if it can be written as $U(\tau)=exp(iG\tau)$ with $G$ a self-adjoint operator. In LQFT such generators act on multiparticle states according to a generalized Leibnitz rule ([*additive*]{} action). This last requirement is intimately related to the notion of locality. To see this we look at how the symmetry generators are characterized in terms of the fundamental field observables.\
Given a local and locally conserved current $j_{\mu}(x)$ one can construct a symmetry generator corresponding to the “formal charge" $Q$. The latter can be defined as the limit $$Q=\lim_{T\rightarrow 0}\lim_{R\rightarrow \infty} j_0(f_R,f_T)$$ of the “partial charge" $$j_0(f_R,f_T)=\int dx f_R(\vec{x}) f_T(x_0) j_0(x)\,.$$ with $f_R$ and $f_T$ appropriate smearing functions. In particular $f_R(\vec{x})$ cuts the tails of the current for large spatial distances and $f_T(x_0)$ averages the current around the point $x_0=0$ (for details see [@Orzalesi:1970tx]). The question is whether or not the formal charge $Q$ defines a symmetry generator $G$. The positive answer to this is given by a fundamental theorem due to Kastler, Robinson and Swieca (KRS) (see [@Orzalesi:1970tx] and references therein) which states that the commutator $[j_0(f_R,f_T),A]$ between the partial charge and any localized or quasi-local operator A is [*independent of $f_R$ and $f_T$*]{} for sufficiently large $R$. In particular this is true for any quasi-local operator $A_{f_i}$ such that $A_{f_i}|\,0>=|\,f_i>$. The KRS theorem allows one to define the action of the generator $G$ associated with the formal charge $Q$ through the [*adjoint action*]{} $$\label{adjas}
GA |\,0>\equiv [Q,A]|\,0>=\lim_{T\rightarrow 0}\lim_{R\rightarrow \infty} [j_0(f_R,f_T),A]|\,0>\, ,$$ as it guarantees that the limit in the last term exist and is independent of the particular choice of smearing functions. One immediate consequence of the definition (\[adjas\]) is that $G|\,0>=0$. Additivity of the action of $G$ immediately follows from the definition (\[adjas\]) and the linearity of the commutator. Such property is also manifest when one writes the generator in terms of the asymptotic creation and annihilation operators $$G=\int d^3\vec{k}\,\eta(\vec{k})\,a^{\dagger}_{\mathrm{in,out}}
(\vec{k})\,a^{\phantom{\dagger}}_{\mathrm{in,out}}(\vec{k})$$ where the kernels $\eta(\vec{k})$ characterize the action of the generator on one-particle states. Indeed the expression above can be derived from the one-particle matrix elements of $G$ $$<\vec{k}|G|\vec{k}'>=\eta(\vec{k})\delta^{(3)}(\vec{k}-\vec{k}')$$ and $$\begin{aligned}
\label{adjo2}
[G,a^{\dagger}_{\mathrm{in,out}}(\vec{k})]&=\eta(\vec{k})a^{\dagger}_{\mathrm{in,out}}(\vec{k})\\
[G,a_{\mathrm{in,out}}(\vec{k})]&=-\eta(\vec{k}) a_{\mathrm{in,out}}(\vec{k})\,.\end{aligned}$$ There is a nice algebraic way to characterize the additivity of a symmetry generator. Let $G$ be an element of the Lie algebra $\mathfrak{g}$ describing the symmetries of the space on which our quantum fields live (in Minkowski space $\mathfrak{g}$ is simply the Poincaré algebra $\mathcal{P}$). The one-particle Hilbert space $\mathcal{H}$ is an irreducible representation of $\mathfrak{g}$. “Multi-particle" (asymptotic) free states are given by appropriately symmetrized tensor products of $\mathcal{H}$. What is the action of $G$ on such states or, in other words, how do we construct representations of $\mathfrak{g}$ on tensor products of $\mathcal{H}$? It turns out that the usual construction of tensor product representation for a Lie algebra $\mathfrak{g}$ is best understood in terms of the universal enveloping (UE) algebra $U(\mathfrak{g})$ associated to $\mathfrak{g}$. In fact, UE algebras are an example of Hopf algebras which in turn are a generalization of standard (unital, associative) algebras. Hopf algebras come equipped with additional structures which, among other things, allow one to properly define tensor product representations of $\mathfrak{g}$. In particular the “co-product" (or co-multiplication) $\Delta$ is a map $\Delta:U(\mathfrak{g})\rightarrow U(\mathfrak{g})\otimes U(\mathfrak{g})$ defined by $$\label{copro}
\Delta(G)=G\otimes 1+1\otimes G$$ where $1$ is the unit element of $U(\mathfrak{g})$. Given two representations of $\mathfrak{g}$, $(\rho_1, \mathcal{H}_1)$ and $(\rho_2, \mathcal{H}_2)$ the tensor product representation $(\rho, \mathcal{H}_1\otimes \mathcal{H}_2)$ is given by $$\rho\equiv (\rho_1\otimes \rho_2) \Delta\, .$$ The co-product (\[copro\]) is just telling us that $G$ acts on a “two-particle“ state of $\mathcal{H}_1\otimes \mathcal{H}_2$ according to the Leibnitz rule i.e. the action of $G$ on such states is [*additive*]{}. An important property of the co-product (\[copro\]) is that it is [*co-commutative*]{} i.e. $$\sigma\circ\Delta=\Delta\circ id$$ with $\sigma: U(\mathfrak{g})\otimes U(\mathfrak{g})\rightarrow U(\mathfrak{g})\otimes U(\mathfrak{g})$ the ”flip" map $\sigma(a\otimes b)=b\otimes a$, $id$ the identity map and $\circ$ the composition of maps. Hopf algebras possessing a co-commutative co-product are called [*trivial*]{}. It is easy to see that co-commutative co-products lead to an additive action of $G$ on multi-particle states[^1]. But that’s not all. The (trivial) Hopf algebra structure of the symmetries is present already at the one-particle level. In fact, the action of $G\in U(\mathfrak{g})$ on the algebra of asymptotic creation and annihilation operators given by (\[adjo2\]) is nothing but the “adjoint action" $$\label{adjact}
\mathrm{ad}_{G}(a_{\mathrm{in,out}})\equiv
((id\otimes S)\Delta(G))\diamond a_{\mathrm{in,out}}
=[G,a_{\mathrm{in,out}}]$$ where $S$ is the antipode map[^2] $S(G)=-G$ and $(F\otimes G)\diamond a=FaG$. This shows how the Hopf algebra structure of the UE $U(\mathfrak{g})$ associated to the Lie algebra of symmetries $\mathfrak{g}$ is hidden behind the familiar “commutator" action of $G$ on linear operators on $\mathcal{F}(\mathcal{H})$. It turns out that there exist “quantum" deformations of UE algebras which lead to non-trivial Hopf algebras which are also known in the literature as [*quantum groups*]{}. In the next section we will discuss how quantum deformations of UE algebras, in the context of quantum field theory, can be related to the presence of an irreducible non-locality.\
Quantum symmetries from quantum fields
======================================
Consider the quantum theory of a massive real scalar field for which a set of asymptotic “in" and “out" states is given. Under the assumption of asymptotic completeness these states span the full Hilbert (Fock) space of the theory $\mathcal{F}(\mathcal{H})$. A unitary $S$-matrix connects the two sets of states. From a “purely" quantum mechanical point of view a symmetry of the theory is a mapping of rays of the Hilbert space which leaves invariant the transition probabilities. According to Wigner’s theorem (see e.g. [@Bargmann:1964zj]) space-time symmetries will be described by unitary operators $U$ on the asymptotic states. Such operators commute with the S-matrix, map one particle states into themselves and leave the vacuum invariant. An infinitesimal transformation will be of the form $U=1+i\tau G$ with $\tau$ an infinitesimal parameter and $G$ the generator of the symmetry. In particular, if we denote the action of the generator $G$ on an operator $A$ defined on $\mathcal{F}(\mathcal{H})$ with $G\vartriangleright A$, one has $<0\,|G\vartriangleright A|\,0>= 0$ [@Bargmann:1964zj]. The properties we described above are the minimal requirements that an external symmetry of our quantum fields has to fulfill.\
We assume now that, according to the results of [@Giddings:2005id; @Dittrich:2006ee; @ArkaniHamed:2007ky], the observables of the theory possess an intrinsic, irreducible, non-locality. In [@Giddings:2005id] it is discussed how, starting from diffeomorphism invariant observables of an effective theory of quantum gravity, one could recover the familiar observables of local quantum field theory. The conclusions reached in [@Giddings:2005id] seem to indicate a fundamental limitation in obtaining such local observables. From a relational point of view in order to “localize" an observable in a diffeo-invariant theory one needs a reference frame given by some dynamical field. The question is whether or not one is able to define a reference frame which in a certain limit reproduces standard local observables of LQFT. It turns out that to do so one has to pick a reference dynamical field which is itself intrinsically non-local [@Dittrich:2006ee]. As discussed in [@ArkaniHamed:2007ky], [*dynamical*]{} gravity is the crucial ingredient which changes the rules of the game. The heuristic argument given in [@ArkaniHamed:2007ky] shows that switching on gravity has the effect of introducing an irreducible error in the measurement of quantum local observables which is [*non-perturbative*]{} in the coupling $G_N$ and is of the order $e^{-r^2/G_N}$ where r is the “size" of the apparatus used in the measurement (or equivalently the spatial separation of two local observables). The non-perturbative nature of the non-local effects discussed in [@ArkaniHamed:2007ky] suggests that, in a quantum gravitational setting, even though a sharp notion of locality is lost, weaker causality properties like those of quasi-local operators can be preserved. Motivated by these considerations we assume that when fluctuations of the background space-time are present [*the only sensible notion of locality in a theory of quantum fields is that of quasi-locality*]{}. As discussed in the previous Section this does not conflict with the construction and existence of asymptotic free states. However the failure of “strict" microscopic causality has deep consequences for the symmetries of the theory. In fact local commutativity is a crucial ingredient in the proof of the KRS theorem (see Section 4.A of [@Orzalesi:1970tx]). The presence of an irreducible non-locality renders void its statement i.e. $[j_0(f_R,f_T),A]$ and consequently $[Q,A]$ are not necessarily independent of $f_R$ and $f_T$ for large $R$. Now, as we saw in the preceding Section, the action of a symmetry generator $G$ is characterized by its associated conserved charge $Q$. The failure of the KRS theorem [*does not*]{} guarantee that (\[adjas\]) consistently defines an operator $G$ associated to the charge $Q$ on the asymptotic states. Once the invariance of the vacuum is taken into account, a necessary condition for (\[adjas\]) to be a consistent definition is that $<0\,|[Q,A]|\,0>=0$ for any quasi-local operator $A$. If the generator of a given symmetry $G$ can not be defined in terms of the “adjoint" action $[Q,A]$ one then has $$\label{nladjo}
0=<0\,|G\vartriangleright A|\,0>\neq <0\,|[Q, A]|\,0>\, .$$ This is somewhat reminiscent of spontaneous symmetry breaking [@Goldstone:1962es] where one has a locally conserved current but for its associated charge $<0\,|[Q,A]|\,0>\neq 0$. The crucial difference is that in our case we want to keep the [*vacuum invariant*]{} under the action of $G$. Thus we see that the presence of an intrinsic non-locality, no matter how mild, requires a generalization of the “adjoint" action $[Q,A]$. Below we will show how [*non-trivial Hopf algebras*]{} naturally provide such a generaliztion.\
Let us consider a charge $Q$ which fails to define an “adjoint" action due to the intrinsic non-locality between the locally conserved current and any quasi-local operator. Taking into account (\[nladjo\]) and specializing to creation operators as in (\[adjo2\]) we can write $$\label{nladj}
<0\,|G\vartriangleright a^{\dagger}(\vec{k})|\,0>\equiv<0\,|[G, a^{\dagger}(\vec{k})]|\,0>+\alpha_1 E^{-1}_p F^{(1)}(\vec{k})+O(E^{-2}_p)$$ where the non-local corrections are given by model-dependent functions of the momentum $\vec{k}$ suppressed by inverse powers of the Planck energy $E_p$. It turns out that [*the “deformed" adjoint action above can be effectively described by the “semiclassical" expansion of the quantum adjoint action of a non-trivial Hopf algebra with deformation parameter $h=E^{-1}_p$*]{}. In particular the non-local behavior in (\[nladj\]) is reproduced by a symmetry generators $G$ belonging to a non-cocommutative Hopf algebra obtained by a deformation of the universal enveloping (UE) algebra $U(\mathfrak{g})$ of a Lie algebra $\mathfrak{g}$. These deformations are known as quantized universal enveloping (QUE) algebras and are one of the most important examples of quantum groups (see e.g [@quantumgr; @Tjin:1991me]). As mentioned at the end of last Section, QUE algebras exhibit non-trivial (non-cocommutative) co-products together with possible additional deformations of the co-algebra sector. The non-trivial co-product of a QUE algebra can be written in “semiclassical" approximation [@Ruegg:1994bk] as $$\Delta(G)=\Delta^{(0)}(G)+h\Delta^{(1)}(G)+O(h^2)$$ with $\Delta^{(0)}(G)=G\otimes1+1\otimes G$, the trivial co-product. Similarly for the deformed antipode one can write $$S(G)=S^{(0)}(G)+h\,S^{(1)}(G)+O(h^2)\, ,$$ with $S^{(0)}(G)=-G$. It is clear now that according to the definition of adjoint action given in (\[adjact\]) the generator belonging to a QUE algebra will act through the “quantum" adjoint action $$\begin{aligned}
%\label{qadjact}
\mathrm{ad}_{G}(a^{\dagger}(\vec{k}))&=((id\otimes S)\Delta(G))\diamond a^{\dagger}(\vec{k})=\\
&[G,a^{\dagger}(\vec{k})]+h\left[((id\otimes S^{(1)})\Delta^{(0)}(G))\diamond a^{\dagger}(\vec{k})+((id\otimes S^{(0)})\Delta^{(1)}(G))\diamond a^{\dagger}(\vec{k})\right]+O(h^2)\end{aligned}$$ which reproduces the “symmetry breaking" of (\[nladj\]) with the leading order terms of the deformed co-product and antipode determined by the model dependent, Planck-scale suppressed, non-local corrections.\
QUE algebras have been studied extensively in recent years as candidate models for “quantum" relativistic symmetries. Two notable examples are the $\kappa$-deformed and $\theta$-“twisted" Poincaré algebras [@Lukierski:1992dt; @Chaichian:2004yh]. Both “quantum algebras" can be viewed as symmetries of different types of non-commutative space-times [@Chaichian:2004yh; @Majid:1994cy; @Agostini:2006nc]. The $\kappa$-Poincaré algebra was originally obtained as a contraction of $U_q(\mathfrak{so}(3,2))$, the quantization of the UE algebra of the Anti-de Sitter algebra, with deformation parameter $q$. In the contraction procedure the deformation parameter acquires dimension of a mass and is denoted by $\kappa$. This type of deformation of the Poincaré algebra has gained popularity as a way to introduce a fundamental (planckian) length $\lambda=1/\kappa$ in a relativistic framework [@AmelinoCamelia:2000mn]. In the last few years it has also been shown how such $\kappa$-symmetries naturally emerge in the description of the low-energy limit of certain $2+1$-dimensional quantum gravity models [@Amelino-Camelia:2003xp; @Freidel:2005me]. The $\kappa$-Poincaré algebra in its most studied version, the so-called “biscrossproduct basis" [@Majid:1994cy], exhibits both deformed co-product and antipode in the boost and translation sector (rotations are left untouched) $$\begin{aligned}
\Delta(P_0)&=&P_0\otimes 1+1\otimes P_0\,\,\,\,\,\,\Delta(P_j)=P_j\otimes 1+e^{-P_0/\kappa}\otimes P_j \nonumber \\
\Delta(N_j)&=&N_j\otimes 1+e^{-P_0/\kappa}\otimes N_j+\frac{\epsilon_{jkl}}{\kappa}P_k\otimes N_l\, . \label{coprod}\end{aligned}$$ and $$\begin{aligned}
S(P_l)&=&-e^{\frac{P_0}{\kappa}}P_l\,\,\,\,\,\,\,S(P_0)=-P_0 \nonumber\\
S(N_l)&=&-e^{\frac{P_0}{\kappa}}N_l+\frac{1}{\kappa}\epsilon_{ljk}e^{\frac{P_0}{\kappa}}P_j M_k \label{anti}\, .\end{aligned}$$ The $\theta$-Poincaré algebra was obtained by “twisting" the co-product of the UE algebra of the Poincaré algebra [@Chaichian:2004yh]. In this case only the co-product for the boost-rotation sector is deformed while the antipodes are the same as in the standard case $$\Delta(M_{\mu\nu})=M_{\mu\nu}\otimes 1+1\otimes M_{\mu\nu}-\frac{1}{2}\theta^{\alpha\beta}[g_{\mu\alpha}(P_{\nu}\otimes P_{\beta}-P_{\beta}\otimes P_{\nu})-g_{\nu\alpha}(P_{\mu}\otimes P_{\beta}-P_{\beta}\otimes P_{\mu})]\, .$$ In the limits $\kappa\rightarrow\infty$ and $\theta\rightarrow 0$ one recovers in both cases the trivial Hopf algebra structure of the UE algebra of the Poincaré algebra. $\theta$ and $\kappa$-deformed quantum fields are currently the subject of active study (see e.g. [@Balachandran:2007vx; @Arzano:2007ef]). Such theories exhibit several non-trivial features, most importantly they seem to lead to interesting behaviors in their multi-particle sectors hinting for possible deviations from usual statistics.
Conclusions
===========
We have discussed how a description of space-time symmetries in terms of quantum groups could arise in quantum field theory when the notion of strict locality is blurred by the effects of dynamical (quantum) gravity. This result provides a physical motivation for the emergence of “non-cocommutative co-products" which characterize the non-trivial Hopf algebra structure of the symmetries of certain non-commutative space-times. Our argument suggests that these frameworks should in principle provide a “finer" resolution than standard effective field theory in describing processes in which the latter ceases to be a good approximation. An important task left for future studies is to investigate the non-local behaviors of different effective quantum gravity models and the relations with their counterparts in terms of space-time quantum group symmetries.
I am indebted to Bianca Dittrich for several stimulating conversations and for comments on a preliminary draft of this letter and to Giovanni Amelino-Camelia for a critical reading of the manuscript. I would also like to thank Florian Koch, Tim Koslowski and Giuseppe Policastro for discussions and useful remarks.\
Research at Perimeter Institute for Theoretical Physics is supported in part by the Government of Canada through NSERC and by the Province of Ontario through MRI.
[99]{}
R. Bousso, Rev. Mod. Phys. [**74**]{}, 825 (2002) \[arXiv:hep-th/0203101\]; A. G. Cohen, D. B. Kaplan and A. E. Nelson, Phys. Rev. Lett. [**82**]{}, 4971 (1999) \[arXiv:hep-th/9803132\]; U. Yurtsever, Phys. Rev. Lett. [**91**]{}, 041302 (2003) \[arXiv:gr-qc/0303023\].
S. B. Giddings, D. Marolf and J. B. Hartle, Phys. Rev. D [**74**]{}, 064018 (2006) \[arXiv:hep-th/0512200\].
N. Arkani-Hamed, S. Dubovsky, A. Nicolis, E. Trincherini and G. Villadoro, JHEP [**0705**]{}, 055 (2007) \[arXiv:0704.1814 \[hep-th\]\].
B. Dittrich and J. Tambornino, Class. Quant. Grav. [**24**]{}, 757 (2007) \[arXiv:gr-qc/0610060\].
R. Haag, “Local quantum physics: Fields, particles, algebras,” [*Berlin, Germany: Springer (1992) 356 p. (Texts and monographs in physics)*]{}
L. J. Landau, Commun. Math. Phys. [**17**]{}, 156 (1970); K. Kraus and L. J. Landau, Commun. Math. Phys. [**24**]{}, 243 (1972).
C. A. Orzalesi, Rev. Mod. Phys. [**42**]{}, 381 (1970).
V. Bargmann, J. Math. Phys. [**5**]{}, 862 (1964).
J. Goldstone, A. Salam and S. Weinberg, Phys. Rev. [**127**]{}, 965 (1962).
H. Ruegg and V. N. Tolstoi, Lett. Math. Phys. [**32**]{}, 85 (1994) \[arXiv:hep-th/9406146\].
V. Chari and A. N. Pressley, “A guide to quantum groups", pp. 667. Cambridge University Press, (1995).
T. Tjin, Int. J. Mod. Phys. A [**7**]{}, 6175 (1992) \[arXiv:hep-th/9111043\].
M. Chaichian, P. P. Kulish, K. Nishijima and A. Tureanu, Phys. Lett. B [**604**]{}, 98 (2004) \[arXiv:hep-th/0408069\]; M. Chaichian, P. Presnajder and A. Tureanu, Phys. Rev. Lett. [**94**]{}, 151602 (2005) \[arXiv:hep-th/0409096\].
J. Lukierski, A. Nowicki and H. Ruegg, Phys. Lett. B [**293**]{}, 344 (1992).
S. Majid and H. Ruegg, Phys. Lett. B [**334**]{}, 348 (1994) \[arXiv:hep-th/9405107\].
A. Agostini, G. Amelino-Camelia, M. Arzano, A. Marciano and R. A. Tacchi, Mod. Phys. Lett. A [**22**]{}, 1779 (2007) \[arXiv:hep-th/0607221\].
G. Amelino-Camelia, Int. J. Mod. Phys. D [**11**]{}, 35 (2002) \[arXiv:gr-qc/0012051\].
G. Amelino-Camelia, L. Smolin and A. Starodubtsev, Class. Quant. Grav. [**21**]{}, 3095 (2004) \[arXiv:hep-th/0306134\].
L. Freidel and E. R. Livine, Phys. Rev. Lett. [**96**]{}, 221301 (2006) \[arXiv:hep-th/0512113\].
A. P. Balachandran, A. Pinzul and B. A. Qureshi, arXiv:0708.1779 \[hep-th\]. A. P. Balachandran, T. R. Govindarajan, G. Mangano, A. Pinzul, B. A. Qureshi and S. Vaidya, Phys. Rev. D [**75**]{}, 045009 (2007) \[arXiv:hep-th/0608179\].
G. Amelino-Camelia and M. Arzano, Phys. Rev. D [**65**]{}, 084044 (2002) \[arXiv:hep-th/0105120\]; M. Arzano and A. Marciano, Phys. Rev. D [**76**]{}, 125005 (2007) \[arXiv:0707.1329 \[hep-th\]\].
[^1]: The definition of an $n$-fold tensor product of representations of $\mathfrak{g}$ can be obtained by simply iterating the definition above.
[^2]: Beside standard multiplication $m$, unit map $\eta$ and the co-product $\Delta$ defined above a Hopf algebra possesses two additional maps, the co-unit $\varepsilon: U(\mathfrak{g})\rightarrow \mathbb{C}$ and the antipode $S: U(\mathfrak{g})\rightarrow U(\mathfrak{g}) $ satisfying the following axioms $$\begin{aligned}
(\Delta\otimes id)\Delta&=(id\otimes \Delta)\Delta~~~~~~~\mathrm{co-associativity}\\
(id\otimes\varepsilon)\Delta&=(\varepsilon\otimes id)\Delta=id~~~~~~~\mathrm{co-unit}\\
m(S\otimes id)\Delta&=m(id\otimes S)\Delta=\eta\circ\varepsilon~~~~~~~\mathrm{antipode}\,.\end{aligned}$$
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We present a method to calculate vertical profiles of particle size distributions in condensation clouds of giant planets and brown dwarfs. The method assumes a balance between turbulent diffusion and sedimentation in horizontally uniform cloud decks. Calculations for the Jovian ammonia cloud are compared with results from previous methods. An adjustable parameter describing the efficiency of sedimentation allows the new model to span the range of predictions made by previous models. Calculations for the Jovian ammonia cloud are consistent with observations. Example calculations are provided for water, silicate, and iron clouds on brown dwarfs and on a cool extrasolar giant planet. We find that precipitating cloud decks naturally account for the characteristic trends seen in the spectra of L- and T-type ultracool dwarfs.'
author:
- 'Andrew S. Ackerman and Mark S. Marley'
title: Precipitating Condensation Clouds in Substellar Atmospheres
---
Introduction
============
The visual appearance and spectrum of every solar system body with an atmosphere depends strongly upon the character and distribution of atmospheric condensates. This is particularly true for the giant planets where optically thick cloud decks dominate the appearance of the planets at most continuum wavelengths in both the reflected solar and the thermal infrared. Condensates also play a role in controlling the spectra of at least some brown dwarfs and most extrasolar giant planets. Indeed one suggested classification scheme (Sudarsky, Burrows & Pinto 2000) for extrasolar planets hinges on the specific atmospheric condensates present. Yet despite the importance of condensates there exists no simple model for predicting the parameters most relevant to radiative transfer: the vertical profile of condensate mass and its distribution over particle size.
Chemical equilibrium models (e.g., Lewis 1969; Fegley & Lodders 1994) predict which species are expected to condense in an atmosphere, yet they provide no guidance as to the expected particle sizes. Other models (e.g., Rossow 1978; Lunine et al. 1989; Carlson, Rossow, & Orton 1988) predict some parameters, but lack a simple, self-consistent recipe for exploring the possible phase space in which clouds might exist.
A single example motivates the need for cloud models in substellar atmospheres. With increasingly later spectral type the warm L-dwarfs become progressively redder in their $J-K$ color (e.g., Kirkpatrick et al. 1999; Mart[í]{}n et al. 1999; Fan et al. 2000). Spectral fitting and models (e.g., Leggett, Allard, & Hauschildt 1998; Chabrier et al. 2000; Marley 2000) demonstrate that this is due to the progressive appearance of more silicate dust in the cooling brown dwarf atmospheres. Yet the cooler T-type brown dwarfs, like Gliese 229B, have blue colors in $J-K$ (e.g., Leggett et al. 1999; Tsvetanov et al. 2000). The spectra and colors of these cool brown dwarfs can only be fit by atmosphere models that assume the silicate dust has settled below the visible atmosphere (e.g., Allard et al. 1996; Marley et al. 1996; Tsuji et al. 1996). Models in which the dust does not settle (Chabrier et al. 2000) produce T-dwarf colors that are at least 2 to 3 magnitudes redder than observed. Marley (2000) has demonstrated that a simple model in which all clouds are a single scale-height thick can explain this behavior, but the assumed distribution was prescribed rather than being calculated from any model physics. Correct modeling of the atmospheres of cooling brown dwarfs and the ultimate assignment of an effective temperature scale to the L-dwarf spectral sequence (e.g., Kirkpatrick et al. 1999; Basri et al. 2000) requires a characterization of clouds. The ideal model would have a small number of free parameters, predict the vertical distribution and particle sizes of the condensates, and yet be simple enough to be included into model atmosphere codes that iteratively search for self-consistent atmospheric structures. No such ideal model yet exists.
We aim to fill this void by presenting a simple model describing precipitating clouds in substellar atmospheres. We limit our treatment to condensation clouds, and hence do not consider photochemically driven hazes likely to appear in illuminated stratospheres. We depart from previous work by explicitly treating the downward transport of raindrops with sizes greater than that predicted from the convective velocity scale. Including rainfall produces clouds of thinner vertical extent, which can better reproduce observations of Jupiter’s ammonia cloud. The resulting model is general enough to be applied to iron and silicate clouds appearing in brown dwarf atmospheres (e.g., objects with effective temperatures, $T_{\rm eff} \sim 1500\ \rm K$) as well as the atmospheres of cool extrasolar giant planets ($T_{\rm eff} \sim 400\,\rm K$) in which water clouds dominate the atmosphere. The few free parameters in the model can produce clouds with dramatically different characteristics; ultimately observations will constrain these parameters and hopefully provide information on the underlying atmospheric dynamics and cloud physics.
In this paper we first summarize previous cloud modeling efforts, then describe the new model. We use the ammonia cloud of Jupiter as a framework for describing the model physics and evaluating the model performance. Finally we illustrate model applications by considering water, silicate, and iron clouds in the atmospheres of brown dwarfs and a cool extrasolar giant planet.
Previous Models {#previous}
===============
A great range of models have been used to represent clouds in the terrestrial atmosphere, which vary in the complexity by which atmospheric dynamics and cloud microphysics are treated. The most detailed models simulate three-dimensional cloud-scale motions and resolve the size distributions of cloud droplets (and the aerosols on which they form) and treat the interactions between dynamics, microphysics, and radiative transfer. The computational demands of such complex models limit their domain sizes to a few kilometers in each dimension. Present global-scale (general circulation) models greatly simplify the representation of clouds by parameterizing cloud-scale motions as well as cloud microphysical processes, and such simplifications lead to profound uncertainties in climate predictions from their simulations. Both types of models, as well as a range of intermediate models, can be considered appropriate for modeling the terrestrial atmosphere by virtue of the wealth of observational data available to constrain them; whether or not the unknowns in such models are uniquely constrained by the data constitutes a debate beyond the scope of this study.
The relative scarcity of observational data for clouds in extraterrestrial atmospheres is far less constraining. Leading uncertainties include the characteristics of atmospheric dynamics and the populations of nuclei upon which cloud droplets form. Hence, we consider it appropriate to model extraterrestrial clouds using much simpler treatments.
Perhaps the simplest approach to modeling clouds is through a Lagrangian parcel model, in which the base of a cloud appears where the adiabatic cooling of an air parcel in an updraft results in saturation (ignoring any supersaturation associated with barriers to cloud droplet formation). Further cooling condenses vapor in excess of saturation onto cloud particles. The particles grow through condensation and coalescence until their sedimentation velocities exceed the updraft speed, and then fall out of the parcel. A number of problems arise in the formulation of updraft parcel models, among them: ignoring parcels in downdrafts, treating the mixing between parcels, treating the source of condensates into a parcel due to sedimentation from above, and determining updraft speeds.
Another simple approach, which we employ here, is through a one-dimensional Eulerian framework, in which turbulent diffusion mixes a condensable vapor upwards, while maintaining a constant mixing ratio (equivalently, mole fraction) below the cloud. Temperature and hence the saturation mixing ratio in the air column decrease with altitude, and the cloud base again appears where the saturation mixing ratio matches the sub-cloud mixing ratio. Above the cloud base, turbulent diffusion works toward maintaining a constant total mixing ratio ($q_{\rm t} = q_{\rm v} + q_{\rm c}$), which is the sum of the vapor ($q_{\rm v}$ = moles of vapor per mole of atmosphere) and condensate ($q_{\rm c}$ = moles of condensate per mole of atmosphere) mixing ratios, while sedimentation reduces $q_{\rm t}$ by transporting condensate downward. Note that by ignoring horizontal variability, any differences between (cloudy) updrafts and (potentially cloud-free) downdrafts are neglected.
A number of models for tropospheric condensation clouds have appeared in the planetary and astrophysical literature. Here we summarize a selection of them that contribute to the present work.
Lewis (1969)
------------
Lewis (1969) represents a foundation in the study of tropospheric clouds in the giant planets. In that work the term “precipitation” is used in the narrow sense used by chemists, in which condensates appear where the local saturation vapor pressure is exceeded by the actual vapor pressure, rather than in the broader sense used by meteorologists, which additionally denotes sedimentation of the condensates (hereafter we use the term in this broader sense). Although there is no mention of sedimentation by Lewis, the treatment does imply certain assumptions. Starting below the cloud base and working upwards, at each computational level the Lewis model assumes that all the condensate remains at the level where it appears. Considered in the framework of a parcel in an updraft, the Lewis model assumes that all condensate rains out with a fallspeed matching the updraft velocity. Were sedimentation slower, condensate would be transported upward (as discussed by Weidenschilling & Lewis 1973); were sedimentation faster, condensate would be transported downward. Hence, the Lewis (1969) assumption regarding sedimentation is an unstated compromise between those two possibilities.
We implement the Lewis model by starting below the cloud base (where $q_{\rm c} = 0$ and $q_{\rm v} = q_{\rm below}$) and condensing all vapor in excess of saturation at each successive level upward: $$\begin{aligned}
%%
\label{lewis_qc}
q_{\rm c}(z) & = & \max( 0, q_{\rm v}(z-\Delta z) - q_{\rm s}(z) ) \\
%%
\label{lewis_qv}
q_{\rm v}(z) & = & \min( q_{\rm v}(z-\Delta z), q_{\rm s}(z) )
%%\end{aligned}$$ where $z$ is altitude and $q_{\rm s}$ is the vapor mole fraction corresponding to the saturation vapor pressure at that altitude. The first and second cases on the right hand side correspond to cloud-free and cloudy conditions, respectively. Note that under all conditions $q_{\rm t}(z) = q_{\rm v}(z-\Delta z)$ in the Lewis model, reflecting the assumption that only vapor is transported upwards.
Beyond this simple model, Lewis (1969) considered the partitioning of chemical species in some detail, and also calculated pseudo-adiabatic lapse rates. Here we simply assume that each condensate results from the saturation of a single condensable, and fix the lapse rate as input from observations or an external model.
For an example, we calculate an ammonia cloud profile from the Lewis model (Figure 1) using the Jovian temperature profile from Voyager (Lindal et al. 1981), the relation for vapor pressure given in Appendix \[pvap\], and a sub-cloud mole fraction of 3 $\times$ $10^{-5}$ (a wide range of abundances below the expected base of the Jovian ammonia cloud have been reported; we adopt the value at 0.6 bars retrieved by Kunde et al. 1982 for the Northern Equatorial Belt, which also agrees with the best-fit values of Carlson, Lacis, & Rossow 1993 and Brooke et al. 1998). The cloud base appears at 0.42 bars, where the temperature is 129 K. Although absent in the figures of Lewis (1969) (likely due to reduced vertical resolution), in our interpretation of that model the vapor is not entirely depleted in the lowest reaches of the cloud (where $q_{\rm c} < q_{\rm t}$), hence $q_{\rm c}$ increases above the cloud base. Such an increase is found in terrestrial clouds of moderate vertical extent, where $q_{\rm c} < q_{\rm t}$, and hence $q_{\rm c}$ increases with altitude throughout their depths. However, at greater altitudes in this deep ammonia cloud, the vapor is so effectively depleted by condensation at the low temperatures that $q_{\rm v} \ll q_{\rm t}$, leading to a cold degeneracy: $q_{\rm c}(z) \approx q_{\rm t}(z) = q_{\rm s}(z-\Delta z)$, in which decreasing temperatures result in $q_{\rm c}$ diminishing with altitude. Note that the condensate abundance drops off rapidly above $\sim$0.13 bars due to increasing temperatures. Hence, the temperature minimum quite reasonably produces a cold-trap in the Lewis model.
Condensate particle sizes, the other ingredient needed for predicting cloud opacity, are not considered by Lewis (1969) or Weidenschilling & Lewis (1973).
Carlson et al. (1988)
---------------------
In their theoretical characterization of cloud microphysics of the giant planets, Carlson et al. (1988) employ the formalism of Rossow (1978) to calculate time constants for droplet condensation within cloudy updrafts (assuming a supersaturation of 10$^{-3}$), droplet coalescence (assuming the mean collision rate is described by particles with a mass ratio of 2), and sedimentation through an atmospheric scale height. From these time constants, estimates are made of the predominant size of cloud particles at cloud base for a number of condensates. For the Jovian ammonia cloud, Carlson et al. estimate a mass-weighted droplet radius of $\sim 10 - 30 \: \mu$m.
Carlson et al. (1988) make no attempt to calculate vertical profiles of condensate mass. For profiles of vapors that condense into multiple forms (such as ammonia, which can also condense onto a cloud of NH$_4$SH below the ammonia cloud), saturation is assumed above the cloud base.
A shortcoming to the approach of Carlson et al. (1988) is that their microphysical time constants strongly depend on a number of uncertain factors, chief among them completely unknown supersaturations, which govern droplet growth rates due to condensation. Supersaturations in a cloudy updraft are determined by balance between the source due to adiabatic cooling, and the sink due to condensation. Uncertainties in updraft speeds and the populations of condensation nuclei (and hence cloud droplets) both contribute to the uncertainty in supersaturations realized in extraterrestrial clouds. Furthermore, the time constants Carlson et al. (1988) use for gravitational coalescence assume that the collection efficiency is unity, and those for sedimentation effectively assume a fixed width of the size distribution. Rather than attempting to constrain the many degrees of freedom using such a detailed approach, we choose instead to reduce the number of assumptions by simplifying the description of cloud microphysics.
Lunine et al. (1989)
--------------------
Lunine et al. (1989) consider a range of possible iron and silicate clouds in brown dwarfs; the possibilities differ in the nature of the balance between sedimentation and turbulent mixing. The framework is based on a theoretical investigation into iron clouds deep in the Jovian atmosphere by Prinn & Olaguer (1981), which in turn draws on an analysis of sulfuric acid clouds on Venus (Prinn 1974). These models represent a fleshing out of the discussion of vertical transport of condensates by Weidenschilling & Lewis (1973).
Two fundamental cloud types are treated by Lunine et al. (1989). The first is “dust-like” (using the terminology of Prinn & Olaguer 1981) in which cloud particles grow and efficiently sediment out, resulting in relatively thin clouds limited by the local vapor pressure, as in the model of Lewis (1969). These dust-like clouds are assumed to prevail in the radiative region (stratosphere), where the temperature profile is stable and convection is suppressed.
The second fundamental type in the Lunine et al. (1989) study is a tropospheric cloud, in which downward transport by sedimentation is opposed by upward transport due to turbulent mixing. For this cloud type, two variations are considered by Lunine et al. (1989). For the first variation, described as “frozen-in,” cloud particles are so small that sedimentation is overwhelmed by upward transport due to turbulent mixing. In this case, the atmosphere is well-mixed with respect to condensate, and hence $q_{\rm c}$ is independent of altitude above the cloud base. For the second variation, which is intermediate to the dust-like and frozen-in cases, particles grow large enough in “convective” clouds to develop appreciable sedimentation velocities, and their downward sedimentation is balanced by their turbulent transport upward. For their calculations of specific brown dwarf models, Lunine et al. (1989) consider only the two endmembers of their cloud spectrum, corresponding to dust-like and frozen-in clouds.
Their intermediate case serves as a starting point for our model of condensate mass profiles. Our interpretation of the convective cloud model of Lunine et al. (1989) as applied to the Jovian ammonia cloud is shown in Figure 1. Note that we have refined that model slightly, allowing the atmospheric properties to vary with height above the cloud base, and relaxing their assumption that $q_{\rm c}$ = $q_{\rm t}$. The condensate mass is seen to be significantly enhanced above the cloud base for that model: at the tropopause (where there is no cold-trap in this case) $q_{\rm c}$ is enhanced a thousand-fold over that computed by the Lewis (1969) model. Thus the treatment in which Lunine et al. (1989) assume particle sedimentation to balance turbulent transport results in a cloud not so different from their frozen-in case (as depicted by the curve in Figure 1 labeled $f_{\rm rain} = 0$). Evidently the sedimentation in this convective cloud model is far less effective than that assumed by Lewis (1969). As described below, for our calculations of condensate mass profiles we extend the Lunine et al. (1989) approach by applying a scale factor to the particle sedimentation.
For radiative calculations, Lunine et al. (1989) assume all particles in the frozen-in and dust-like clouds are 1 and 10 $\mu$m in radius, respectively.
Marley et al. (1999)
--------------------
The Marley et al. (1999) model of water and silicate clouds in extrasolar giant planets represents a variation on the Lewis (1969) model. As in the Lewis model, the calculation of vapor pressure (or equivalently, $q_{\rm v}$) assumes that any supersaturation is quenched locally by condensation (Equation \[lewis\_qv\]). However, the calculation of the condensate mole fraction ($q_{\rm c}$) represents a departure: instead of calculating it from the vapor pressure in the underlying layer from Equation \[lewis\_qc\], Marley et al. scale it to the local saturation vapor pressure with the following assumption: $$q_{\rm c}(z) = \cases{0 & if $q_{\rm v}(z-\Delta z) < q_{\rm s}(z)$ \cr
f_{\rm s} q_{\rm s}(z) & otherwise \cr }$$ The parameter $f_{\rm s}$ corresponds to the potential supersaturation prior to condensation. Marley et al. (1999) treat $f_{\rm s}$ as an adjustable parameter, ranging from a baseline value of 0.01 to an extreme value of 1. The baseline model as applied to the Jovian ammonia cloud is shown in Figure 1. The condensate mass is seen to be diminished by a factor of $\sim$100 relative to the Lewis (1969) model.
Increasing $f_{\rm s}$ to 1 results in a hundredfold enhancement of $q_{\rm c}$ throughout the cloud compared to the baseline case, as shown in Figure 2*a*. The principle difference between that extreme and the Lewis (1969) condensate model is that for the former there is no regime near the cloud base akin to shallow terrestrial clouds, in which $q_{\rm c}$ increases with altitude. This difference is attributable to a discontinuity of $q_{\rm t}$ in the treatment of Marley et al.: below the cloud base $q_{\rm t}(z) = q_{\rm v}(z-\Delta z)$ as in the Lewis (1969) model, but above the cloud base $q_{\rm t}(z) = (1+f{\rm s}) \: q_{\rm s}(z)$.
For their calculations of cloud particle sizes, which are decoupled from their calculation of condensate mass, Marley et al. (1999) apply the formalism of Rossow (1978) to two atmospheric endmembers: first, a quiescent atmosphere, in which the mean particle size is determined from the condition that the sedimentation rate matches the faster of coagulation and condensation (at an assumed supersaturation of 0.01); and second, a turbulent atmosphere in which mixing is balanced by sedimentation.
The first endmember is subject to a similar catalog of unconstrained assumptions as required by the treatment of Carlson et al. (1988), the most notable among them being the great uncertainty in the supersaturation driving droplet condensation. Also, the model physics underlying this quiescent atmosphere seems to be self-contradictory, on the one hand explicitly assuming that there is too little convection to regulate the maximum size of the droplets, yet on the other hand implicitly assuming that there is enough convection to supply the vapor necessary to drive condensational growth.
However, the second case of Marley et al. (1999) requires significantly fewer assumptions, and is also appropriate to tropospheric condensation clouds. This second case serves as a starting point for the calculation of the cloud particle sizes in our model.
The Present Model
=================
We model all condensation clouds as horizontally homogeneous (globally averaged) structures whose vertical extent is governed by a balance between the upward turbulent mixing of condensate and vapor ($q_{\rm t} = q_{\rm c} + q_{\rm v}$) and the downward transport of condensate due to sedimentation: $$\begin{aligned}
\label{K_pde}
-K \frac{ \partial q_{\rm t}}{\partial z} - f_{\rm rain} w_* q_{\rm c} = 0\end{aligned}$$ where $K$ is the vertical eddy diffusion coefficient and $f_{\rm rain}$ is a new parameter that we have introduced, defined as the ratio of the mass-weighted droplet sedimentation velocity to $w_*$, the convective velocity scale. We solve Equation \[K\_pde\] for each condensate independently, and hence ignore any microphysical interactions between clouds. Equation \[K\_pde\] is an extension of Lunine et al.’s convective cloud model, relaxing their implicit assumptions $f_{\rm rain}$ = 1 and $q_{\rm c} = q_{\rm t}$.
The product $f_{\rm rain} w_*$ represents an average sedimentation velocity for the condensate, which offsets turbulent mixing and thereby leads to $q_{\rm t}$ decreasing with altitude. The extreme case with no sedimentation to offset turbulent mixing ($f_{\rm rain}$ = 0) is equivalent the frozen-in endmember of Lunine et al. (1989) and the “dusty” models of the Lyon group (e.g., Chabrier et al. 2000). In this case the solution to Equation \[K\_pde\] is a well-mixed atmosphere ($q_{\rm t}$ independent of altitude), which is seen in Figure 1 to loft even more condensate than the convective cloud of Lunine et al. (1989).
We adopt $f_{\rm rain}$ as an adjustable input parameter, which together with $q_{\rm c}$ constrains the droplet size distributions. First we describe our calculation of $q_{\rm c}$, then the size distributions.
Condensate Mass Profiles
------------------------
The eddy diffusion coefficient ($K$) for $q_{\rm t}$ is assumed to be the same as that for heat as derived for free convection (Gierasch and Conrath, 1985): $$\begin{aligned}
\label{K_equals}
K = \frac{H}{3} \left( \frac{L}{H} \right)^{4/3}
\left( \frac{R F}{\mu \rho_{\rm a} c_{\rm p}} \right)^{1/3}\end{aligned}$$ where the atmospheric scale height is given by $H = R T / \mu g$ (for Jupiter we use $g$ = 25 m s$^{-2}$), $L$ is the turbulent mixing length, $R$ the universal gas constant, $\mu$ the atmospheric molecular weight (2.2 g mol$^{-1}$ assumed here), $\rho_{\rm a}$ the atmospheric density, and $c_{\rm p}$ the specific heat of the atmosphere at constant pressure (ideal gas assumed). Here we assume all the interior heat to be transported through the convective heat flux: $F = \sigma T_{\rm eff}^4$, where $\sigma$ is the Stefan-Boltzmann constant and the effective temperature for Jupiter is $T_{\rm eff}$ = 124 K. In the general case, a profile of $F$ is specified by an external model, which partitions the transport of interior heat between radiative and convective fluxes. The convective heat flux can be reduced further by other heat fluxes, such as para hydrogen conversion or latent heat release, as discussed by Gierasch and Conrath (1985). Beyond any uncertainty in the convective heat flux, the constant coefficient scaling the eddy diffusion coefficient is only loosely constrained by observations. For our baseline model we use a coefficient of 1/3 based on previous modeling studies of the Jovian atmosphere (D. M. Hunten 1996, private communication), and consider the sensitivity of our model results to its value in a subsequent section.
For freely convecting atmospheres, the mixing length is typically assumed to be the pressure scale height. However, in stable atmospheric regions, the mixing length will be diminished. We account for this reduction by scaling the mixing length to the local stability: $$\begin{aligned}
\label{L_equals}
L = H \max( \Lambda, \Gamma/\Gamma_{\rm adiab})\end{aligned}$$ where $\Gamma$ and $\Gamma_{\rm adiab}$ are the local and dry adiabatic lapse rates, respectively, and $\Lambda$ is the minimum scaling applied to $L$ (we assume a value of 0.1). In the general case, convective heat fluxes are diminished in radiative regions; hence, we also assume an eddy diffusion coefficient no less than a prescribed minimum value ($K_{\rm min} =
10^5$ cm$^2$ s$^{-1}$ in our baseline model), which represents residual turbulence due to breaking buoyancy waves (Lindzen 1981) and such. A list of prescribed/adjustable parameters is provided in Table 1.
The remaining parameter in Equation \[K\_equals\] is the convective velocity scale from mixing-length theory: $w_* = K / L$. Our baseline values for the turbulent mixing parameters just below the Jovian ammonia cloud are $H = L = 20 \rm \, km$, $K = 2 \times 10^8 \rm \, cm^2 s^{-1}$, and $w_* = 1.1 \rm \, m \, s^{-1}$.
To compute the vertical distributions of condensate and vapor, we proceed upwards from the sub-cloud conditions, requiring all excess vapor to condense and solving Equation \[K\_pde\] at each level. If we heuristically assume that $q_{\rm c} / q_{\rm t}$ and $L$ are constant in a cloud, the solution is an exponential decline of total mixing ratio with height above cloud base (where we define $z = 0$): $$\label{heuristic_soln}
q_{\rm t}(z) =
q_{\rm below} \exp\left( - f_{\rm rain}
\frac{q_{\rm c}}{q_{\rm t}}
\frac{z}{L} \right)$$ Note that by using the sub-cloud mixing ratio as a lower boundary condition, any moistening due to rain evaporating below the cloud base is ignored.
Comparing our adaption of the Lunine et al. (1989) profile with our calculation using $f_{\rm rain} = 1$ isolates the effect of reducing the mixing length due to atmospheric stability. The cumulative effect of the progressive reduction in mixing length due to the stability of the Voyager temperature profile above the cloud base is seen to result in a cold-trap in the lower stratosphere. Tripling $f_{\rm rain}$ further reduces the cloud density and lowers the cold-trap to the tropopause; increasing it to 10 results in a cloud with less condensate than the Lewis (1969) model.
We assume that $f_{\rm rain}$ is independent of altitude. Yet specifying an appropriate value of $f_{\rm rain}$ at the cloud base, let alone any vertical dependence, poses a significant challenge. For guidance, first we turn to *in situ* measurements and detailed simulations of terrestrial water clouds, and then consider constraints provided by values retrieved through remote sensing of Jovian ammonia clouds.
For terrestrial stratocumulus clouds capping well-mixed planetary boundary layers, we find that $f_{\rm rain} < 1$ in the cloud deck and increases with distance below cloud top. An assortment of *in situ* measurements indicates that $f_{\rm rain}$ increases with decreasing droplet concentrations ($N$), as fewer and hence larger droplets more efficiently produce drizzle and thereby decrease cloud water. For example, $f_{\rm rain} \sim$ 0.2 for a case study over the North Sea, where $N$ = 100 cm$^{-3}$ (Nicholls, 1984), while in California stratocumulus it increased from 0.3 to 0.5 as $N$ decreased from 40 cm$^{-3}$ in clouds contaminated by ship exhaust, to 10 cm$^{-3}$ in clean ambient air (from Figure 3*b* of Ackerman et al. 2000b). We note that in clean marine stratocumulus, reduced rain is observed (e.g., Taylor and Ackerman 2000) and predicted (e.g., Ackerman, Toon, and Hobbs 1993; Stevens et al. 1998) to result in deeper cloud layers, which is consistent with our simple model. However, any changes in cloud cover associated with precipitation changes are not represented in our model.
We have also calculated profiles of $f_{\rm rain}$ for deeper convection, using large-eddy simulations of trade cumulus clouds (Ackerman et al. 2000a), which are twice as deep as stratocumulus clouds ($\sim$1000 m compared to $\sim$500 m). As in the stratocumulus clouds, we find that $f_{\rm rain} < 1$ in the stratiform anvils at cloud top. However, $f_{\rm rain}$ is significantly enhanced throughout the bulk of the trade cumulus clouds, where it ranges from $\sim$ 2 to 6. Vertical winds are more symmetrically distributed in stratocumulus, and hence the convective velocity scale is representative of the mean updraft velocity. The circulation is more skewed in trade cumulus, with narrow updrafts opposing the broad subsidence: for the Ackerman et al. (2000a) trade-cumulus simulations the mean vertical velocity in cloudy updrafts is three times the convective velocity scale. Such a skewness is consistent with an enhancement of the rain factor in deep convection.
We recommend a more systematic analysis of $f_{\rm rain}$ in terrestrial clouds. Perhaps more pertinent to the much deeper clouds expected in gas giants and brown dwarfs, we also recommend consideration of much deeper convection than considered here. For now we treat $f_{\rm rain}$ as an adjustable parameter, leaning toward values $>$ 1, as we expect the deep convection in substellar atmospheres (on the order of an atmospheric scale height and deeper) to more closely resemble cumuliform than stratiform convection in the terrestrial atmosphere.
For observations of Jovian ammonia clouds, we first turn to the retrievals obtained from the Voyager IRIS (InfraRed Interferometer Spectrometer) instrument by Carlson, Lacis, & Rossow (1994), who considered latitudinal variations among zones and belts in the Jovian tropics. Although they did not provide profiles of condensed ammonia, we can compare our results to their ratios of condensate to atmospheric scale height ($H_{\rm p}/H_{\rm g}$ in their notation). Carlson et al. (1994) retrieve ratios of 0.35 and 0.40 ($\pm$ 0.10) for ammonia clouds in the Equatorial and Northern Tropical Zones, respectively, which are seen in Figure 3 to span a range of $f_{\rm rain}$ values between $\sim$ 1 and 3. Retrievals of ammonia cloud properties for the Jovian tropics from ISO (Infrared Space Observatory) measurements by Brooke et al. (1998) indicate a scale height ratio of 0.3, which is consistent with our model results for $f_{\rm rain} \sim 2$.
Were the ratio $q_{\rm c} / q_{\rm t}$ fixed in our model, as assumed heuristically for Equation \[heuristic\_soln\], the condensate height should vary as $f_{\rm rain}^{-1}$. However, as seen in Figure 3, the model dependence is not nearly that steep, and for $f_{\rm rain} > 3$ the dependence nearly vanishes. The dependence is moderated by a negative feedback in which $q_{\rm c} / q_{\rm t}$ decreases with increasing $f_{\rm rain}$. Our assumption of zero supersaturation within a cloud is equivalent to $q_{\rm c} / q_{\rm t} = S - 1$, where $S$ is the potential supersaturation before condensation eliminates it. Increased $f_{\rm rain}$ offsets more of the turbulent mixing of vapor and condensate, thereby reducing the potential supersaturation and decreasing the ratio $q_{\rm c} / q_{\rm t}$. The negative feedback is thereby due to the reduction of $q_{\rm c} / q_{\rm t}$, which diminishes the dependence of the condensate scale height on $f_{\rm rain}$ (Equation \[heuristic\_soln\]).
The assumption that all vapor in excess of saturation condenses can be relaxed in our model by replacing Equation \[lewis\_qc\] with $$q_{\rm c}(z) = \max( 0, q_{\rm v}(z-\Delta z) -
( S_{\rm cloud} + 1) \: q_{\rm s}(z) )$$ where $S_{\rm cloud}$ is the supersaturation that persists after accounting for condensation. Allowing $S_{\rm cloud} > 0$ represents conditions in which there is a significant barrier to the formation of cloud droplets, and/or condensation is too slow to effectively offset the supersaturation driven by cooling in updrafts. In shallow terrestrial water clouds, such as stratocumulus, neither of these conditions holds, as there are typically abundant condensation nuclei upon which droplets form at low supersaturations, and the concentration and diffusivity of water vapor are sufficient to allow condensation to balance the modest dynamic forcing at low supersaturations ($\sim$10$^{-3}$). However, in cirrus clouds that form directly from the vapor phase (as opposed to the freezing of water droplets from deep convection), there are typically few effective ice nuclei available, and hence the barrier to nucleation can result in supersaturation building to $\sim$0.5 before ice crystals form, even in moderate updrafts. High supersaturations ($\sim$0.3) can be maintained after nucleation in cirrus clouds because the concentration and diffusivity of water vapor are greatly reduced at the cold temperatures of the upper troposphere (Jensen et al. 2000).
As extreme cases, in Figure 2*b* the condensate profile for $S_{\rm cloud}$ = 0 is compared to that for $S_{\rm cloud}$ = 1. The barrier to condensation results in a lifting of the cloud base and enhanced lofting of vapor, but the altitude of the cold-trap is unchanged (recall that the temperature profile is fixed here). The greater lofting enhances condensate mass above $\sim$0.4 bar, which tends to increase the column of condensate. However, the tendency is more than offset by the decrease in atmospheric mass density at the elevated cloud base, and hence the condensate column decreases from 52 to 39 g m$^{-2}$ in response to increasing $S_{\rm cloud}$ from 0 to 1. This 25$\%$ reduction of condensate column contrasts markedly with the effect of increasing $f_{\rm s}$ from 0.01 to 1 in the Marley et al. (1999) model, in which the condensate column increases a hundredfold (Figure 2*a*). These opposite responses (to comparable changes in maximum relative humidities: from 100 to 200$\%$ for the present model, and 100.01 to 200$\%$ for the Marley et al. model) arise from the distinct definitions of the supersaturation factor: $S_{\rm cloud}$ corresponds to the supersaturation *after* condensation is treated in the present model, whereas $f_{\rm s}$ corresponds to the potential supersaturation *before* condensation is treated in the Marley et al. model. Marley et al.’s $f_{\rm s}$ is more comparable to $f_{\rm rain}^{-1}$ than to $S_{\rm cloud}$ in the present model.
Droplet Size Distributions
--------------------------
Size spectra of cloud particles in terrestrial condensation clouds are commonly observed as bimodal number distributions, with a condensational mode (of radius $\sim$10 $\mu$m in stratocumulus) resulting from condensational growth at modest supersaturations, and a precipitation mode at larger radii resulting from coalescence due to dispersion in sedimentation velocities. The bimodal structure is evident in droplet size distributions measured in stratocumulus clouds off the coast of California, as shown in Figure 4*a*. At the cloud base, the mean size of the condensation mode depends on factors such as the updraft velocity and the composition and size distribution of the condensation nuclei upon which droplets form. Above the cloud base, the mean size of the condensation mode increases with altitude in an updraft.
The production of precipitation in terrestrial clouds is generally observed to increase with the mean size and spectral width of the condensation mode. The mean size of droplets in the precipitation mode[^1] is found to increase with distance from cloud top in marine stratocumulus (e.g., Nicholls 1984) (the opposite tendency of the condensation mode), typically explained in microphysical terms as due to “fortunate” collector drops sweeping up smaller droplets and growing as they fall. Such a profile is consistent, as it must be, with the observed decrease in $f_{\rm rain}$ with height above the cloud base. We note in passing that measured profiles of $f_{\rm rain}$ are also consistent with our assumption that the size of precipitation particles decreases as convective velocity decreases above the cloud base (approaching the temperature inversion that caps the boundary layer), though we assume a uniform value of $f_{\rm rain}$ in our model calculations.
We make no attempt to model the complexity of cloud processes here, as such detailed computations are prohibitively demanding, and the parameter space of unknowns is overwhelming for the range of condensates expected in substellar atmospheres (including such basic issues as whether condensates are solid or liquid). Instead, we simply prescribe a single, broad lognormal size distribution of condensate particles at each level (Figure 4), thereby halving the number of parameters required for a bimodal distribution. The lognormal size distribution is given by $$\frac{dn}{dr} = \frac{N}{r \sqrt{2\pi} \ln \sigma_{\rm g}}
\exp\left( -\frac{ \ln^2(r/r_{\rm g})}{2\ln^2 \sigma_{\rm g}} \right)$$ where $n$ is the number concentration of particles smaller than radius $r$, and the three parameters to be constrained appear on the right side: $N$ is the total number concentration of particles, $r_{\rm g}$ the geometric mean radius, and $\sigma_{\rm g}$ the geometric standard deviation.
Marley et al. (1999) also prescribe a lognormal size distribution of condensate particles, in which $\sigma_{\rm g}$ is fixed at 1.5 and $r_{\rm g}$ is determined from the particle sedimentation velocity corresponding to the convective velocity scale ($w_*$). We choose to use $\sigma_{\rm g}$ as an adjustable parameter, and determine $N$ and $r_{\rm g}$ through $q_{\rm c}$ and $f_{\rm rain}$. The rain factor is rigorously defined through $$\begin{aligned}
\label{frain_equals}
f_{\rm rain} = \frac{\int_0^\infty v_{\rm f} \: (dm/dr) \: dr}
{\epsilon \rho_{\rm a} w_* q_{\rm c}}\end{aligned}$$ where $v_{\rm f}$ is the particle sedimentation velocity (described in Appendix \[vfall\]), $m$ is particle mass, and $\epsilon$ is the ratio of condensate to atmospheric molecular weights. To close the system analytically we fit a power-law dependence for particle fallspeed about its value at $v_{\rm f}( r_{\rm w} ) = w_*$ through $$v_{\rm f} = w_* ( r / r_{\rm w} )^\alpha$$ where the exponent $\alpha$ is calculated from a fit to the fallspeeds between $r_{\rm w}/\sigma$ and $r_{\rm w}$ when $f_{\rm rain} > 1$, and between $r_{\rm w}$ and $r_{\rm w} \, \sigma$ otherwise ($\sigma$ is constrained to be $\geq 1.1$ for the fit). The power-law approximation allows Equation \[frain\_equals\] to be expressed as $$\begin{aligned}
\label{frain_analytic}
f_{\rm rain} = \frac{\int_0^\infty r^{3+\alpha} \: (dn/dr) \: dr}
{r_{\rm w}^\alpha \int_0^\infty r^3 \: (dn/dr) \: dr }\end{aligned}$$ Integration of the lognormal distribution then leads to $$\begin{aligned}
\label{rg_equals}
r_{\rm g} = r_{\rm w} f_{\rm rain}^{1/\alpha}
\exp\left (-\frac{\alpha+6}{2} \ln^2\sigma_{\rm g} \right)\end{aligned}$$ $$\begin{aligned}
\label{N_equals}
N = \frac{3 \epsilon \rho_{\rm a} q_{\rm c}}{4 \pi \rho_{\rm p} r_{\rm g}^3}
\exp \left( -\frac{9}{2} \ln^2 \sigma_{\rm g} \right)\end{aligned}$$ where $\rho_{\rm p}$ is the density of a condensed particle (see Appendix \[vfall\]). The parameter $f_{\rm rain}$ can be interpreted in terms of microphysics by identifying the radius of mass-weighted sedimentation flux in Equation \[frain\_analytic\]: $$\begin{aligned}
r_{\rm sed} = \left(
\frac{\int_0^\infty r^{3+\alpha} \: (dn/dr) \: dr}
{\int_0^\infty r^3 \: (dn/dr) \: dr }
\right)^{1/\alpha}\end{aligned}$$ which leads to $f_{\rm rain} = ( r_{\rm sed} / r_{\rm w})^\alpha$. In Figure 4*b* the size distribution is weighted by precipitation, where it is seen that $r_{\rm sed} < r_{\rm w}$, consistent with $f_{\rm rain} < 1$. In contrast, for the large-eddy simulations of trade cumulus mentioned above, and as implied by the retrievals from the Jovian ammonia clouds, the droplets grows sufficiently large to satisfy $r_{\rm sed} > r_{\rm w}$.
It is seen from Equations \[frain\_analytic\]$-$\[N\_equals\] that droplet sizes are decoupled from condensate mass (though both depend on $f_{\rm rain}$); the condensate mass simply scales the distribution through $N$. Ignoring any vertical dependence of atmospheric stability (Equation \[L\_equals\]), vertical variations in droplet sizes are due to the height dependence of convective velocity, leading to $r_{\rm w} \propto \rho_{\rm a}^{1/3\alpha}$, which yields a mild vertical dependence of approximately $r_{\rm w} \propto \rho_{\rm a}^{1/4}$ for our baseline Jovian ammonia cloud (for which $\alpha = 1.3$, corresponding to a moderately turbulent sedimentation regime in which fallspeeds are reduced from those in viscous flow).
Of greater interest than the mild vertical dependence are the sensitivities of droplet sizes on $f_{\rm rain}$ and $\sigma_{\rm g}$ at the base of our Jovian ammonia cloud, where $r_{\rm w} = 35~\mu$m. For a given value of $\sigma_{\rm g}$, more efficient rain implies larger droplets: $r_{\rm g} \propto f_{\rm rain}^{1/\alpha}$. For our baseline droplet distribution (with $\sigma_{\rm g} = 2$), a value of $f_{\rm rain} = 3$ leads to $r_{\rm g} = 14~\mu$m (compare to $\sim$ 10 $\mu$m for shallow terrestrial clouds), while $f_{\rm rain} = 5$ yields $r_{\rm g} = 20~\mu$m.
For efficient precipitation ($f_{\rm rain} > 1$), reducing the width of the size distribution requires an increase in $r_{\rm g}$, as the narrower distribution is centered on a larger radius. For example, a monodisperse distribution ($\sigma_{\rm g} = 1$) with $f_{\rm rain} = 3$ results in $r_{\rm g} = 77~\mu$m. Such narrow distributions of large particles could result from condensational growth at high supersaturations, reminiscent of methane “rain without clouds” suggested by Toon et al. (1988) for Titan’s atmosphere.
In the complete model, we calculate spectrally-resolved profiles of condensate opacity by integrating the scattering and absorption coefficients (from Mie calculations) over the particle size distributions. Here we simply present opacities for geometric scatterers, which for a model layer of thickness $\Delta z$ is given by $$\label{tau_equals}
\Delta\tau = \frac{3}{2} \frac{ \epsilon \rho_{\rm a} q_{\rm c} }
{ \rho_{\rm p} r_{\rm eff} } \Delta z$$ where the effective (area-weighted) droplet radius is evaluated from the lognormal size distribution: $$\label{reff_equals}
r_{\rm eff} = r_{\rm w} f_{\rm rain}^{1/\alpha}
\exp \left(-\frac{\alpha+1}{2} \ln^2\sigma_{\rm g} \right)$$ The effective radius can be greater or less than $r_{\rm w}$, depending on the combination of $f_{\rm rain}$, $\alpha$, and $\sigma_{\rm g}$. At the base of our baseline Jovian ammonia cloud $r_{\rm eff} = 46~\mu$m, which is 11 $\mu$m greater than $r_{\rm w}$.
The computation of cloud optical depth depends on vertical grid resolution due to the exponential temperature dependence of saturation vapor pressures. To reduce such resolution dependence we progressively subdivide each model layer until its optical depth converges to 1% precision. For the Jovian ammonia cloud our calculations converge at a minimum sub-layer thickness of $\sim$ 30 m.
Equations \[tau\_equals\] and \[reff\_equals\] show that increased precipitation reduces opacity not only by decreasing $q_{\rm c}$ but also by increasing $r_{\rm eff}$. For heuristic purposes the column optical depth can be estimated from the cloud base properties by ignoring any height dependence of the mixing length and atmospheric scale height and assuming $q_{\rm c} = q_{\rm t}$ within the cloud, in which case $$\tau = \frac{3}{2} \frac{ \epsilon p q_{\rm below}}
{ g r_{\rm eff}(1+f_{\rm rain})}$$ where $p$ and $r_{\rm eff}$ are atmospheric pressure and droplet effective radius at the cloud base. Note that $\tau$ is more than linearly dependent on $q_{\rm below}$, due to its dependence on the cloud base pressure. For the following comparisons with observations of the Jovian ammonia cloud, we sum optical depths from Equation \[tau\_equals\] over the model layers and compaute a cloud average $r_{\rm eff}$ from the cloud optical depth and vertical column of condensate.
Comparisons with Observations
=============================
West et al. (1986) attempt to reconcile among a vast array of observations of the Jovian ammonia cloud, and conclude that its optical depth at visible wavelengths is between $\sim$ 2 and 10, comprised of a population of small particles ($r \sim 1\:\mu$m) reaching the tropopause in both belts and zones, underlain in zones by a population of larger particles ($r \sim 3 - 100\:\mu$m) concentrated near the cloud base.
More recently, Banfield et al. (1998), who make no attempt to retrieve particle sizes, assume a particle effective radius of 0.2 $\mu$m and retrieve optical depths from Galileo imaging data at a wavelength of 0.756 $\mu$m that cluster in the range $\sim1 - 4.5$ (their Figure 9). For comparison with the optical depths of ours and West et al.’s (at a mid-visible wavelength of 0.55 $\mu$m), we scale Banfield et al.’s $\tau$ by the ratio of extinction efficiencies at 0.55 to 0.756 $\mu$m for 0.2 $\mu$m particles (using their refractive index of 1.4), resulting in a mid-visible $\tau$ range of $\sim2 - 10$. Hence, the West et al. (1986) and Banfield et al. (1998) results are effectively identical, and are hereafter lumped together as “West et al.”
Recall from Figure 3 that the condensate scale height retrievals of Carlson et al. (1994) are consistent with our baseline model for $f_{\rm rain}$ between $\sim$ 1 and 3. This entire $f_{\rm rain}$ range for our baseline model (in which $\sigma_{\rm g} = 2$) is seen in Figure 5[*a*]{} to overlap with the data reported by West et al. (1986). The overlap corresponds to $r_{\rm eff}$ ranging from $\sim$ 18 to 45 $\mu$m. A narrower size distribution results in larger droplet effective radii and therefore smaller optical depths; for the monodisperse case the data mutually overlap at $r_{\rm eff} \sim$ 34 to 60 $\mu$m.
Carlson et al. (1994) use Voyager IRIS spectra to retrieve cloud optical depths[^2] and droplet sizes for the Northern Equatorial Belt ‘hot-spots’ (note that the infrared data are insensitive to the submicron particles reported by West et al. 1986). Their best-fit particle distribution is a mixture of small ($r_{\rm eff} = 3\:\mu$m, $\tau$ = 0.16) and larger ($r_{\rm eff} = 100\:\mu$m, $\tau$ = 0.38) particles, resulting in a combined $r_{\rm eff}$ = 72 $\mu$m and $\tau$ = 0.54. However, hot spots are anomalous features of reduced cloudiness associated with pronounced dynamical forcings (e.g., Showman & Ingersoll 1998), and our one-dimensional model is intended to represent horizontally averaged conditions, which would seem more comparable to the Equatorial and Northern Tropical Zones, where Carlson et al. retrieve $\tau$ of $\sim 1.2$ and 2, respectively. Carlson et al. do not present separate retrievals of droplet sizes for the zones, but do state that their optical depths are dominated by larger particles, which we interpret as an observed range of $70\:\mu{\rm m} \leq r_{\rm eff} \leq 100\:\mu{\rm m}$.
Given that our model results overlap with the entire range of the West et al. (1986) optical depths, which in turn do not overlap with those of Carlson et al. (1994), it should not be surprising that there is no mutual overlap between our baseline model results and the retrievals of $\tau$ and $r_{\rm eff}$ by Carlson et al. (1994). However, our model results with $f_{\rm rain} = 3$ just overlap with the Carlson et al. (1994) retrievals if a monodisperse size distribution is assumed, which suggests that the Carlson et al. data are dominated by particles in the precipitation mode. Alternatively, multiplying our eddy diffusion coefficients by a factor of 3 also results in larger particles and therefore reduced optical depths, which leads to overlap with the Carlson et al. (1994) retrievals for $f_{\rm rain} \sim 2$ (Figure 5[*b*]{}).
The above comparisons show that reasonable choices of model parameters produce agreement with observations of tropical Jovian ammonia clouds. The few unknown model parameters in our simple model are not uniquely constrained, befitting the incompleteness and uncertainty in the observations and the ambiguity in comparing a model for globally averaged clouds with measurements that resolve large-scale horizontal variability in the clouds. To compare our model with observations averaged over a wider area we next consider measurements of the Jovian tropics obtained from Earth orbit.
Brooke et al. (1998) use a 3-$\mu$m ISO spectrum to retrieve microphysical properties of the Jovian ammonia cloud and find best fits for two possibilities: first, a monomodal distribution of 10-$\mu$m ammonia particles with a visible optical depth of 1.1; and second, a bimodal distribution of 1 and 10-$\mu$m ammonia particles with an optical depth of 1.3 (equivalent to an effective radius of 7 $\mu$m). Both fits include an additional optical depth of 0.1 from grey particles. In their analysis, a fit for 10-$\mu$m particles indicates a superior fit in comparison to those for 1 and 30 $\mu$m particles, which we intepret as allowing a size range of 5 to 20 $\mu$m. Recalling from the discusson of Figure 3 that the Brooke et al. (1998) condensate scale height (a single value with no uncertainty) is consistent with $f_{\rm rain} = 2$. Figure 5[*b*]{} indicates that mutual consistency with the effective radii retrieved by Brooke et al. (1998) requires increasing the width of our size distrubutions and/or reducing our eddy diffusion coefficients. Simultaneously matching the Brooke et al. (1998) optical depths requires a substantial reduction of the sub-cloud ammonia abundance. However, the baseline ammonia abundance we use is reported by Brooke et al. (1998) (as a single value with no uncertainty) to best fit their 3-$\mu$m spectrum. Hence, our calculations are evidently inconsistent with that baseline ammonia abundance and the combination of small particles and small optical depths retrieved by Brooke et al. (1998).
We have already noted the shortcoming that our model excludes the possibility of horizontal variability. For the case of modeling emitted radiative flux, this simplification will of course result in an underestimate at some wavelengths, since any flux leaking out through the clearings between patchy clouds is not treated. Horizontal variability is also ignored in the retrievals of Brooke et al. (1998), which results in their underestimating cloud optical depth due to a plane-parallel albedo bias (e.g., Calahan et al. 1994): the area-weighted albedo of a cloud deck calculated from a single column with an optical depth $\tau$ is always greater than the albedo averaged over a variety of columns with the same area-weighted average $\tau$. Inverted for the Brooke et al. (1994) retrievals of optical depth from reflected spectra, this bias indicates that the optical depth from an area-weighted average radiance underestimates the area-weighted optical depth. Hence, the actual area-weighted optical depth is greater than reported by Brooke et al. (1994) and closer to the optical depths calculated by our model. We are unable to quantify the magnitude of this error without reproducing their retrievals, which is beyond the scope of this study.
Applications to Substellar Atmospheres
======================================
Vertical Cloud Structure
------------------------
We use our baseline model (Table 1) to calculate profiles of condensed water, silicate (as enstatite, MgSiO$_3$), and iron in theoretical atmospheres of brown dwarfs and a giant planet (see Figure 6 for gravities and effective temperatures). The temperature profiles are calculated for cloud-free conditions (Marley 2000). The L-dwarf-like atmosphere (Figure 6*a*) is too warm for water to condense. Between 1 and 10 bars, within the convective region, the silicate and iron clouds are seen to overlap, suggesting the possibility of microphysical interactions between them, which are ignored by our model. The silicate particles in this case are about twice as large as the iron particles because the assumed density of a silicate particle is about half that of an iron particle (Appendix \[vfall\]).
Although the temperature profiles in Figure 6 were not calculated self-consistently to include the effects of clouds, it is clear that for objects near $T_{\rm eff} \sim 1500$ K, clouds are an important opacity source. The silicate and iron clouds in the L-dwarf-like model (Figure 6*a*) appear in the visible atmosphere and therefore play an important role in controlling opacity and the temperature structure of the atmosphere. Nevertheless these clouds are confined to a relatively thin cloud deck, which does not reach the upper regions of the atmosphere as do condensates in the well-mixed profiles also shown. The cloud particles are also fairly large ($r_{\rm eff} \sim 40 - 80\,\rm \mu m$) and will have a substantially different spectral opacity than smaller particles. The Lyon group (Chabrier et al. 2000) employs an “astrophysical dust” size distribution of sub-micron particles to model dust opacity in such atmospheres. The cloud model presented here, with larger particles confined to a discrete cloud deck, represents a substantial departure from the previous work. In a future publication we will discuss the spectral and color properties of atmospheres with these new cloud models.
Figure 6*b* presents the cloud model applied to a T-dwarf-like atmosphere with $T_{\rm eff}=900\,\rm K$, which is again too warm for water to condense. (No iron cloud is shown in Figures 6*b* and *c* because the cloud base is below the bottom of the model domain.) Although the silicate cloud and the omitted iron cloud may be important to the atmospheric temperature structure, they no longer represent significant opacity sources to an observer.
The changing role of cloud opacity with effective temperature is more clearly shown in Figure 7, which illustrates the brightness temperature spectra of several radiative-equilibrium models for brown dwarf atmospheres as well as the atmospheric temperature range over which most of the cloud opacity is found. In a model with $T_{\rm eff} = 1800\,\rm K$ (Figure 7[*a*]{}) the silicate cloud deck forms in the model stratosphere and is relatively thin. Comparison of the solid and dotted curves, which respectively include and exclude silicate and iron cloud opacities, shows little difference between the two cases. Since the cloud optical depth is only a few tenths, flux is efficiently transported from levels deeper than the base of the cloud. The iron cloud (not shown) adds a few more tenths of optical depth. Hence, the clear and cloudy models are very similar. Such a model would be appropriate for an early type L-dwarf.
Figure 7[*b*]{} shows the results for a cooler atmosphere, with $T_{\rm eff} =
1400\,\rm K$, appropriate for a late L-dwarf (Kirkpatrick et al. 1999; Stephens et al. 2001). Here the cloud is much more optically thick, and flux originates no deeper than the middle of the cloud layer. In the clear atmosphere, flux originates from deeper, hotter levels. As a result the band depths of the cloudy model are shallower, a result shown for dusty M-dwarfs by Jones and Tsuji (1997). Since flux is conserved for $T_{\rm eff}$ fixed, the cloudy model emits more flux beyond about $2\,\rm\mu m$ than the clear model. Regions of strong molecular absorption, including the depths of the water bands shortwards of 2 $\mu$m as well as most of the 2 to 5 $\mu$m region evidence higher brightness temperatures in the cloudy case since the atmosphere above the cloud must warm to produce the same total emitted flux as in the cloud free calculation. Regions of stronger molecular opacities are sensitive to these warmer temperatures higher in the atmosphere.
In Figure 7[*c*]{}, for which $T_{\rm eff} = 900\,\rm K$ the silicate cloud forms well below the region in which most flux originates and again the clear and cloudy models are similar. However in the regions in which the molecular opacity is lowest, near 1.1 and 1.3 $\mu m$, flux originates from deeper regions in the clear atmosphere than for the cloudy case. As a result the peak-to-trough variation in emitted flux is again somewhat smaller for the cloudy model. We note that clear atmosphere models for T-dwarfs like Gl229 B and GD165B typically over predict the water band depths (e.g., Marley et al. 1996; Allard et al. 1996; Tsuji et al. 1996; Saumon et al. 2000; Geballe et al. 2001), and suggest that the attenuation of flux by the top of the silicate cloud deck may be responsible for this effect.
Notably, these model atmospheres illustrate the origin of the curious change in infrared colors of the L- and T-dwarfs (e.g., Kirkpatrick et al. 1999; Mart[í]{}n et al. 1999; Fan et al. 2000). The cloud-free cases shown in the figure monotonically vary in $J-K$ from 1.39 to -0.17 from warmest to coolest. In contrast the cloudy models initially become redder with falling $T_{\rm eff}$ (moving from 1.6 to 1.7) before they move to the blue ($J-K = 0.38$ for the case in Figure [*c*]{}) when the silicate and iron clouds begin to disappear below optical depth unity in the gas. Thus our precipitating condensation cloud model qualitatively reproduces the color variation of the L- and T-dwarfs, consistent with previous arguements based on interpretation of spectra (Allard et al. 1996; Marley et al. 1996; Tsuji et al. 1996). In contrast, the pure chemical equilibrium model (Chabrier et al. 2000) predicts the presence of substantial dust opacity well to the top of the atmosphere, which is clearly excluded by the data. A more complete treatment of color changes will be given in a future study.
For the cooler atmosphere representative of a cool extrasolar giant planet (Figure 6*c*), water condenses in the radiative region, in essence a stratospheric cloud. The ice particles are seen to be larger than the silicate particles, chiefly due to the lesser densities assumed for the individual ice particles. The reduced gravity in the atmosphere of the less massive extrasolar giant planet requires larger silicate particles to match the mean sedimentation velocity than does the more massive T-dwarf. Note again that the well-mixed assumption produces a profoundly different vertical structure, in which the silicate cloud is so deep that it significantly overlaps the water cloud.
As did Marley et al. (1999), Sudarsky et al. (2000) have computed water cloud profiles in order to estimate extrasolar giant planet albedos. Sudarsky et al. (2000) essentially assume $f_{\rm rain}=0$ and limit the cloud to be no more than 1 scale height thick. Such a model would be similar to the well-mixed water cloud in Figure 6*c* with a flat cloud top at $\sim 4 \times 10^{-3}$ bars.
The emergence of water clouds in substellar atmospheres with $T_{\rm eff}$ below about 500 K will reshape the vertical temperature profile and emergent spectra of these objects. Preliminary models computed with this cloud profile suggest that such cool objects will again move to the red in $J-K$ after the blueward excursion caused by the sinking of the silicate cloud below the visible atmosphere and the emergence of $\rm CH_4$ as a dominant opacity source in $K$ band. Hence the near-IR colors of very cool objects computed from cloud-free atmosphere models (e.g., Burrows et al. 1997) are likely to differ substantially from actual objects.
Non-uniform Clouds
------------------
Variable brightness in $I$ band has been detected for some L-dwarfs by Bailer-Jones & Mundt (2001), who attribute the variability to evolution of dust clouds. They find some evidence that variability may be more common in later-type L-dwarfs. Although we do not model horizontally variable clouds, these observations are consistent with the model presented here. As clouds form in progressively cooler objects they become more optically thick and form deeper within the convective region of the atmosphere. Thus global scale tropospheric weather patterns, as seen on Jupiter and predicted for brown dwarfs (Schubert & Zhang 2000), can more easily produce photometric variability since the turbulent motions are greater, making local clearings more likely, and enhancing the potential contrast between clear and cloudy air. Indeed the great red spot of Jupiter produces a photometric signal in both reflected sunlight and emitted thermal radiation (Gelino & Marley 2000).
Horizontally varying silicate clouds, even if not of the appropriate scale to produce a varying photometric signal, may play an important role in the transition from the dusty L-dwarfs to the relatively cloud-free T-dwarfs. The change in $J-K$ color from the latest red L-dwarfs ($J-K\sim2$) to the blue T-dwarfs ($J-K\sim0$) is quite abrupt. Four L8 dwarfs with known or estimated absolute magnitudes are only 1 magnitude brighter in $J$ band (Reid et al. 2001) than Gl229B. Reid et al. argue that this implies the L8 dwarfs are only about 250 K warmer than Gl229B. Even with the silicate cloud deck forming at progressively deeper levels with falling $T_{\rm eff}$, it may be difficult to account for such rapid color variation. In fact the rapid transition may be a signature of horizontally varying clouds. Once tropspheric convective patterns begin to produce substantial horizontal variability, the flux from the more cloud-free regions will begin to dominate the total emitted flux, even if large fractions of the object are still cloudy. For example, Jupiter’s 5-$\mu$m flux is dominated by the relatively cloud-free ‘hot-spots’ (Westphal et al. 1974) that typically cover about 1% of the surface area of the planet (Orton et al. 1996). Thus the apparent rapid change from cloudy L-dwarfs to clear T-dwarfs may be due to a gradual change in cloud coverage in the visible atmosphere, with the larger flux from the clear regions quickly dominating.
Summary
=======
We have developed a simple cloud model for substellar atmospheres that includes precipitation by condensate particles larger than that set by the convective velocity scale, which permits us to reproduce the properties retrieved from Jovian ammonia clouds. Effective precipitation also produces cloud profiles in theoretical brown dwarf and extrasolar giant planet atmospheres that are broadly consistent with observations.
As in the solar system, real clouds in the atmospheres of substellar objects will likely be neither uniform nor homogeneous, however we hope that this model will provide a framework for evaluating the globally-averaged role such clouds play in controlling the thermal radiative transfer and spectra of brown dwarfs and extrasolar giant planets.
We thank Sarah Beckmann for detecting anomalous behavior in an early version of the model. We also thank Robert West and Kevin Zahnle for providing helpful comments on the manuscript. M.S.M. acknowledges support from NASA grants NAG58919 and NAG59273 as well as NSF grants AST 9624878 and AST 0086288.
Saturation Vapor Pressures {#pvap}
==========================
For the saturation vapor pressure of ammonia ($e_{\rm s}$, in dyne cm$^{-2}$) we fit the measurements tabulated in the CRC handbook (Weast, 1971) with $$\begin{aligned}
e_{\rm s}({\rm NH_3}) & = & \exp \left(
10.53 - \frac{2161}{T} - \frac{86596}{T^2} \right)\end{aligned}$$ where the temperature is in K.
For the vapor pressure of water we use the expressions of Buck (1981), over ice for $T~<~273.16$ K, and over liquid water at greater temperatures:
$$\begin{aligned}
e_{\rm s}({\rm H_2O,ice}) & = & 6111.5\exp \left(
\frac{ 23.036 T_{\rm c} - T_{\rm c}^2/333.7} { T_{\rm c} + 279.82 } \right) \\
e_{\rm s}({\rm H_2O,liquid}) & = & 6112.1\exp \left(
\frac{ 18.729 T_{\rm c} - T_{\rm c}^2/227.3} { T_{\rm c} + 257.87 } \right)\end{aligned}$$
where $T_{\rm c}$ is the temperature in degrees Celsius. These expressions are unsuitable at $T > 1048$ K, leading to vapor pressures that decrease with with increasing temperatures. Hence, at greater temperatures we simply fix $e_{\rm s} \rm (H_2O) = 6\times10^8 \: \rm dyne \: cm^{-2}$, which is its value at $T = 1048$ K.
The vapor pressures for iron and enstatite are taken from Barshay and Lewis (1976). For iron below and above its melting point of 1800 K, we use, respectively
$$\begin{aligned}
e_{\rm s}({\rm Fe,solid}) & = & \exp \left(15.71 - \frac{47664}{T} \right) \\
e_{\rm s}({\rm Fe,liquid}) & = & \exp \left(9.86 - \frac{37120}{T} \right)\end{aligned}$$
and for enstatite we use $$\begin{aligned}
e_{\rm s}({\rm MgSiO_3}) & = & \exp \left( 25.37 - \frac{58663}{T} \right)\end{aligned}$$
Sedimentation Velocities {#vfall}
========================
Droplet terminal fallspeeds are calculated by first assuming viscous flow around spheres corrected for gas kinetic effects: $$v_{\rm f} = \frac {2}{9} \frac { \beta g r^2 \Delta \rho} {\eta}$$ where $g$ is the gravitational acceleration, $r$ is the droplet radius, and $\Delta \rho = \rho_{\rm p} - \rho_{\rm a}$ is the difference between the densities of the condensate and the atmosphere. The Cunningham slip factor, $\beta~=~(1~+~1.26 N_{\rm Kn})$, accounts for gas kinetic effects, in which the Knudsen number ($N_{\rm Kn}$) is the ratio of the molecular mean free path to the droplet radius. The dynamic viscosity of the atmosphere is given by Rosner (2000): $$\eta = \frac{5}{16}
\frac{ \sqrt{\pi m k_{\rm B} T} }
{ \pi d^2}
\frac{ \left( k_{\rm B} T / \epsilon \right)^{0.16} }
{ 1.22 }$$ where $d$ is the molecular diameter and $\epsilon$ is the depth of the Lennard-Jones potential well for the atmosphere ($2.827\times10^{-8}$ cm and $59.7 k_{\rm B}$ K, respectively, for H$_2$) and $k_{\rm B}$ is the Boltzmann constant.
For turbulent flow, at Reynolds numbers ($N_{\rm Re} = 2 r \rho_{\rm a} v_{\rm f} / \eta $) between 1 and 1000, we use a standard trick to solve the drag problem. Noting that $C_{\rm d} N_{\rm Re}^2 = 32 \rho_{\rm a} g r^3 \Delta \rho / 3 \eta^2$ is independent of fall velocity, we fit $y = \log( N_{\rm Re} )$ as a function of $x = \log( C_{\rm d} N_{\rm Re}^2 )$ to the following data: at $N_{\rm Re}$ = 1 we assume viscous flow, with $C_{\rm d} = 24$; for intermediate Reynolds numbers we use the data for rigid spheres from Table 10-1 of Pruppacher & Klett (1978); and at $N_{\rm Re}$ = 1000 we assume an asymptote of $C_{\rm d} = 0.45$. This asymptote is appropriate to moderately oblate spheroids (Figure 10-36 in Pruppacher & Klett 1978), which are more appropriate to unknown condensates than the extreme case of smooth spheres. Our fit to the data is $y = 0.8x - 0.01x^2$, which allows us to evaluate the droplet terminal fall velocity from $N_{\rm Re}$.
At Reynolds numbers $>$ 1000 we assume the drag coefficient is fixed at its asymptotic value ($C_{\rm d}~=~0.45$), which leads to $$v_{\rm f} = \beta \sqrt{ \frac{ 8 g r \Delta \rho } {3 C_{\rm d} \rho_{\rm a}} }$$
We assume rigid particles and thereby ignore breakup, for instance by liquid droplets due to hydrodynamic instability. For the density of ammonia ice particles, we use 0.84 g cm$^{-3}$ (Manzhelii & Tolkachev 1964); for water we use 0.93 g cm$^{-3}$ (corresponding to ice at a temperature of 200 K, using Equation 4-17 from Pruppacher & Klett 1978); and for enstatite and iron we use 3.2 and 7.9 g cm$^{-3}$, respectively (Table 1.18 of Lodders & Fegley, 1998).
Ackerman, A. S., Toon, O. B., & Hobbs, P. V., 1993, Science, 262, 226 Ackerman, A. S., Toon, O. B., Stevens, D. E., Heymsfield, A. J., Ramanathan, V., & Welton, E. J. 2000a, Science, 288, 1042 Ackerman, A. S., Toon, O. B., Taylor, J. P., Johnson, D. W., Hobbs, P. V., & Ferek, R. J. 2000b, J. Atmos. Sci., 57, 2684 Allard, F., Hauschildt, P. H., Baraffe, I. & Chabrier, G. 1996, , 465, L123 Bailer-Jones, C. A. L. & Mundt, R. 2001, , 367, 218 Barshay, S., S., & Lewis, J. S. 1976, , 14, 81 Banfield, D., Gierasch, P. J., Bell, M., Ustinov, E., Ingersoll, A. P., Vasavada, A. R., West, R. A., & Belton, M. J. S. 1998, Icarus, 135, 230 Brooke, T. Y., Knacke, R. F., Encrenaz, Th., Drossart, P., Crisp, D., & Feuchtgruber, H., 1998, Icarus, 136, 1 Basri, G., Mohanty, S., Allard, F., Hauschildt, P. H., Delfosse, X., Mart[í]{}n, E. L., Forveille, T. & Goldman, B. 2000, , 538, 363 Buck, 1981, J. Atmos. Sci. 20, 1527 Burrows, A. et al. 1997, , 491, 856 Calahan, R. F., Ridgeway, W., Wiscombe, W. J., Bell, T. L., & Snider, J. B. 1994, J. Atmos. Sci., 51, 2434 Carlson, B. E., Rossow, W. B., & Orton, G. S. 1988, J. Atmos. Sci., 45, 2066 Carlson, B. E., Lacis, A. A., & Rossow, W. B. 1993, J. Geophys. Res., 98, 5251 Carlson, B. E., Lacis, A. A., & Rossow, W. B. 1994, J. Geophys. Res., 99, 14623 Chabrier, G., Baraffe, I., Allard, F. & Hauschildt, P. 2000, , in press Fan, X. et al. 2000, , 119, 928 Fegley, B. J. & Lodders, K. 1994, Icarus, 110, 117 Geballe, T., Saumon, D., Leggett, S., Knapp, G., Marley, M. & Lodders, K. 2001 , in press Gelino, C. & Marley, M. 2000, in ASP Conf. Ser. 212, From Giant Planets to Cool Stars, ed. C. A. Griffith & M. S. Marley, (Flagstaff: ASP), 322 Gierasch, P. J. & Conrath, B. J. 1985, in Recent Advances in Planetary Meteorology, ed. G. E. Hunt, (New York: Cambridge University Press), 121 Jensen, E. J. et al. 2000, J. Geophys. Res., in press Jones, H. R. A. & Tsuji, T. 1997, , 480, L39 Kirkpatrick, J. D. et al. 1999, , 519, 802 Kunde, V. et al., 1982, , 263, 443 Leggett, S. K., Allard, F. & Hauschildt, P. H. 1998, , 509, 836 Leggett, S. K., Toomey, D. W., Geballe, T. R. & Brown, R. H. 1999, , 517, L139 Leggett, S. K. et al. 2000, , 536, L35 Lewis, J. S. 1969, Icarus, 10, 365 Lindal, G. F. et al. 1981, , 86, 8721 Lindzen, R. S. 1981, J. Atmos. Sci., 86, 9707 Lodders, K. & Fegley, B., Jr., 1998, The Planetary Scientist’s Companion, New York: Oxford Lunine, J. I., Hubbard, W. B., Burrows, A., Wang, Y-P., & K. Garlow 1989, , 338, 314 Manzhelii, V. G., & Tolkachev, A. M. 1964, Sov. Phys. Solid State, 5, 2506 Marley, M. S., Saumon, D., Guillot, T., Freedman, R. S., Hubbard, W. B., Burrows, A. & Lunine, J. I. 1996, Science, 272, 1919 Marley, M. S., Gelino, C., Stephens, D., Lunine, J. I., & Freedman, R. 1999, , 513, 879 Marley, M. S. 2000, in ASP Conf. Ser. 212, From Giant Planets to Cool Stars, ed. C. A. Griffith & M. S. Marley, (Flagstaff: ASP), 152 Mart[í]{}n, E. L., Delfosse, X., Basri, G., Goldman, B., Forveille, T., & Zapatero Osorio, M. R. 1999, , 118, 2466 Nicholls, S. 1984, Q. J. Roy. Met. Soc., 112, 431 Orton, G. et al. 1996, Science, 272, 839 Prinn, R. G. 1974, J. Atmos. Sci., 31, 1691 Prinn, R. G., & Olaguer, E. P. 1981, J. Geophys. Res., 86, 9895 Pruppacher, H. R., & Klett, J. D. 1978, Microphysics of Clouds and Precipitation, Dordrecht: Kluwer Reid, I. N., Gizis, J. E., Kirkpatrick, J. D., & Koerner, D. W. 2001, , 121, 489 Rosner, D. E., 2000, Transport Processes in Chemically Reacting Flow Systems, Dover: Mineola Rossow, W. B. 1978, Icarus, 36, 1 Saumon, D., Geballe, T. R., Leggett, S. K., Marley, M. S., Freedman, R. S., Lodders, K., Fegley, B. & Sengupta, S. K. 2000, , 541, 374 Schubert, G. & Zhang K. 2000, in ASP Conf. Ser. 212, From Giant Planets to Cool Stars, ed. C. A. Griffith & M. S. Marley, (Flagstaff: ASP), 210 Showman, A. P., D., & A. P. Ingersoll 1998, Icarus, 132, 205 Stephens, D., Marley, M., Noll, K., & Chanover, N. 2001, , in press Stevens, B., Cotton, W. R., Feingold, G., Moeng, C.-H., 1998 J. Atmos. Sci., 55, 3616 Sudarsky, D., Burrows, A. & Pinto, P. 2000, , 538, 885 Taylor, J. P, & Ackerman, A. S., 1999 Q. J. Roy. Met. Soc., 125, 2643 Toon, O. B., McKay, C. P., Courtin, R., and Ackerman, T. P. 1988, Icarus, 75, 255 Tsuji, T., Ohnaka, K., Aoki, W. & Nakajima, T. 1996, , 308, L29 Tsvetanov, Z. I. et al. 2000, , 531, L61 Weidenschilling, S. J., & Lewis, J. S. 1973, Icarus, 20, 465 Weast, R. C., ed. 1971, Handbook of Chemistry and Physics. 52nd ed., Boca Raton: CRC Press West, R. A., D. F. Strobel, & Tomasko, M. G. 1986, Icarus, 65, 161 Westphal, J. A., Matthews, K., & Terrile, R. J. 1974, , 188, L111
[ccc]{}
$f_{\rm rain}$ & 3 & Ratio of mass-weighted sedimentation velocity to convective velocity scale\
$K_{\rm min}$ & $10^5$ cm$^2$ s$^{-1}$ & Minimum value of eddy diffusion coefficient\
$\Lambda$ & 0.1 & Minimum ratio of turbulent mixing length to atmospheric scale height\
$S{\rm cloud}$ & 0 & Supersaturation that persists after accounting for condensation\
$\sigma_{\rm g}$ & 2 & Geometric standard deviation in lognormal size distributions of condensates
[^1]: The maximum size of precipitation particles is set by their breakup due to hydrodynamic instability, which for terrestrial raindrops occurs at a radius of $\sim$3 mm.
[^2]: Carlson et al. (1994) report cloud optical depths for an assumed extinction efficiency of 1, whereas by treating cloud particles as geometric scatterers, we assume an extinction efficiency of 2. Hence, we multiply their optical depths by 2 for comparison with ours.
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'R. Carrera'
- 'E. Pancino'
date: 'Received September 15, 2110; accepted March 16, 2220'
title: 'Chemical abundance analysis of the open clusters Berkeley 32, NGC 752, Hyades, and Praesepe[^1]'
---
[Open clusters are ideal test particles for studying the chemical evolution of the Galactic disc. However, the number and accuracy of existing high-resolution abundance determinations, not only of \[Fe/H\], but also of other key elements, remains largely insufficient.]{} [We attempt to increase the number of Galactic open clusters that have high quality abundance determinations, and to gather all the literature determinations published so far.]{} [Using high-resolution (R$\sim$30000), high-quality (S/N$\geq$60 per pixel), we obtained spectra for twelve stars in four open clusters with the fibre spectrograph FOCES, at the 2.2 Calar Alto Telescope in Spain. We employ a classical equivalent-width analysis to obtain accurate abundances of sixteen elements: Al, Ba, Ca, Co, Cr, Fe, La, Mg, Na, Nd, Ni, Sc, Si, Ti, V, and Y. We derived oxygen abundances derived by means of spectral synthesis of the 6300 Å forbidden line.]{} [We provide the first determination of abundance ratios other than Fe for NGC 752 giants, and ratios in agreement with the literature for the Hyades, Praesepe, and Be 32. We use a compilation of literature data to study Galactic trends of \[Fe/H\] and \[$\alpha$/Fe\] with Galactocentric radius, age, and height above the Galactic plane. We find no significant trends, but some indication for a flattening of \[Fe/H\] at large $R_{gc}$, and for younger ages in the inner disc. We also detect a possible decrease in \[Fe/H\] with $|$z$|$ in the outer disc, and a weak increase in \[$\alpha$/Fe\] with $R_{gc}$.]{}
Introduction {#sec1}
============
Open clusters (hereafter OC) are ideal [*test particles*]{} for studying the evolution of metallicity with time, inferring the so-called [*age-metallicity relation*]{}, and with Galactocentric radius, the [*metallicity gradient*]{}, measuring in the Galactic disc. Their properties can be determined with smaller uncertainties than for field stars, since they are coeval group of stars at the same distance that have a homogeneous chemical composition. Unfortunately, of the $\simeq$1700 known OC [e.g. @dias2002], only $\simeq$140 possess some metallicity determination, mostly obtained from photometric indicators, such as Washington or Strömgren photometry [see @twarog1997; @chen2003 and references therein] and low-resolution spectroscopy [e.g. @friel1993; @friel02].
The most accurate way to determine the chemical abundances is to analyse high-resolution spectroscopy. It allows us to investigate not only metallicity, but also abundance ratios – with respect to iron or hydrogen – of other chemical species such as $\alpha$-elements, $s$-process elements, and $r$-process elements, which are synthesised in different environments and on different timescales (e.g. SNe Ia, SNe II, giants, supergiants, etc). In the past few years, a number of research groups have addressed the challenge of increasing the number of OC with chemical abundances determined from high-resolution spectroscopy [e.g. @sestito2004; @dorazi2006; @sestito2006; @bragaglia2008; @pace2008; @dorazi2009; @friel2010; @pace2010; @pancino2010a; @jacobson2011]. However, the number of OC with chemical abundances determined with this technique is still small (see Section \[sec6\]), and significant uncertainties remain in the determinations of both the metallicity gradient and the age-metallicity relation, which are the fundamental ingredients of chemical evolution models.
--------------- ---------------------- ----------------- ----------------- ------- ------- ------- ------- --------- ----------- ----------------- ------------- --
Cluster Star $\alpha_{2000}$ $\delta_{2000}$ B V R I[^2] K$_{S}$ n$_{exp}$ t$^{tot}_{exp}$ S/N$^{tot}$
(hrs) (deg) (mag) (mag) (mag) (mag) (mag) (sec)
Be 32[^3] 0456 06:58:08.2 +06:24:19.6 14.76 13.67 — 12.53 11.03 7 18900 60
1948 06:58:04.2 +06:27:17.1 14.50 13.37 — 12.20 10.68 6 16200 70
NGC 752[^4] 001 01:55:12.6 +37:50:14.6 10.47 9.51 — — 7.23 4 2400 160
208 01:57:37.6 +37:39:38.1 10.04 8.97 — — 6.41 4 2400 180
213 01:57:38.9 +37:46:12.5 10.08 9.07 — — 6.68 3 1800 80
311 01:58:52.9 +37:48:57.3 10.11 9.07 — — 6.64 4 2400 100
Hyades[^5] 028 ($\gamma$ tau) 04:19:47.6 +15:37:39.5 4.64 3.65 2.92 2.45 1.52 2 120 560
(Mel 25) 041 ($\delta$ tau) 04:22:56.1 +17:32:33.0 4.75 3.76 3.03 2.56 1.64 3 180 450
070 ($\epsilon$ tau) 04:28:37.0 +19:10:49.5 4.55 3.54 2.81 2.31 1.42 3 180 270
Praesepe [^6] 212 08:39:50.7 +19:32:27.0 7.53 6.58 5.87 5.38 4.39 4 240 165
(NGC 2632) 253 08:40:06.4 +20:00:28.1 7.35 6.38 5.67 5.20 4.20 4 240 215
(M 44) 283 08:40:22.1 +19:40:11.9 7.42 6.41 5.68 5.21 4.18 2 120 150
--------------- ---------------------- ----------------- ----------------- ------- ------- ------- ------- --------- ----------- ----------------- ------------- --
In this paper, the second of a series initiated by @pancino2010a [hereafter Paper I], we present high quality and homogeneous measurements of chemical abundances for red clump stars in four OC: Be 32, NGC 752, Hyades, and Praesepe. The Hyades is the nearest OC and its four known red giants have been widely studied [@schuler2009; @mishenina2007; @fulbright2007; @schuler2006; @mishenina2006; @boyarchuk2000; @luck1995], hence it provides a very good reference frame to compare our abundances with the literature. Both NGC 752 and Praesepe have been well-studied, but all information about their chemical composition is based mainly on their main-sequence stars [e.g. @pace2008; @an2007; @sestito2004; @burkhart1998; @hobbs1992]. To our knowledge, there have been no recent measurements of the chemical abundance of their giants from high-resolution spectroscopy. Finally, Be 32 has been the subject of some studies [e.g. @Richtler2001; @friel2010; @bragaglia2008; @dorazi2006]. The properties and previous studies of each cluster is described in more depth in Section \[sec5\].
This paper is structured as follows: observations and data reduction are described in Section \[sec2\]; equivalent-width measurements are presented in Section \[sec3\], together with the abundance analysis and its uncertainties; results are compared with the literature in Sections \[sec5\], \[sec6\], and \[sec7\]; and finally our main conclusions are summarised in Section \[sec8\].
Observational material {#sec2}
======================
A total of twelve stars spread in the four OC were observed. They were selected from the WEBDA[^7] database [@mermilliod1995], and the 2MASS[^8] survey [@2mass; @2mass2]. Table \[obslog\] summarizes the identifications, coordinates, and magnitudes of each target star. Their position in the color-magnitude diagram taken from @dorazi2006, @johnson1953, @johnson1955, and @johnson1952 for Berkeley 32, NGC 752, Hyades, and Preasepe, respectively, are shown in Figure \[CMDs\].
Observations were carried out with the fibre echelle spectrograph FOCES [@foces] attached at the 2.2 m Calar Alto Telescope (Almeria, Spain) between the 1 and 3 of January 2005. The chosen set-up provides a spectral resolution (R=$\lambda/\delta\lambda$) of about 30000. In summary, all stars were observed in 2–7 exposures lasting 10–30 min each, depending on their magnitudes, until a global signal-to-noise ratio (S/N) of at least 60 per pixel was reached around 6000 Å. Exposures with S/N$<$20 were neglected because they were too noisy. Finally, sky exposures as long as our longest exposures (30 min) were taken, but the levels were sufficiently low for us to avoid sky subtraction (as in Paper I). The number of useful exposures, the total integration time, and the global S/N for each star are listed in the last three columns of Table \[obslog\].
![Location of target stars (large black dots with star ID labels) in the color-magnitude diagrams of their respective parent clusters (small grey dots).[]{data-label="CMDs"}](fig_dcm.eps){height="16cm" width="\columnwidth"}
Data reduction
--------------
Various steps of data reduction were performed exactly as in Paper I. Briefly, the frames were de-trended with the IRAF[^9] tasks *ccdproc* and *apflatten*. The spectra were then extracted, wavelength-calibrated, normalized, and the echelle orders were merged using tasks in the IRAF *echelle* package. Finally, the noisy ends of each combined spectrum were cut, allowing for an effective wavelength coverage from 5000 to 9000 Å.
Before combining all exposures of each star, we removed sky absorption features (telluric bands of O$_2$ and H$_2$O) with the help of the IRAF task *telluric*. The same two hot, rapidly rotating stars, HR 3982 and HR 8762, of Paper I were used. The strong O$_2$ band around 7600 Å had been saturated and therefore could not be properly removed. This spectral region was not used for the abundance analysis, in addition to the small gaps between echelle orders that appeared after $\lambda\simeq$8400 Å.
Radial velocities
-----------------
-------------------------- ---------- ----------------------------- ---------------------------- -- --
Cluster Star $(V_r\pm\delta V_r)_{here}$ $(V_r\pm\sigma V_r)_{lit}$
(km s$^{-1}$) (km s$^{-1}$)
Be 32[^10] 0456 105.59$\pm$0.54 110.0$\pm$1.2
1948 104.78$\pm$0.35 105.5$\pm$4.9
NGC 752[^11] 001 5.49$\pm$0.44 4.79$\pm$0.15
208[^12] 1.10$\pm$0.23 4.86$\pm$0.06
213 5.11$\pm$0.42 5.50$\pm$0.10
311 6.00$\pm$0.30 5.28$\pm$0.08
Hyades[^13] 028 38.15$\pm$0.43 39.28$\pm$0.12
041 38.56$\pm$0.36 39.65$\pm$0.08
070 38.26$\pm$0.35 39.37$\pm$0.07
Praesepe[^14] (NGC 2632) 212 35.96$\pm$0.36 34.81$\pm$0.21
253 34.39$\pm$0.27 33.67$\pm$0.22
283 34.67$\pm$0.39 34.35$\pm$0.20
-------------------------- ---------- ----------------------------- ---------------------------- -- --
: Heliocentric radial velocity measurements and 1$\sigma$ errors ($V_r \pm \delta V_r$)$_{here}$ for each programme star. Literature measurements are also reported with their uncertainties ($V_r \pm \delta V_r$)$_{lit}$.[]{data-label="radvel"}
We used DAOSPEC [@daospec] to measure the observed radial velocities for each individual exposure with S/N$\geq$20, using $\simeq$ 300 absorption lines of different elements, with typical uncertainties of about 0.1 km s$^{-1}$ (see Paper I for details). We used the same linelist as the one used for abundance determinations (see Section \[sec3\] for details). Heliocentric corrections were obtained with the IRAF task [*rvcorrect*]{}, with a negligible uncertainty of smaller than 0.005 km s$^{-1}$. We also used DAOSPEC to determine the absolute zero-point of the radial velocity determinations, using a list of telluric absorption lines as the input linelist, obtained from the GEISA database [@geisa1; @geisa2]. The resulting zero-point corrections, based on $\simeq$250 telluric lines, are generally no larger than $\pm$1 km s$^{-1}$, with a typical error of about $\simeq$0.5 km s$^{-1}$.
The final values, computed as the weighted mean of heliocentric velocities resulting from each exposure of the same star, are listed in Table \[radvel\]. Our determinations are generally in close agreement with literature values to within 3$\sigma$, except for star 208 in NGC 752, which has a slightly smaller44 radial velocity than other objects in this cluster. The fact that this star was recognised as a spectroscopic binary [see @pourbaix2004; @mermilliod2007] explains the disagreement. According to its radial velocity curve [@mermilliod2007], we observed this binary near minimum, which implies that we observed only one of the components of the system. For this reason, and because derived abundances are in good agreement with those of other stars in the same cluster, we retained this object in our final sample. In summary, we considered all the observed targets as likely members of their respective clusters.
Photometric parameters {#secphot}
----------------------
First guesses of the atmospheric parameters effective temperature ($T_{\rm{eff}}$), logarithmic gravity ($\log g$), and microturbulent velocity ($v_t$), for our target stars were derived from a photometric data listed in Table \[obslog\], as described in Paper I. In brief, $T_{\rm{eff}}$ were obtained using the @alonso99 and @montegriffo98 colour-temperature relations, both theoretical and empirical, and the dereddened colours (B-V)$_0$, (V-I$_J$)$_0$, (V-R)$_0$, and (V-K$_S$)$_0$. We assumed the E(B-V) values listed in Table \[phot\] and the reddening laws of @cardelli89. In the case of Be 32, we have I$_C$ magnitudes instead of I$_J$ ones, so we dereddened (V-I$_C$) with the law of @dean1978, and converted it into (V-I$_J$)$_0$ with the transformations by @bessell1979. The 1$\sigma$ errors in each $T_{\rm{eff}}$ estimate were computed using the magnitude and reddening uncertainties together with the standard deviation in the colour-temperature relationships used. The photometric $T_{\rm{eff}}$ estimates, listed in Table \[pars\], are the weighted mean of the different values obtained from each considered colour and colour-temperature relations.
Photometric gravity estimates were derived from the above $T_{\rm{eff}}$ and the bolometric corrections, BC$_V$, derived using the @alonso99 prescriptions and the fundamental relationships
$$\begin{aligned}
\nonumber
\log \frac{g}{g_\odot}=\log \frac{M}{M_\odot}+2\log
\frac{R_\odot}{R},\\
\nonumber
0.4 (M_{\rm{bol}}-M_{\rm{bol},\odot})=-4\log
\frac{T_{\rm{eff}}}{T_{\rm{eff},\odot}}+2\log \frac{R_\odot}{R},\end{aligned}$$
where red clump masses, listed in the last column of Table \[pars\], were extrapolated from Table 1 of @girardi2001. We assumed that $\log
g_{\odot}=4.437$, $T_{\rm{eff},\odot}=5770~\rm{K}$ and $M_{\rm{bol},\odot}=4.75$, in conformity with the IAU recommendations [@andersen1999]. As above, we averaged all our estimates to obtain $\log g^{(phot)}$, listed in column 5 of Table \[pars\].
As discussed in Paper I, the photometric estimate of the microturbulent velocity, $v_t$, was obtained using the prescriptions both of @ramirez2003, $v_t=4.08-5.01~10^{-4}~T_{\rm{eff}}$, and of @carretta2004, $v_t=1.5-0.13~\log g$. The latter velocity, which takes into account the effect described by @magain1984[^15], is on average lower by $\Delta v_t=0.50 \pm 0.03$ km s $^{-1}$ than the @ramirez2003 estimate. Therefore, we chose not to average the two estimates, but to use them as an indication of the $v_t$ range to explore in our abundance analysis (see Section \[sec4\]).
-------------- --------------------- -------------------- --------------------
Cluster E(B–V) (m-M)$_o$ Age
(mag) (mag) (Gyr)
Be 32[^16] 0.15$\pm$0.05 12.62$\pm$0.18 4.8$\pm$1.5
NGC 752[^17] 0.038$\pm$0.002 8.04$\pm$0.23 1.59$\pm$0.45
Hyades $\leq$0.001[^18] 3.34$\pm$0.01[^19] 0.70$\pm$0.07[^20]
Praesepe 0.027$\pm$0.004$^c$ 6.22$\pm$0.02[^21] 0.65[^22]$\pm$0.25
-------------- --------------------- -------------------- --------------------
: Adopted cluster parameters. When more than one determination exists, the average is shown with 1 $\sigma$ errors.[]{data-label="phot"}
---------- ------ ------------------------- ------------------------- ------------------- ------------------- ----------------------------- ---------------- ------------- --
Cluster Star T$_{\rm{eff}}^{(phot)}$ T$_{\rm{eff}}^{(spec)}$ $\log g^{(phot)}$ $\log g^{(spec)}$ $v_t^{(phot)}$ $v_t^{(spec)}$ M$_{clump}$
(K) (K) (cgs) (cgs) (km s$^{-1}$) (km s$^{-1}$) (M$_\odot$)
Be 32 0456 4759$\pm$92 4650 2.61$\pm$0.14 2.1 1.70$\pm$0.30/1.16$\pm$0.10 1.4 1.2$\pm$0.1
1948 4706$\pm$99 4700 2.47$\pm$0.14 2.3 1.72$\pm$0.30/1.18$\pm$0.10 1.5 1.2$\pm$0.1
NGC 752 001 4949$\pm$80 5050 3.02$\pm$0.14 3.1 1.60$\pm$0.30/1.11$\pm$0.10 1.3 1.9$\pm$0.2
208 4698$\pm$110 4600 2.73$\pm$0.14 2.9 1.73$\pm$0.31/1.15$\pm$0.10 1.2 1.9$\pm$0.2
213 4841$\pm$86 4900 2.81$\pm$0.14 3.0 1.65$\pm$0.30/1.13$\pm$0.10 1.4 1.9$\pm$0.2
311 4793$\pm$74 4800 2.80$\pm$0.14 3.2 1.68$\pm$0.30/1.14$\pm$0.10 1.2 1.9$\pm$0.2
Hyades 028 4865$\pm$73 4750 2.67$\pm$0.04 2.7 1.64$\pm$0.30/1.15$\pm$0.15 1.4 2.5$\pm$0.1
041 4871$\pm$79 4800 2.71$\pm$0.05 2.8 1.64$\pm$0.30/1.15$\pm$0.15 1.4 2.5$\pm$0.1
070 4858$\pm$95 4800 2.62$\pm$0.05 2.8 1.65$\pm$0.30/1.16$\pm$0.15 1.6 2.5$\pm$0.1
Praesepe 212 4901$\pm$35 4900 2.70$\pm$0.07 2.8 1.62$\pm$0.30/1.15$\pm$0.14 1.5 2.6$\pm$0.3
253 4869$\pm$23 4900 2.60$\pm$0.07 2.8 1.64$\pm$0.30/1.16$\pm$0.14 1.6 2.6$\pm$0.3
283 4841$\pm$29 4800 2.61$\pm$0.07 2.9 1.65$\pm$0.30/1.16$\pm$0.14 1.4 2.6$\pm$0.3
---------- ------ ------------------------- ------------------------- ------------------- ------------------- ----------------------------- ---------------- ------------- --
...
----------- ------ ------------------ ----------- ------- ------------ ------- ------- ------------ ------- ----- ------- ------------ -------
$\lambda$ Elem $\chi_{\rm{ex}}$ $\log gf$ EW $\delta$EW Q EW $\delta$EW Q ... EW $\delta$EW Q
(A) (eV) (dex) (mÅ) (mÅ) (mÅ) (mÅ) ... (mÅ) (mÅ)
6696.79 AL1 4.02 -1.42 16.5 1.9 0.361 25.3 3.9 0.481 ... 30.0 5.0 0.585
6698.67 AL1 3.14 -1.65 47.8 2.7 0.432 43.4 2.6 0.501 ... 59.2 1.8 0.321
7361.57 AL1 4.02 -0.90 33.8 3.9 0.633 33.2 6.5 0.946 ... 52.1 2.1 0.173
7362.30 AL1 4.02 -0.75 48.6 4.3 0.346 39.7 2.9 0.737 ... 72.0 6.9 0.797
7835.31 AL1 4.02 -0.65 53.7 3.7 0.393 48.4 3.9 0.611 ... 95.8 9.8 1.399
7836.13 AL1 4.02 -0.49 66.0 7.8 1.279 68.4 5.1 0.783 ... 94.0 6.0 0.665
8772.86 AL1 4.02 -0.32 ... ... ... ... ... .... ... 102.7 8.7 0.748
8773.90 AL1 4.02 -0.16 115.0 8.6 1.151 111.9 8.3 1.111 ... ... ... ...
5853.67 BA2 0.60 -1.00 119.7 3.9 1.098 102.0 2.6 0.343 ... 122.1 3.4 0.676
------------------- ----------------- ---------------- ----------------- ----------------- ----------------- ---------------- -------------
External
Ratio Star 456 Star 1948 Star 001 Star 208 Star 213 Star 311 Uncertainty
$[$FeI/H$]$ –0.33$\pm$0.02 –0.27$\pm$0.02 +0.07$\pm$0.01 +0.07$\pm$0.01 +0.04$\pm$0.01 +0.14$\pm$0.01 $\pm$0.03
$[$FeII/H$]$ –0.30$\pm$ 0.06 –0.29$\pm$0.06 +0.02$\pm$0.03 +0.06$\pm$0.03 +0.05$\pm$0.04 +0.18$\pm$0.12 $\pm$0.03
$[$$\alpha$/Fe$]$ –0.29$\pm$0.21 –0.25$\pm$0.09 +0.07$\pm$0.04 +0.05$\pm$0.12 +0.07$\pm$0.12 +0.14$\pm$0.09 $\pm$0.07
$[$Al/Fe$]$ +0.15$\pm$0.06 +0.08$\pm$0.07 –0.11$\pm$0.04 –0.12$\pm$0.03 –0.06$\pm$0.06 –0.21$\pm$0.06 $\pm$0.05
$[$Ba/Fe$]$ +0.52$\pm$0.05 +0.35$\pm$0.17 +0.55$\pm$0.04 +0.52$\pm$0.04 +0.51$\pm$0.01 +0.57$\pm$0.06 $\pm$0.04
$[$Ca/Fe$]$ –0.06$\pm$0.08 –0.05$\pm$0.04 –0.02$\pm$0.03 –0.12$\pm$0.02 –0.09$\pm$0.03 –0.17$\pm$0.05 $\pm$0.06
$[$Co/Fe$]$ +0.02$\pm$0.05 +0.09$\pm$0.04 –0.03$\pm$0.03 +0.06$\pm$0.04 +0.00$\pm$0.03 +0.05$\pm$0.05 $\pm$0.04
$[$Cr/Fe$]$ –0.25$\pm$0.07 +0.04$\pm$0.07 +0.02$\pm$0.03 +0.00$\pm$0.03 –0.01$\pm$0.03 –0.01$\pm$0.04 $\pm$0.05
$[$La/Fe$]$ –0.14$\pm$0.02 –0.04$\pm$0.08 +0.14$\pm$0.06 +0.18$\pm$0.03 +0.18$\pm$0.09 +0.32$\pm$0.13 $\pm$0.04
$[$Mg/Fe$]$ +0.38$\pm$0.12 +0.24$\pm$0.16 +0.13$\pm$0.06 +0.16$\pm$0.05 +0.20$\pm$0.04 +0.06$\pm$0.03 $\pm$0.09
$[$Na/Fe$]$ –0.14$\pm$0.08 –0.08$\pm$0.10 +0.05$\pm$0.01 –0.07$\pm$0.02 –0.03$\pm$0.05 -0.10$\pm$0.05 $\pm$0.04
$[$Nd/Fe$]$ –0.05$\pm$0.13 +0.04$\pm$0.03 +0.29$\pm$0.14 +0.27$\pm$0.23 +0.34$\pm$0.11 +0.46$\pm$0.18 $\pm$0.13
$[$Ni/Fe$]$ –0.04$\pm$0.03 –0.01$\pm$0.03 –0.04$\pm$0.02 +0.00$\pm$0.02 –0.02$\pm$0.02 +0.03$\pm$0.03 $\pm$0.02
$[$O/Fe$]$ –0.16$\pm$0.13 +0.15$\pm$0.11 +0.15$\pm$0.06 –0.06$\pm$0.05 +0.02$\pm$0.08 +0.00$\pm$0.06 $\pm$0.08
$[$Sc/Fe$]$ +0.02$\pm$0.05 –0.02$\pm$0.05 –0.02$\pm$0.05 +0.04$\pm$0.06 +0.05$\pm$0.06 +0.09$\pm$0.08 $\pm$0.05
$[$Si/Fe$]$ +0.18$\pm$0.04 +0.11$\pm$0.04 –0.03$\pm$0.03 +0.04$\pm$0.03 +0.04$\pm$0.03 +0.01$\pm$0.04 $\pm$0.04
$[$TiI/Fe$]$ –0.10$\pm$0.05 –0.04$\pm$0.05 +0.00$\pm$0.02 –0.03 $\pm$0.02 –0.08$\pm$0.02 –0.13$\pm$0.03 $\pm$0.03
$[$TiII/Fe$]$ –0.17$\pm$0.05 +0.01$\pm$0.07 +0.03$\pm$0.02 +0.08$\pm$0.07 +0.07$\pm$0.06 +0.15$\pm$0.13 $\pm$0.03
$[$V/Fe$]$ –0.14$\pm$0.10 –0.07$\pm$0.05 +0.00$\pm$0.02 +0.16$\pm$0.05 –0.04$\pm$0.03 +0.05$\pm$0.06 $\pm$0.06
$[$Y/Fe$]$ –0.41$\pm$N.A. –0.09$\pm$N.A. –0.12$\pm$0.06 –0.03$\pm$0.10 +0.03$\pm$0.07 +0.05$\pm$0.09 $\pm$0.04
Ratio Star 28 Star 41 Star 70 Star 212 Star 253 Star 283
$[$FeI/H$]$ +0.12$\pm$0.01 +0.10$\pm$0.01 +0.11$\pm$0.01 +0.11$\pm$0.01 +0.18$\pm$0.01 +0.21$\pm$0.01 $\pm$0.03
$[$FeII/H$]$ +0.13$\pm$0.03 +0.13$\pm$0.03 +0.09$\pm$0.03 +0.10$\pm$0.03 +0.16$\pm$0.03 +0.23$\pm$0.04 $\pm$0.03
$[$$\alpha$/Fe$]$ +0.13$\pm$0.15 +0.11$\pm$0.12 +0.09$\pm$0.11 +0.11$\pm$0.15 +0.18$\pm$0.14 +0.19$\pm$0.13 $\pm$0.07
$[$Al/Fe$]$ –0.01$\pm$0.05 +0.00$\pm$0.05 +0.02$\pm$0.05 +0.01$\pm$0.04 +0.02$\pm$0.06 –0.04$\pm$0.05 $\pm$0.05
$[$Ba/Fe$]$ +0.37$\pm$0.05 +0.39$\pm$0.05 +0.31$\pm$0.05 +0.30$\pm$0.08 +0.27$\pm$0.06 +0.37$\pm$0.05 $\pm$0.04
$[$Ca/Fe$]$ –0.07$\pm$0.03 –0.06$\pm$0.03 –0.07$\pm$0.02 –0.07$\pm$0.02 –0.08$\pm$0.03 –0.11$\pm$0.03 $\pm$0.06
$[$Co/Fe$]$ +0.00$\pm$0.04 +0.01$\pm$0.03 +0.06$\pm$0.03 +0.05$\pm$0.03 +0.01$\pm$0.03 +0.05$\pm$0.05 $\pm$0.04
$[$Cr/Fe$]$ +0.02$\pm$0.03 +0.03$\pm$0.03 +0.08$\pm$0.04 +0.06$\pm$0.03 +0.04$\pm$0.04 +0.04$\pm$0.04 $\pm$0.05
$[$La/Fe$]$ –0.12$\pm$0.06 –0.08$\pm$0.05 –0.05$\pm$0.05 –0.07$\pm$0.05 –0.04$\pm$0.05 –0.04$\pm$0.04 $\pm$0.04
$[$Mg/Fe$]$ +0.13$\pm$0.05 +0.06$\pm$0.04 +0.21$\pm$0.07 +0.31$\pm$0.06 +0.27$\pm$0.05 +0.22$\pm$0.06 $\pm$0.09
$[$Na/Fe$]$ +0.19$\pm$0.02 +0.18$\pm$0.02 +0.18$\pm$0.02 +0.23$\pm$0.02 +0.30$\pm$0.03 +0.18$\pm$0.05 $\pm$0.04
$[$Nd/Fe$]$ +0.04 $\pm$0.29 +0.08$\pm$0.30 +0.08$\pm$0.28 +0.00$\pm$0.21 +0.05$\pm$0.25 +0.10$\pm$0.31 $\pm$0.13
$[$Ni/Fe$]$ +0.02 $\pm$0.02 +0.04$\pm$0.02 +0.03$\pm$0.02 +0.01$\pm$0.02 +0.01$\pm$0.02 +0.04$\pm$0.03 $\pm$0.02
$[$O/Fe$]$ –0.35$\pm$0.07 –0.25$\pm$0.05 –0.22$\pm$ 0.07 –0.11$\pm$0.09 –0.14$\pm$0.07 –0.09$\pm$0.06 $\pm$0.08
$[$Sc/Fe$]$ –0.04 $\pm$0.05 +0.00$\pm$0.05 –0.02$\pm$0.06 –0.10$\pm$0.06 +-0.03$\pm$0.05 +0.00$\pm$0.06 $\pm$0.05
$[$Si/Fe$]$ +0.09 $\pm$0.03 +0.09$\pm$0.02 +0.10$\pm$0.03 +0.06$\pm$0.03 +0.07$\pm$0.03 +0.04$\pm$0.03 $\pm$0.04
$[$TiI/Fe$]$ –0.12 $\pm$0.02 –0.11$\pm$0.02 –0.06$\pm$0.02 –0.05$\pm$0.03 –0.08$\pm$0.02 –0.09$\pm$0.02 $\pm$0.03
$[$TiII/Fe$]$ –0.03 $\pm$0.06 +0.00$\pm$0.07 –0.02$\pm$0.11 –0.05$\pm$0.10 –0.02$\pm$0.08 +0.05$\pm$0.08 $\pm$0.03
$[$V/Fe$]$ +0.02 $\pm$0.04 +0.00$\pm$0.03 +0.09$\pm$0.03 +0.06$\pm$0.04 +0.04$\pm$0.03 +0.10$\pm$0.04 $\pm$0.06
$[$Y/Fe$]$ –0.12 $\pm$0.05 –0.06$\pm$0.06 –0.07$\pm$0.05 –0.11$\pm$0.10 –0.12$\pm$0.07 –0.11$\pm$0.09 $\pm$0.04
------------------- ----------------- ---------------- ----------------- ----------------- ----------------- ---------------- -------------
Equivalent widths and abundance analysis {#sec3}
========================================
We used the same linelist as that described in Paper I. In brief, all lines and their atomic data were extracted from the VALD database [@vald], with a few exceptions (see Paper I for details). Briefly, for some highly discrepant Mg lines, we used the NIST log$gf$ values; we used the @johansson03 log$gf$ for the Ni line that contaminates the \[O I\] line at 6300 Å, and provides oxygen abundances more in line with the other $\alpha$-elements; we used the Nd log$gf$ values by @denhartog2003, which minimize the spread in the Nd abundance. Finally, we tried both the VALD and the NIST values for Ca, finding an average difference of 0.17 dex (see paper I). There is no special reason for choosing NIST over VALD (or vice-versa), so we kept the VALD values to help maintain some homogeneity, but we note that the Ca log$gf$ values carry a large uncertainty of the order of 0.2 dex.
![Comparison of our EW measurements with those by @bragaglia2008 for star 0456 in Be 32, and by @boyarchuk2000 for three Hyades giants. Dotted lines mark perfect agreement (zero difference), while dashed lines are linear fits to the data.[]{data-label="compew"}](fig_comp_ew.eps){width="\columnwidth"}
Ratio Be 32 NGC 752 Hyades Praesepe (M 44)
----------------- --------------------------- --------------------------- --------------------------- --------------------------- --
$[$Fe/H$]$ –0.30$\pm$0.02($\pm$0.03) +0.08$\pm$0.04($\pm$0.03) +0.11$\pm$0.01($\pm$0.03) +0.16$\pm$0.05($\pm$0.03)
$[\alpha$/Fe$]$ –0.04$\pm$0.14($\pm$0.07) +0.02$\pm$0.06($\pm$0.07) +0.00$\pm$0.12($\pm$0.07) +0.00$\pm$0.14($\pm$0.07)
$[$Al/Fe$]$ +0.12$\pm$0.05($\pm$0.05) –0.12$\pm$0.06($\pm$0.05) +0.00$\pm$0.02($\pm$0.05) +0.00$\pm$0.03($\pm$0.05)
$[$Ba/Fe$]$ +0.51$\pm$0.12($\pm$0.04) +0.52$\pm$0.03($\pm$0.04) +0.36$\pm$0.04($\pm$0.04) +0.33$\pm$0.05($\pm$0.04)
$[$Ca/Fe$]$ –0.05$\pm$0.01($\pm$0.06) –0.09$\pm$0.06($\pm$0.06) –0.07$\pm$0.01($\pm$0.06) –0.08$\pm$0.02($\pm$0.06)
$[$Co/Fe$]$ +0.07$\pm$0.05($\pm$0.04) +0.01$\pm$0.04($\pm$0.04) +0.03$\pm$0.03($\pm$0.04) +0.04$\pm$0.02($\pm$0.04)
$[$Cr/Fe$]$ –0.11$\pm$0.21($\pm$0.05) +0.00$\pm$0.01($\pm$0.05) +0.04$\pm$0.03($\pm$0.05) +0.05$\pm$0.01($\pm$0.05)
$[$La/Fe$]$ –0.14$\pm$0.07($\pm$0.04) +0.18$\pm$0.08($\pm$0.04) –0.08$\pm$0.04($\pm$0.04) –0.05$\pm$0.02($\pm$0.04)
$[$Mg/Fe$]$ +0.33$\pm$0.10($\pm$0.09) +0.12$\pm$0.06($\pm$0.09) +0.10$\pm$0.08($\pm$0.09) +0.27$\pm$0.05($\pm$0.09)
$[$Na/Fe$]$ –0.12$\pm$0.04($\pm$0.04) +0.01$\pm$0.07($\pm$0.04) +0.18$\pm$0.01($\pm$0.04) +0.25$\pm$0.06($\pm$0.04)
$[$Nd/Fe$]$ +0.03$\pm$0.06($\pm$0.13) +0.34$\pm$0.09($\pm$0.13) +0.07$\pm$0.02($\pm$0.13) +0.04$\pm$0.05($\pm$0.13)
$[$Ni/Fe$]$ –0.03$\pm$0.02($\pm$0.02) –0.01$\pm$0.03($\pm$0.02) +0.03$\pm$0.01($\pm$0.02) +0.02$\pm$0.02($\pm$0.02)
$[$O/Fe$]$ +0.00$\pm$0.16($\pm$0.08) +0.03$\pm$0.04($\pm$0.08) –0.27$\pm$0.04($\pm$0.08) –0.11$\pm$0.03($\pm$0.08)
$[$Sc/Fe$]$ +0.00$\pm$0.03($\pm$0.05) +0.03$\pm$0.05($\pm$0.05) –0.02$\pm$0.02($\pm$0.05) –0.04$\pm$0.05($\pm$0.05)
$[$Si/Fe$]$ +0.14$\pm$0.05($\pm$0.04) +0.02$\pm$0.03($\pm$0.04) +0.09$\pm$0.01($\pm$0.04) +0.06$\pm$0.02($\pm$0.04)
$[$Ti/Fe$]$ –0.08$\pm$0.07($\pm$0.03) –0.03$\pm$0.06($\pm$0.03) –0.09$\pm$0.04($\pm$0.03) –0.07$\pm$0.03($\pm$0.07)
$[$V/Fe$]$ –0.08$\pm$0.05($\pm$0.06) +0.01$\pm$0.09($\pm$0.06) +0.04$\pm$0.05($\pm$0.06) +0.06$\pm$0.03($\pm$0.06)
$[$Y/Fe$]$ –0.23$\pm$0.23($\pm$0.04) –0.03$\pm$0.08($\pm$0.04) –0.09$\pm$0.03($\pm$0.04) –0.11$\pm$0.01($\pm$0.04)
Equivalent widths with DAOSPEC
------------------------------
The task DAOSPEC [@daospec] was used to automatically find and measure equivalent widths (hereafter EW), by performing a Gaussian fitting of the identified lines. DAOSPEC provides a formal error in the Gaussian fit, $\delta EW$, and a quality parameter, $Q$ [see @daospec and Paper I, for more details]. The relative error $\delta EW/EW$ and the quality parameter $Q$ can be used to distinguish good and bad lines, and they were indeed used to select the highest quality lines for the abundance analysis, as described in detail in Paper I. The measured EW for our program stars are shown in the electronic version of Table \[tab-ew\] along with the $\delta EW$ and $Q$ parameter estimated by DAOSPEC.
Four of our target stars have published EW measurements from high-resolution spectra. These consist of three stars (namely, 028, 041, and 070) observed in the Hyades by @boyarchuk2000 with $R\sim45000$, and star 0456 in Be 32 studied by @bragaglia2008 with $R\sim40000$. We have a total of 100, 92, and 51 lines in common for stars 028, 041, and 070 in the Hyades, respectively, and 51 lines for star 0456 in Be 32. Figure \[compew\] compares the comparison between the EW determined with DAOSPEC with the values published by @bragaglia2008 and @boyarchuk2000. The differences (see Figure \[compew\]) are negligible within the uncertainties; we find a small offset of 5.6 mÅ in the case of star 041 in the Hyades, which is however still within 1$\sigma$. We can therefore consider our measurements in good agreement with similar studies.
Abundance analysis {#sec4}
------------------
Abundance calculations and spectral synthesis (for oxygen) were performed using the latest version of the abundance calculation code originally described by @spite1967. We used the MARCS model atmospheres developed by @edvardson1993. We also used of ABOMAN, a tool developed by E. Rossetti at the INAF, Bologna Observatory, Italy, which allows the semi-automatic processing of data for several objects, using the aforementioned abundance calculation code. The tool ABOMAN performs all the steps needed to choose the best-fit model automatically (see below) and compute abundance ratios for all elements, and provides all the graphical tools required to analyse the results.
The detailed procedure followed to derive the chemical abundances is described in depth in Paper I. In brief, we calculated Fe I and Fe II abundances for a set of models with parameters extending $\pm$3$\sigma$ around the photometric estimates of Table \[pars\]. We chose the model that satisfied simultaneously the following conditions: [*(i)*]{} the abundance of Fe I lines should not vary with excitation potential $\chi_{ex}$; [*(ii)*]{} the abundance of Fe I lines should not vary significantly with EW, i.e., strong and weak lines should infer the same abundance[^23]; [*(iii)*]{} the abundance of Fe I lines should not differ significantly from the abundance of Fe II lines; and [*(iv)*]{} the abundance of Fe I lines should not vary significantly with wavelength.
Once the best-fit model has been found, abundance ratios of all the measured elements were determined, as shown in Table \[abotab\], as the average of abundances given by single lines. The internal (random) errors were then computed as $\sigma
/ \sqrt(n_{lines})$. Oxygen abundances were determined by means of spectral synthesis of the region around the \[O I\] forbidden line at 6300 Å. In this case, the internal uncertainty was estimated using the average abundance difference between the best-fit spectrum and two spectra placed approximately 1$\sigma$ (of the Poissonian noise) above and below it. Average cluster abundances (Tables \[tab-cluster\]) were computed as weighted averages of abundance ratios of single stars.
Comparison of our results with available literature is discussed in details in Section \[sec5\].
Abundance uncertainties and the Sun {#sec-err}
-----------------------------------
The internal (random) uncertainty described above includes uncertainties related to the measurement of EW and to the atomic parameters (dominated by log$gf$ determinations). We must consider other sources of uncertainty (see Paper I for details) such as: the uncertainty owing to the choice of atmospheric parameters; the uncertainty owing to the continuum normalization procedure; the uncertainty in the reference solar abundance values.
Uncertainties due to the choice of stellar parameters were evaluated with the method proposed by @cayrel2004. In brief, we altered the predominant atmospheric parameter, i.e., by altering one atmospheric parameter, T$_{\rm{eff}}$, within its uncertainty ($\sim$100 K) and re-optimizing the other parameters for the hottest and coolest stars in our sample. We re-calculated abundances with the procedure described in the previous Section. The external uncertainties, listed in the last column of Table \[abotab\], are estimated by averaging errors calculated with the higher and lower temperatures for the warmest and coolest stars in our sample (namely, stars 001 and 208 in NGC 752).
Uncertainties due to the continuum normalization procedure might also affect the obtained EW and, therefore, the derived abundances. Their contribution is estimated by averaging the differences between the EW obtained with the “best-fit” continuum and those derived by lowering and raising the continuum level by the continuum placement uncertainty. This is calculated from Equation 7 of @daospec. The typical uncertainty caused by the continuum placement is $\Delta EW\sim$1 mÅ and almost independent of the EW. This small uncertainty has a negligible impact on the derived abundances in comparison with other sources of uncertainty described above. Therefore, they have not been explicitly included in the error budget.
[*Here*]{} F10[^24] B08[^25]
--------------------------- ---------------- ---------------- ----------------
R=$\lambda/\delta\lambda$ 30000 28000 40000
S/N 50–100 100 45-100
$[$Fe/H$]$ –0.30$\pm$0.02 –0.30$\pm$0.02 –0.29$\pm$0.04
$[$Al/Fe$]$ +0.12$\pm$0.05 +0.19$\pm$0.06 +0.11$\pm$0.10
$[$Ba/Fe$]$ +0.51$\pm$0.12 — +0.29$\pm$0.10
$[$Ca/Fe$]$ –0.05$\pm$0.01 –0.07$\pm$0.01 +0.07$\pm$0.04
$[$Co/Fe$]$ +0.07$\pm$0.05 +0.00$\pm$0.01 —
$[$Cr/Fe$]$ –0.11$\pm$0.21 –0.16$\pm$0.11 –0.05$\pm$0.04
$[$La/Fe$]$ –0.14$\pm$0.07 — —
$[$Mg/Fe$]$ +0.33$\pm$0.10 +0.13$\pm$0.01 +0.27$\pm$0.08
$[$Na/Fe$]$ –0.12$\pm$0.04 +0.20$\pm$0.01 +0.13$\pm$0.02
$[$Ni/Fe$]$ –0.03$\pm$0.02 –0.02$\pm$0.01 +0.00$\pm$0.04
$[$O/Fe$]$ +0.00$\pm$0.16 –0.01$\pm$0.03 —
$[$Sc/Fe$]$ +0.00$\pm$0.03 — —
$[$Si/Fe$]$ +0.14$\pm$0.05 +0.27$\pm$0.05 +0.12$\pm$0.04
$[$Ti/Fe$]$ –0.08$\pm$0.07 –0.17$\pm$0.01 +0.11$\pm$0.06
$[$V/Fe$]$ –0.08$\pm$0.05 — —
$[$Y/Fe$]$ –0.23$\pm$0.23 — —
: High-resolution average Be 32 abundances.[]{data-label="be32"}
--------------------- ---------------- ---------------- ---------------- ---------------- ---------------- ----------------
Parameter Here S09/S06 M07/M06 F07 Bo00 LC95
Resolution 30000 60000 42000 30000 45000 30000
S/N 300–600 $\sim$500 100-350 175 100–300 $>$100
Star
T$_eff$ (K) 4750 4965 4955 4823 4956 4900
log$g$ (dex) 2.7 2.63 2.7 2.43 2.83 2.6
$v_t$ (km s$^{-1}$) 1.4 1.32 1.4 1.57 1.35 2.0
$[$FeI/H$]$ +0.12$\pm$0.01 +0.14$\pm$0.08 +0.11 +0.16$\pm$0.05 +0.11$\pm$0.01 +0.13$\pm$0.02
$[$FeII/H$]$ +0.13$\pm$0.03 +0.22$\pm$0.16 +0.10 +0.09$\pm$0.08 +0.11$\pm$0.02 +0.12$\pm$0.02
$[$Al/Fe$]$ -0.01$\pm$0.05 +0.20$\pm$0.01 — +0.19$\pm$0.07 +0.12$\pm$0.00 –0.17$\pm$0.03
$[$Ba/Fe$]$ +0.37$\pm$0.05 — –0.07 — +0.09$\pm$0.05 –0.04$\pm$0.00
$[$Ca/Fe$]$ –0.07$\pm$0.03 — +0.10$\pm$0.12 –0.01$\pm$0.11 +0.01$\pm$0.04 –0.28$\pm$0.04
$[$Co/Fe$]$ +0.00$\pm$0.04 — — — +0.02$\pm$0.02 +0.04$\pm$0.04
$[$Cr/Fe$]$ +0.02$\pm$0.03 — — — –0.01$\pm$0.02 –0.03$\pm$0.11
$[$La/Fe$]$ –0.12$\pm$0.06 — –0.23 — –0.03$\pm$0.02 —
$[$Mg/Fe$]$ +0.13$\pm$0.05 +0.43$\pm$0.01 –0.08 +0.03$\pm$0.07 +0.16$\pm$0.02 +0.18$\pm$0.05
$[$Na/Fe$]$ +0.19$\pm$0.02 +0.49$\pm$0.05 +0.22 +0.05$\pm$0.11 +0.32$\pm$0.01 +0.22$\pm$0.05
$[$Nd/Fe$]$ +0.04$\pm$0.29 — –0.15 — –0.02$\pm$0.03 +0.07$\pm$0.00
$[$Ni/Fe$]$ +0.02$\pm$0.02 +0.12$\pm$0.07 –0.04$\pm$0.12 — +0.00$\pm$0.03 +0.06$\pm$0.03
$[$O/Fe$]$ –0.35$\pm$0.07 –0.09$\pm$0.06 –0.09 –0.04$\pm$0.11 — —
$[$Sc/Fe$]$ –0.04$\pm$0.05 — — — +0.00$\pm$0.02 +0.01$\pm$0.08
$[$Si/Fe$]$ +0.09$\pm$0.03 — +0.07$\pm$0.12 +0.09$\pm$0.09 +0.09$\pm$0.03 +0.21$\pm$0.03
$[$TiI/Fe$]$ –0.12$\pm$0.02 — — –0.05$\pm$0.10 –0.01$\pm$0.01 –0.14$\pm$0.03
$[$TiII/Fe$]$ –0.03$\pm$0.06 — — –0.04$\pm$0.15 — —
$[$V/Fe$]$ +0.02$\pm$0.04 — — — +0.01$\pm$0.02 —
$[$Y/Fe$]$ –0.12$\pm$0.05 — –0.11 — +0.01$\pm$0.01 +0.10$\pm$0.00
Star
T$_eff$ (K) 4800 4938 4975 — 4980 4875
log$g$ (dex) 2.8 2.69 2.65 — 2.83 2.4
$v_t$ (km s$^{-1}$) 1.4 1.40 1.4 — 1.25 2.0
$[$FeI/H$]$ +0.10$\pm$0.01 +0.14$\pm$0.07 +0.11 — +0.19$\pm$0.01 +0.07$\pm$0.01
$[$FeII/H$]$ +0.13$\pm$0.03 +0.26$\pm$0.16 +0.07 —- +0.18$\pm$0.03 +0.04$\pm$0.02
$[$Al/Fe$]$ +0.00$\pm$0.05 +0.16$\pm$0.01 — — +0.08$\pm$0.01 –0.14$\pm$0.01
$[$Ba/Fe$]$ +0.39$\pm$0.05 — –0.02 —- +0.15$\pm$0.01 –0.13$\pm$0.00
$[$Ca/Fe$]$ –0.06$\pm$0.03 — +0.08$\pm$0.12 — +0.00$\pm$0.05 –0.17$\pm$0.06
$[$Co/Fe$]$ +0.01$\pm$0.03 — — — +0.01$\pm$0.03 +0.10$\pm$0.05
$[$Cr/Fe$]$ +0.03$\pm$0.03 — — — –0.04$\pm$0.02 –0.06$\pm$0.09
$[$La/Fe$]$ –0.08$\pm$0.05 — –0.33 — –0.05$\pm$0.09 —
$[$Mg/Fe$]$ +0.06$\pm$0.04 +0.36$\pm$0.02 –0.10 — +0.15$\pm$0.01 +0.33$\pm$0.07
$[$Na/Fe$]$ +0.18$\pm$0.02 +0.44$\pm$0.05 +0.16 — +0.32$\pm$0.01 +0.28$\pm$0.05
$[$Nd/Fe$]$ +0.08$\pm$0.30 — –0.17 — +0.02$\pm$0.02 –0.17$\pm$0.00
$[$Ni/Fe$]$ +0.04$\pm$0.02 –0.03$\pm$0.06 +0.06$\pm$0.09 — +0.09$\pm$0.06 +0.11$\pm$0.03
$[$O/Fe$]$ –0.25$\pm$0.05 –0.03$\pm$0.06 –0.21 — — —
$[$Sc/Fe$]$ +0.00$\pm$0.05 — — — +0.01$\pm$0.02 –0.02$\pm$0.07
$[$Si/Fe$]$ +0.09$\pm$0.02 — +0.07$\pm$0.11 — +0.06$\pm$0.02 +0.23$\pm$0.03
$[$TiI/Fe$]$ –0.11$\pm$0.02 — — — –0.04$\pm$0.02 –0.07$\pm$0.03
$[$TiII/Fe$]$ –0.00$\pm$0.07 — — — — —
$[$V/Fe$]$ +0.00$\pm$0.03 — — — +0.02$\pm$0.02 —
$[$Y/Fe$]$ –0.06$\pm$0.06 — –0.10 — –0.04$\pm$0.03 +0.19$\pm$0.00
Star
T$_eff$ (K) 4800 4911 4925 4838 4880 —
log$g$ (dex) 2.8 2.57 2.55 2.52 2.50 —
$v_t$ (km s$^{-1}$) 1.6 1.47 1.4 1.63 1.46 —
$[$FeI/H$]$ +0.11$\pm$0.01 +0.20$\pm$0.08 +0.11 +0.21$\pm$0.07 +0.11$\pm$0.01 —
$[$FeII/H$]$ +0.09$\pm$0.03 +0.22$\pm$0.16 +0.11 +0.18$\pm$0.10 +0.05$\pm$0.03 —
$[$Al/Fe$]$ +0.02$\pm$0.05 +0.15$\pm$0.01 — +0.17$\pm$0.08 +0.20$\pm$0.01 —
$[$Ba/Fe$]$ +0.31$\pm$0.05 — –0.02 — +0.09$\pm$0.01 —
$[$Ca/Fe$]$ –0.07$\pm$0.02 — +0.11$\pm$0.12 +0.01$\pm$0.10 +0.09$\pm$0.03 —
$[$Co/Fe$]$ +0.06$\pm$0.03 — — — –0.01$\pm$0.04 —
$[$Cr/Fe$]$ +0.08$\pm$0.04 — — — +0.00$\pm$0.01 —
$[$La/Fe$]$ –0.05$\pm$0.05 — -0.20 — -0.17$\pm$0.00 —
$[$Mg/Fe$]$ +0.21$\pm$0.07 +0.37$\pm$0.02 –0.08 –0.03$\pm$0.08 — —
$[$Na/Fe$]$ +0.18$\pm$0.02 +0.41$\pm$0.04 +0.23 +0.04$\pm$0.11 +0.40$\pm$0.04 —
$[$Nd/Fe$]$ +0.08$\pm$0.28 — –0.21 — –0.10$\pm$0.05 —
$[$Ni/Fe$]$ +0.03$\pm$0.02 +0.06$\pm$0.08 +0.09$\pm$0.11 — +0.00$\pm$0.02 —
$[$O/Fe$]$ –0.22$\pm$0.07 –0.13$\pm$0.06 –0.01 –0.04$\pm$0.13 — —
$[$Sc/Fe$]$ –0.04$\pm$0.05 — — — — —
$[$Si/Fe$]$ +0.10$\pm$0.03 — +0.09$\pm$0.11 +0.05$\pm$0.11 +0.09$\pm$0.03 —
$[$TiI/Fe$]$ –0.06$\pm$0.01 — — –0.01$\pm$0.08 –0.05$\pm$0.03 —
$[$TiII/Fe$]$ –0.02$\pm$0.11 — — –0.14$\pm$0.11 — —
$[$V/Fe$]$ +0.09$\pm$0.03 — — — –0.04$\pm$0.03 —
$[$Y/Fe$]$ –0.07$\pm$0.05 — –0.11 — –0.05$\pm$0.03 —
--------------------- ---------------- ---------------- ---------------- ---------------- ---------------- ----------------
To validate the whole procedure used here, in Paper I we performed an abundance analysis of the ESO HARPS solar spectrum reflected by Ganymede. We used the same line list, model atmospheres, and abundance calculation code that we used on our OC target stars, and found solar values for all elements, with the only marginal exceptions of barium and aluminium (see also Section \[sec6\]). While the details of this analysis can be found in Paper I, we mention here that our reference solar abundances are taken from @gre96.
Cluster-by-cluster discussion {#sec5}
=============================
Berkeley 32
-----------
Berkeley 32 ($\alpha_{2000}=06^h58^m07^s$ and $\delta_{2000}=+06^o25'43``$) is a distant OC (R$_{gc}$=11.6 kpc) located towards the Galactic anticentre and situated 260 pc above the disc plane. Its distance makes it one of the crucial clusters for a correct determination of the metallicity gradient along the Galactic disc, and therefore one of the key OC to the understanding of disc formation and evolution. The color-magnitude diagram of this cluster [e.g. @dorazi2006], contaminated by disc stars, shows a clear main sequence turn-off with a sparsely populated red giant branch. Determinations of its age, mainly using morphological indicators, yield a value of $\simeq$5 Gyr [e.g. @dorazi2006; @salaris2004; @Richtler2001; @carraro1994; @kaluzny1991].
Given its large distance, it has not been well-studied spectroscopically, but we could compare our results with two recent high-resolution studies of @bragaglia2008 and @friel2010. We found a very close agreement of our abundance ratios with those studies (see Table \[be32\]). The exceptions are Ba and Na. It is well-known that Ba abundances are enhanced by HFS [e.g. @dorazi2009] effects that should explain the differences from @bragaglia2008. The \[Na/Fe\] ratio is lower than the values reported by @bragaglia2008 and @friel2010 by -0.25 and -0.32 dex, respectively. The difficulty in measuring Na lines, which suffer from NLTE effects, could easily explain this controversy. Moreover, different model atmospheres, stellar and atomic parameters, etc., between different studies may also play a role (and remove this discrepancy).
NGC 752
-------
NGC 752 ($\alpha_{2000}=01^h57^m41^s$, $\delta_{2000}=+37^o47'06``$) is an old ($\sim$1.6 Gyr) OC located in the solar neighbourhood at a distance of $\simeq$400 pc. This cluster has a low central concentration and contains a relatively small number of members. Its color-magnitude diagram [e.g. @johnson1953] has a still poorly understood morphology. The turn-off area is well-populated by early F-type stars, while the low main-sequence appears to be sparsely populated (Figure \[CMDs\]). This, together with the age of this cluster, may be an indication of the dynamic escape of low mass stars. Stellar evolution models also predict a well-populated red giant branch, which is not observed. All the known red giants are located in the red clump region [@bartasiute2007], which has a peculiar morphology because it has a faint extension slightly to the blue of its main concentration, which cannot be reproduced by stellar evolution models [@girardi2000].
[*Here*]{} S09/S06[^26] Bo00[^27] V99[^28]
--------------------------- ---------------- ---------------- ---------------- ----------------
R=$\lambda/\delta\lambda$ 30000 60000 45000 30–65000
S/N 50–100 100–200 100–300 $\sim$200
$[$Fe/H$]$ +0.11$\pm$0.01 +0.21$\pm$0.04 +0.12$\pm$0.06 –0.05$\pm$0.03
$[$Al/Fe$]$ +0.00$\pm$0.02 +0.17$\pm$0.03 +0.13$\pm$0.06 —
$[$Ba/Fe$]$ +0.36$\pm$0.04 — +0.10$\pm$0.03 –0.03$\pm$0.07
$[$Ca/Fe$]$ –0.07$\pm$0.01 — +0.03$\pm$0.05 +0.03$\pm$0.04
$[$Co/Fe$]$ +0.03$\pm$0.03 — +0.00$\pm$0.02 —
$[$Cr/Fe$]$ +0.04$\pm$0.03 — –0.02$\pm$0.02 —
$[$La/Fe$]$ –0.08$\pm$0.04 — –0.10$\pm$0.08 —
$[$Mg/Fe$]$ +0.10$\pm$0.08 +0.38$\pm$0.03 +0.16$\pm$0.01 +0.17$\pm$0.04
$[$Na/Fe$]$ +0.18$\pm$0.01 +0.45$\pm$0.04 +0.37$\pm$0.05 +0.10$\pm$0.05
$[$Ni/Fe$]$ +0.03$\pm$0.01 +0.05$\pm$0.07 +0.05$\pm$0.06 +0.04$\pm$0.04
$[$O/Fe$]$ –0.27$\pm$0.04 –0.08$\pm$0.05 — +0.11$\pm$0.04
$[$Sc/Fe$]$ –0.02$\pm$0.02 — +0.00$\pm$0.01 +0.03$\pm$0.07
$[$Si/Fe$]$ +0.09$\pm$0.01 — +0.07$\pm$0.01 +0.13$\pm$0.03
$[$Ti/Fe$]$ –0.09$\pm$0.04 — –0.04$\pm$0.02 —
$[$V/Fe$]$ +0.04$\pm$0.05 — –0.01$\pm$0.03 —
$[$Y/Fe$]$ –0.09$\pm$0.03 — –0.04$\pm$0.02 –0.04$\pm$0.07
: High-resolution average Hyades abundances from giants.[]{data-label="hyades_cl"}
Photometry and low/medium resolution spectroscopy studies [see @bartasiute2007 and references therein] determined a slightly subsolar metallicity [i.e. [\[]{}Fe/H[\]]{}=–0.16$\pm$0.05, @friel1993]. A similar result was found with high-resolution spectroscopy (R$\simeq$40000, S/N$\simeq$80–150) in eight F-type stars around the main sequence turn-off [[\[]{}Fe/H[\]]{}=–0.09$\pm$0.05, @hobbs1992]. However, an investigation based on high-resolution spectroscopy (R $\simeq$57000, S/N$\simeq$30–80) of 18 G giant stars obtained a solar \[Fe/H\] ratio [[\[]{}Fe/H[\]]{}=+0.01$\pm$0.04, @sestito2004] in closer agreement with the value determined here. To our knowledge, we are the first to publish abundance ratios of elements other than \[Fe/H\] for this cluster.
Hyades
------
The Hyades cluster (Melotte 25; $\alpha_{2000}=04^h26^m54^s$ and $\delta_{2000}=+15^o52'00``$) is the closest OC to the Sun ($\sim$45 pc) located in the constellation of Taurus. Its proximity has motivated an extensive study lasting more than a century [starting with @herttzsprung1909]. The OC is embedded into a moving group with the same name, which suggests that it would be part of a dynamical stream coming from the inner Galaxy or a disrupting cluster [@famaey2007].
Being one of the most studied clusters, both photometrically and spectroscopically, it is the ideal cluster for abundance analysis comparisons. The color-magnitude diagram of this young OC [$\sim$0.7 Gyr, see Table \[phot\]; e.g. @johnson1955] contains only four red giant stars that have been confirmed as members from their parallaxes, proper motions, and radial velocities. Most of the existing abundance studies are focused on main sequence stars [see e.g. @paulson2003; @burkhart2000; @varenne1999 and references therein]. A comparison of the Hyades average abundances determined from some (or all) of the known four red giants are shown in Table \[hyades\_cl\]. The averages of the abundances compiled until 1999 by @varenne1999 are shown in the last column of Table \[hyades\_cl\] for reference. In general, \[Fe/H\] appears slightly supersolar, while all other abundance ratios are solar, and our abundance ratios agree well with literature values.
[*Here*]{} P08[^29] A07[^30] Bu98[^31]
--------------------------- ---------------- ---------------- ---------------- ---------------- --
R=$\lambda/\delta\lambda$ 30000 100000 55000 90000
S/N 50–100 $\simeq$80 $\simeq$100 $\sim$200
$[$Fe/H$]$ +0.16$\pm$0.05 +0.27$\pm$0.10 +0.11$\pm$0.03 +0.40$\pm$0.14
$[$Al/Fe$]$ +0.00$\pm$0.03 –0.05$\pm$0.12 — –0.19$\pm$0.17
$[$Ba/Fe$]$ +0.33$\pm$0.05 +0.22$\pm$0.06 — —
$[$Ca/Fe$]$ –0.08$\pm$0.02 +0.00$\pm$0.11 — +0.04$\pm$0.16
$[$Co/Fe$]$ +0.04$\pm$0.02 — — —
$[$Cr/Fe$]$ +0.05$\pm$0.01 –0.01$\pm$0.08 — —
$[$La/Fe$]$ –0.05$\pm$0.02 — — —
$[$Mg/Fe$]$ +0.27$\pm$0.05 — — —
$[$Na/Fe$]$ +0.25$\pm$0.06 –0.04$\pm$0.12 — —
$[$Ni/Fe$]$ +0.02$\pm$0.02 –0.02$\pm$0.12 — +0.21$\pm$0.17
$[$O/Fe$]$ –0.11$\pm$0.03 –0.40$\pm$0.20 — —
$[$Sc/Fe$]$ –0.04$\pm$0.05 — — —
$[$Si/Fe$]$ +0.06$\pm$0.02 –0.01$\pm$0.12 — —
$[$Ti/Fe$]$ –0.07$\pm$0.03 –0.04$\pm$0.12 — —
$[$V/Fe$]$ +0.06$\pm$0.03 — — —
$[$Y/Fe$]$ –0.11$\pm$0.01 — — —
: High-resolution average Praesepe (NGC 2632) abundances.[]{data-label="m44"}
The three late-type Hyades giants (028, 041, and 070) have been widely studied [e.g. @luck1995; @boyarchuk2000; @schuler2006; @schuler2009; @mishenina2006; @mishenina2007; @fulbright2007]. In Table \[tab\_hyades\_individual\] we compiled available literature data. Our temperatures are slightly lower (by $\sim$100 K) than the literature ones, whereas our values of $\log g$ and $v_t$ are similar. These marginal differences appear to have no significant impact on the derived abundance ratios, which agree very well with literature ones. Exceptions are Al, Ba, and O, which suffer from technical measurement problems (not strictly related to the Hyades cluster) and are discussed in Sections \[sec-err\] and \[sec6\].
Praesepe (NGC 2632)
-------------------
The cluster popularly known as Praesepe or Beehive (also called M 44, NGC 2632 or Melotte 88; $\alpha_{2000}=08^h40^m24^s$ and $\delta_{2000}=+19^o40'00``$ ), is an old OC (0.65 Gyr, see Table \[phot\]) well known from the antiquity. It is located in the Cancer constellation at a distance of $\simeq$175 pc, computed from Hipparcos parallaxes.
Its metal content was derived with different methods [e.g. @friel1992; @komarov1993; @claria1996; @huibonhoa1998; @burkhart1998; @burkhart2000; @dias2002; @pace2008]. In general, all the quoted studies obtained a metallicity either barely or definitely supersolar. Of these, the high-resolution abundance determinations were derived mainly for dwarfs or early-type giants [e.g. @friel1992; @burkhart1998; @an2007; @pace2008]. Surprisingly, to our knowledge, there are no recent high-resolution abundance determinations of late-type giants in this cluster.
Table \[m44\] shows a comparison of our results with some of the most recent high-resolution studies. In general, the \[Fe/H\] we derived in our late-type giants lies in-between those of @pace2008 and @an2007, suggesting that the proposed dichotomy of literature values (barely supersolar versus definitely supersolar) should be interpreted rather as an above average uncertainty. This larger than usual uncertainty could naturally arise from the different spectral types and abundance analysis methods employed in the literature. The \[Fe/H\] ratio derived by [@burkhart1998], based on Am stars, is on average $\simeq$0.3 dex larger than the values obtained in other works using different spectral-type stars. Although these stars should in principle reflect the chemical composition of the cluster, Am stars always have overabundant Fe abundances relative to other objects in the same clusters, without a clear explanation appearing in the literature. As in the case of the Hyades, Na and O abundances derived by us appear marginally discrepant with those by @pace2008, and will be discussed in more detail in Section \[sec6\].
Discussion of abundance ratios {#sec6}
==============================
As in Paper I, we compared our abundance ratios (and those from Paper I) with both others in the literature and the abundances of the Galactic disc field stars from @reddy2006 [@reddy2003] in Figures \[fig\_fe\_peak\] to \[fig\_naal\]. We extended the open cluster abundance compilation of Paper I (see Table \[tab-hiresnew\]) with both recent published works and old studies that were not included in the previous version. In both cases, as in Paper I, we included only studies based on high-resolution (R$\gtrsim$18000) spectroscopy. When more than one determination was available for one cluster, we simply plotted them all to give a realistic idea of the uncertainties involved in the compilation, and we did not attempt to correct for differences between the abundance analysis procedures (log$gf$, solar reference, and so on), because this would be beyond the scope of the present article.
![Comparison between our iron-peak abundance ratios (large black dots), those of Paper I (large black open circles), high-resolution measurements listed in Table \[tab-hiresnew\] (large dark grey dots), field stars belonging to the thin disc [light grey dots, @reddy2003], and to the thick disc [tiny light grey dots, @reddy2006]. Errorbars in our measurements are the quadratic sum of all uncertainties discussed in Sections \[sec4\] and \[sec-err\].[]{data-label="fig_fe_peak"}](fig_fe_peak.eps){width="\columnwidth"}
Iron-peak element ratios {#iron-abo}
------------------------
Figure \[fig\_fe\_peak\] shows the abundance ratios of iron-peak elements. Our OC with abundances close to solar (i.e., Hyades, Praesepe, and NGC 752) are in very good agreement with the results obtained in other OC studied with high-resolution spectra and in disc stars of similar metallicity. A larger scatter or marginal discrepancies are sometimes observed for the odd elements Sc, V, and Co, but this is because of the well-known hyperfine structure (HFS) of the lines usually employed in the analysis. The element that appears to suffer more from these effects is vanadium. This scatter is due, at least in part, to the different procedures used in the literature for treating the HFS splitting. We stress that in our case, we do not attempt any HFS correction.
The most metal-poor and oldest OC in our sample, Be 32, has a puzzling behaviour. While all its iron-peak abundance ratios are still compatible with the literature values for OC and field stars of similar metallicity (uncertainties are large), some underlying discrepancy could be present. For example, HFS should cause an overestimate (and not an underestimate) of vanadium. In addition, chromium appears to be lower than solar. We note that (see Table \[be32\]) the literature Co and Cr determinations by @friel2010, @sestito2006, and @bragaglia2008 are very similar to ours. In the case of our \[Cr/Fe\] measurement for Be 32, we must note that our two giants appear to exhibit quite different \[Cr/Fe\] abundances, resulting in a large scatter in the cluster average value. This large scatter is most probably due to a measurement uncertainty, and should not be considered significant.
In Paper I, we noticed a peculiar behaviour in the Ni abundance ratios of literature OC determinations: they appear to be slightly richer in Ni than field stars by roughly 0.05 dex. Our \[Ni/Fe\] ratios are in closer agreement with the field star determinations than with the OC ones. Although this difference is small (within the uncertainties), it appears systematic in nature, and we were unable to find any easy explanation, such as the choice of either solar reference abundances or the log$gf$ system, of this discrepancy.
Alpha-element ratios {#alfa-abo}
--------------------
![Comparison of our $\alpha$-elements ratios with literature values. Symbols are the same as in Figure \[fig\_fe\_peak\].[]{data-label="fig_alfas"}](fig_alfa.eps){width="\columnwidth"}
Figure \[fig\_alfas\] shows the abundance ratios of $\alpha$-elements. As for iron-peak elements, our measurements are always compatible with the literature values, within their uncertainties. Generally speaking, all our OC show roughly solar $\alpha$-enhancements, even Be 32, which has a lower metallicity.
However, some elements deserve some more discussion, as was noted in Paper I. For example, the log$gf$ of calcium are quite uncertain, and we chose the VALD reference atomic data, which explains why our \[Ca/Fe\] ratios are slightly lower than the bulk of literature determinations for cluster and disc stars. A similar problem affects the Mg lines, as can clearly be appreciated from the large spread of literature values. Our \[Mg/Fe\] determinations tend to lie on the upper envelope of literature ratios for OC. A deeper discussion of Mg abundances can be found in Paper I.
In the case of oxygen, the problem is instead in the difficulty in measuring its small lines. The forbidden \[O I\] line at 6300 Å, which we used in this paper, suffers from contamination by a Ni line and by telluric absorption features, while the O triplet around 7770 Å (used by some other studies) suffers from NLTE effects. This is reflected by the large scatter in literature values.
![Comparison of our s-process elements ratios with the literature ones. Symbols are the same as in Figure \[fig\_fe\_peak\], except for the black star-like symbols in the top \[Ba/Fe\] panel, which represent the revision of Ba abundances with spectral synthesis performed by @dorazi2009.[]{data-label="fig_esse"}](fig_esse.eps){width="\columnwidth"}
Heavy element ratios {#heavy-abo}
--------------------
We determined abundances for three heavy s-process elements: Ba, La, and Nd; and one light s-process element: Y (Figure \[fig\_esse\]). Literature determinations for these elements are not numerous. @dorazi2009 measured Ba in several OC using spectral synthesis to take into account HFS. The \[Ba/Fe\] abundances derived by @dorazi2009 taking into account HFS do not differ significantly from other literature determinations (including ours). The \[Ba/Fe\] ratios are clearly above solar for most of the clusters and they show a scatter larger than $\sim$0.5. @dorazi2009 found this scatter to be due to age: the Ba content appears to increase for younger clusters. The available lanthanum and neodymium lines were unfortunately relatively small, and we were able to find fewer published studies to compare with. As a result, the solar clusters (Hyades, Praesepe, and NGC 752) have La and Nd ratios in good agreement with the literature, while Be 32 appears to have lower \[La/Fe\] and \[Nd/Fe\] than the few studied OC at a similar metallicity, which are Mel 66 [@gratton1994] and NGC 2243 [@smith1987]. However, our \[Nd/Fe\] agrees well with the field star solar ratios. The only light s-process element we could measure, Y, relies on a couple of weak lines that provide uncertain abundances (see the large errorbar in Figure \[fig\_esse\]). Our Y ratio appears to be lower than all literature estimates, although still compatible with the solar values of field stars of similar metallicity, within the large uncertainties.
In summary, we can say that all the studied clusters appear to have roughly solar s-process enhancements, but it would be extremely interesting to attempt a more detailed study of s-process elements in OC, as done by @dorazi2009 for barium.
Ratios of Na and Al and anticorrelations
----------------------------------------
![Comparison between our \[Na/Fe\] and \[Al/Fe\] ratios and the literature ones. Symbols are the same as in Figure \[fig\_fe\_peak\].[]{data-label="fig_naal"}](fig_naal.eps){width="\columnwidth"}
As discussed in Paper I, the study of light elements in OC is quite interesting. The elements Al and Na, together with Mg, O, C, and N, show puzzling (anti-)correlations in almost all of the studied globular clusters, in the Milky Way [see, e.g., @carretta2010; @pancino2010b and references therein] and outside [e.g. @mucciarelli2009; @letarte2006]. No (anti-)correlations were observed in either field stars [but see @martell2010] or OC [@martell2009; @desilva2009; @smiljanic2009 and Paper I] so far. This suggests that metallicity, cluster size and age, or the environment must play a rôle, and therefore finding (anti-)correlations in some OC would be of enormous importance to put tighter constraints on the phenomenon.
We determined abundances of Al and Na and compared them with published results in Figure \[fig\_naal\]. While in the case of aluminium the agreement with literature values is good, we find a significantly lower \[Na/Fe\] ratio for Be 32 than for other clusters or field stars of similar metallicity. Generally speaking, the large scatter in Na determinations could be due to the difficulties in measuring Na lines, often affected by NLTE effects [@gratton1999], although no such scatter is observed among field stars. However, a few clusters have \[Na/Fe\] lower than our Be 32 determination, and NLTE corrections [@gratton1999] could make the discrepancy of our Be 32 Na determination even worse. Unfortunately, given the large scatter and the difficulty of measurement, it is difficult to either confirm or exclude the presence of some (small) intrinsic \[Na/Fe\] scatter in this clusters.
In Figure \[fig\_anticorr\], all the studied stars occupy the “normal stars" loci, which is around solar for Na and Al, and slightly $\alpha$-enhanced for O and Mg (see Section \[alfa-abo\]). There is a hint of correlation between \[Al/Fe\] and \[Na/Fe\], which was also observed for objects studied in Paper I. Of course, small variations in T$_{\rm{eff}}$ could induce artificial correlations between element pairs, so the observed trend is most probably not-significant. However, we again note that the Na spread is very large (see above), suggesting that a small degree of chemical anomalies (barely hidden within the present observational uncertainties) cannot be completely excluded.
![A search for (anti)-correlations of Al, Mg, Na, and O among our target stars. The four panels show different planes of abundance ratios, where stars belonging to each cluster are marked with different symbols. Dotted lines show solar values, solid lines show linear regressions and the typical uncertainty ($\sim$0.1 dex) is marked at the lower right corner of each panel.[]{data-label="fig_anticorr"}](fig_anticorr.eps){width="\columnwidth" height="\columnwidth"}
Galactic trends {#sec7}
===============
The existence of trends in the chemical abundances with Galactocentric distance, $R_{gc}$, vertical distance to the Galactic plane, $z$, and age, are key to understanding Galactic disc formation and evolution because they provide fundamental constraints on chemical evolution models. Different tracers have been used to investigate trends in the Galactic disc: OB stars [e.g. @daflon2004], Cepheids [e.g. @lemasle2008], H II regions [e.g. @deharveng2000], and planetary nebulae [e.g. @costa2004]. However, as coeval groups of stars at the same distance and with a homogeneous chemical composition, OC are the ideal test particles to investigate the existence of radial and vertical gradients and of an age-metallicity relation in the disc.
We complement the small sample of abundance ratios obtained here and in Paper I with a revised version of the literature data first presented in Paper I (Table \[tab-hiresnew\]). When a cluster had two or more abundance determinations available in the literature, we averaged them to make the figures easily readable and the error bars are, simply, calculated as the standard deviation. For those clusters with only one abundance determination, the error bars are the uncertainties in those determinations. The heliocentric distances compiled in the updated version of the @dias2002 database were used to obtain $R_{gc}$ and $z$ for each cluster, assuming $R_{GC_{\odot}}$=8.5 kpc. Ages were obtained from the same source, which is a compilation of different values available in the literature, hence might still be quite inaccurate for some clusters. In spite of its heterogeneity, our compilation contains a total of 89 clusters and is, to the best of our knowledge, the largest available in the literature, based on high-resolution spectroscopic abundances. Any attempt to homogenize this sample, for which abundances, distances, and ages have been derived from very different techniques, is clearly beyond the scope of this paper. This prevents us from a detailed analysis of the Galactic trends of all elements. For this reason, we focus only on \[Fe/H\] and \[$\alpha$/Fe\] ratios. In spite of this heterogeneity, this analysis is still very useful owing to the number of clusters, and the large range of ages, and vertical and radial distances covered, even if the heterogeneity of the sample forces us to be extremely cautions when drawing any conclusion from the data.
![Trends of \[Fe/H\] (top panel) and \[$\alpha$/Fe\] (bottom panel) with galactocentric radius. Grey dots are OC compiled in Table \[tab-hiresnew\], while black dots are the ones analysed here and in Paper I. A global linear fit is drawn in both panels (long-dashed line). Two separate linear fits of OC inside and outside 12.5 kpc are also shown (solid lines).[]{data-label="fig_trend_rgc"}](fig_trends_rgc.eps){width="\columnwidth" height="\columnwidth"}
Trends with Galactocentric radius
---------------------------------
Radial gradients may arise when the disc forms, and different mechanisms can produce them: for example, different timescales of star formation at different distances [e.g., @schaye2004]; a radial variation in the infall of gas; or a change in the yield as a function of the radius [e.g., @molla1996]. This initial radial gradient can be either amplified (steepened) or washed out (flattened) with time by radial mixing [e.g., @roskar2008].
Since the pioneering work of @janes1979, OC have been widely used to investigate the gradient in metallicity with radius in the Galactic disc [e.g., @twarog2003; @friel02; @magrini2009; @friel2010; @jacobson2011; @jacobson2011b]. @friel1995 reviewed the firsts investigations in this field. Since then, a great effort have been performed to obtain both homogeneous [e.g., @friel02; @sestito2008; @friel2010] and/or larger samples [e.g., @twarog1997; @jacobson2011; @jacobson2011b]. All these investigations agree on the fact that the iron content decreases with increasing radius [e.g., @friel02]. This behaviour has been generally considered linear with a slope between –0.05 and –0.09 dex kpc$^{-1}$, depending on the cluster sample used. Similar trends were obtained for other different tracers of the disc [e.g., @andrievsky2004; @lemasle2008]. Most of these works were limited to the inner $R_{gc}\simeq$15 kpc. However, investigations based on samples containing clusters at larger distances [e.g., @twarog1997; @yong2005; @sestito2008] found that the \[Fe/H\] ratio decreases as a function of increasing radius to $R_{gc}\simeq$12.5 kpc and appears to flatten from there outwards.
The variation in \[Fe/H\] with $R_{gc}$ in our compilation has been plotted in the top panel of Figure \[fig\_trend\_rgc\]. The whole sample is well fitted by a line with a slope of –0.046$\pm$0.005 dex kpc$^{-1}$ (long-dashed line), in concordance with the result obtained in Paper I from a $\simeq$20% smaller sample (–0.05$\pm$0.01 dex kpc$^{-1}$) and in other investigations in the literature [e.g. –0.06$\pm$0.02 dex kpc$^{-1}$; @friel02]. The sample used here contains more clusters with distances larger than $R_{gc}\geq$12 kpc. This allows us to investigate the discontinuity observed by some authors at $R_{gc}\simeq$12-13 kpc. At first sight, no clear discontinuity in slope appears, partly because of the large range of \[Fe/H\] at this radius ($\simeq$0.5 dex) and partly as a possible consequence of the heterogeneity of our sample. However, when we fit separately clusters inwards and outwards of 12.5 kpc, we find two significantly different slopes: the metallicity in the inner disc decreases with a slope of –0.07$\pm$0.01 dex kpc$^{-1}$, while in the outer disc the slope is –0.01$\pm$0.01 dex kpc$^{-1}$. The obtained slopes change within the uncertainties if the cut radius varies between 11.5 and 13.5 kpc. This is also in very good agreement with the recent results by @andreuzzi2011, who find –0.07 dex kpc$^{-1}$ in the inner 12 kpc. This bimodal behaviour can be explained by a different chemical enrichment and star formation in the inner and outer disc; [e.g. @chiappini2001; @magrini2009] however, a sharp discontinuity between the inner and outer disc is not expected theoretically.
The ratio \[$\alpha$/Fe\] reflects the relative contributions of Type Ia and II supernovae: chemical evolution models predict an increase of this ratio with $R_{gc}$ [e.g. @chiappini2001; @magrini2009]. This tendency was indeed observed in OC by, e.g., @yong2005, @magrini2009, and in Paper I. The bottom panel of Figure \[fig\_trend\_rgc\] shows the variation in \[$\alpha$/Fe\] with $R_{gc}$ for our compilation: a weak increase in $\alpha$-element abundances with radius is apparent. However, the slope is still compatible with a flat distribution at the 1$\sigma$ level, as in Paper I, especially if the two outermost clusters are removed. The discontinuity observed for \[Fe/H\] is not evident at all in \[$\alpha$/Fe\].
An accretion of a satellite into the outer disc could also explain the trend observed [e.g. @chiappini2001; @yong2005]. In this case, we would expect to find some inhomogeneities corresponding to the trajectory of the merger. @carraro2009 indeed found evidence that two OC, Berkeley 29 and Saurer 1, are related to the Sagittarius dwarf galaxy. Our compiled sample unfortunately do not allow us to investigate this question in depth.
![Gradient in \[Fe/H\] as a function of $R_{gc}$ in four different age bins (labeled in top–right corner). A linear fit is performed for the OC within a radius of $R_{gc}$=12.5 kpc, and the slope indicated on each panel. A flatter and roughly constant slope is found outside a radius of $R_{gc}$=12.5 kpc.[]{data-label="fig_trend_slope"}](fig_trends_slope.eps){width="\columnwidth" height="\columnwidth"}
Time evolution of the radial gradient
-------------------------------------
Chemical evolution models of the Galactic disc predict a variation in the metallicity gradient with time, but they disagree about the direction of this gradient variation [see @maciel2007 for a recent review], some predicting a steepening and some a flattening of the gradient with time. Studies based on metallicities derived from low-resolution spectroscopy found that old OC ($\gtrsim$1 Gyr) followed a steeper radial gradient, $\sim$-0.08 dex kpc$^{-1}$, than the younger ones, $\sim$-0.02 dex kpc$^{-1}$ [@friel02; @chen2003]. Only recently have chemical abundances been derived from high-resolution spectroscopy for a sufficient number of OC to significantly investigate the variation in the radial gradient with time. As for studies based on low-resolution spectra, they agree that the gradient was steeper in the past and has flattened with time [@magrini2009; @andreuzzi2011]. For example, on the basis of a sample of $\sim$70 OC @andreuzzi2011 found that all objects younger than 4 Gyr display a similar gradient with a slope -0.07 dex kpc$^{-1}$ in the inner 12 kpc, while the one for older objects is steeper, -0.15 dex kpc$^{-1}$.
Other tracers have been used to study the time variation in radial gradients. Studies based on planetary nebulae found more puzzling results: while @maciel2003 found a flattening of the gradient with time, as generally observed for OC, @stanghellini2010 found that the gradient steepens with time. At the moment, there is no explanation of this contradictory result. Comparisons among the slopes of the radial gradients described by populations of different ages also show that the gradient has flattened out in the past few Gyr [see @maciel2009 for a recent review].
To investigate the behaviour of the radial gradient in our compiled sample of high-resolution abundances, we plotted in Figure \[fig\_trend\_slope\] the gradient in \[Fe/H\] as a function of $R_{gc}$ in four different age bins. We obtained a linear fit in each age bin for the inner 12.5 kpc, and for the outer range we simply used the same fit as in Figure \[fig\_trend\_rgc\], owing to the paucity of OC after age binning in this region. We found that the slope of the \[Fe/H\] gradient increases as we go back in time from –0.02$\pm$0.01 dex kpc$^{-1}$ for objects younger than 0.1 Gyr to –0.10$\pm$0.01 dex kpc$^{-1}$ for clusters older than 2.5 Gyr.
![Trends in \[Fe/H\] with $|z|$ in four radial annuli as indicated on the top-right corner of each panel, moving outwards from the top to the bottom panel. Symbols are the same as in Figure \[fig\_trend\_rgc\]. As a reference, we plotted dashed lines in each panel, representing the median metallicity of clusters in each radial annulus.[]{data-label="fig_trend_z"}](fig_rgc_z.eps){width="\columnwidth" height="\columnwidth"}
Trends with the disc scale–height
---------------------------------
Another interesting trend that could be investigated is the behaviour of \[Fe/H\] with the vertical scale-height of the disc $z$, i.e., the vertical \[Fe/H\] gradient. Although the formation of the thick discs remains an open question, the existence of vertical gradients can help us to discriminate among the mechanisms proposed to their formation. No vertical chemical gradients are expected in thick discs formed by heating caused by accretion events or major mergers. In contrast, vertical gradients may exist in discs thickened by gradual heating of the thin disc or before the gas has settled to form a thin disc [see @mould2005 for a review]. Up to now, there is no conclusive agreement about the existence of a vertical metallicity gradient in the Galactic disk. The existence of a vertical gradient for field stars have been claimed by several authors, although they cover only about 1 kpc above and below the disc plane [@bartasiute2003; @marsakov2005; @marsakov2006; @soubiran2008]. Studies covering large ranges of $|z|$ do not find any evidence of a vertical gradient [@gilmore1995; @soubiran2005; @navarro2011] among the field populations. Studies using OCs have found a vertical gradient of $\sim$-0.3dex kpc$^{-1}$[@piatti1995; @carraro1998; @chen2003], although, these studies do not distinguish the effects of the radial gradient, which can mask any vertical trend. This effects were taken into account by @jacobson2011 who found no evidence of a vertical gradient.
To investigate the presence of trends with $z$ in our compilation, we firstly had to remove the contribution of the radial metallicity gradient. We plotted in Figure \[fig\_trend\_z\] the variation in \[Fe/H\] with $|z|$ in four different annuli of $R_{gc}$. We note that OC with high $|z|$ are preferentially located at large $R_{gc}$; this is not unexpected because the disc thickens in its external regions. Moreover, an intrinsic bias caused by obscuration in the plane appears: clusters at large Galactocentric radii are found and observed preferentially higher above the plane. This could explain why the two outermost annuli studied uncover a possible weak decrease in \[Fe/H\] as $z$ increases. This trend is however still compatible with no gradient at the 1$\sigma$ level and, once again, larger samples of homogeneous data are necessary to investigate this result in detail.
![The evolution of \[Fe/H\] with age in the same four radial annuli as in Figure \[fig\_trend\_z\]. Again, dashed lines representing the median metallicity of clusters in each radial annulus have been plotted as reference.[]{data-label="fig_trend_age"}](fig_rgc_age.eps){width="\columnwidth" height="\columnwidth"}
Is there an age–metallicity relation for open clusters?
-------------------------------------------------------
Another important prediction of the chemical evolution models is the existence of an age-metallicity relation for disc populations. It is still unclear whether or not the field disc stars follow an age-metallicity relation. Some works find it [e.g. @reddy2003; @bensby2004; @reid2007], but others do not [e.g. @feltzing2001; @nordstrom2004; @karatas2005]. Again, no clear trend of chemical abundances with age has been clearly observed in the case of Galactic OC [e.g. @friel1995]. Although @friel2010 notices a trend of \[Al/Fe\] and \[O/Fe\] ratios with age, again larger and homogeneous samples are necessary to confirm this result. If an age-metallicity relation is confirmed for the field population but not for OC, this would imply that they might have followed a different chemical evolution [@yong2005].
The evolution of the radial gradient as a function of time, described above, indicates that the chemical enrichment of OC is modulated by their location in the Galaxy and not by the moment at which they formed. To investigate whether an age-metallicity relationship exits at a given $R_{gc}$, we plotted in Figure \[fig\_trend\_age\] the evolution of \[Fe/H\] with age in four different radial annuli. There is no clear trend in any of the studied annuli, although not all of them contain clusters covering the same age range. Only in the outermost annulus is a weak trend observed, although it is still not very significant. Again, we conclude that a larger sample of homogeneous data are necessary to investigate this point in depth.
Summary and conclusions {#sec8}
=======================
We have enlarged our sample of homogeneous high-resolution abundance measurements from the five clusters of Paper I to a total of nine, analysing here spectra of red clump giants in the Hyades, Praesepe, NGC 752, and Be 32. Our main results can be summarized as follows:
- [We provide the first high-resolution based abundance ratios [other than [\[]{}Fe/H[\]]{}, see @sestito2008] for NGC 752, which turned out to be mostly of solar composition;]{}
- [We have presented the abundance ratios of Praesepe red clump giants, which appear to solve a puzzling dichotomy of literature determinations for stars of different evolutionary stages;]{}
- [We have found that our abundance ratios for the Hyades and Berkeley 32 are in good agreement with other literature determinations;]{}
- [We have confirmed the absence of light elements (anti-)correlations in the OC studied so far.]{}
We have updated our compilation of previous literature data for 57 clusters of Paper I to a total of 89 clusters presented here. With this updated compilation and our homogeneous measurements in hand, we have investigated Galactic trends in \[Fe/H\] (and \[$\alpha$/Fe\]) with age, Galactocentric radius, and height above the Galactic plane. Our findings are in substantial agreement with other similar investigations, where the abundance gradient appears to indeed flatten out outside $R_{gc}\simeq$12.5 kpc, and the inner disc slope appears to flatten for younger ages as well, although the age bins are not too well-sampled. At the same time, \[$\alpha$/Fe\] shows a weak increase with $R_{gc}$. No significant gradients are observed with $|$z$|$ or age, except for a weak tendency of \[Fe/H\] to decrease with increasing $|$z$|$ and decrease with age in the outermost disc annulus studied. None of our measured weak trends have any significance above 1$\sigma$. Larger samples of homogeneous data are still necessary to investigate the existence of any dependence on age and $|$z$|$ in the Galactic disc.
We acknowledge the anonymous referee for helping us to improve this paper. R. C acknowledge the support from the Spanish Ministry of Science and Technology (Plan Nacional de Investigación Científica, Desarrollo, e Investigación Tecnológica, AYA2004-06343 and AYA2007-3E3507). R. C. also acknowledges the funds by the Spanish Ministry of Science and Innovation under the Juan de la Cierva and MEC/Fullbrigth fellowships, and by the Centro de Investigaciones de Astronomía (Venezuela) under its postdoctoral fellowship programe.
Alonso, A., Arribas, S., & Mart[í]{}nez-Roger, C. 1999, A&AS, 140, 261
An, D., Terndrup, D. M., Pinsonneault, M. H., Paulson, D. B., Hanson, R. B., & Stauffer, J. R. 2007, , 655, 233
Andersen, J. 1999, Transactions of the IAU, Vol. XXIVA, p.36 (1999) , 24, A36
Andreuzzi, G., Bragaglia, A., Tosi, M., & Marconi, G. 2011, , 412, 1265
Andrievsky, S. M., Luck, R. E., Martin, P., & L[é]{}pine, J. R. D. 2004, , 413, 159
Arp, H. 1962, , 136, 66
Barry, D. C., Cromwell, R. H., & Hege, E. K. 1987, , 315, 264
Barta[š]{}i[= u]{}t[ė]{}, S., Aslan, Z., Boyle, R. P., Kharchenko, N. V., Ossipkov, L. P., & Sperauskas, J. 2003, Baltic Astronomy, 12, 539
Barta[š]{}i[= u]{}t[ė]{}, S., Deveikis, V., Strai[ž]{}ys, V., & Bogdanovi[č]{}ius, A. 2007, Baltic Astronomy, 16, 199
Bensby, T., Feltzing, S., & Lundstr[ö]{}m, I. 2004, , 421, 969
Bessell, M. S. 1979, PASP, 91, 589
Blake, R. M. 2002, Ph.D. Thesis,
Blake, R. M., & Rucinski, S. M. 2004, Bulletin of the American Astronomical Society, 36, 1483
Boesgaard, A. M. 1989, , 336, 798
Boesgaard, A. M., & Friel, E. D. 1990, , 35
Boesgaard, A. M. 1991, , 370, L95
Boesgaard, A. M., Jensen, E. E. C., & Deliyannis, C. P. 2009, , 137, 4949
Bouvier, J., et al. 2008, , 481, 661
Boyarchuk, A. A., Antipova, L. I., Boyarchuk, M. E., & Savanov, I. S. 2000, Astronomy Reports, 44, 76
Bragaglia, A., et al. 2001, , 121, 327
Bragaglia, A., Sestito, P., Villanova, S., Carretta, E., Randich, S., & Tosi, M. 2008, , 480, 79
Brown, J. A., Wallerstein, G., Geisler, D., & Oke, J. B. 1996, , 112, 1551
Burkhart, C., & Coupry, M. F. 1998, , 338, 1073
Burkhart, C., & Coupry, M. F. 2000, , 354, 216
Cardelli, J. A., Clayton, G. C., & Mathis, J. S. 1989, ApJ, 345, 245
Carraro, G., Bertelli, G., Bressan, A., & Chiosi, C. 1993, , 101, 381
Carraro, G., & Chiosi, C. 1994, , 287, 761
Carraro, G., Ng, Y. K., & Portinari, L. 1998, , 296, 1045
Carraro, G., Bresolin, F., Villanova, S., Matteucci, F., Patat, F., & Romaniello, M. 2004, , 128, 1676
Carraro, G., Villanova, S., Demarque, P., McSwain, M. V., Piotto, G., & Bedin, L. R. 2006, , 643, 1151
Carraro, G., de La Fuente Marcos, R., Villanova, S., Moni Bidin, C., de La Fuente Marcos, C., Baumgardt, H., & Solivella, G. 2007a, , 466, 931
Carraro, G., Geisler, D., Villanova, S., Frinchaboy, P. M., & Majewski, S. R. 2007b, , 476, 217
Carraro, G., Villanova, S., Demarque, P., Moni Bidin, C., & McSwain, M. V. 2008, , 386, 1625
Carraro, G., & Bensby, T. 2009, , 397, L106
Carretta, E., Bragaglia, A., Gratton, R. G., & Tosi, M. 2004, , 422, 951
Carretta, E., Bragaglia, A., Gratton, R. G., & Tosi, M. 2005, , 441, 131
Carretta, E., Bragaglia, A., & Gratton, R. G. 2007, , 473, 129
Carretta, E., Bragaglia, A., Gratton, R. G., Recio-Blanco, A., Lucatello, S., D’Orazi, V., & Cassisi, S. 2010, , 516, A55
Cayrel, R., et al. 2004, , 416, 1117
Chen , L., Hou, J. L. & Wang, J. J. 2003, AJ, 125, 1397
Chiappini, C., Matteucci, F., & Romano, D. 2001, , 554, 1044
Claria, J. J., Piatti, A. E., & Osborn, W. 1996, , 108, 672
Coleman, L. A. 1982, , 87, 369
Costa, R. D. D., Uchida, M. M. M., & Maciel, W. J. 2004, , 423, 199
Crawford, D. L., & Barnes, J. V. 1970, , 75, 946
Cutri, R.M., Skrutskie, M.F., Van Dyk, et al. 2003, Explanatory Supplement to the 2MASS All Sky Data Release Daflon, S., & Cunha, K. 2004, , 617, 1115
Daniel, S. A., Latham, D. W., Mathieu, R. D., & Twarog, B. A. 1994, , 106, 281
Dean, J. F., Warren, P. R., & Cousins, A. W. J. 1978, MNRAS, 183, 569
Deharveng, L., Pe[ñ]{}a, M., Caplan, J., & Costero, R. 2000, , 311, 329
Den Hartog, E. A., Lawler, J. E.,Sneden, C., Cowan, J. J. 2003, ApJS, 148, 543
De Silva, G. M., Sneden, C., Paulson, D. B., Asplund, M., Bland-Hawthorn, J., Bessell, M. S., & Freeman, K. C. 2006, , 131, 455
De Silva, G. M., Freeman, K. C., Asplund, M., Bland-Hawthorn, J., Bessell, M. S., & Collet, R. 2007, , 133, 1161
de Silva, G. M., Gibson, B. K., Lattanzio, J., & Asplund, M. 2009, , 500, L25
Dias, W. S., Alessi, B. S., Moitinho, A., & L[é]{}pine, J. R. D. 2002, , 389, 871
Dinescu, D. I., Demarque, P., Guenther, D. B., & Pinsonneault, M. H. 1995, , 109, 2090
D’Orazi, V., Bragaglia, A., Tosi, M., Di Fabrizio, L., & Held, E. V. 2006, , 368, 471
D’Orazi, V., Magrini, L., Randich, S., Galli, D., Busso, M., & Sestito, P. 2009, , 693, L31
D’Orazi, V., & Randich, S. 2009, , 501, 553
Dutra, C. M., & Bica, E. 2000, , 359, 347
Dzervitis, U., & Paupers, O. 1993, , 199, 77
Edvardsson, B., Andersen, J., Gustafsson, B., Lambert, D. L., Nissen, P. E., Tomkin, J., 1993, A&A, 275, 101
Edvardsson, B., Pettersson, B., Kharrazi, M., & Westerlund, B. 1995, , 293, 75
Eggen, O. J. 1963, , 138, 356
Eggen, O. J. 1989, , 101, 54 Eggen, O. J. 1998, , 116, 284
Famaey, B., Jorissen, A., Luri, X., Mayor, M., Udry, S., Dejonghe, H., & Turon, C. 2005, , 430, 165
Famaey, B., Pont, F., Luri, X., Udry, S., Mayor, M., & Jorissen, A. 2007, , 461, 957 Feltzing, S., Holmberg, J., & Hurley, J. R. 2001, , 377, 911
Ford, A., Jeffries, R. D., & Smalley, B.2005, , 364, 272
Fossati, L., Folsom, C. P., Bagnulo, S., Grunhut, J. H., Kochukhov, O., Landstreet, J. D., Paladini, C., & Wade, G. A. 2011, , 413, 1132
Francic, S. P. 1989, , 98, 888
Friel, E. D., & Boesgaard, A. M. 1992, , 387, 170
Friel E.D., Janes K.A. 1993 A&A, 267, 75
Friel, E. D. 1995, , 33, 381
Friel, E. D., Janes, K. A., Tavarez, M., Scott, J., Katsanis, R., Lotz, J., Hong, L., & Miller, N. 2002, , 124, 2693
Friel, E. D., Jacobson, H. R., Barrett, E., Fullton, L., Balachandran, S. C., & Pilachowski, C. A. 2003, , 126, 2372
Friel, E. D., Jacobson, H. R., & Pilachowski, C. A. 2005, , 129, 2725
Friel, E. D., Jacobson, H. R., & Pilachowski, C. A. 2010, , 139, 1942
Frinchaboy, P. M., Marino, A. F., Villanova, S., Carraro, G., Majewski, S. R., & Geisler, D. 2008, , 391, 39
Fulbright, J. P., McWilliam, A., & Rich, R. M. 2007, , 661, 1152 G[á]{}sp[á]{}r, A., Rieke, G. H., Su, K. Y. L., Balog, Z., Trilling, D., Muzzerole, J., Apai, D., & Kelly, B. C. 2009, , 697, 1578
Gebran, M., Monier, R., & Richard, O. 2008, , 479, 189
Gebran, M., & Monier, R. 2008, , 483, 567
Giardino, G., Pillitteri, I., Favata, F., & Micela, G. 2008, , 490, 113
Gilmore, G., Wyse, R. F. G., & Jones, J. B. 1995, , 109, 1095
Girardi, L., Mermilliod, J.-C., & Carraro, G. 2000, , 354, 892
Girardi, L., & Salaris,M. 2001, , 323, 109
Gonzalez, G., & Lambert, D. L. 1996, , 111, 424
Gonzalez, G., & Wallerstein, G. 2000, , 112, 1081
Gratton, R. G., & Contarini, G.1994, , 283, 911
Gratton, R. G., Carretta, E., Eriksson, K., Gustafsson, B. 1999, A&A, 350, 955
Grevesse, N., Noels, A., & Sauval, A. J. 1996, ASP Conf. Ser. 99: Cosmic Abundances, 99, 117
Griffin, R. F., Griffin, R. E. M., Gunn, J. E., & Zimmerman, B. A. 1988, , 96, 172
Hamdani, S., North, P., Mowlavi, N., Raboud, D., & Mermilliod, J.-C. 2000, , 360, 509
Hardy, E. 1979, , 84, 319
Heinemann, K. 1926, Astronomische Nachrichten, 227, 193
Hernandez, M. M., Perez Hernandez, F., Michel, E., Belmonte, J. A., Goupil, M. J., & Lebreton, Y. 1998, , 338, 511
Hertzsprung, E. 1909, , 30, 135
Hill, V., & Pasquini, L. 1999, , 348, L21
Hobbs, L. M., & Thorburn, J. A. 1992, , 104, 669
H[ø]{}g, E., et al. 2000, , 355, L27
Hui-Bon-Hoa, A., & Alecian, G. 1998, , 332, 224 Jacobson, H. R., Friel, E. D., & Pilachowski, C. A. 2007, , 134, 1216
Jacobson, H. R., Friel, E. D., & Pilachowski, C. A. 2008, , 135, 2341
Jacobson, H. R., Friel, E. D., & Pilachowski, C. A. 2009, , 137, 4753
Jacobson, H. R., Friel, E. D., & Pilachowski, C. A. 2011a, , 141, 58
Jacobson, H. R., Pilachowski, C. A., & Friel, E. D. 2011b, , 142, 59
Jacquinet-Husson, N., Ari, E., Ballard, J., Barbe, A., Bjoraker, G. et al. 1999, JQSRT, 62, 205
Jacquinet-Husson, N., Scott, N. A., Garceran, K., Armante, R., Chédin, A. 2005, JQSRT, 95, 429
Jameson, R. F., Lodieu, N., Casewell, S. L., Bannister, N. P., & Dobbie, P. D. 2008, , 385, 1771
Janes, K. A. 1979, , 39, 135
Jennens, P. A., & Helfer, H. L.1975, , 172, 681
Johansson, S., Litz[é]{}n, U., Lundberg, H., & Zhang, Z. 2003, , 584, L107
Johnson, H. L. 1952, , 116, 640
Johnson, H. L. 1953, , 117, 356
Johnson, H. L., & Knuckles, C. F. 1955, , 122, 209
Johnson, H. L. 1961, Lowell Observatory Bulletin, 5, 133
Johnson, H. L., Iriarte, B., Mitchell, R. I., & Wisniewskj, W. Z. 1966, Communications of the Lunar and Planetary Laboratory, 4, 99
Kaluzny, J., & Mazur, B. 1991, Acta Astronomica, 41, 167
Karata[ş]{}, Y., Bilir, S., & Schuster, W. J. 2005, , 360, 1345 King, J. R., Soderblom, D. R., Fischer, D., & Jones, B. F. 2000, , 533, 944
Klein Wassink, W. J. 1927, Publications of the Kapteyn Astronomical Laboratory Groningen, 41, 1
Komarov, N. S., & Basak, N. Y. 1993, , 70, 111
Kraus, A. L., & Hillenbrand, L. A. 2007, , 134, 2340
Kupka, F., Piskunov, N., Ryabchikova, T. A., Stempels, H. C., & Weiss, W. W. 1999, , 138, 119
Lata, S., Pandey, A. K., Sagar, R., & Mohan, V. 2002, , 388, 158
Lemasle, B., Fran[ç]{}ois, P., Piersimoni, A., Pedicelli, S., Bono, G., Laney, C. D., Primas, F., & Romaniello, M. 2008, , 490, 613
Letarte, B., Hill, V., Jablonka, P., Tolstoy, E., Fran[ç]{}ois, P., & Meylan, G. 2006, , 453, 547
Loktin, A. V. 2000, Astronomy Letters, 26, 657
Loktin, A. V., & Beshenov, G. V. 2001, Astronomy Letters, 27, 386
Luck, R. E. 1994, , 91, 309
Luck, R. E., & Challener, S. L. 1995, , 110, 2968
Maciel, W. J., Costa, R. D. D., & Uchida, M. M. M. 2003, , 397, 667
Maciel, W. J., Quireza, C., & Costa, R. D. D. 2007, , 463, L13
Maciel, W. J., & Costa, R. D. D. 2009, IAU Symposium, 254, 38P
Maeder, A. 1971, , 10, 354
Magain, P. 1984, , 134, 189
Magrini, L., Sestito, P., Randich, S., & Galli, D. 2009, , 494, 95
Magrini, L., Randich, S., Zoccali, M., Jilkova, L., Carraro, G., Galli, D., Maiorca, E., & Busso, M. 2010, , 523, 11
Mallik, S. V. 1998, , 338, 623
Marsakov, V. A., & Borkova, T. V. 2005, Astronomy Letters, 31, 515
Marsakov, V. A., & Borkova, T. V. 2006, Astronomy Letters, 32, 376
Martell, S. L., & Smith, G. H. 2009, , 121, 577
Martell, S. L., & Grebel, E. K. 2010, , 519, A14
Mathieu, R. D., & Mazeh, T. 1988, , 326, 256
Mazzei, P., & Pigatto, L. 1988, , 193, 148
Mendoza, E. E. 1967, Boletín de los Observatorios Tonantzintla y Tacubaya, 4, 149
Mermilliod, J.-C. 1995, ASSL Vol. 203: Information & On-Line Data in Astronomy, 127
Mermilliod, J.-C., Mathieu, R. D., Latham, D. W., & Mayor, M. 1998, , 339, 423
Mermilliod, J.-C., Andersen, J., Latham, D. W., & Mayor, M. 2007, , 473, 829
Meynet, G., Mermilliod, J.-C., & Maeder, A. 1993, , 98, 477
Mikolaitis, [Š]{}., Tautvai[š]{}ien[ė]{}, G., Gratton, R., Bragaglia, A., & Carretta, E. 2010, , 407, 1866
Mikolaitis, [Š]{}., Tautvai[š]{}ien[ė]{}, G., Gratton, R., Bragaglia, A., & Carretta, E. 2011a, , 413, 2199
Mikolaitis, [Š]{}., Tautvai[š]{}ien[ė]{}, G., Gratton, R., Bragaglia, A., & Carretta, E. 2011b, arXiv:1105.4047
Milone, E. F., Stagg, C. R., Sugars, B. A., McVean, J. R., Schiller, S. J., Kallrath, J., & Bradstreet, D. H. 1995, , 109, 359
Mishenina, T. V., Bienaym[é]{}, O., Gorbaneva, T. I., Charbonnel, C., Soubiran, C., Korotin, S. A., & Kovtyukh, V. V. 2006, , 456, 1109
Mishenina, T. V., Gorbaneva, T. I., Bienaym[é]{}, O., Soubiran, C., Kovtyukh, V. V., & Orlova, L. F. 2007, Astronomy Reports, 51, 382
Molla, M., Ferrini, F., & Diaz, A. I. 1996, , 466, 668
Montegriffo, P., Ferraro, F. R., Origlia, L., & Fusi Pecci, F. 1998, , 297, 872
Monroe, T. R., & Pilachowski, C. A. 2010, , 140, 2109
Mould, J. 2005, , 129, 698
Mucciarelli, A., Origlia, L., Ferraro, F. R., & Pancino, E. 2009, , 695, L134
Mucciarelli, A. 2011, , 528, 44
Narayanan, V. K., & Gould, A. 1999, , 515, 256
Navarro, J. F., Abadi, M. G., Venn, K. A., Freeman, K. C., & Anguiano, B. 2011, , 412, 1203
Nicolet, B. 1981, , 104, 185
Nissen, P. E. 1988, , 199, 146 Nordstr[ö]{}m, B., et al. 2004, , 418, 989
Origlia, L., Valenti, E., Rich, R. M., & Ferraro, F. R. 2006, , 646, 499
Pace, G., Pasquini, L., & Fran[ç]{}ois, P. 2008, , 489, 403
Pace, G., Danziger, J., Carraro, G., Melendez, J., Fran[ç]{}ois, P., Matteucci, F., & Santos, N. C. 2010, , 515, A28
Pancino, E., Carrera, R., Rossetti, E., & Gallart, C. 2010, , 511, 56. Paper I
Pancino, E., Rejkuba, M., Zoccali, M., & Carrera, R. 2010, , 524, A44
Pasquini, L., Randich, S., Zoccali, M., Hill, V., Charbonnel, C., & Nordstr[ö]{}m, B. 2004, , 424, 951
Patenaude, M. 1978, , 66, 225
Paulson, D. B., Sneden, C., & Cochran, W. D. 2003, , 125, 3185
Pereira, C. B., & Quireza, C. 2010, IAU Symposium, 266, 495
Percival, S. M., Salaris, M., & Kilkenny, D. 2003, , 400, 541
Perryman, M. A. C., et al. 1997, , 323, L49
Perryman, M. A. C., et al. 1998, , 331, 81
Peterson, R. C., & Green, E. M. 1998, , 502, L39
Pfeiffer, M. J., Frank, C., Baumueller, D., Fuhrmann, K., & Gehren, T. 1998, , 130, 381
Piatti, A. E., Claria, J. J., & Abadi, M. G. 1995, , 110, 2813
Pinsonneault, M. H., Stauffer, J., Soderblom, D. R., King, J. R., & Hanson, R. B. 1998, , 504, 170
Platais, I., Melo, C., Fulbright, J. P., Kozhurina-Platais, V., Figueira, P., Barnes, S. A., & M[é]{}ndez, R. A. 2008, , 391, 1482
Pourbaix, D., et al. 2004, , 424, 727 Ram[í]{}rez, S. V.,& Cohen, J. G. 2003, , 125, 224
Randich, S., Pallavicini, R., Meola, G., Stauffer, J. R., & Balachandran, S. C. 2001, , 372, 862
Randich, S., Sestito, P., & Pallavicini, R. 2003, , 399, 133
Randich, S., Sestito, P., Primas, F., Pallavicini, R., & Pasquini, L. 2006, , 450, 557
Reddy, B. E., Tomkin, J., Lambert, D. L., & Allende Prieto, C. 2003, , 340, 304
Reddy, B. E., Lambert, D. L., & Allende Prieto, C. 2006, , 367, 1329
Reid, I. N., Turner, E. L., Turnbull, M. C., Mountain, M., & Valenti, J. A. 2007, , 665, 767
Richtler, T., & Sagar, R. 2001, Bulletin of the Astronomical Society of India, 29, 53
Roman, N. G. 1955, , 121, 454 Rohlfs, K., & Vanysek, V. 1962, Astronomische Abhandlungen der Hamburger Sternwarte, 5, 341
Ro[š]{}kar, R., Debattista, V. P., Quinn, T. R., Stinson, G. S., & Wadsley, J. 2008, , 684, L79
Salaris, M., Weiss, A., & Percival, S. M. 2004, , 414, 163
Santos, N. C., Lovis, C., Pace, G., Melendez, J., & Naef, D. 2009, , 493, 309
Schuler, S. C., King, J. R., Fischer, D. A., Soderblom, D. R., & Jones, B. F. 2003, , 125, 2085
Schuler, S. C., Hatzes, A. P., King, J. R., K[ü]{}rster, M., & The, L.-S. 2006, , 131, 1057
Schuler, S. C., King, J. R., & The, L.-S. 2009, , 701, 837
Sestito, P., Randich, S., Mermilliod, J.-C., & Pallavicini, R. 2003, , 407, 289
Sestito, P., Randich, S., & Pallavicini, R. 2004, , 426, 809
Sestito, P., Bragaglia, A., Randich, S., Carretta, E., Prisinzano, L., & Tosi, M. 2006, , 458, 121
Sestito, P., Randich, S., & Bragaglia, A.2007, , 465, 185
Sestito, P., Bragaglia, A., Randich, S., Pallavicini, R., Andrievsky, S. M., & Korotin, S. A. 2008, , 488, 943
Schaye, J. 2004, , 609, 667
Shen, Z.-X., Jones, B., Lin, D. N. C., Liu, X.-W., & Li, S.-L. 2005, , 635, 608
Skrutskie, M. F., et al. 2006, , 131, 1163
Smiljanic, R., Gauderon, R., North, P., Barbuy, B., Charbonnel, C., & Mowlavi, N. 2009, , 502, 267
Smith, V. V., & Suntzeff, N. B. 1987, , 93, 359
Soderblom, D. R., Laskar, T., Valenti, J. A., Stauffer, J. R., & Rebull, L. M. 2009, , 138, 1292
Soubiran, C., & Girard, P. 2005, , 438, 139
Soubiran, C., Bienaym[é]{}, O., Mishenina, T. V., & Kovtyukh, V. V. 2008, , 480, 91
Spite, M. 1967, Annales d’Astrophysique, 30, 211
Stanghellini, L., & Haywood, M. 2010, , 714, 1096
Stetson, P. B., & Pancino, E. 2008, , 120, 1332
Tadross, A. L. 2001, New Astronomy, 6, 293
Tautvaiŝiene, G., Edvardsson, B., Tuominen, I., Ilyin, I. 2000, A&A, 360, 499
Tautvai[š]{}ien[ė]{}, G., Edvardsson, B., Puzeras, E., & Ilyin, I. 2005, , 431, 933
Taylor, B. J. 2006, , 132, 2453
Taylor, B. J. 2007, , 134, 934
Terndrup, D. M., Pinsonneault, M., Jeffries, R. D., Ford, A., Stauffer, J. R., & Sills, A. 2002, , 576, 950
Tosi, M., Bragaglia, A., & Cignoni, M. 2007, , 378, 730
Tsvetkov, T. G. 1993, , 203, 247
Twarog, B. A. 1983, , 267, 207
Twarog, B. A., Ashman, K. M., & Anthony-Twarog, B. J. 1997, AJ, 114, 2556
Twarog, B. A., Anthony-Twarog, B. J., & De Lee, N. 2003, , 125, 1383
van Bueren, H. G. 1952, , 11, 385
van den Heuvel, E. P. J. 1969, , 81, 815
van Leeuwen, F. 1999, , 341, L71
Varenne, O., & Monier, R. 1999, , 351, 247
Villanova, S., Carraro, G., Bresolin, F., & Patat, F. 2005, , 130, 652
Villanova, S., Baume, G., & Carraro, G. 2007, , 379, 1089
Villanova, S., Carraro, G., & Saviane, I. 2009, , 504, 845
Villanova, S., Geisler, D., & Piotto, G. 2010, , 722, L18
Yong D., Carney B.W., & Texeira de Almeida M.L. 2005, AJ, 130, 597
[^1]: Based on observations collected with the fiber spectrograph FOCES at the 2.2m Calar Alto Telescope. Also based on data from 2MASS survey and the WEBDA, VALD, NIST, and GEISA online database.
[^2]: All I magnitudes are in the Johnson system (I$_J$) with the exception of those of the Be 32 stars which are in the Cousins system (I$_C$).
[^3]: Star names from @Richtler2001; Coordinates, B, V & I$_C$ magnitudes from @dorazi2006; K$_S$ magnitudes from 2MASS.
[^4]: Star names from @Heinemann1926; Coordinates from @hog2000; B & V magnitudes from @jennens75; K$_S$ magnitudes from 2MASS.
[^5]: Star names from @vanBueren1952; Coordinates from @perryman1997; B, V, R & I$_J$ magnitudes from @Johnson1966; K$_S$ magnitudes from 2MASS.
[^6]: Star names from @kleinWassink1927; Coordinates from @perryman1997; B, V & R magnitudes from @coleman82; I$_J$ magnitudes from @mendoza1967 [@Johnson1966], K$_S$ magnitudes from 2MASS.
[^7]: [http://www.univie.ac.at/webda]{}
[^8]: [http://www.ipac.caltech.edu/2mass]{}. 2MASS (Two Micron All Sky Survey) is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation.
[^9]: Image Reduction and Analysis Facility, IRAF is distributed by the National Optical Astronomy Observatories, which are operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation.
[^10]: @dorazi2006.
[^11]: @mermilliod98.
[^12]: Spectroscopic binary according to @pourbaix2004.
[^13]: @griffin88.
[^14]: @Famaey2005.
[^15]: However, see the discussion by @mucciarelli11 about the pros and cons of the @magain1984 correction, which depends heavily on data quality and line selection effects.
[^16]: Averages of measurements by @kaluzny1991, @carraro1994, @dutra2000, @Richtler2001, @tadross2001, @lata2002, @salaris2004, @dorazi2006, and @tosi2007.
[^17]: Averages of measurements by @johnson1953, @roman1955, @johnson1961, @rohlfs1962, @arp1962, @eggen1963, @crawford1970, @patenaude1978, @hardy1979, @nicolet1981, @twarog1983, @barry1987, @nissen1988, @mazzei1988, @eggen1989, @francic1989, @boesgaard1991, @dzervitis1993, @carraro1993, @meynet1993, @daniel1994, @piatti1995, @dinescu1995, @milone1995, @claria1996, @dutra2000, @loktin2001, @blake2002, @blake2004, @salaris2004, @bartasiute2007, @taylor2007, and @giardino2008.
[^18]: Derived by @taylor2006 from a review of published values.
[^19]: Averages of measurements obtained from the Hipparcos parallaxes by @pinsonneault1998, @perryman1998, @narayanan1999, @loktin2001, and @percival2003.
[^20]: Averages of measurements by @eggen1998, @loktin2001, @salaris2004, @jameson2008, and @bouvier2008.
[^21]: Averages of measurements obtained from the Hipparcos parallaxes by @pinsonneault1998, @perryman1998, @vanleeuwen1999, @loktin2000, @loktin2001, and @percival2003.
[^22]: Averages of measurements obtained from the Hipparcos parallaxes by @vandenheuvel1969, @maeder1971, @mathieu1988, @mazzei1988, @boesgaard1989, @tsvetkov1993, @piatti1995, @claria1996, @hernandez1998, @loktin2001, @salaris2004, @kraus2007, and @gaspar2009
[^23]: We decided not to use the @magain1984 effect, because we prefer to have internally consistent abundances from each line, and because of the additional effects described by @mucciarelli11.
[^24]: @friel2010, from 2 stars.
[^25]: @bragaglia2008 and @sestito2006, from 10 red clump and RGB stars.
[^26]: @schuler2009 and @schuler2006, from the same 3 K giants studied here.
[^27]: @boyarchuk2000, from the same 3 K giants studied here.
[^28]: @varenne1999, from 29 F dwarfs.
[^29]: @pace2008, from 6 G and 1 F main sequence stars.
[^30]: @an2007, from 4 G dwarfs stars.
[^31]: @burkhart1998, from 10 Am stars.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
A non-uniform hypergraph $H=(V,E)$ consists of a vertex set $V$ and an edge set $E\subseteq 2^V$; the edges in $E$ are not required to all have the same cardinality. The set of all cardinalities of edges in $H$ is denoted by $R(H)$, the set of edge types. For a fixed hypergraph $H$, the Turán density $\pi(H)$ is defined to be $\lim_{n\to\infty}\max_{G_n}h_n(G_n)$, where the maximum is taken over all $H$-free hypergraphs $G_n$ on $n$ vertices satisfying $R(G_n)\subseteq R(H)$, and $h_n(G_n)$, the so called Lubell function, is the expected number of edges in $G_n$ hit by a random full chain. This concept, which generalizes the Turán density of $k$-uniform hypergraphs, is motivated by recent work on extremal poset problems. The details connecting these two areas will be revealed in the end of this paper.
Several properties of Turán density, such as supersaturation, blow-up, and suspension, are generalized from uniform hypergraphs to non-uniform hypergraphs. Other questions such as “Which hypergraphs are degenerate?" are more complicated and don’t appear to generalize well. In addition, we completely determine the Turán densities of $\{1,2\}$-hypergraphs.
author:
- 'Travis Johnston [^1]'
- 'Linyuan Lu [^2]'
title: 'Turán Problems on Non-uniform Hypergraphs'
---
Introduction
============
A hypergraph $H$ is a pair $(V,E)$; $V$ is the vertex set, and $E\subseteq 2^V$ is the edge set. If all edges have the same cardinality $k$, then $H$ is a $k$-uniform hypergraph. Turán problems on $k$-uniform hypergraphs have been actively studied for many decades. However, Turán problems on non-uniform hypergraphs are rarely considered (see [@MZ; @Lemons] for two different treatments). Very recently, several groups of people have started actively studying extremal families of sets avoiding given sub-posets. Several new problems have been established. One of them is the diamond problem:
[**The diamond conjecture: [@GriLu]**]{} Any family ${\mathcal{F}}$ of subsets of $[n]:=\{1,2,\ldots,n\}$ with no four sets $A,B,C,D$ satisfying $A\subseteq B\cap C$, $B\cup C \subseteq D$ can have at most $(2+o(1)){\binom{n}{\lfloor \frac{n}{2}\rfloor}}$ subsets.
This conjecture, along with many other problems, motivates us to study Turán-type problems on non-uniform hypergraphs. The details of this connection will be given in the last section.
We briefly review the history of Turán Problems on uniform hypergraphs. Given a positive integer $n$ and a $k$-uniform hypergraph $H$ on $n$ vertices (or $k$-graph, for short), the Turán number ${{\rm ex}}(n,H)$ is the maximum number of edges in a $k$-graph on $n$ vertices that does not contain $H$ as a subgraph; such a graph is called $H$-*free*. Katona et al. [@KNS] showed that $f(n,H)={{\rm ex}}(n,H)/{n\choose k}$ is a decreasing function of $n$. The limit $\displaystyle \pi(H)=\lim_{n\to \infty} f(n,H)$, which always exists, is called the [*Turán density*]{} of $H$.
For $k=2$, the graph case, Erdős-Stone-Simonovits proved $\pi(G)=1-\frac{1}{\chi(G)-1}$ for any graph $G$ with chromatic number $\chi(G)\geq 3$. If $G$ is bipartite, then ${{\rm ex}}(n,G)=o(n^2)$. The magnitude of ${{\rm ex}}(n,G)$ is unknown for most bipartite graphs $G$.
Let $K^r_k$ denote the complete $r$-graph on $k$ vertices. Turán determined the value of ${{\rm ex}}(n,K^2_{k})$ which implyies that $\pi(K^2_k)=1-\frac{1}{k-1}$ for all $k\geq 3$. However, no Turán density $\pi(K^r_k)$ is known for any $k>r\geq 3$. The most extensively studied case is when $k=4$ and $r=3$. Turán conjectured [@Tu] that $\pi(K_4^3)= 5/9$. Erdős [@Er81] offered \$500 for determining any $\pi(K^r_k)$ with $k>r\geq 3$ and \$1000 for answering it for all $k$ and $r$. The upper bounds for $\pi(K_4^3)$ have been sequentially improved: $0.6213$ (de Caen [@dC94]), $0.5936$ (Chung-Lu [@ChLu]), $0.56167$ (Razborov [@Razborov], using flag algebra method.) There are a few uniform hypergraphs whose Turán density has been determined: the Fano plane [@FS05; @KS05a], expanded triangles [@KS05b], $3$-books, $4$-books [@FMP], $F_5$ [@FrFu83], extended complete graphs [@Pik], etc. In particular, Baber [@baber] recently found the Turán density of many $3$-uniform hypergraphs using flag algebra methods. For a more complete survey of methods and results on uniform hypergraphs see Peter Keevash’s survey paper [@KeevashSurvey].
A non-uniform hypergraph $H=(V,E)$ consists of a vertex set $V$ and an edge set $E\subseteq 2^V$. Here the edges of $E$ could have different cardinalities. The set of all the cardinalities of edges in $H$ is denoted by $R(H)$, the set of edge types. For a fixed hypergraph $H$, the Turán density $\pi(H)$ is defined to be $\lim_{n\to\infty}\max_{G_n}h_n(G_n)$, where the maximum is taken over all $H$-free hypergraphs $G_n$ on $n$ vertices satisfying $R(G_n)\subseteq R(H)$. $h_n(G_n)$, the so called Lubell function, is the expected number of edges in $G_n$ hit by a random full chain. The Lubell function has been a very useful tool in extremal poset theory, in particular it has been used in the study of the diamond conjecture.
In section 2, we show that our notion of $\pi(H)$ is well-defined and is consistent with the usual definition for uniform hypergraphs. We also give examples of Turán densities for several non-uniform hypergraphs. In section 3, we generalize the supersaturation Lemma to non-uniform hypergraphs. Then we prove that blowing-up will not affect the Turán density. Using various techniques, we determine the Turán density of every $\{1,2\}$-hypergraph in section 4. Remarkably, the Turán densities of $\{1,2\}$-hypergraphs are in the set $$\bigg\{1,\frac{9}{8}, \frac{5}{4}, \frac{3}{2}, \frac{5}{3},
\ldots, 2-\frac{1}{k},\ldots \bigg \}.$$
Among $r$-uniform hypergraphs, $r$-partite hypergraphs have the smallest possible Turán density. Erdős proved that any $r$-uniform hypergraph forbidding the complete $r$-uniform $r$-partite hypergraphs can have at most $O(n^{r-1/\delta})$ edges. We generalize this theorem to non-uniform hypergraphs. A hypergraph is degenerate if it has the smallest possible Turán density. For $r$-uniform hypergraphs, a hypergraph $H$ is degenerate if and only if $H$ is the subgraph of a blow-up of a single edge. Unlike the degenerate $r$-uniform hypergraphs, the degenerate non-uniform hypergraphs are not classified. For non-uniform hypergraphs, chains–one natural extension of a single edge–are degenerate. Additionally, every subgraph of a blow-up of a chain is also degenerate. However, we give an example of a degenerate, non-uniform hypergraph not contained in any blow-up of a chain. This leaves open the question of which non-uniform hypergraphs are degenerate.
In section 6, we consider the suspension of hypergraphs. The suspension of a hypergraph $H$ is a new hypergraph, denoted by $S(H)$, with one additional vertex, $\ast$, added to every edge of $H$. In a hypergraph Turán problem workshop hosted by the AIM Research Conference Center in 2011, the following conjecture was posed: $\displaystyle \lim_{t\to\infty}\pi(S^t(K^{r}_n))=0$. We conjecture $\displaystyle\lim_{t\to\infty} \pi(S^t(H))=|R(H)|-1$ holds for any hypergraph $H$. Some partial results are proved. Finally in the last section, we will point out the relation between the Turán problems on hypergraphs and extremal poset problems.
Non-uniform hypergraphs
=======================
Notation
--------
Recall that a hypergraph $H$ is a pair $(V,E)$ with the vertex set $V$ and edge set $E\subseteq 2^{V}$. Here we place no restriction on the cardinalities of edges. The set $R(H):=\{|F|\colon F\in E\}$ is called the set of its [*edge types*]{}. A hypergraph $H$ is $k$-uniform if $R(H)=\{k\}$. It is non-uniform if it has at least two edge types. For any $k\in R(H)$, the [*level hypergraph*]{} $H^k$ is the hypergraph consisting of all $k$-edges of $H$. A uniform hypergraph $H$ has only one (non-empty) level graph, i.e., $H$ itself. In general, a non-uniform hypergraph $H$ has $|R(H)|$ (non-empty) level hypergraphs. Throughout the paper, for any finite set $R$ of non-negative integers, we say, $G$ is an $R$-graph if $R(G)\subseteq R$. We write $G^R_n$ for a hypergraph on $n$ vertices with $R(G)\subseteq R$. We simplify it to $G$ if $n$ and $R$ are clear under context.
Let $R$ be a fixed set of edge types. Let $H$ be an $R$-graph. The number of vertices in $H$ is denoted by $v(H):=|V(H)|$. Our goal is to measure the edge density of $H$ and be able to compare it (in a meaningful way) to the edge density of other $R$ graphs. The standard way to measure this density would be: $$\mu(H) = \frac{|E(H)|}{\sum_{k\in R(H)}\binom{v(H)}{k}}.$$ This density ranges from 0 to 1 (as one would expect)–a complete $R$-graph having a density of 1. Unfortunately, this is no longer a useful measure of density since the number of edges with maximum cardinality will dwarf the number of edges of all other sizes. Specifically, one could take $k$-uniform hypergraph (where $k=\max\{r:r\in R(H)\}$) on enough vertices and make its density as close to 1 he likes. The problem is that this $k$-uniform hypergraph is quite different from the complete $R$-graph (when $|R|>1$) with the same number of vertices. Instead, we use the Lubell function to measure the edge density. This is adapted from the use of the Lubell function studying families of subsets.
For a non-uniform hypergraph $G$ on $n$ vertices, we define the Lubell function of $G$ as $$\label{eq:lubell}
h_{n}(G):=\sum_{F\in E(G)}\frac{1}{\binom{n}{|F|}}=\sum_{k\in R(G)}\frac{|E(H^{k})|}{\binom{n}{k}}.$$
The Lubell function is the expected number of edges hit by a random full chain. Namely, pick a uniformly random permutation $\sigma$ on $n$ vertices; define a random full chain $C_\sigma$ by $$\{ \{\emptyset\}, \{\sigma(1)\}, \{\sigma(1), \sigma(2)\}, \cdots, [n]\}.$$ Let $X=|E(G)\cap C_{\sigma}|$, the number of edges hit by the random full chain. Then $$\label{eq:X}
h_n(G)={{\rm E}}(X).$$
Given two hypergraphs $H_1$ and $H_2$, we say $H_1$ is a subgraph of $H_2$, denoted by $H_1\subseteq H_2$, if there exists a 1-1 map $f\colon V(H_1)\to V(H_2)$ so that $f(F)\in E(H_2)$ for any $F\in E(H_1)$. Whenever this occurs, we say the image $f(H_1)$ is an [*ordered copy*]{} of $H_2$, written as $H_1\stackrel{f}{\hookrightarrow} H_2$. A necessary condition for $H_1\subseteq H_2$ is $R(H_1)\subseteq R(H_2)$. Given a subset $K\subseteq V(H)$ and a subset $S\subseteq R(H)$, the [*induced subgraph*]{}, denoted by $H^{[S]}[K]$, is a hypergraph on $K$ with the edge set $\{F\in E(H)\colon F\subseteq K \mbox{ and } |F|\in S\}$. When $S=R(H)$, we simply write $H[K]$ for $H^{[S]}[K]$.
Given a positive integer $n$ and a subset $R\subseteq [n]$, the complete hypergraph $K^R_{n}$ is a hypergraph on $n$ vertices with edge set $\bigcup_{i\in R} \binom{[n]}{i}$. For example, $K^{\{k\}}_n$ is the complete $k$-uniform hypergraph. $K^{[k]}_n$ is the non-uniform hypergraph with all possible edges of cardinality at most $k$.
(0,0)–(2,1)–(2,-1)–cycle; at (0,0) \[vertex\_open\] (v1) \[label=above:[1]{}\] ; at (2,1) \[vertex\_open\] (v2) \[label=above:[2]{}\] ; at (2,-1) \[vertex\_open\] (v3) \[label=below:[3]{}\] ; (v1)–(v2); (v1)–(v3); (v2)–(v3); at (1,-2) [Illustration of $K_{3}^{\{2,3\}}$]{} ;
(0,0)–(2,0)–(3.5,2)–cycle; (2,0)–(3.5,2)–(2,4)–cycle; (3.5,2)–(2,4)–(0,4)–cycle; (2,4)–(0,4)–(-1.5,2)–cycle; (0,4)–(-1.5,2)–(0,0)–cycle; (-1.5,2)–(0,0)–(2,0)–cycle;
at (0,0) \[vertex\_open\] (v1) \[label=below:[1]{}\] ; at (2,0) \[vertex\_open\] (v2) \[label=below:[2]{}\] ; at (3.5,2) \[vertex\_open\] (v3) \[label=right:[3]{}\] ; at (2,4) \[vertex\_open\] (v4) \[label=above:[4]{}\] ; at (0,4) \[vertex\_open\] (v5) \[label=above:[5]{}\] ; at (-1.5,2) \[vertex\_open\] (v6) \[label=left:[6]{}\] ; (v1)–(v2)–(v3)–(v4)–(v5)–(v6)–(v1); at (1,-1) [A (tight) cycle $C_{6}^{\{2,3\}}$]{};
Given a family of hypergraphs ${\mathcal{H}}$ with common set of edge-types $R$, we define $$\pi_n({\mathcal{H}}):= \max \left\{h_{n}(G)\colon v(G)=n, G\subseteq K^{R}_{n}, \text{ and } G \mbox{ contains no subgraph in }{\mathcal{H}}\right\}.$$
A hypergraph $G:=G_n^R$ is [*extremal*]{} with respect to the family ${\mathcal{H}}$ if
1. $G$ contains no subgraph in ${\mathcal{H}}$.
2. $h_n(G)=\pi_n({\mathcal{H}})$.
The Turán density of ${\mathcal{H}}$ is defined to be $$\begin{aligned}
\pi(\mathcal{{\mathcal{H}}}):&= \lim_{n\to \infty} \pi_n({\mathcal{H}}) \\
&= \lim_{n\to \infty} \max \left\{\sum_{F\in E(G)} \frac{1}{\binom{n}{|F|}}\colon v(G)=n, G\subseteq K^{R}_{n},
\text{ and } G \mbox{ contains no subgraph in }{\mathcal{H}}\right\}\end{aligned}$$ when the limit exists; we will soon show that this limit always exists. When ${\mathcal{H}}$ contains one hypergraph $H$, then we write $\pi(H)$ instead of $\pi(\{H\})$.
Throughout, we will consider $n$ growing to infinity, and $R$ to be a fixed set (not growing with $n$). Note that $\pi(\mathcal{H})$ agrees with the *usual* definition of $$\displaystyle \pi(\mathcal{H})=\lim_{n\to\infty}\frac{\text{ex}(\mathcal{H},n)}{\binom{n}{k}}$$ when $\mathcal{H}$ is a set of $k$-uniform hypergraphs. The following result is a direct generalization Katona-Nemetz-Simonovit’s theorem [@KNS].
For any family ${\mathcal{H}}$ of hypergraphs with a common edge-type $R$, $\pi(\mathcal{H})$ is well-defined, i.e. the limit $\displaystyle \lim_{n\to\infty}\pi_{n}({\mathcal{H}})$ exists.
It suffices to show that $\pi_n({\mathcal{H}})$, viewed as a sequence in $n$, is decreasing.
Write $R=\{k_1,k_2,...,k_{r}\}$. Let $G_{n}\subseteq K^{R}_{n}$ be a hypergraph with $v(G_{n})=n$ not containing $\mathcal{H}$ and with Lubell value $h_n(G_{n})=\pi_n({\mathcal{H}})$. For any $\ell< n$, consider a random subset $S$ of the vertices of $G$ with size $|S|=\ell$.
Let $G_{n}[S]$ be the induced subgraph of $G_{n}$ (whose vertex set is restricted to $S$). Clearly $$\pi_\ell({\mathcal{H}}) \geq
\mathbb{E}(h_{\ell}(G_{n}[S])).$$ Write $E(G_{n})=E_{k_1}\bigcup
E_{k_2}\bigcup ...\bigcup E_{k_r}$ where $E_{k_i}$ contains all the edges of size $k_{i}$. Note that the expected number of edges of size $k_i$ in $G_{n}[S]$ is precisely $\frac{\binom{\ell}{k_i}}{\binom{n}{k_i}}|E_{k_i}|$. It follows that $$\begin{aligned}
\pi_\ell({\mathcal{H}}) &\geq \mathbb{E}(h_{\ell}(G_{n}[S])) \\
&=\sum_{i=1}^{r} \frac{\mathbb{E}(|E_{k_i} \bigcap \binom{S}{k_i}|)}{\binom{\ell}{k_{i}}} \\
&=\sum_{i=1}^{r} \frac{ \frac{\binom{\ell}{k_{i}}}{\binom{n}{k_i}}|E_{k_i}|}{\binom{\ell}{k_{i}}} \\
&=\sum_{i=1}^{r} \frac{ |E_{k_i}| }{\binom{n}{k_i}} \\
&= h_{n}(G_{n})\\
&=\pi_n({\mathcal{H}}).\end{aligned}$$ The sequence $\pi_n({\mathcal{H}})$ is non-negative and decreasing; therefore it converges.
For a fixed set $R:=\{k_1,k_2,\ldots, k_r\}$ (with $k_1<k_2<\cdots < k_r$), an [*$R$-flag*]{} is an $R$-graph containing exactly one edge of each size. The chain $C^R$ is a special $R$-flag with the edge set $${{\rm E}}(C^R)=\{[k_1], [k_2],\ldots, [k_r]\}.$$
\[p1\] For any hypergraph $H$, the following statements hold.
1. $|R(H)|-1\leq \pi_n(H)\leq |R(H)|$.
2. For subgraph $H'$ of $H$, we have $\pi(H')\leq \pi(H)- |R(H)|+|R(H')|$.
3. For any $R$-flag $L$ on $m$ vertices and any $n\geq m$, we have $\pi_n(L)=|R|-1$.
[**Proof:**]{} Pick any maximal proper subset $R'$ of $R(H)$. Consider the complete graph $K_n^{R'}$. Since $K_n^{R'}$ misses one type of edge in $R(H)\setminus R'$, it does not contain $H$ as a subgraph. Thus $$\pi_n(H)\geq h_n(K_n^{R'})=|R'|=|R(H)|-1.$$ The upper bound is due to the fact $h_n(K_n^{R(H)})=|R(H)|$.
Proof of item 2 is similar. Let $S=R(H')$ and $G^S_n$ be an extremal hypergraph for $\pi_n(H')$. Extend $G^S_n$ to $G^{R(H)}_n$ by adding all the edges with cardinalities in $R(H)\setminus S$. The resulting graph $G^{R(H)}_n$ is $H$-free. We have $$\pi_n(H)\geq \pi_n(G^{R(H)}_n)=\pi_{n}(G^S_n)+ |R(H)|-|S|= |R(H)|-|R(H')|+ \pi_n(H').$$ Taking the limit as $n$ goes to infinity, we have $$\pi(H)\geq |R(H)|-|R(H')|+ \pi(H').$$
Finally, for item 3, consider $L$-free hypergraph $G_n^R$. Pick a random $n$-permutation $\sigma$ uniformly. Let $X$ be the number of edges of $G^R_n$ hit by a random flag $\sigma(L)$. Note that each edge $F$ has probability $\frac{1}{{n\choose |F|}}$ of being hit by $\sigma(L)$. We have $$\label{eq:ef}
{{\rm E}}(X)=\sum_{F\in E(G)}\frac{1}{{n\choose |F|}}=h_n(G).$$ Since $G^R_n$ is $L$-free, we have $X\leq r-1$. Taking the expectation, we have $$h_n(G^R_n)={{\rm E}}(X)\leq r-1.$$ Hence, $\pi_n(H)\leq r-1$. The result is followed after combining with item 1. $\square$
A hypergraph $H$ is called [*degenerate*]{} if $\pi(H)=|R(H)|-1$.
By Proposition \[p1\], flags, and specifically chains, are degenerate hypergraphs. A necessary condition for $H$ to be degenerate is that every level hypergraph $H^{k_i}$ is $k_i$-partite. The following examples will show that the converse is not true.
[**Example 1:**]{} The complete hypergraph $K^{\{1,2\}}_2$ has three edges $\{1\}$, $\{2\}$, and $\{1,2\}$. We claim $$\label{eq:k12}
\pi(K^{\{1,2\}}_2)=\frac{5}{4}.$$ The lower bound is from the following construction. Partition $[n]$ into two parts $A$ and $B$ of nearly equal size. Consider the hypergraph $G$ with the edge set $$E(G)={A\choose 1} \cup \left({[n]\choose 2}\setminus {A\choose 2}
\right).$$ It is easy to check $h_n(G)=\frac{5}{4}+O(\frac{1}{n})$ and that $G$ contains no copy of $K_{2}^{\{1,2\}}$.
Now we prove the upper bound. Consider any $K^{\{1,2\}}_2$-free hypergraph $G$ of edge-type $\{1,2\}$ on $n$ vertices. Let $A$ be the set of all singleton edges. For any $x,y\in A$, $xy$ is not a 2-edge of $G$. We have $$\begin{aligned}
h_n(G)&\leq \frac{|A|}{n} + 1-\frac{{|A|\choose 2}}{{n\choose 2}} \\
&= 1+ \frac{|A|}{n} -\frac{|A|^2}{n^2} + O\left(\frac{1}{n}\right) \\
&\leq 1+ \frac{1}{4} + O\left(\frac{1}{n}\right).\end{aligned}$$ In the last step, we use the fact that $f(x)=1+x-x^2$ has the maximum value $\frac{5}{4}$. Combining the upper and lower bounds we have $\pi(K^{\{1,2\}}_2)=\frac{5}{4}$.
The argument is easily generalized to the complete graph $K^{\{1,k\}}_k$ (for $k>1$). We have $$\label{eq:k1k}
\pi(K^{1,k}_k)=1+ \frac{k-1}{k^{\frac{k}{k-1}}}.$$
Let $H$ be a hypergraph. The [*suspension*]{} of $H$, denoted by $S(H)$, is a new hypergraph with the vertex set $V(H)\cup \{\ast\}$ and the edge set $\{F\cup \{\ast\}\colon F\in E(H)\}$. Here $\ast$ is a new vertex not in $H$.
Let $H$ be a hypergraph. The $k$-degree of a vertex $x$, denoted $d_{k}(x)$, is the number of edges of size $k$ containing $x$.
[**Example 2:**]{} Consider $H:=S(K^{\{1,2\}}_2)$. The edges of $H$ are $\{1,2\}$, $\{2,3\}$, $\{1,2,3\}$. We claim $$\pi(S(K^{\{1,2\}}_2))= \frac{5}{4}.$$
The lower bound is from the following construction. Partition $[n]$ into two parts $A$ and $B$ of nearly equal size. Consider the hypergraph $G$ with the edge set $E=E_{2}\bigcup E_{3}$ where $E_{2}=\binom{A}{2}\bigcup \binom{B}{2}$ and $E_{3}=\binom{[n]}{3}\setminus \left(\binom{A}{3} \bigcup \binom{B}{3}\right)$. It is easy to check $h_n(G)=\frac{5}{4}+O\left(\frac{1}{n}\right)$ and that $G$ is $H$-free.
Now we prove the upper bound. Consider any $H$-free hypergraph $G$ of edge-type $\{2,3\}$ on $n$ vertices. Recall that $d_{2}(v)$ denotes the number of 2-edges that contain $v$. For each pair of 2-edges that intersect $v$ there is a unique 3-set containing those two pairs. This 3-set cannot appear in the edge set of $G$ since $G$ is $H$-free. We say that the edge is forbidden. Note that each 3-edge may be forbidden up to 3 times in this manner–depending on which of the three vertices we call $v$. Hence there are at least $\frac{1}{3} \sum_{v\in [n]} \binom{d_{2}(v)}{2}$ 3-edges which are not in $G$. Note that this is true for any $H$-free graph $G$ with number of vertices. Hence $$\begin{aligned}
h_n(G) &\leq \frac{ \binom{n}{3}-\frac{1}{3}\sum_{v\in [n]} \binom{d_{2}(v)}{2} }{\binom{n}{3}} + \frac{ \frac{1}{2}\sum_{v\in [n]} d_{2}(v)}{\binom{n}{2}} \\
&= 1-\frac{\sum_{v\in [n]} d_2(v)^{2}}{6\binom{n}{3}} +\left(\frac{1}{6\binom{n}{3}}+\frac{1}{2\binom{n}{2}}\right)\sum_{v\in [n]} d_2(v) +1.\end{aligned}$$ Applying Cauchy-Schwarz Inequality and letting $m:=\sum_vd_2(v)$, we have $$\begin{aligned}
h_n(G) &\leq \frac{-\left(\sum_{v\in [n]} d_2(v)\right)^2}{6n\binom{n}{3}}
+\left(\frac{1}{6\binom{n}{3}}+\frac{1}{2\binom{n}{2}}\right)\sum_{v\in [n]} d_2(v) + 1 \\
&=-\frac{m^2}{n^4} + \frac{m}{n^2}+1 +O\left(\frac{1}{n}\right)\\
&\leq \frac{5}{4} +O\left(\frac{1}{n}\right).\end{aligned}$$ In the last step, we use the fact that $f(x)=1+x-x^2$ has the maximum value $\frac{5}{4}$. Taking the limit, we get $\pi(S(K_{2}^{\{1,2\}}))\leq \frac{5}{4}.$
We can generalize this construction, giving the following lower bound for $S^k(K_2^{\{1,2\}})$ (the $k$-th suspension of $K_2^{\{1,2\}}$). The details of the computation are omitted. $$\label{eq:sk122}
\pi(S^k(K_2^{\{1,2\}}))\geq 1+ \frac{1}{2^{k+1}}.$$
\[conj:1\] For any $k\geq 2$, $\pi(S^k(K_2^{\{1,2\}}))= 1+ \frac{1}{2^{k+1}}.$
Supersaturation and Blowing-up
==============================
Supersaturation Lemma [@ErSi] is an important tool for uniform hypergraphs. There is a natural generalization of the supersaturation lemma and blowing-up in non-uniform hypergraphs.
[**(Supersaturation)**]{} For any hypergraph $H$ and $a>0$ there are $b$, $n_0>0$ so that if $G$ is a hypergraph on $n>n_0$ vertices with $R(G)=R(H)$ and $h_n(G)>\pi(H)+a$ then $G$ contains at least $b{n\choose v(H)}$ copies of $H$.
[**Proof:**]{} Let $R:=R(H)$ and $r:=|R|$. Since $\displaystyle \pi(H)=\lim_{n\to\infty}\pi_n(H)$, there is an $n_0$ so that for $m\geq n_0$, $\pi_m(H)\leq \pi(H)+\frac{a}{2}$. For any $n_0\leq m\leq n$, there must be at least $\frac{a}{2r}{n\choose m}$ $m$-sets $M\subset V(G)$ inducing a $R$-graph $G[M]$ with $h(G[M])>\pi(H)+\frac{a}{2}$. Otherwise, we have $$\sum_{M}h_m(G[M])\leq \left(\pi(H)+\frac{a}{2}\right){n\choose m} + \frac{a}{2r}{n\choose m}r=(\pi(H)+a){n\choose m}.$$ But we also have $$\begin{aligned}
\sum_Mh_m(G[M]) &=\sum_M \sum_{\stackrel{F\in E(G)}{F\subseteq M}}\frac{1}{{m\choose |F|}} \\
&=\sum_{F\in E(G)}\sum_{M\supseteq F}\frac{1}{{m\choose |F|}} \\
&= \sum_{F\in E(G)} \frac{{n-|F| \choose m-|F|}}{{m\choose |F|}} \\
&=\sum_{F\in E(G)} \frac{{n\choose m}}{{n\choose |F|}}\\
&= {n\choose m}h_n(G).\end{aligned}$$ This is a contradiction to the assumption that $h_n(G)>\pi(H)+a$. By the choice of $m$, each of these $m$-sets contains a copy of $H$, so the number of copies of $H$ in $G$ is at least $\frac{a}{2r}{n\choose m}/{{n-v(H)\choose m-v(H)}}=b {n\choose v(H)}$ where $b:=\frac{a}{2r}{m\choose v(H)}^{-1}$. $\square$
Supersaturation can be used to show that “blowing up” does not change the Turán density $\pi(H)$ just like in the uniform cases.
For any hypergraph $H_n$ and positive integers $s_1,s_2,\ldots, s_n$, the [*blowup*]{} of $H$ is a new hypergraph $(V,E)$, denoted by $H_n(s_1,s_2,\ldots, s_n)$, satisfying
1. $V:=\sqcup_{i=1}^n V_i$, where $|V_i|=s_i$.
2. $E=\cup_{F\in {{\rm E}}(H)} \prod_{i\in F} V_i$.
When $s_1=s_2=\cdots=s_n=s$, we simply write it as $H(s)$.
Consider the following simple example. Take $H$ to be the hypergraph with vertex set $[3]$ and edge set $E=\{\{1,2\}, \{1,2,3\}\}$, a chain. Consider the blow-ups $H(2,1,1)$ and $H(1,1,2)$ illustrated below.
(0,1)–(2,0)–(4,0)–cycle; at (0,1) \[vertex\] (v1) \[label=above:[1]{}\] ; at (2,0) \[vertex\] (v2) \[label=below:[2]{}\] ; at (4,0) \[vertex\] (v3) \[label=right:[3]{}\] ; (v1)–(v2); at (2,-2) (empty) [$H$]{};
(0,1)–(2,0)–(4,0)–cycle; (0,-1)–(2,0)–(4,0)–cycle; at (0,1) \[vertex\] (v11) \[label=above:[$v_{1,1}$]{}\] ; at (0,-1) \[vertex\] (v12) \[label=below:[$v_{1,2}$]{}\] ; at (2,0) \[vertex\] (v2) \[label=left:[$v_2$]{}\] ; at (4,0) \[vertex\] (v3) \[label=right:[$v_3$]{}\] ; (v11)–(v2); (v12)–(v2); at (2,-2) [$H(2,1,1)$]{};
(0,0)–(2,0)–(4,1)–cycle; (0,0)–(2,0)–(4,-1)–cycle; at (0,0) \[vertex\] (v1) \[label=above:[$v_{1}$]{}\] ; at (2,0) \[vertex\] (v2) \[label=right:[$v_{2}$]{}\] ; at (4,1) \[vertex\] (v31) \[label=above:[$v_{3,1}$]{}\] ; at (4,-1) \[vertex\] (v32) \[label=below:[$v_{3,2}$]{}\] ; (v1)–(v2); at (2,-2) [$H(1,1,2)$]{};
In the blow-up $H(2,1,1)$ vertex 1 splits into vertices $v_{1,1}$ and $v_{1,2}$; vertex 2 becomes $v_2$ and vertex 3 becomes $v_3$. In the blow-up $H(1,1,2)$ vertex 3 splits into vertices $v_{3,1}$ and $v_{3,2}$; vertex 1 becomes $v_1$ and vertex 2 becomes $v_2$.
[**(Blowing up)**]{} \[blowup\] Let $H$ be any hypergraph and let $s\geq 2$. Then $\pi(H(s))=\pi(H)$.
[**Proof:**]{} Let $R:=R(H)$. By the supersaturation lemma, for any $a>0$ there is a $b>0$ and an $n_0$ so that any $R$-graph $G$ on $n\geq n_0$ vertices with $h_n(G)>\pi(H)+a$ contains at least $b{n \choose v(H)}$ copies of $H$. Consider an auxiliary $v(H)$-uniform hypergraph $U$ on the same vertex set as $G$ where edges of $U$ correspond to copies of $H$ in $G$. For any $S>0$, there is a copy of $K=K^{v(H)}_{v(H)}(S)$ in $U$. This follows from the fact that $\pi(K^{v(H)}_{v(H)}(S))=0$ since it is $v(H)$-partite, and $h_{n}(U)=b>0$. Now color each edge of $K$ by one of $v(H)!$ colors, each color corresponding to one of $v(H)!$ possible orders the vertices of $H$ are mapped to the parts of $K$. The pigeon-hole principle gives us that one of the color classes contains at least $S^{v(H)}/v(H)!$ edges. For large enough $S$ there is a monochromatic copy of $K^{v(H)}_{v(H)}(s)$, which gives a copy of $H(s)$ in $G$. $\square$
[**(Squeeze Theorem)**]{} Let $H$ be any hypergraph. If there exists a hypergraph $H^{\prime}$ and integer $s\geq 2$ such that $H^{\prime}\subseteq H\subseteq H^{\prime}(s)$ then $\pi(H)=\pi(H^{\prime})$.
[**Proof:**]{} One needs only observe that for any hypergraphs $H_{1}\subseteq H_{2}\subseteq H_{3}$ that $\pi(H_{1})\leq \pi(H_{2})\leq \pi(H_{3})$. If $H_{3}=H_{1}(s)$ for some $s\geq 2$ then $\pi(H_{1})=\pi(H_{3})$ by the previous theorem. $\square$
Turán Densities of $\{1,2\}$-hypergraphs
========================================
In this section we will determine the Turán density for any hypergraph $H$ with $R(H)=\{1,2\}$. We begin with the following more general result.
Let $H=H^{1}\cup H^{k}$ be a hypergraph with $R(H)=\{1,k\}$ and $E(H^{1})=V(H^{k})$. Then $$\pi(H) = \begin{cases} 1+\pi(H^{k}) & \text{if } \pi(H^{k})\geq 1-\frac{1}{k}; \\
1+\left(\frac{1}{k(1-\pi(H^{k}))}\right)^{1/(k-1)}\left(1-\frac{1}{k}\right) & \text{otherwise.} \end{cases}$$
[**Proof:**]{} For each $n\in \mathbb{N}$, let $G_{n}$ be any $H$-free graph $n$ vertices with $h_{n}(G_{n})=\pi_{n}(H)$. Partition the vertices of $G_{n}$ into $X_{n}=\{v\in V(G_{n}):\{v\}\in E(G)\}$ and $\bar{X_{n}}$ containing everything else. Say that $|X_{n}|=x_{n}n$ and $|\bar{X_{n}}|=(1-x_{n})n$ for some $x_{n}\in [0,1]$. Since $(x_{n})$ is a sequence in $[0,1]$ it has a convergent subsequence. Consider $(x_{n})$ to be the convergent subsequence, and say that $x_{n}\to x\in [0,1]$. With the benefit of hindsight, we know that $x>0$, however, for the upper bound portion of this proof we will not assume this knowledge.
Since there is no copy of $H$ in $G_{n}$, it follows that $G_{n}[X_{n}]$ contains no copy of $H^{k}$. We have that $$\begin{aligned}
\pi(H) &=\lim_{n\to\infty} h_{n}(G_{n}) \\
&=\lim_{n\to\infty} \sum_{F\in H^{1}}\frac{1}{\binom{n}{1}} + \sum_{F\in H^{k}} \frac{1}{\binom{n}{k}} \\
&\leq \lim_{n\to\infty} \frac{x_{n}n}{\binom{n}{1}} + \frac{\binom{n}{k}-(1-\pi_{x_{n}n}(H^{k}))\binom{x_{n}n}{k}}{\binom{n}{k}} \\
&=\lim_{n\to\infty} 1 + x_{n} - (1-\pi_{x_{n}n}(H^{k}))\frac{\binom{x_{n}n}{k}}{\binom{n}{k}} \\
&\leq \lim_{n\to\infty} \begin{cases} 1+\frac{1}{\sqrt{n}} & \text{if } x_{n}n \leq \sqrt{n}, \\
1+ x_{n} - (1-\pi_{x_{n}n}(H^{k}))x_{n}^{k} & \text{if } x_{n}n>\sqrt{n}, \end{cases} \\
&\leq \max\{1, 1+x-(1-\pi(H^{k}))x^{k}\}.\end{aligned}$$ Let $f(x)=1+x-(1-\pi(H^{k}))x^{k}$ and then note that $$\pi(H)= \lim_{n\to\infty} h_{n}(G) \leq \max_{x\in [0,1]} f(x).$$ An easy calculus exercise shows that $f^{\prime\prime}(x)<0$ for all $x>0$, and $f^{\prime}(x)=0$ when $x=\left(\frac{1}{k(1-\pi(H^{k}))}\right)^{\frac{1}{k-1}}.$ If $\frac{1}{k(1-\pi(H^{k}))}\geq 1$ then $f^{\prime}(x)>0$ when $x\in [0,1)$ and hence $f(x)$ is maximized when $x=1$. Note that $f(1)=1+\pi(H^{k})$. If, on the other hand, $\frac{1}{k(1-\pi(H^{k}))}< 1$ it follows that $f(x)$ is maximized at $x=\left(\frac{1}{k(1-\pi(H^{k}))}\right)^{1/(k-1)}$. Together, this gives us $$\pi(H) \leq \begin{cases} 1+\pi(H^{k}) & \text{if } \pi(H^{k})\geq 1-\frac{1}{k}; \\
1+\left(\frac{1}{k(1-\pi(H^{k}))}\right)^{1/(k-1)}\left(1-\frac{1}{k}\right) & \text{otherwise}. \end{cases}$$ To get equality, take $x$ that maximizes $f(x)$ as above. For any $n\in \mathbb{N}$ (thinking of $n\to \infty$) partition $[n]$ into two sets $X$ and $\bar{X}$ with $|X|=xn$ and $|\bar{X}|=(1-x)n$. Let $E(G^{1})=\{\{v\}:v\in X\}$ and let $g^{k}$ be a $k$-uniform graph on $xn$ vertices attaining $|E(g^{k})|=\text{ex}(xn,H^{k})$ and $g^{k}$ is $H^{k}$-free. Then $$E(G^{k})=\{F\in \binom{[n]}{k}:\text{either } F\in E(g^{k}) \text{ or } F\cap \bar{X}\neq \emptyset\}.$$ Then $G=G^{1}\cup G^{k}$ is $H$-free and (by choice of $x$) we have that $\displaystyle \lim_{n\to\infty}h_{n}(G)$ attains the upper bound of $\pi(H)$. $\square$
Let us now return to the task of determining $\pi(H)$ when $H=H^{1}\cup H^{2}$.
Let $H=H^{1}\cup H^{2}$. If $H^{2}$ is not bipartite, then $$\pi(H)=1+\pi(H^{2})= 1+\left(1-\frac{1}{\chi(H^{2})-1}\right)=2-\frac{1}{\chi(H^{2})-1}.$$
[**Proof:**]{} First, $\pi(H)\geq 1+\pi(H^{2})$ since one can construct an $H$-free graph $G_{n}$ by letting $$E(G_{n})=\{\{v\}:v\in V(G_{n})\}\cup E(G^{\prime}_{n})$$ where $G^{\prime}_{n}$ attains $h_{n}(G^{\prime}_{n})=\pi_{n}(H^{2})$ and $G^{\prime}_{n}$ is $H^{2}$-free. Then $$\pi(H)\geq \lim_{n\to\infty} h_{n}(G_{n}) = \lim_{n\to\infty} 1+\pi_{n}(H^{2}) = 1+\pi(H^{2}).$$
To get the upper-bound, first add every missing $1$-edge into $H$, call the new graph $H^{\prime}$. Note that $\pi(H)\leq \pi(H^{\prime})$. Note that we didn’t change the edge set $H^{2}$. The Erdős-Stone-Simonovits theorem states that if $H^{2}$ is not bipartite, then $\pi(H^{2})=1-\frac{1}{\chi(H^{2})-1}$. Also, if $H^{2}$ is not bipartite, then $\chi(H^{2})\geq 3$. With the added vertices, taking $k=2$, we apply the previous theorem. Since $$\pi(H^{2})=1-\frac{1}{\chi(H^{2})-1}\geq 1-\frac{1}{2}$$ we may conclude that $\pi(H)\leq \pi(H^{\prime})=1+\pi(H^{2})$. $\square$
It remains to investigate the cases when $H^{2}$ is bipartite.
Let $H=H^{1}\cup H^{2}$. If $H^{2}$ is bipartite and $K_{2}^{\{1,2\}}\subseteq H$ then $\pi(H)=\frac{5}{4}$.
[**Proof:**]{} First, in example 1, we computed $\pi(K_{2}^{\{1,2\}})=\frac{5}{4}$. Second, $H$ must be contained in some blow-up of $K_{2}^{\{1,2\}}$ since $H^{2}$ is bipartite, i.e. there exists some $s>2$ such that $H\subseteq K_{2}^{\{1,2\}}(s)$. So, by the squeeze theorem we have $$\frac{5}{4}=\pi(K_{2}^{\{1,2\}}) \leq \pi(H) \leq \pi(K_{2}^{\{1,2\}}(s))=\frac{5}{4}.$$ Hence $\pi(H)=\frac{5}{4}$ as claimed. $\square$
We will say that $H=H^{1}\cup H^{2}$ is a ***closed path*** (from $x_{1}$ to $x_{k}$) of length $k$ if $V(H)=\{x_{1}, x_{2},...,x_{k}\}$ and $E(H^{1})=\{\{x_{1}\}, \{x_{k}\}\}$ and $E(H^{2})=\{ \{x_{i}, x_{i+1}\}:1\leq i\leq k-1\}$. We will denote a closed path of length $k$, or a closed $k$-path, by $\bar{P}_{k}$.
Pictorially, we view a closed path of length $k$ as follows:
at (0,0) \[vertex\_closed\] (v1) \[label=below:[$x_{1}$]{}\] ; at (1,0) \[vertex\_open\] (v2) \[label=below:[$x_{2}$]{}\] ; at (2,0) \[vertex\_open\] (v3) \[label=below:[$x_{3}$]{}\] ; at (3,0) \[vertex\_open\] (v4) \[label=below:[$x_{k-2}$]{}\] ; at (4,0) \[vertex\_open\] (v5) \[label=below:[$x_{k-1}$]{}\] ; at (5,0) \[vertex\_closed\] (v6) \[label=below:[$x_{k}$]{}\] ; (v1)–(v2); (v2)–(v3); (v4)–(v5); (v5)–(v6); (v3)–(2.3, 0); (2.7, 0)–(v4); at (2.5, 0) [$\dots$]{};
Let $H=H^{1}\cup H^{2}$. If $H^{2}$ is bipartite and $H$ does not contain a copy of $K_{2}^{\{1,2\}}$ and $H$ contains a closed path of length $2k$, then $\pi(H)=\frac{9}{8}$.
[**Proof:**]{} First, we will give a construction giving us the lower bound. For any $n\in \mathbb{N}$ let $G_{n}$ have vertex set $[n]$. Partition the vertices of $G_{n}$ into two sets $X$ and $\bar{X}$ where $|X|=\frac{3n}{4}$ and $|\bar{X}|=\frac{n}{4}$. Let $$E(G)=\{\{x\}:x\in X\} \cup \{\{x,\bar{x}\}: x\in X \text{ and } \bar{x}\in \bar{X}\}.$$ It is clear that $G_{n}$ contains no closed paths of length $2k$ when $k\geq 1$. Also, $$\begin{aligned}
\lim_{n\to\infty} h_{n}(G_{n}) &= \lim_{n\to\infty} \frac{|X|}{\binom{n}{1}}+\frac{|X|\cdot |\bar{X}|}{\binom{n}{2}} \\
&= \lim_{n\to\infty} \frac{3}{4}+\frac{\frac{3}{16} n^{2}}{\binom{n}{2}} \\
&= \frac{3}{4}+\frac{3}{8} =\frac{9}{8}.\end{aligned}$$ Thus $\pi(H)\geq \frac{9}{8}$ for any $H$ containing a closed path of length $2k$ for any $k\geq 1$.
Since $H^{2}$ is bipartite, and $H^{2}$ does not contain a copy of $K_{2}^{\{1,2\}}$, then $H$ is contained in a blow-up of a closed $4$-path. To see this, note that there is a bipartition of the vertices of $H$, $V(H)=A\cup B$, (with respect to the $2$-edges in $H$). Furthermore, we can partition $A$ into $A_{1}\cup A_{2}$ where $v\in A_{1}$ if $\{v\}\in E(H)$ and $v\in A$, $v\in A_{2}$ if $v\in A\setminus A_{1}$. And similarly partition $B$ into $B_{1}\cup B_{2}$ with $v\in B_{1}$ if $\{v\}\in E(H)$ and $v\in B$. Then note that there are no edges from $A_{1}$ to $B_{1}$ since $H$ contains no copy of $K_{2}^{\{1,2\}}$. So $H\subset \bar{P}_{4}(\max\{|A_{1}|, |A_{2}|, |B_{1}|, |B_{2}|)$–a blow-up of $\bar{P}_{4}$. Below is a graphical representation of $H$, illustrating that $H$ is contained in a blow-up of $\bar{P}_{4}$.
(0, 0) circle (1 cm); (0,-3) circle (1 cm); (3, 0) circle (1 cm); (3,-3) circle (1 cm); at (-1.5, 0) [$A_{1}$]{}; at (-1.5,-3) [$A_{2}$]{}; at (4.5, 0) [$B_{2}$]{}; at (4.5, -3) [$B_{1}$]{};
at (0, .5) \[vertex\_closed\] (v1) ; at (-.5, -.5) \[vertex\_closed\] (v2) ; at (.35,-.35) \[vertex\_closed\] (v3) ;
at (3, .5) \[vertex\_open\] (v4) ; at (2.65,-.35) \[vertex\_open\] (v5) ; at (3.5, -.5) \[vertex\_open\] (v6) ;
at (0, -3.5) \[vertex\_open\] (v7) ; at (-.5, -2.65) \[vertex\_open\] (v8) ;
at (3, -3.5) \[vertex\_closed\] (v9) ; at (3.5, -2.65) \[vertex\_closed\] (v10) ;
(.6,-.2)–(2.75, 0); (.35, .2)–(2.5, .5); (3,-.4)–(0,-2.65); (3.15, -.5)–(.35, -3); (.5, -3.5)–(2.5, -3.25); (.65,-3.25)–(2.7, -2.8);
Since $\pi(H)\leq \pi(\bar{P}_{4}(s))=\pi(\bar{P}_{4})$ we need only show that $\pi(\bar{P}_{4})\leq \frac{9}{8}$. Let $G_{n}$ be a family of $\bar{P}_{4}$-free graphs such that $h_{n}(G_{n})=\pi_{n}(\bar{P}_{4})$. Partition the vertices of $G_{n}$ as follows: $$\begin{aligned}
X_{n} &= \{v:\{v\}\in E(G_{n})\}, \\
Y_{n} &= \{v: \{v\}\notin E(G_{n}) \text{ and } \exists x_{1}\neq x_{2}\in X_{n} \text{ with } \{x_{1}, v\}, \{x_{2}, v\} \in E(G_{n})\}, \\
Z_{n} &= V(G)\setminus (X_{n}\cup Y_{n}).\end{aligned}$$ Let us say that $|X_{n}|=xn$, $|Y_{n}|=yn$ and hence $|Z_{n}|=(1-x-y)n$.
First, note that $E(G)\cap \binom{Y_{n}}{2}=\emptyset$. Otherwise, since each vertex in $Y_{n}$ has at least 2 neighbors in $X_{n}$, $G_{n}$ would contain a closed path of length $4$. Also, each vertex in $Z_{n}$ has at most 1 neighbor in $X_{n}$. It follows that $$\begin{aligned}
\pi(\bar{P}_{4}) &=\lim_{n\to\infty} \pi_{n}(\bar{P}_{4}) \\
&= \lim_{n\to \infty} h_{n}(G_{n}) \\
&\leq \lim_{n\to\infty} \frac{|X_{n}|}{\binom{n}{1}} + \frac{|X_{n}|\cdot |Y_{n}|}{\binom{n}{2}} + \frac{|Y_{n}|\cdot |Z_{n}|}{\binom{n}{2}} + \frac{\binom{|Z_{n}|}{2}}{\binom{n}{2}} + \frac{|Z_{n}|}{\binom{n}{2}} \\
&\leq \lim_{n\to\infty} \frac{xn}{\binom{n}{1}} + \frac{xyn^{2}}{\binom{n}{2}} + \frac{y(1-x-y)n^{2}}{\binom{n}{2}} + \frac{\frac{(1-x-y)^{2}n^{2}}{2}}{\binom{n}{2}} + \frac{(1-x-y)n}{\binom{n}{2}} \\
&\leq \max_{\substack {0\leq x\leq 1 \\ 0\leq y\leq 1-x}} x + 2xy + 2y(1-x-y)+ (1-x-y)^{2} \\
&=\frac{9}{8}.\end{aligned}$$ The last inequality is an easy multivariate calculus exercise. One can also verify it with software, such as *Mathematica*, the syntax being:
Maximize[{x^2-x-y^2+2*x*y+1, 0<=x<=1, 0<=y<=1-x}, {x,y}].
It may be of interest to note that the maximum value of the function is obtained when $x=\frac{3}{4}$ and $y=\frac{1}{4}$. In this case $Z_{n}$ is empty. Since our upper bound matches the lower bound, we have the desired result. $\square$
Let $H=H^{1}\cup H^{2}$. If $H^{2}$ is bipartite and $H^{2}$ does not contain a closed $2k$-path for any $k\geq 1$, then $\pi(H)=1$.
[**Proof:**]{} First, since $|R(H)|=2$ we have, trivially, that $\pi(H)\geq 1$. Since $H$ contains no path of length $2k$ for any $k\geq 1$ it must be the case that $H$ is contained in a blow-up of a chain $C^{\{1,2\}}=\{\{x\}, \{x,y\}\}$. This is most clearly seen by again, considering the previous illustration. The difference is, in this case, $B_{1}$ (or $A_{1}$) is empty.
(0, 0) circle (1 cm); (0,-3) circle (1 cm); (3, 0) circle (1 cm); at (-1.5, 0) [$A_{1}$]{}; at (-1.5,-3) [$A_{2}$]{}; at (4.5, 0) [$B_{2}$]{};
at (0, .5) \[vertex\_closed\] (v1) ; at (-.5, -.5) \[vertex\_closed\] (v2) ; at (.35,-.35) \[vertex\_closed\] (v3) ;
at (3, .5) \[vertex\_open\] (v4) ; at (2.65,-.35) \[vertex\_open\] (v5) ; at (3.5, -.5) \[vertex\_open\] (v6) ;
at (0, -3.5) \[vertex\_open\] (v7) ; at (-.5, -2.65) \[vertex\_open\] (v8) ;
(.6,-.2)–(2.75, 0); (.35, .2)–(2.5, .5); (3,-.4)–(0,-2.65); (3.15, -.5)–(.35, -3);
at (1.5, -4) [$H$]{};
at (7,0) \[vertex\_closed\] (w1) ; at (8,0) \[vertex\_open\] (w2) ; at (7,-1) \[vertex\_open\] (w3) ; (w1)–(w2)–(w3); at (7.5, -1.5) [$K$]{};
It is clear that $H$ is contained in a blow-up of $K$ where $$K=\{\{x\}, \{x,y\}, \{y,z\}\}\subseteq C^{\{1,2\}}(2,1)=\{\{x\}, \{z\}, \{x, y\}, \{z, y\}\}.$$ It follows that $\pi(H)\leq \pi(C^{\{1,2\}})=1$. $\square$
The combination of these propositions completely determines $\pi(H)$ when $R(H)=\{1,2\}$. The results are summarized by the following theorem.
\[t12\] For any hypergraph $H$ with $R(H)=\{1,2\}$, we have $$\pi(H) = \begin{cases} 2-\frac{1}{\chi(H^{2})-1} & \text{if } H^{2} \text{ is not bipartite}; \\
\frac{5}{4} & \text{if } H^{2} \text{ is bipartite and } \min \{k:\bar{P}_{2k}\subseteq H\}=1; \\
\frac{9}{8} & \text{if } H^{2} \text{ is bipartite and } \min \{k:\bar{P}_{2k}\subseteq H\}\geq 2; \\
1 & \text{if } H^{2} \text{ is bipartite and } \bar{P}_{2k}\nsubseteq H \text{ for any }k\geq 1. \end{cases}$$
Degenerate hypergraphs
======================
Recall that a hypergraph $H$ is [*degenerate*]{} if $\pi(H)=|R(H)|-1$. For $k$-uniform hypergraph $H$, $H$ is degenerate if and only $H$ is $k$-partite. From Proposition \[p1\] and Theorem \[blowup\], we have the following proposition.
Suppose $H$ is a degenerate hypergraph. Then the following properties hold.
- Every subgraph of $H$ is degenerate.
- Every blowup of $H$ is degenerate.
- Any subgraph of the blowup of a flag is degenerate.
Note that every flag is a subgraph of some blowup of a chain with the same edge type. Is every degenerate hypergraph a subgraph of some blowup of a chain? The answer is yes for uniform hypergraphs and $\{1,2\}$-hypergraphs. This follows from Theorem \[t12\], which completely determined $\pi(H)$ when $R(H)=\{1,2\}$, and from the fact that a $k$-uniform hypergraph is degenerate if and only if it is $k$-partite (a subgraph of a blowup of a single edge). However, the answer in general is false. We will show that the following hypergraph $H_1$ with edge set $E(H_1)=\{\{1,2\}, \{1,3\}, \{2,3,4,\}\}$ is degenerate.
(2,1)–(4,0)–(2,-1)–cycle; at (0,0) \[wvertex\] (v1) \[label=above:[1]{}\] ; at (2,1) \[wvertex\] (v2) \[label=above:[2]{}\] ; at (2,-1) \[wvertex\] (v3) \[label=below:[3]{}\] ; at (4,0) \[wvertex\] (v4) \[label=above:[4]{}\] ; (v1)–(v2); (v1)–(v3);
\
$H_{1}$: A degenerate hypergraph not contained in the blowup of a chain.
This result is a special case of the following theorem.
Let $H$ be a hypergraph containing some $2$-edges. The $2$-subdivision of $H$ is a new hypergraph $H'$ obtained from $H$ by subdividing each $2$-edge simultaneously. Namely, if $H$ contains $t$ $2$-edges, add $t$ new vertices $x_1,x_2,\ldots,x_t$ to $H$ and for $i=1,2,\ldots, t$ replace the $2$-edge $\{u_{i},v_{i}\}$ with $\{u_i,x_i\}$ and $\{x_i,v_i\}$.
\[t:sd\] Let $H'$ be the $2$-subdivision of $H$. If $H$ is degenerate, then so is $H'$.
For example, $H_1$ can be viewed as the $2$-division of the chain $C^{\{2,3\}}$. Since any chain is degenerate, so is $H_1$. To prove this theorem, we need a Lemma on graphs, which has independent interest.
Let $G$ be any simple graph. Then $G^{(2)}$, a variation of the square of $G$, will be defined as follows:
- $V(G^{(2)}):=V(G)$,
- $E(G^{(2)}):=\{ \{u,v\}| \exists w\in V(G) \text{ with } \{u,w\},\{v,w\}\in E(G)\}$.
Note that an edge of $G$ may or may not be an edge of $G^{(2)}$. For example, if $G$ is the complete graph, then $G^{(2)}$ is also the complete graph. However, if $G$ is a complete bipartite graph with partite set $V_1\cup V_2$, then $G^{(2)}$ is the disjoint union of two complete graphs on $V_1$ and $V_2$. In this case, $G^{(2)}$ is the complement graph of $G$! We also note that $G^{(2)}$ is the empty graph if $G$ is a matching. Surprisingly, we have the following Lemma on the difference of the number of edges in $G$ and $G^{(2)}$.
\[l:square\] For any simple graph $G$ on $n$ vertices, $$\label{eq:5.1}
|E(G)|-|E(G^{(2)})| \leq \left\lfloor
\frac{n}{2}\right\rfloor.$$ Furthermore, equality holds if and only if $G$ is the vertex-disjoint union of complete bipartite graphs of balanced part-size with at most one component having odd number of vertices, i.e. $$G= K_{t_1,t_1}\cup K_{t_2,t_2} \cup \cdots \cup K_{t_k,t_k} \cup K_{\lfloor \frac{n}{2}\rfloor-\sum_{i=1}^kt_i, \lceil \frac{n}{2}\rceil-\sum_{i=1}^kt_i},$$ for some positive integers $t_1,t_2,\ldots, t_k$ satisfying $\sum_{i=1}^kt_i= \lfloor \frac{n}{2}\rfloor$.
First, we will show that Equation holds for any forest. Let $G$ be a forest. Since $G$ is a forest, if $\{a,b\}\in E(G^{(2)})$ then $a$ and $b$ have a unique common neighbor in $G$. Furthermore, given any vertex $c\in V(G)$, it follows that any pair of neighbors of $c$ is in $E(G^{(2)})$. Thus we have $$\begin{aligned}
|E(G)|-|E(G^{(2)})| &= \frac{1}{2}\sum_{v\in V(G)} \deg(v) - \sum_{v\in
V(G)} \binom{\deg(v)}{2} \\
&=\sum_{v\in V(G)} \frac{1}{2}\deg(v)-\binom{\deg(v)}{2} \\
&=\sum_{v\in V(G)}-\frac{1}{2}\deg(v)^{2} + \deg(v) \\
&\leq \sum_{v\in V(G)} \frac{1}{2} \\
&=\frac{n}{2}.\end{aligned}$$ The inequality above comes from the fact that $-\frac{1}{2}x^{2}+x \leq \frac{1}{2}$, attaining its maximum when $x=1$. Since $|E(G)|-|E(G^{(2)})|$ is an integer, we have that $$\label{eq:5.2}
|E(G)|-|E(G^{(2)})|\leq \left\lfloor \frac{n}{2}\right\rfloor$$ as claimed.
Now we will prove the statement $|E(G)|-|E(G^{(2)})|\leq\left\lfloor \frac{|V(G)|}{2}\right\rfloor$ for general graphs using induction on the number of vertices. It holds trivially for $n=1,2$.
Assume that the statement holds for all graphs with at fewer than $n$ vertices. Consider a graph $G$ with $n$ vertices. If $G$ is a forest, then the statement holds. Otherwise, $G$ contains a cycle. Choose $C_g$ to be a minimal cycle in $G$, i.e. one with no chords. If $G=C_g$, then ${{\rm E}}(C_g)-{{\rm E}}(C_g^{(2)})=0$ if $g\not=4$ or $2$ if $g=4$. The statement holds.
Now assume $V(C)\subsetneq V(G)$. Let $V_1:=V(C)=\{x_1,x_2,...,x_{g}\}$, where $x_{i}$ is adjacent to $x_{i+1}$, and let $V_2:=V(G)\setminus V_1=\{v_1,v_2,...,v_{n-g}\}$.
The edges of $G$ can be partitioned into three parts: the induced graph $G[V_1]=C_g$, the induced graph $G[V_2]$, and the bipartite graph $G[V_1,V_2]$. Similarly, the edges of $G^{(2)}$ can be partitioned into three parts: $G^{(2)}[V_1]$, the induced graph $G^{(2)}[V_2]$, and the bipartite graph $G^{(2)}[V_1,V_2]$. Now we compare term by term.
1. Note $|E(G^{(2)}[V_1])|\geq |E(C_g^{(2)})|$, and $|E(C_g^{(2)})|=g$ if $g\neq 4$ or $2$. We have $$\label{eq:5.3}
|E(G[V_1])|-|E(G^{(2)}[V_1])|\leq |E(C_g)|-|E(C_g^{(2)})|\leq \left\lfloor \frac{g}{2}\right\rfloor.$$
2. By inductive hypothesis, we have $|E(G[V_2])|-|E((G[V_2])^{(2)})|\leq \left\lfloor \frac{n-g}{2}\right\rfloor$. Combining with the fact $|E(G^{(2)}[V_2])\geq |E((G[V_2])^{(2)})|$, we have $$\label{eq:5.4}
|E(G[V_2]))|-|E(G^{(2)}[V_2])|\leq \left\lfloor \frac{n-g}{2}\right\rfloor.$$
3. We claim $|E(G[V_1,V_2])| \leq |E(G^{(2)}[V_1,V_2])|$. We define a map $$f\colon E(G[V_1,V_2]) \to E(G^{(2)}[V_1,V_2])$$ as follows. For any edge $x_iv\in E(G)$ with $v\in V_2$ and $x_i\in V_1$, define $f(vx_i)=vx_{i+1}$ (with the convention $x_{g+1}=x_1$). Since $x_iv\in E(G)$ and $x_ix_{i+1}\in E(G)$, we have $vx_{i+1}\in E(G^{(2)})$. The map $f$ is well-defined. We also observe that $f$ is an injective map. Thus $$\label{eq:5.5}
|E(G[V_1,V_2])| \leq |E(G^{(2)}[V_1,V_2])|.$$
Combining equations , , and , we get $$|E(G)|-|E(G^{(2)})|\leq \left\lfloor\frac{n-g}{2}\right\rfloor + \left\lfloor\frac{g}{2}\right\rfloor \leq \left\lfloor\frac{n}{2}\right\rfloor.$$
The inductive step is finished.
Now we check when equality holds. It is straightforward to verify the sufficient condition; we omit the computation here.
Now we prove the necessary condition. Assume that $G$ has $k+1$ connected components $G_1, G_2,\ldots, G_{k+1}$. Then we have $$|E(G)|-|E(G^{(2)})|\leq \sum_{i=1}^{k+1} (|E(G_i)|-|E(G^{(2)}_i)|) \leq \sum_{i=1}^{k+1} \left\lfloor \frac{|V(G_i)|}{2} \right\rfloor \leq \left\lfloor \frac{n}{2}\right\rfloor.$$ If equality holds, then all but possibly one component has an even number of vertices. It remains to show each component is a balanced complete bipartite graph.
Without loss of generality, we assume $G$ is connected. If $G$ is a tree, then equality in Equation either forces the degree of every vertex to be $1$, or all the degrees are $1$ with a single exceptional vertex of degree $2$. Since $G$ is assumed to be connected, $G$ is either $P_2=K_{1,1}$ or $P_3=K_{1,2}$.
Suppose that $G$ contains cycles, and the equalities hold in Equations , , and . First we show that $C_4$ is the only possible chordless cycle in $G$. Suppose not; let $C_g$ ($g\not=4$) be a cordless cycle. We have $|E(C_g)|-|E(C^{(2)}_g)|=0$; which contradicts the assumption that equality holds in Equation . Thus $G$ is a bipartite graph. Furthermore,the equality in forces each vertex $v$ to be connected to at least $2$ vertices of $C_4$. Hence $G$ is 2-connected. Now $G$ must be a complete bipartite graph. Otherwise, say $uv$ is a nonedge crossing the partite sets. Since $G$ is $2$-connected, there exists a cycle containing both $u$ and $v$. Let $C$ be such a cycle with minimum length; $C$ is cordless but not a $C_4$. Contradiction. Finally we show $G=K_{st}$ is balanced. Note that $$|E(G)|-|E(G^{(2)})|=st-{s\choose 2}-{t\choose 2}=\frac{n}{2}-\frac{(s-t)^2}{2}\leq \left\lfloor \frac{n}{2}\right\rfloor.$$ The equality holds only if $|s-t|\leq 1$. So $G$ is balanced.
[**Proof of Theorem \[t:sd\]:**]{} We will prove by contradiction. Let $R:=R(H)=R(H')$ be the common set of edge types of $H$ and $H'$. Suppose that $H'$ is not degenerate, then $\pi(H')>|R|-1 +\epsilon$ for some $\epsilon>0$. Thus, there exists an $n_0$ satisfying $\pi_n(H')> |R|-1 +\epsilon/2$ for any $n\geq n_0$. Let $G_n^R$ be a $H'$-free hypergraph with $\pi_n(G)> |R|-1 +\epsilon/2$. Define a new hypergraph $G'_n$ over the same vertex set of $G$ with a new edge set $E(G'_n)=E(G_n)\setminus E(G_n^{2}) \cup E((G_n^{2})^{(2)})$. The hypergraph $G'_n$ is obtained for $G_n$ by replacing all $2$-edges by the edges in its square graph while keeping other type of edges. By Lemma \[l:square\], we have $$\label{eq:6}
\pi_n(G'_n)\geq \pi_n(G) - \frac{\lfloor
\frac{n}{2}\rfloor}{{n\choose 2}}
\geq |R|-1 +
\epsilon/2 -\frac{1}{n}.$$
Suppose that $H$ has $t$ $2$-edges. Since $H$ is degenerate, so is the blowup hypergraph $H(t+1)$. For sufficiently large $n$, $G'_n$ contains a subhypergraph $H(t+1)$. By the definition of $G'$, for every copy of $H\subseteq H(t+1)$ and every $2$-edge $u_iv_i$ (for $1\leq i\leq t$) of $H$, there exists a vertex $x_i:=x_i(u_i,v_i)$ satisfying $u_ix_i$ and $v_ix_i$ are $2$-edges of $G$. Our goal is to force that $x_1,x_2,\ldots, x_t$ are distinct from the vertices of $H$ and from each other. This can be done by a greedy algorithm. Suppose that the vertices of $H$ are listed by $y_1,y_2,y_3,\ldots,$ and so on. Each vertex has $y_i$ has $t+1$ copies in $H(t+1)$. For $i=1,2,3,\ldots$, select a vertex $y_i'$ from the $t+1$ copies of $y_i$ so that $y_i'$ is not the same vertex as $x_j(u_j,v_j)$ for some $2$-edge $u_jv_j$ where $u_j, v_j$ have been selected. This is always possible since $H$ has only $t$ $2$-edges. Thus, we found a copy of $H'$ as a subgraph of $G$. Contradiction! $\square$
It remains an open question to classify all non-degenerate hypergraphs.
In the remainder of this section, we generalize the following theorem due to Erdős [@Erdos64] on the Turán density of complete $k$-partitite $k$-uniform hypergraphs.
[**Theorem (Erdős [@Erdos64]):**]{} [ Let $K^{(k)}_k(s_1,\ldots, s_k)$ be the complete $k$-partite $k$-uniform hypergraph with partite sets of size $s_1,\ldots,
s_k$. Then any $K^{(k)}_k (s_1, \ldots, s_k)$-free $r$-uniform hypergraph can have at most $O(n^{k-\delta})$ edges, where $\delta = \left(\prod^{k-1}_{i=1}
s_i\right)^{-1}$. ]{}
We have the following theorem.
\[tflag\] Let $L(s_1,s_2,\ldots, s_{v(L)})$ be a blowup of a flag $L^R$, we have $$\pi_n(L(s_1,s_2,\ldots, s_{v(L)}))=r-1+O(n^{-\delta}),$$ where $\delta= \frac{\max\{s_i\colon 1\leq i\leq v(L)\}}{\prod_{i=1}^{v(L)}s_i}$.
Using the concept of $H$-density, we can say a lot more about avoiding a blowup of any hypergraph $H$.
Given two hypergraphs $H$ and $G$ with the same edge-type $R(H)=R(G)$, the [*density*]{} of $H$ in $G$, denoted by $\mu_H(G)$, is defined as the probability that a random injective map $f\colon V(H)\to V(G)$ satisfies $H\stackrel{f}{\hookrightarrow} G$ (i.e. $f$ maps $H$ to an ordered copy of $H$ in $G$). We have the following theorem.
For a fixed hypergraph $H$ on $m$ vertices and $m$ positive integers $s_1,s_2,\ldots, s_m$, let $H(s_1,s_2,\ldots, s_m)$ be the blowup of $H$. For sufficiently large $n$ and any hypergraph $G$ on $n$ vertices with edge type $R(G)=R(H)$, if $G$ contains no subgraph $H(s_1,s_2,\ldots, s_m)$, then $$\mu_H(G)=O(n^{-\delta}),$$ where $\delta= \frac{\max\{s_i\colon 1\leq i\leq m\}}{\prod_{i=1}^{m}s_i}$.
[**Proof:**]{} We will prove by contradiction. We assume $\mu_H(G)\geq C n^{-\delta}$ for some constant $C$ to be chosen later. By reordering the vertices of $H$, we can assume $s_1\leq s_2\leq \cdots \leq s_m$. Without loss of generality, we assume $n$ is divisible by $m$. Consider a random $m$-partition of $V(G)=V_1\cup V_2\cup \cdots \cup V_{m}$ where each part has size $\frac{n}{m}$. For any $m$-set $S$ of $V(G)$ , we say $S$ is a [*transversal*]{} (with respect to this partition of $V(G)$) if $S$ intersects each $V_i$ exactly once. The probability that $S$ is transversal is given by $\frac{(\frac{n}{m})^{m}}{{n\choose m}m!}$.
We say, an ordered copy $f(H)$ is a [*transversal*]{} if $f(V(H))$ is a transversal. By the definition of $\mu_H(G)$, $G$ contains $\mu_H(G){n\choose m}m!$ ordered copies of $H$. Thus, the expected number of transversal ordered copies of $H$ is $$\mu_H(G)\left(\frac{n}{m}\right)^{m}\geq \frac{C}{m^m} n^{m-\delta}.$$ There exists a partition so that the number of crossing maximum chains in $E(H)$ is at least $Cm^{-\delta} n^{m-{\delta}}$. Now we fix this partition $[n]=V_1\cup \cdots \cup V_m$.
For $t_i \in \{1, s_i\}$ with $i=1,2\ldots, m$, we would like to estimate the number of monochromatic (ordered) copies in $H'$, denoted by $f(t_1, t_2, \ldots, t_{m})$, of $H(t_1,\ldots, t_{m})$ so that the first $t_1$ vertices in $V_{\tau_1}$, the second $t_2$ vertices in $V_{\tau_2}$, and so on.
[**Claim a:**]{} For $0\leq l\leq m-1$, we have $$f(s_{1},\ldots, s_{l}, 1,\ldots, 1)\geq
\left(1+o(1)\right)
\frac{\left(Cn^{-\delta}\right)^{\prod_{j=1}^l s_{j}}}
{\prod_{j=1}^l \left(s_{j}! \right)^{\prod_{u=j+1}^ls_{u}}}
\left(\frac{n}{m}\right)^{m -l+ \sum_{j=1}^ls_j}.$$ We prove claim (a) by induction on $l\in [0,m]$. For the initial case $l=0$, the claim is trivial since $f(1,1,\ldots, 1)$ counts the number of transversal ordered copies of $H$. We have $$f(1,1,\ldots,1)\geq Cn^{-\delta} \left(\frac{n}{m}\right)^{m}.$$ The statement holds for $l=0$. Now we assume claim (a) holds for $l>0$. Consider the case $l+1$, for some $l\geq 0$. For any $S\in {V_{1}\choose s_{1}}\times \cdots \times {V_{l}\choose s_{}}\times V_{l+2}\times \cdots \times V_{m}$, let $d_S$ be the number of vertices $v$ in $V_{l+1}$ such that the induced subgraph of $H'$ on $S\times \{v\}$ is $H(s_{1},\ldots, s_{l}, s_{l+1},\ldots, 1)$. We have $$\begin{aligned}
\label{eq:si}
f(s_1,\ldots, s_l,1,1, \ldots, 1) &= \sum_{S}d_S; \\
f(s_1,\ldots, s_l, s_{l+1},1,\ldots,1)&=\sum_{S} {d_S\choose s_{l+1}}.
\label{eq:sil}\end{aligned}$$
Let $\bar d_l$ be the average of $d_S$. By equation and the inductive hypothesis, we have $$\label{eq:dl}
\bar d_l\geq \frac{\sum_{S} d_S}{(\frac{n}{m})^{m -l-1+ \sum_{j=1}^ls_j}}
\geq \left(1+o(1)\right)
\frac{(\frac{n}{m})\left(Cn^{-\delta}\right)^{\prod_{j=1}^l s_j}}
{\prod_{j=1}^l \left(s_j!\right)^{\prod_{u=j+1}^ls_u}}.$$ Applying the convex inequality, we have $$\begin{aligned}
f(s_1,\ldots, s_l, s_{l+1},1,\ldots,1) &=\sum_{S} {d_S\choose s_{l+1}} \\
&\geq \left(\frac{n}{m}\right)^{m-1 +\sum_{j=1}^l(s_j-1)} {\bar d_l\choose s_{l+1}} \\
&=\left(1+O\left(\frac{1}{\bar d_l}\right)\right)\frac{\bar d_l^{s_{l+1}}}{s_{l+1}!} \left(\frac{n}{m}\right)^{m +\sum_{j=1}^{l+1}(s_j-1)}.\end{aligned}$$ Combining with equation , we get $$f(s_1,\ldots, s_l, s_{l+1},1,\ldots,1) \geq \left(1+o(1)\right) \frac{\left(Cn^{-\delta}\right)^{\prod_{j=1}^{l+1} s_j}}
{\prod_{j=1}^{l+1} \left(s_j!\right)^{\prod_{u=j+1}^{l+1}s_u}} \left(\frac{n}{m}\right)^{m -l-1+ \sum_{j=1}^{l+1}s_j}.$$ In the last step, we used the assumption $s_{m}:=\max\{s_i\colon 1\leq i\leq m\}$.
Applying claim (a) with $l=m-1$, we get $$\begin{aligned}
\nonumber
f(s_1,s_2,\ldots, s_{m-1},1) &\geq
\left(1+o(1)\right)
\frac{\left(Cn^{-\delta}\right)^{\prod_{j=1}^{m-1} s_j}}
{\prod_{j=1}^{m-1} \left(s_j!\right)^{\prod_{u=j+1}^{m-1}s_u}}
\left(\frac{n}{m}\right)^{1+ \sum_{j=1}^{m-1}s_j} \\
&=\left(1+o(1)\right) \frac{m^{-1}C^{\prod_{j=1}^{m-1} s_j} (\frac{n}{m})^{\sum_{j=1}^{m-1}s_j}}
{\prod_{j=1}^{m-1} \left(s_j! \right)^{\prod_{u=j+1}^{m-1}s_u}}.
\label{eq:lb}\end{aligned}$$
For any $S\in {V_1\choose s_1}\times \cdots \times {V_{m-1}\choose s_{m-1}}$, let $d_S$ be the number of vertices $v$ in $V_{l+1}$ such that the edges in the induced subgraph of $H'$ on $S\times \{v\}$ are monochromatic. Since $H'$ contains no monochromatic copy of $C^R(s_1,\ldots, s_{m})$, we have $d_S\leq s_{m}$. It implies $$\label{eq:ub}
f(s_1,s_2,\ldots, s_{m-1},1)=\sum_S d_S\leq s_{m} \left(\frac{n}{m}\right)^{\sum_{j=1}^{m-1}s_j}.$$
Choosing $C$ so that $C>\left((ms_{m})^{\frac{1}{\prod_{u=1}^{m}s_u}}\right)\cdot
\prod_{j=1}^{m-1} \left(s_j!\right)^{\frac{1}{\prod_{u=1}^{j}s_u}}$, equations and contradict each other. $\square$
[**Proof of Theorem \[tflag\]:**]{} Consider a hypergraph $G:=G^R_n$ with $$h_n(G)=r-1+Cn^{-\delta}.$$ Let $r=|R|$. It suffices to show that $\mu_H(G)\geq C'n^{-\delta}$.
Given a random permutation $\sigma$, let $X$ be the number of edges on a random full chain $\sigma(L)$. By the definition of the Lubell function, we have $h_n(G)={{\rm E}}(X)$. Note $X$ only takes integer values $0,1,\ldots, r$. Since ${{\rm E}}(X)>r-1$, there is non-zero probability that $X=r$. In fact, we have $$\begin{aligned}
{{\rm E}}(X)&=\sum_{i=0}^r i{{\rm Pr}}(X=i)\\
&\leq r{{\rm Pr}}(X=r) + (r-1)(1-{{\rm Pr}}(X=r))\\
&=r-1 + {{\rm Pr}}(X=r). \end{aligned}$$
Thus, we get $$\label{eq:xr}
{{\rm Pr}}(X=r)\geq \frac{C}{n^{\delta}}.$$
Every flag $\sigma(L)$ contributes an equal share of the probability of the event that $X=r$, namely, $$\label{eq:pchain}
\frac{|Aut(L)|}{{n\choose v(L)}v(L)!}.$$ Here $Aut(L)$ is the automorphism of $L$. Thus, the number of such flags is at least $$\label{eq:nchain}
\frac{C}{|Aut(L)|n^{\delta}}{n\choose v(L)} v(L)!.$$ It follows that $\mu_H(G)\geq Cn^{-\delta}$. $\square$
Suspensions
===========
The [*suspension*]{} of a hypergraph $H$, denoted $S(H)$, is the hypergraph with $V=V(H) \bigcup \{\ast\}$ where $\{\ast\}$ is an element not in $V(H)$, and edge set $E=\{F\bigcup \{\ast\}\colon F\in E(H)\}$. We write $S^{t}(H)$ to denote the hypergraph obtained by iterating the suspension operation $t$-times, i.e. $S^{2}(H)=S(S(H))$ and $S^{3}(H)=S(S(S(H)))$, etc.
In this section we will investigate the relationship between $\pi(H)$ and $\pi(S(H))$ and look at limits such as $\lim_{t\to \infty}\pi(S^{t}(H))$.
Given a graph $G$ with vertex set $v_1,...,v_n$ the [*link*]{} hypergraph $G^{v_i}$ is the hypergraph with vertex set $V(G)\setminus \{v_i\}$ and edge set $E=\{F\setminus \{v_i\}\colon v_i\in F \text{ and } F\in E(G)\}$.
For any hypergraph $H$ we have that $\pi(S(H))\leq \pi(H)$.
Let $G_{n}$ be a graph on $n$ vertices containing no copy of $S(H)$ such that $h_{n}(G_{n})=\pi_{n}(S(H))$. Say $V(G_{n})=\{v_1,v_2,...,v_{n}\}$. Note that for any $v_i\in V(G_n)$, we have that Lubell value of the corresponding link graph is $$h_{n-1}(G^{v_i}_{n})=\sum_{F\in G_n, v_i\in F} \frac{1}{\binom{n-1}{|F|-1}}.$$ Also, note that $G^{v_i}_n$ contains no copy of $H$. If it did, then $S(H)\subset S(G^{v_i}_n)\subseteq G_n$; but $S(H)$ is not contained in $G_n$. Thus $h_{n-1}(G^{v_i}_{n})\leq \pi_{n-1}(H)$. We then have the following: $$\begin{aligned}
\pi_{n}(S(H)) &= h_n(G_n) \\
&=\sum_{F\in E(G_n)} \frac{1}{\binom{n}{|F|}} \\
&=\sum_{F\in E(G_n)} \frac{1}{|F|} \sum_{v_i\in F} \frac{1}{\binom{n}{|F|}} \\
&=\sum_{i=1}^{n} \sum_{F\in E(G_n), v_i\in F} \frac{1}{|F|}\cdot \frac{1}{\binom{n}{|F|}} \\
&=\sum_{i=1}^{n} \sum_{F\in E(G_n), v_i\in F} \frac{1}{n}\cdot \frac{1}{\binom{n-1}{|F|-1}} \\
&=\frac{1}{n} \sum_{i=1}^{n} h_{n-1}(G^{v_i}_{n}) \\
&\leq \frac{1}{n} \sum_{i=1}^{n} \pi_{n-1}(H) \\
&=\pi_{n-1}(H).\end{aligned}$$ Thus, for any $n$, $\pi_{n}(S(H))\leq \pi_{n-1}(H)$; taking the limit as $n\to\infty$ we get the result as claimed.
If $H$ is degenerate, so is $S(H)$.
For all $H$, $\displaystyle \lim_{t\to\infty}\pi(S^{t}(H))=|R(H)|-1$.
To conclude our paper, we prove a special case of this conjecture.
\[t8\] Suppose that $H$ is a subgraph of the blowup of a chain. Let $k_1$ be the minimum number in $R(H)$. Suppose $k_1\geq 2$, and $H'$ is a new hypergraph obtained by adding finitely many edges of type $k_1-1$ arbitrarily to $H$. Then $$\lim_{t\to\infty} \pi(S^{t}(H'))=|R(H')|-1.$$
Without loss of generality, we can assume that $H$ is a blowup of a chain and $V(H')=V(H)$. (This can be done by taking blowup of $H$ and adding more edges.)
Suppose that $H$ has $v$ vertices and its edge type is $R(H):=\{k_1,k_2,\ldots, k_r\}$. Set $k_0:=k_1-1$ so that $R(H'):=\{k_0,k_1,\ldots, k_r\}$. For convenience, we write $R$ for $R(H)$ and $R'$ for $R(H')$, and $$\begin{aligned}
R+t &:=\{k_1+t,k_2+t,\ldots, k_r+t\},\\
R'+t &:=\{k_0+t,k_1+t,\ldots, k_r+t\}.\end{aligned}$$
For any small $\epsilon>0$, let $n_0=\lfloor \epsilon^{-t}\rfloor$. For any $n\geq n_0$ and any hypergraph $G_n^{R+t}$ with $$\pi_n(G)>|R(H)|-1+\epsilon=r+\epsilon,$$ we will show $G$ contains a subhypergraph $S^t(H')$.
Take a random permutation $\sigma\in S_n$ and let $X$ be the number of edges in $G$ hit by the random full chain $C_\sigma$: $$\emptyset \subset \{\sigma(1)\} \subset \{\sigma(1),\sigma(2)\}
\cdots \subset \{\sigma(1), \sigma(2),\ldots, \sigma(i)\}
\subset \cdots \subset [n].$$ We have $${{\rm E}}(X)=\pi_n(G)>r+\epsilon.$$ Since $X\leq r+1$, we have $${{\rm E}}(X)=\sum_{i=0}^{r+1}i {{\rm Pr}}(X=i)\leq (r+1){{\rm Pr}}(X=r+1) +r.$$ Thus, we get $$\label{eq:X=r+1}
{{\rm Pr}}(X=r+1)\geq \frac{{{\rm E}}(X)-r}{r+1}>\frac{\epsilon}{r+1}.$$
Recall that the density $\mu_H(G)$ is the probability that a random injective map $f\colon V(H)\to V(G)$ such that $H\stackrel{f}{\hookrightarrow}G$. Applying to $H=C^{R'+t}$, we have $$\mu_{C^{R'+t}}(G)={{\rm Pr}}(X=r+1)>\frac{\epsilon}{r+1}.$$ Every copy of the chain $C^{R'+t}$ will pass through a set $A_1\in E^{k_1+t}(G)$. Let $\mu_{C^{R+t}, A_1}(G)$ be the conditional probability that a random injective map $f\colon V(C^{R+t}) \to V(G)$ satisfies $C^{R+t}\stackrel{f}{\hookrightarrow}G$ given that the chain $C^{R+t}$ passes through $A_1$. Let $d_-(A_1)$ be the number of sets $A_0$ satisfying $A_0\in E^{k_0+t}(G)$ and $A_0\subset A_1$. Then, we have $$\mu_{C^{R'+t}}(G)=\frac{1}{{n\choose k_1+t}}\sum_{A_1\in
E^{k_1+t(G)}}\mu_{C^{R+t}, A_1}(G)
\cdot \frac{d_-(A_1)}{k_1+t}.$$ Setting $\eta=\frac{\epsilon}{2(r+1)}$, define a family $${\mathcal{A}}=\{A_1\in E^{k_1+t}(G) \colon \mu_{C^{R+t}, A_1}(G)> \eta \mbox{ and } d_-(A_{1})> \eta (k_1+t)\}.$$ We claim $|{\mathcal{A}}|>\eta {n\choose k_1+t}.$ Otherwise, we have $$\begin{aligned}
\mu_{C^{R'+t}}(G)&=\frac{1}{{n\choose k_1+t}}\sum_{A_1\in
E^{k_1+t(G)}}\mu_{C^{R+t}, A_1}(G)
\cdot \frac{d_-(A_1)}{k_1+t}\\
&= \frac{1}{{n\choose k_1+t}}\sum_{A_1\in{\mathcal{A}}}\mu_{C^{R+t}, A_1}(G)
\cdot \frac{d_-(A_1)}{k_1+t} + \frac{1}{{n\choose k_1+t}}\sum_{A_1\not\in{\mathcal{A}}}\mu_{C^{R+t}, A_1}(G)
\cdot \frac{d_-(A_1)}{k_1+t}\\
&\leq\eta+\eta<\frac{\epsilon}{t+1}.\end{aligned}$$ Contradiction!
A $k_1$-configuration is a pair $(S,A_1)$ satisfying $A_1\in {\mathcal{A}}$, $S=A_1\setminus \{i_1,i_2,\ldots, i_{k_1}\}$, and $A_1\setminus \{i_j\}\in E^{k_0+t}(G)$ for any $1\leq j\leq k_1$.
For any $A_1\in {\mathcal{A}}$, the number of $S$ such that $(S,A_1)$ forms a $k_1$-configuration is at least $${d_-(A_1)\choose k_1}\geq {\eta(k_1+t)\choose k_1}
>\left(\frac{\eta}{2}\right)^{k_1} {k_1+t\choose k_1}.$$ In the above inequality, we use the assumption $t>\frac{2}{\eta}k_1$.
By an averaging argument, there exists an $S$ so that the number of $k_1$-configurations $(S, \bullet)$ is at least $$\frac{|{\mathcal{A}}|\left(\frac{\eta}{2}\right)^{k_1} {k_1+t\choose k_1}}
{{n\choose t}}
\geq \frac{\eta^{k_1+1}}{2^{k_1}}{{n-t\choose k_1}}.$$ Now consider the link graph $G^S$. The inequality above implies $$\mu_{C^R}(G^S)\geq \frac{\eta^{k_1+2}}{2^{k_1}}.$$ This implies $G^S$ contains a blow up of $C^R$. Thus $G^S$ has a subhypergraph $H$. By the definition of $k_1$-configuration, this $H$ can be extended to $H'$ in $G^S$. In another words, $G$ contains $S^t(H')$.
Connections to extremal poset problems
======================================
As stated earlier, the Turán density of non-uniform hypergraphs is motivated by the extremal subset/poset problems.
Let ${\mathcal{B}}_n=(2^{[n]},\subseteq)$ be the $n$-dimensional Boolean lattice. Under the partial relation $\subseteq$, any family ${\mathcal{F}}\subseteq 2^{[n]}$ can be viewed as a subposet of ${\mathcal{B}}_n$.
For posets $P=(P,\le)$ and $P'=(P',\le')$, we say $P'$ is a [*weak subposet of $P$*]{} if there exists an injection $f\colon P'\to P$ that preserves the partial ordering, meaning that whenever $u\le'
v$ in $P'$, we have $f(u)\le f(v)$ in $P$. If $P'$ is not a weak poset of $P$, we say $P$ is $P'$-free. The following problems originate from Sperner’s theorem, which states that the largest antichain of ${\mathcal{B}}_n$ is ${\binom{n}{\lfloor \frac{n}{2}\rfloor}}$.
[**Extremal poset problems:**]{} Given a fixed poset $P$, what is the largest size of a $P$-free family ${\mathcal{F}}\subset {\mathcal{B}}_n$?
Let ${{\rm La}}(n,P)$ be the largest size of a $P$-free family ${\mathcal{F}}\subseteq {\mathcal{B}}_n$. The value of ${{\rm La}}(n,P)$ is known for only a few posets $P$. Let ${\mathcal{P}}_k$ be the (poset) chain of size $k$. Then ${{\rm La}}(n,{\mathcal{P}}_2)={\binom{n}{\lfloor \frac{n}{2}\rfloor}}$ by Sperner’s theorem. Erdős [@Erdos45] proved that ${{\rm La}}(n,{\mathcal{P}}_k)=\Sigma(n,k)$, where $\Sigma(n,k)$ is the sum of $k$ largest binomial coefficients. De Boinis-Katona-Swanepoel [@DebKatSwa] proved ${{\rm La}}(n,{\mathcal{O}}_{4})=\Sigma(n,2)$. Here ${\mathcal{O}}_4$ is the butterfly poset ($A,B \; \subset \; C,D$), or the crown poset of size $4$.
The asymptotic value of ${{\rm La}}(n, P)$ has been discovered for various posets (see Table \[tab:1\]). Let $e(P)$ be the largest integer $k$ so that the family of $k$ middle layers of ${\mathcal{B}}_n$ is $P$-free. Griggs and Lu [@GriLu] first conjecture $\lim_{n\to\infty}\frac{{{\rm La}}(n,P)}{{\binom{n}{\lfloor \frac{n}{2}\rfloor}}}$ exists and is an integer, and it slowly involves into the following conjecture.
\[conj:2\] For any fixed poset $P$, $\lim_{n\to\infty}\frac{{{\rm La}}(n,P)}{{\binom{n}{\lfloor \frac{n}{2}\rfloor}}}=e(p)$.
We overload the notation $\pi(P)$ for the limit $\lim_{n\to\infty}\frac{{{\rm La}}(n,P)}{{\binom{n}{\lfloor \frac{n}{2}\rfloor}}}$, where $P$ is a poset. The conjecture is based on the observation of several previous known results, which are obtained by Katona and others [@CarKat; @DebKat; @DebKatSwa; @GriKat; @KNS; @KatTar; @Tha]. We summarize the known poset $P$, for which the conjecture has been verified in Table \[tab:1\].
[|l|c|c|c|c|]{} $P$ & meaning & & $e(P)$ &References\
& $A<B_1,\ldots,B_r$ & \[5mm\]\[2mm\]
at (0,0) \[vertex\] (v0) ; at (-0.5,1) \[vertex\] (v1) ; at (-0.3,1) \[vertex\] (v2) ; at (0.3,1) \[vertex\] (v4) ; at (0.5,1) \[vertex\] (v5) ; (v0) – (v1); (v0) – (v2); (v2) – (v4); (v0) – (v4); (v0) – (v5);
&1&\
${\mathcal{N}}$ & $A<B$, $B>C$, and $C<D$. & \[5mm\]\[2mm\]
at (0,0) \[vertex\] (v0) ; at (0,1) \[vertex\] (v1) ; at (1,0) \[vertex\] (v2) ; at (1,1) \[vertex\] (v3) ; (v0) – (v1); (v1) – (v2); (v2) – (v3);
&1&\
& $A<B$, $B>C$, $C<D$, and $A<D$. & \[5mm\]\[2mm\]
at (0,0) \[vertex\] (v0) ; at (0,1) \[vertex\] (v1) ; at (1,0) \[vertex\] (v2) ; at (1,1) \[vertex\] (v3) ; (v0) – (v1); (v1) – (v2); (v2) – (v3); (v0) – (v3);
&2&\
& & \[6mm\]\[3mm\]
at (0,0) \[vertex\] (v0) ; at (-0.5,1) \[vertex\] (v1) ; at (-0.3,1) \[vertex\] (v2) ; at (0.3,1) \[vertex\] (v3) ; at (0.5,1) \[vertex\] (v4) ; at (0,2) \[vertex\] (v5) ; (v0) – (v1); (v0) – (v2); (v0) – (v3); (v0) – (v4); (v5) – (v1); (v5) – (v2); (v5) – (v3); (v5) – (v4);
&2&\
& & \[8mm\]\[5mm\]
at (0,0) \[vertex\] (b0) ; at (0,0.5) \[vertex\] (b1) ; at (0,1) \[vertex\] (b2) ; at (-0.5,-0.5) \[vertex\] (a1) ; at (-0.3,-0.5) \[vertex\] (a2) ; at (0.3,-0.5) \[vertex\] (a3) ; at (0.5,-0.5) \[vertex\] (a4) ; at (-0.5,1.5) \[vertex\] (c1) ; at (-0.3,1.5) \[vertex\] (c2) ; at (0.3,1.5) \[vertex\] (c3) ; at (0.5,1.5) \[vertex\] (c4) ; (b0) – (b1); (b1) – (b2); (b0) – (a1); (b0) – (a2); (b0) – (a3); (b0) – (a4); (b2) – (c1); (b2) – (c2); (b2) – (c3); (b2) – (c4); (c2) – (c3); (a2) – (a3);
&k-1&\
& & \[5mm\]\[2mm\]
at (0,0) \[vertex\] (b0) ; at (1,0) \[vertex\] (b1) ; at (2,0) \[vertex\] (b2) ; at (-0.5,1) \[vertex\] (c0) ; at (0.5,1) \[vertex\] (c1) ; at (1.5,1) \[vertex\] (c2) ; at (-0.5,-1) \[vertex\] (a0) ; at (0.5,-1) \[vertex\] (a1) ; at (1.5,-1) \[vertex\] (a2) ; (a0) – (b0); (a1) – (b1); (a2) – (b2); (b0) – (c0); (b0) – (c1); (b1) – (c1); (b1) – (c2); (a2) – (c1);
&$h(T)-1$&\
& & \[6mm\]\[4mm\]
at (0,0) \[vertex\] (b0) ; at (0.2,0) \[vertex\] (b1) ; at (0.4,0) \[vertex\] (b2) ; at (0.8,0) \[vertex\] (b4) ; at (1,0) \[vertex\] (b5) ; at (0,1) \[vertex\] (c0) ; at (0.2,1) \[vertex\] (c1) ; at (0.4,1) \[vertex\] (c2) ; at (0.8,1) \[vertex\] (c4) ; at (1,1) \[vertex\] (c5) ; (b0) – (c0); (b1) – (c0); (b1) – (c1); (b2) – (c1); (b2) – (c2); (b2) – (b4); (c2) – (c4); (b4) – (c4); (b5) – (c4); (b5) – (c5); (b0) – (c5);
&$1$&\
The posets in Table \[tab:1\] are far from complete. Let $\lambda_n(P)=\max\{h_n({\mathcal{F}})\colon {\mathcal{F}}\subseteq 2^{[n]}, P\mbox{-free}\}.$ A poset $P$ is called [*uniform-L-bounded*]{} if $\lambda_n(P)\leq e(P)$ for all $n$. Griggs-Li [@GriLi; @GriLi2] proved ${{\rm La}}(n,P)
=\sum_{i=\lfloor \frac{n+e(p)-1}{2}\rfloor}^{i=\lfloor
\frac{n-e(p)+1}{2}\rfloor}{n\choose i}$ if $P$ is uniform-L-bounded. The uniform-L-bounded posets include ${\mathcal{P}}_k$ (for any $k\geq 1$), diamonds ${\mathcal{D}}_k$ (for $k\in [2^{m-1}-1,
2^m-{\binom{m}{\lfloor \frac{m}{2}\rfloor}}-1]$ where $m:=\lceil \log_2(k+2)\rceil$), and harps ${\mathcal{H}}(l_1,l_2,\ldots,l_k)$ (for $l_1>l_2>\cdots>l_k$), and other posets. Noticeably, Griggs-Li [@GriLi2] provides a method to construct large uniform-L-bounded posets from smaller uniform-L-bounded posets. There are infinitely many posets $P$ so that $\pi(P)=e(P)$ holds.
Although there is no counter example found yet for Conjecture \[conj:2\], some posets have resisted efforts to determine their $\pi$ value. The most studied, yet unsolved, poset is the diamond poset ${\mathcal{D}}_2$ (or ${\mathcal{B}}_2$, $Q_2$ in some papers) as shown in Figure \[fig:d2c6c10\]. Griggs and Lu first observed $\pi({\mathcal{D}}_2)\in [2,2.296]$. Axenovich, Manske, and Martin [@AxeManMar] came up with a new approach which improves the upper bound to $2.283$. Griggs, Li, and Lu [@GriLiLu] further improved the upper bound to $2.27\dot 3=2\frac{3}{11}$. Very recently, Kramer-Martin-Young [@KMY] recently proved $\pi({\mathcal{D}}_2)\leq 2.25$. While it seems to be hard to prove the conjecture $\pi({\mathcal{D}}_2)=2$, several groups of researchers have considered restricting the problem to three consecutive layers. Let ${{\rm La}}^{c}(n,P)$ be the largest size a $P$-free family ${\mathcal{F}}\subseteq{\mathcal{B}}_n$ such that ${\mathcal{F}}$ is in $e(p)+1$ consecutive layers. Let $\pi^c(P)=\lim_{n\to\infty}\frac{{{\rm La}}^c(n,p)}{{\binom{n}{\lfloor \frac{n}{2}\rfloor}}}$, if the limit exists. Here is a weaker conjecture (of consecutive layers).
\[conj:3\] For any fixed poset $P$, $\pi^c(P)=e(p)$.
at (0,0) \[vertex\] (v0) ; at (-0.5,1) \[vertex\] (v1) ; at (0.5,1) \[vertex\] (v2) ; at (0,2) \[vertex\] (v3) ; (v0) – (v1); (v0) – (v2); (v1) – (v3); (v2) – (v3); at (0,-0.5) (label) [${\mathcal{D}}_2$]{};
at (0,0) \[vertex\] (b0) ; at (1,0) \[vertex\] (b1) ; at (2,0) \[vertex\] (b2) ; at (0,2) \[vertex\] (c0) ; at (1,2) \[vertex\] (c1) ; at (2,2) \[vertex\] (c2) ; (b0) – (c0); (b1) – (c0); (b1) – (c1); (b2) – (c1); (b2) – (c2); (b0) – (c2); at (1,-0.5) (label) [${\mathcal{O}}_6$]{};
at (0,0) \[vertex\] (b0) ; at (1,0) \[vertex\] (b1) ; at (2,0) \[vertex\] (b2) ; at (3,0) \[vertex\] (b3) ; at (4,0) \[vertex\] (b4) ; at (0,2) \[vertex\] (c0) ; at (1,2) \[vertex\] (c1) ; at (2,2) \[vertex\] (c2) ; at (3,2) \[vertex\] (c3) ; at (4,2) \[vertex\] (c4) ; (b0) – (c0); (b1) – (c0); (b1) – (c1); (b2) – (c1); (b2) – (c2); (b3) – (c2); (b3) – (c3); (b4) – (c3); (b4) – (c4); (b0) – (c4); at (2,-0.5) (label) [${\mathcal{O}}_{10}$]{};
Axenovich-Manske-Martin [@AxeManMar] first proved $\pi^c({\mathcal{D}}_2)\leq
2.207$; it was recently improved to $2.1547$ (Manske-Shen [@MS]) and $2.15121$ (Balogh-Hu-Lidický-Liu [@BHLL]).
We say a hypergraph $H$ [*represents*]{} a poset $P$ if the set of edges of $H$ (as a poset) is isomorphic to $P$. For any fixed finite poset $P$, by the definition of $e(P)$, there exists a hypergraph $H\subseteq {\mathcal{B}}_{n_0}$ with $|R(H)|=e(P)+1$ representing a superposet of $P$.
\[t9\] Suppose that a hypergraph $H$ with $|R(H)|=e(P)+1$ represents a superposet of $P$. Then, for any integer $t\geq 0$, we have $$\pi^c(P)\leq \pi(S^t(H)).$$
[**Proof:**]{} Let $x:=\pi(S^t(H))$. For any $\epsilon>0$, there exists an $n_1$ so that $\pi_{n}(S^{t}(H))
\leq x+\epsilon$ for all $n\geq n_1$. We claim $$\pi^c(P)\leq x+2\epsilon.$$ Otherwise, for any sufficiently large $n$, there exists a family ${\mathcal{F}}\subset {\mathcal{B}}_n$ which is in $e(p)+1$ consecutive layers with $|{\mathcal{F}}|>(x+\epsilon){\binom{n}{\lfloor \frac{n}{2}\rfloor}}$. Let $k_0$ be the smallest size of edges in $S^{t}(H)$. Let $k_1$ be the integer that ${\mathcal{F}}$ is in $k_1$-th to $(k_1+e(P))$-th layer. Since $e(P)\leq x < e(P)+1$, we have $$|{\mathcal{F}}|>(x+\epsilon){\binom{n}{\lfloor \frac{n}{2}\rfloor}}\geq (e(P)+\epsilon){\binom{n}{\lfloor \frac{n}{2}\rfloor}}.$$ Note any layer below $\frac{n}{2}-2\sqrt{n\ln n}$ can only contribute $\frac{2^n}{n^2}$, which is less than $\epsilon {\binom{n}{\lfloor \frac{n}{2}\rfloor}}$ for sufficiently large $n$. We get $$k_1\geq \frac{n}{2}-2\sqrt{n\ln n}.$$ Choose $n$ large enough so that $k_1\geq k_0$ and $n-k_1+k_0\geq n_1$. We observe that $$h_n({\mathcal{F}})\geq \frac{|F|}{{\binom{n}{\lfloor \frac{n}{2}\rfloor}}}> x+2\epsilon.$$ By the property of Lubell function, $h_n({\mathcal{F}})$ is the average of $h_{n+k_0-k_1}({\mathcal{F}}_S)$ over all $S\in {[n]\choose k_1-k_0}$, where ${\mathcal{F}}_S$ is the link hypergraph over $S$. Therefore, there exists a set $S\in {[n]\choose k_1-k_0}$ so that $h_{n+k_0-k_1}({\mathcal{F}}_S)>x+2\epsilon$. Thus, ${\mathcal{F}}_S$ contains a subhypergraph $S^{t_0}(H)$. In particular, ${\mathcal{F}}$ contains a subposet $P$.
Thus, we have $$\pi^c(P)\leq x+2\epsilon.$$ Since this holds for any $\epsilon>0$, we have $\pi^c(P)\leq x.$ $\square$
Conjecture \[conj:2\] implies Conjecture \[conj:3\].
In particular, from Theorem \[t8\], we get a new family of posets $P$ so that $\pi^c(P)=e(p)$. A special example is the crown ${\mathcal{O}}_{2t}$, where $t=4$ and $t\geq 6$. The idea can be traced back from Conlon’s concept [*$k$-representation*]{} of bipartite graphs [@Conlon]. Theorem \[t8\] can be viewed as a natural generalization of Conlon’s theorem. It is easy to generate more examples of posets in this family. However, a complete description of these posets is tedious; thus it is omitted here.
Note that the complete hypergraph $K_2^{\{0,1,2\}}$ has $4$ edges $\emptyset, \{1\}, \{2\}, \{1,2\}$; which form the diamond poset ${\mathcal{D}}_2$. In particular, for any $t\geq 0$, we have $$\pi^c({\mathcal{D}}_2)\leq \pi(S^t(K_2^{\{0,1,2\}})).$$ This provides a possible way to improve the bounds of $\pi^c({\mathcal{D}}_2)$.
[30]{} M. Axenovich, J. Manske, and R. Martin, [$Q_2$-free families in the Boolean lattice]{}, [*Order*]{} published online: 15 March 2011.
Rahil Baber and John Talbot, [New Turán densities for 3-graphs]{}, [*Elect. J. Combin.*]{}, (2012), P22, 21p.
J. Balogh, P. Hu, B. Lidický, and H. Liu Upper bounds on the size of 4- and 6-cycle-free subgraphs of the hypercube, arxiv:1201.0209 \[math.CO\].
B. Bukh, [Set families with a forbidden poset]{}, [*Elect. J. Combin.*]{} [**16**]{} (2009), R142, 11p.
T. Carroll and G. O. H. Katona, Bounds on maximal families of sets not containing three sets with $A\cup B\subset C, A \not\subset B$, [*Order*]{} [**25**]{} (2008) 229–236.
D. de Caen, The current status of Turán’s problem on hypergraphs. [*Extremal problems for finite sets*]{} (Visegrád, 1991), 187–197, Bolyai Soc. Math. Stud., 3, János Bolyai Math. Soc., Budapest, 1994.
F. Chung, L. Lu, An upper bound for the Turán number $t\sb 3(n,4)$. [*J. Combin. Theory Ser. A*]{} [**87**]{} (1999), no. 2, 381–389.
D. Conlon, An extremal theorem in the hypercube. [*Electron. J. Combin.*]{} [**17**]{} (2010), \#R111.
A. De Bonis and G. O. H. Katona, [Largest families without an $r$-fork]{}, [*Order*]{} [**24**]{} (2007), 181–191.
A. De Bonis, G. O.H. Katona and K. J. Swanepoel, [Largest family without $A\cup B \subset C\cap D$]{}, [*J. Combin. Theory (Ser. A)*]{} [**111**]{} (2005), 331–336.
P. Erdős, On a lemma of Littlewood and Offord, [*Bull. Amer. Math. Soc.*]{} [**51**]{} (1945), 898–902.
P. Erdős, On extremal problems of graphs and generalized g raphs, [*Israel J. Math.*]{} [**2**]{} (1964), 183–190.
P. Erdős, On the combinatorial problems which I would like to see solved, [*Combinatorica*]{} [**1**]{} (1981), 25-42.
P. Erdős, M. Simonovits, Supersaturated graphs and hypergraphs. [*Combinatorica*]{} [**3**]{} (1983), no. 2, 181–192.
P. Frankl, Z. Füredi, A new generalization of the Erdős-Ko-Rado theorem, [*Combinatorica*]{} [**3**]{} (1983), 341-349.
Z. Füredi and M. Simonovits, Triple systems not containing a Fano configuration, *Combin. Probab. Comput.* **14** (2005), 467–484.
Z. Füredi, D. Mubayi and O. Pikhurko, Quadruple systems with independent neighborhoods, **115** (2008), 1552–1560.
J. R. Griggs and G. O. H. Katona, [No four subsets forming an $N$]{}, [*J. Combinatorial Theory (Ser. A)*]{} [**115**]{} (2008), 677–685.
J. R. Griggs and W.-T. Li, [The partition method for poset-free families]{}, accepted by Journal of Combinatorial Optimization.
J. R. Griggs and W.-T. Li, [Uniformly L-bounded posets]{}, preprint (2011).
J. R. Griggs and L. Lu, [On families of subsets with a forbidden subposet]{}, [*Combinatorics, Probability, and Computing*]{} [**18**]{} (2009), 731–748.
J. R. Griggs, W.-T. Li, and L. Lu, Diamond-free Families, [*Journal of Combinatorial Theory Ser. A*]{}, [**119**]{} (2012) 310-322.
J. R. Griggs and L. Lu, [On families of subsets with a forbidden subposet]{}, [*Combinatorics, Probability, and Computing*]{} [**18**]{} (2009), 731–748.
Gy Katona, T. Nemetz, M. Simonovits, On a problem of Turán in the theory of graphs. [*Mat. Lapok*]{} [**15**]{} (1964) 228–238.
G. O. H. Katona and T. G. Tarján, [Extremal problems with excluded subgraphs in the $n$-cube]{}, in: M. Borowiecki, J. W. Kennedy, and M. M. Sysło (eds.) [**Graph Theory**]{}, Łagów, 1981, [*Lecture Notes in Math.*]{}, [**1018**]{} 84–93, Springer, Berlin Heidelberg New York Tokyo, 1983.
P. Keevash, Hypergraph Turán Problems, [*Surveys in Combinatorics*]{}, Cambridge University Press, 2011, 80-140.
P. Keevash and B. Sudakov, The Turán number of the Fano plane, *Combinatorica* **25** (2005), 561-574.
P. Keevash and B. Sudakov, The Turán number of the Fano plane,*Combinatorica* **25** (2005), 561-574.
L. Kramer, R. Martin, M. Young, On diamond-free subposets of the Boolean lattice, `http://arxiv.org/abs/1205.1501`.
N. W. Lemons, [*Turán Problems for Hypergraphs*]{}, dissertation, Central European University, Budapest, Hungary, (2008).
Linyuan Lu, On crown-free families of subsets, arxiv:1206.6258v1 \[math.CO\].
Manske and Shen, Three Layer $Q_2$-Free Families in the Boolean Lattice, arxiv:1108.4373 \[math.CO\].
D. Mubayi and Y. Zhao, Non-Uniform Turán-Type problems, [*J. Comb. Th. A*]{}, [**111**]{} (2004), 106–110.
O. Pikhurko, Exact computation of the hypergraph Turán function for expanded complete 2-graphs, Accepted by *J. Combin. Theory Ser. B* publication suspended for an indefinite time, see http://www.math.cmu.edu/$\sim$pikhurko/Copyright.html.
A. A. Razborov, On 3-hypergraphs with forbidden 4-vertex configurations, *SIAM J. Disc. Math.* **24** (2010), 946–963.
H. T. Thanh, An extremal problem with excluded subposets in the Boolean lattice, [*Order*]{} [**15**]{} (1998), 51–57.
P. Turán, On an extremal problem in graph theory. *Mat. Fiz. Lapok* **48** (1941), 436–452.
[^1]: University of South Carolina, Columbia, SC 29208, ([[email protected]]{}).
[^2]: University of South Carolina, Columbia, SC 29208, ([[email protected]]{}). This author was supported in part by NSF grant DMS 1000475.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'This is a written version of the review talk given at the meeting on “Interface of Gravitational and Quantum Realms" at IUCAA, Pune during December 2001. The talk reviewed the recent work of Martin Bojowald on Loop Quantum Cosmology.'
address: 'The Institute of Mathematical Sciences, CIT Campus, Chennai-600 113, INDIA.'
author:
- 'G. Date [^1]'
title: Quantum Geometric Description of Cosmological Models
---
6.2in 9.0in
0.5cm
Introduction
============
Canonical quantization of gravity formulated in terms of connection variable is now at a stage where one can address questions of physical interest. At the interpretational level, one still has the outstanding issues of understanding the classical limit and development of a systematic semiclassical expansion.[@thiemann]\
There are at least two classes of situations in classical GR that call for a quantum elucidation: (i) situations where classical GR predicts ‘singularities’ eg early universe and collapse, and (ii) situations where GR predicts horizons possessing entropy. In a sense, these can be thought of as ‘requiring’ a quantum theory of gravity and hence any such theory should lead to better understanding of these situations.\
There is another context in which one can look for signatures of quantum nature of space-time. This refers to matter wave propagation in a quantum geometrical background. Looking for such effects experimentally actually seems possible thanks to the GRB sources at cosmological distances. This context is very different from the first two in that QG effects are being probed in very ordinary, highly classical, non-extreme situations.[@urrutia]\
The horizon context has already been analyzed and quantum geometry framework does provide a microscopic understanding of entropy. Hawking effect is however is not yet clear.[@entropy]\
Of the two singular context, Martin Bojowald has recently analyzed the cosmological context in detail. As this is a first application of the quantum geometry framework with very interesting results, I decided to review this work in some details.
Standard Quantum Cosmology
==========================
A version of quantum gravity in the cosmological context was already attempted in the sixties. This was again a canonical quantization approach but using the geometro-dynamical variables - metric $g_{ij}$ and a symmetric second rank tensor $k_{ij}$ defined on a three manifold. This is a constrained system with diffeomorphism constraint implementing spatial diffeomorphism and the Hamiltonian constraint implementing the space-time diffeomorphisms. Since one does not know how to carry out the Dirac (or otherwise ) quantization of this constrained system, one looks for highly symmetric class of space-times, does the symmetry reduction classically and quantizes the left over finitely many degrees of freedom. In the context of so called Bianchi models, the gravitational phase space is at most six dimensional. This is a ‘minisuperspace’ quantization and has become synonymous with ‘standard quantum cosmology’.[@std-cosmology] Here are its main conclusions.\
(i) The Hamiltonian constraint (quadratic in momenta) can be interpreted as an evolution equation which is second order in a suitably chosen ‘time’ variable. This is intimately connected with the ‘problem of time’. In a 4-diffeomorphism invariant theory one has to interpret dynamics in terms of evolution relative to a ‘clock’ degree of freedom. For FRW models, the scale factor (a gravitational degree of freedom) is a natural choice of a ‘clock’. The evolution equation - Wheeler-De Witt equation - being second order in time has two independent solutions i.e. a non-unique wave function of the universe.\
(ii) When the scale factor vanishes, the inverse scale factor and hence the curvatures blow up, implying persistence of classical singularity.\
(iii) All geometrical quantities such as areas, volumes etc have continuous spectra.\
Note that what is quantized is a [*finite*]{} dimensional phase space as nothing else could be done. One was forced to do the symmetry reduction classically and then proceed to quantization. The Ashtekar reformulation in terms of connection variables offers a different alternative. One can quantize the infinite dimensional kinematical phase space and explore the possibility of doing symmetry reduction after quantization. Further more generically the spectra of areas , volumes etc turn out to be discrete indicating that this quantization is qualitatively different.
Quantum Geometry Framework
==========================
Let me briefly recall some of the basic steps.\
(a) Choose a classical phase space: This involves choosing a three manifold, $\Sigma$, with or without boundary and/or asymptotic regions. The basic variables (for the gravitational sector) are $su(2)$ (Lie algebra of $SU(2)$ group) valued connection 1-forms, $A_{a}^{i}\tau_{i}$ and a densitised triad, $E_{i}^{a}$ (‘dual’ to a 2-form). The classical configuration space is the space of all smooth (real analytic) connections satisfying appropriate boundary conditions, $\cal{A}$. The space modulo gauge transformations is denoted as $\cal{A/G}$. The phase space is the cotangent bundle of the configuration space. On this phase space we have the usual Gauss law, the diffeomorphism and the Hamiltonian constraints.\
(b) There is a natural class of functions on $\cal{A/G}$ namely, $Tr(P
exp \int_{\gamma} A )$. These are labeled by closed loops in $\Sigma$, are gauge invariant and diffeomorphism invariant. One now chooses to look for a Hilbert space on which these functions are represented by multiplicative operators. This is achieved by using the commutative $C^*$ algebra of these functions and choosing one of the representations of this algebra. This gives $H_{kin} =
L^{2}(\overline{\cal{A/G}},
\mu_{AL})$. In other words, the Hilbert space is the space of square integrable complex functions on the ‘quantum configuration space’ $\bar{\cal{A/G}}$ square integrable with respect to the Ashtekar-Lewandowski measure constructed from the Haar measure on $SU(2)$. A convenient description of the Hilbert space is obtained in terms of the so called spin-network functions.[@AL]\
In $\Sigma$ consider closed graphs $\gamma$. Associate with each edge $e_i$ a representation $\pi_i$ of $SU(2)$ and associate with each vertex $v_{\alpha}$ an intertwiner (contractor/invariant tensor), $C_{\alpha}$. For each $A \in
\bar{\cal{A}}$ one has, by definition, $g_i = A(e_i)$, the generalized holonomy. Define, $$\psi_{\gamma}(A) = ``Tr" \Pi_{edges} \pi_i(g_i) \ \Pi_{vertices} C_{\alpha}$$
These are the spin network functions - functions on $\cal{A/G}$ - labeled by graphs, representations and contractors. The set of all such functions forms an orthonormal set which is dense in $H_{kin}$. In practice one defines various operators by their actions on these and extends them to $H_{kin}^{\Sigma}$. This dense space is also called the space of cylindrical function.\
(c) In order to accommodate the possibility that the zero eigenvalue of constraints could be in a continuum (i.e. a generalized eigenvalue) one introduces a ‘rigging’ - $\Omega \subset H_{kin} \subset \Omega^*$. $\Omega$ is the dense space above while $\Omega^*$ is the space of continuous linear functional on $\Omega$. The important point is that the physical states generically belong to $\Omega^*$. One says that physical states are distributional.[@diff-inv]\
(d) For matter sector analogous constructions based on suitable $C^*$ algebras are made. These are available for all the usual scalar, spinor, gauge fields. The kinematical Hilbert spaces constructed here differ from the usual Fock spaces. The crucial point of these constructions is to have no dependence on background space-time geometry.[@std-model]
Quantum Symmetry Reduction
==========================
One would now like to specialize this frame work to the cosmological context of highly symmetric space-times. Since the configuration space variable are now connections, the notion of symmetry requires that the connection transformed under a symmetry diffeomorphism of $\Sigma$ to be gauge equivalent to the original connection. The first task is to characterize such symmetric connections precisely. This has already been done and can be summarized as (simplifying a bit for brevity):[@class; @bk]\
If $S$ is a symmetry group (compact Lie group) and $F$ a (Lie) subgroup of $S$, then\
(a) $\Sigma \sim B \times S/F \ , B \sim \Sigma/S $ is the space of orbits of S-action on $\Sigma$ ;\
(b) Symmetric connections on $\Sigma$ are completely characterized by a (reduced) connection on $B$ together with a set of (Higgs) scalars on $B$, possibly satisfying further constraints.\
For example, for the Bianchi class A models, $S$ is one of the Bianchi groups with structure constants satisfying $C^I_{JI} = 0$ while $F =
\{e\}$. $\Sigma$ is the group manifold of $S$ while $B$ is a single point of $\Sigma$. In terms of the Maurer-Cartan 1-forms $\omega^I$ and the corresponding left invariant vector fields $X_J$ one has, $$A^i_a = \Phi^i_I \omega^I_a \ \ \ \ E^a_i = P^I_i X^a_I
\ \ \ \ \ \ \{\Phi^i_I, P^J_j\} = 8 \pi G \gamma \delta^i_j \delta^J_I$$
$\Phi_I^i$ are the scalars, $P_i^I$ are the conjugate momenta, $\gamma$ is the Barbero-Immirzi parameter and the indices $i$ ($su(2)$), $I$ (Lie algebra of Bianchi group) both take three values. If in addition to spatial homogeneity one also has isotropy then the scalars satisfy further conditions whose solution is $\Phi_I^i = c \delta_I^i$ while the conjugate momenta satisfy $P_i^I = p
\delta_i^I$. The constraint expressions can be likewise simplified and expressed in terms of the scalars and their conjugates. This is the classical symmetry reduction. If one proceeds with quantization in a traditional manner then this is very similar to the usual minisuperspace quantization apart from the phase space variables being different. The results are also similar.[@kodama]\
However an alternative quantization is possible. Instead of requiring the scalars to be well defined operators one can take their exponentials, the so called point holonomies, as well defined operators. Then, similar to the general framework of quantum geometry (polymer representation), one constructs a new $H_{kin}^B$. One can immediately ask why one should do this? Does this Hilbert have anything to do with the $H_{kin}^{\Sigma}$ of the full theory? The answer turns out to be yes! Bojowald and Kastrup show that the (cylindrical) states in $H_{kin}^B$ can be identified with those distributions in $\Omega^*_{\Sigma}$ whose support consists of precisely the classical (smooth) symmetric connections.[@bk] Recall that the physical states of the full theory are supposed to reside in $\Omega^*_{\Sigma}$. A natural identification of symmetric (distributional) states of the full theory is via the properties of their support. The result shows that such quantum states can be alternatively be dealt with by working with the cylindrical states of $H_{kin}^B$. This provides a justification for the alternative (holonomy based) quantization of the symmetry reduced theory. It also makes available the tools of the general framework in a simplified context. This is what is referred to as ‘quantum symmetry reduction’. Note that this is quite general and not restricted to a cosmological context.
Bianchi Class A models
======================
The strategy now is step by step adaptation of the general framework. While majority of steps are identical, there are also crucial new inputs needed particularly when additional symmetry such as isotropy is at work. I am including only the minimal details necessary to communicate the final results.\
For classical configurations, point holonomies are just the group elements obtained by exponentiating the Lie algebra valued scalars, $u(\Phi_I) \equiv exp\{ \Phi_I^i \tau_i\} \in SU(2)$. Distributional scalars do just that, they associate with each vertex, an $SU(2)$ group element. For general anisotropic case the kinematical Hilbert space turns out to be $H_{kin} = L^2( SU(2)^3, d\mu_{Haar}^3 )$.[@b1] There are three copies since $I$ takes three values. For gauge invariant functions, a convenient basis is provided by the usual spin-network functions constructed from graphs $\gamma$ in the group manifold of $S$ with a single vertex of order 6 and three (closed) edges corresponding to the three left invariant vector fields. Using these one defines the operators corresponding to the conjugate momenta and then proceeds to build the constraint operators. Following the Thiemann approach, the Hamiltonian constraint is expressed in terms of the volume operator together with various commutators of holonomies with the volume operator. So the main problem is to define a volume operator and obtain its spectrum. The full spectrum is not available in the general case. However, for the homogeneous and isotropic models (Bianchi I and IX), spectrum of volume operator has been determined.\
When isotropic case is considered, the scalars have to satisfy further conditions. Naively one would expect that since we now have a single scalar, spin-network functions associated with graphs with a single closed edge (and a single vertex) should suffice. This turns out to be false. Although the $H_{kin}$ in this case does turn out to be $L^2(SU(2),d\mu_{Haar})$, it contains more gauge invariant functions than what a graph with single edge could supply. Bojowald determines the extra functions needed thereby obtaining an explicit orthonormal basis for the kinematical Hilbert space. A volume operator is now defined explicitly and is spectrum obtained.[@b2] The eigenfunctions and eigenvalues are given by (in the gauge invariant sector), $$\begin{aligned}
\chi_j(c) & = & \frac{sin((j + \frac{1}{2})c)}{sin(c/2)} \ \ \ \ \ j =
0, 1/2, 1, ... \nonumber \\
\zeta_j(c) & = & \frac{cos((j + \frac{1}{2})c)}{sin(c/2)} \ \ \ \ \ j =
0, 1/2, 1, ... \nonumber \\
\zeta_{-\frac{1}{2}} & = & \frac{1}{\sqrt{2} sin( c/2 )} \nonumber \\
V_j & = & (\gamma \ell_p^2)^{3/2} \sqrt{\frac{j (j + \frac{1}{2}) (j + 1)}{27}}\end{aligned}$$
It turns out to be more convenient to choose a slightly different orthonormal basis which we now denote as:[@absense] $$|n\rangle \equiv \frac{e^{i \frac{n}{2} c}}{\sqrt{2} sin(c/2)}, \ \ \ \ \ n \in
{\bf Z}$$
These are also eigenstates of the volume operator with eigenvalues $V_{\frac{|n| - 1}{2}}$. Now the Hamiltonian constraint can be defined by its action on these states. In general it has the form: $$\hat{H} | n \rangle = \sum_{k = -L}^{L} A_{n - k}^k | n - k \rangle .$$
The physical states, $| s \rangle = \sum_n s_n |n\rangle $, which belong to the kernel of $\hat{H}$, have conditions on the coefficients $s_n$ of the form, $$\sum_{k = -L}^{L} A_n^k s_{n + k} = 0 .$$
This is a difference equation for the $s_n$ coefficients. The order of the equation, $2 L$, depends on the Bianchi type while the coefficients $A_n^k$ depend on the details of the Hamiltonian constraint and the factor ordering chosen.[@b3; @b4; @isotropic] For the flat isotropic model, the order of the equation is 16. The ([*non-*]{}symmetric) ordering chosen is such that the coefficient of $s_0$ always vanishes. One can introduce matter sector and add its contribution to the Hamiltonian constraint. Denoting matter symbolically by $\varphi$ the equations are modified by putting in a $\varphi$ dependence in $s_n$ and adding a term of the form $\hat{H}_{\varphi}(n) s_n (\varphi)$. At this stage, particular form of matter and its couplings are not detailed. It is sufficient to note that the matter part of the Hamiltonian operator is diagonal with respect to the (gravitational) states $|n\rangle$. This specifies the physical states of the loop quantum cosmology.[@absense; @unique]
Dynamical Interpretation, Absence of Singularity and Uniqueness of Solution
===========================================================================
In order to interpret the physical states, in particular to explore what ‘happens’ to the classical singularity, one needs to view the physical states obtained above as solutions of a ‘time evolution’ equation. In a generally covariant theory of space-time there is no external, inert ‘time’ variable. One can at best single out one of the degrees of freedoms as a ‘clock’ and view functions of this (gravitational or matter) degree of freedom as functions of ‘time’. While there is no a priori given clock, in a given context there can be a natural choice. For instance, for the FRW cosmologies, the scale factor in the metric is one such choice. It is in terms of this choice that one says that there is a singularity at the vanishing value of the scale factor (curvatures blow up). In the present formulation, absolute value of the conjugate momentum, $p$, corresponds to the square of scale factor. The spectrum of $|\hat{p}|$ consists of eigenvalues $(j + 1/2)$ with eigenfunctions $\chi_j(c), \zeta_j(c)$. The zero eigenvalue of $|\hat{p}|$ occurs for $j = -1/2$ which is a non-degenerate eigenvalue. (The volume operator consists of this operator together with another commuting operator with eigenvalues $j(j + 1)$. This makes the zero eigenvalue of the volume operator three fold degenerate.) In terms of the $|n\rangle$ basis, $n = 0$ corresponds to vanishing scale factor. One therefore takes, the label $n \in
{\bf Z}$ as a time label. The physical state condition can now be viewed as specifying a discrete time evolution.[@b4; @absense; @unique]\
Restricting to the isotropic, flat models, the evolution equation is of order 16 (and consists of the $n \pm 8, n \pm 4$ and $n$ terms only). The coefficients are such that $A_n^k = 0$ if and only if $n + k = 0$. Furthermore, $\hat{H}_{\varphi}(n = 0) = 0$ as well. This has two consequences. The $s_0(\varphi)$ never appears in the equation. Therefore it is neither determined by nor determines any other $s_n(\varphi)$. The state $|0\rangle$ is thus orthogonal to all other physical states (evolving solutions). This is precisely the state which corresponds to the zero eigenvalue of the scale factor operator. Thus one sees that unlike the classical case where evolution through vanishing scale factor is not possible, the quantum evolution equation does not suffer such a breakdown. In this sense the classical singularity is absent in the evolving states. The second consequence is that one gets a conditions on the initial data - the choice of 16 coefficients $s_{-16}, s_{-15}, ... s_{-1}$ (say). This is because the equation can not determine $s_0$ since its coefficient vanishes. Thus one can conclude that the classical singularity is avoided by the evolving solutions and that there are 15 instead of 16 independent solutions.[@absense; @unique]\
This seems worse than the standard quantum cosmology with only two independent solutions! Quite independently, one has also to understand the discrete time evolution in terms of more familiar continuous time evolution, at least when one hopes to be in classical regime. Clearly, the volume and the scale factor eigenvalues become large compared to the Planck scale values when $n \gg 1$. One expects a solution to be classically interpretable when for large $n$, $s_{n + m}(\varphi)$ can vary significantly from $s_n(\varphi)$ when $m$ is also large compared to 1 but remains almost the same when $m$ is comparable to 1. One can now ask: how many solutions exhibit this behavior? In a precise formulation of such a behavior, the Barbero-Immirzi parameter $\gamma$ come very handy. Note that for large $n = 2j + 1$, $V_j \to
(\gamma \ell_p^2 |n|/6 )^{3/2} \sim (a^2)^{3/2}$. We can use this to define the scale factor as a function of $n$ as: $a^2(n) \equiv (n \gamma)
\ell_p^2 /6 $. Now change in $a^2(n)$ as $n$ changes by $1$ will be infinitesimal if $\gamma \ell_p^2 \to 0$. Thus we can mimic continuous evolution with respect to the scale factor by considering the formal limit $\gamma \to 0 , n \to \infty$ keeping $\gamma n$ a constant. We are keeping $\ell_p$ fixed in this and so are in the quantum domain still. If a solution $\{s_n(\varphi)\}$ has a limiting value in the above limit then it is said to be ‘pre-classical’.[@unique] The question now becomes as to how many evolving solutions are pre-classical? For large $n$, the equations itself becomes an equation with constant coefficients and is easy to analyze. The result is that, of the sixteen solutions, only two are pre-classical. There is still the constraint on the initial conditions which reduces these two to a single solution. Thus while there are many evolving solutions (all avoiding the singularity), only one of these can mimic a classical evolution.[@unique]\
From the black hole entropy computations, $\gamma$ is fixed to be a constant of order one. The $\gamma \to 0$ limit noted above is to be thought of as a formal device to single out solutions mimicking classical evolution. The presence of the Barbero-Immirzi parameter however offers a new limit to be explored apart from the classical limit with $\ell_p \to 0$. Bojowald has further shown that in the $\gamma \to
0$ limit, referred to as ‘continuum limit’ in the sense that the discrete structure of quantum geometry can be ignored, one recovers the standard quantum cosmology.[@semiclassical] This shows also that it is the specific, discrete structure of quantum geometry that is responsible for avoiding the singularity and selecting a unique (pre-classical) solution and not just any quantization procedure.
Concluding Remarks
==================
This is a first explicit application of the highly abstract framework of quantum geometry to a context of physical interest, particularly addressing the issue of classically indicated singularity. Not only does it meet the expectation that in a quantum theory of gravity, classically indicated singularities should be absent, it provides a quantitative means to estimate how rapidly quantum geometry picture goes over to the classical picture. In the quantization procedure, many conceptual and technical assumptions have gone in (eg choice of polymer representation). In a sense the classical space-time picture has been ‘mutilated’ quite a bit. Therefore it could have happened that the none of final set of solutions could be interpreted in classical terms thereby recovering the classical theory. Not only this does [*[not]{}*]{} happen but one gets a unique solution displaying the classical picture. It also brings out an intriguing role played by the Barbero-Immirzi parameter.\
It is a pleasure to thank the organizers for the invitation and IUCAA for the excellent atmosphere and hospitality. I would like to thank Martin Bojowald for his useful comments on an earlier draft.
T. Thiemann, gr-qc/0110034.\
A good introduction to the subject is in this review article by Thiemann.
See talk by L.F. Urrutia in this volume.
A. Ashtekar, J. Baez and K. Krasnov, [*Adv. Theor. Math. Phys.*]{} [**4**]{}, 1 (2000) and references therein, gr-qc/0005126.
There is a very vast literature on this subject. The one I largely followed is:\
A. Ashtekar, R.S. Tate and C. Uggla, [*Int. Jour. Mod. Phys.*]{} [**D2**]{}, 15 (1993), gr-qc/9302027 and references therein.
A. Ashtekar and J. Lewandowski, [*Knots and Quantum Gravity*]{} (ed. J. Baez, Oxford Univ. Press, 1995), gr-qc/9311010.
A. Ashtekar, J. Lewandowski, D. Marolf, J. Mourao and T. Thiemann, [*Jour. Math. Phys.*]{} [**36**]{}, 6456 (1995), gr-qc/9504018.
T. Thiemann, [*Class. Quant. Grav.*]{} [**15**]{}, 1487 (1998), gr-qc/9705021.
For the classification of invariant connections see\
S. Kobayashi and K. Nomizu, [*Foundations of Differential Geometry*]{} vol I, II (John Wiley & Sons, New York, 1963);\
I found a paper by Forgacs and Manton also useful.\
P. Forgacs, and N.S. Manton, [*Comm. Math. Phys.*]{} [**72**]{}, 15 (1980)
M. Bojowald and H.A. Kastrup, [*Class. Quant. Grav.*]{} [**17**]{}, 3009 (2000), hep-th/9907042.
H. Kodama, [*Prog. Theo. Phys.*]{} [**80**]{}, 1024 (1988).
M. Bojowald, [*Class. Quant. Grav.*]{} [**17**]{}, 1489 (2000), gr-qc/9910103.
M. Bojowald, [*Class. Quant. Grav.*]{} [**17**]{}, 1509 (2000), gr-qc/9910104.
M. Bojowald, [*Phys. Rev. Lett.*]{} [**86**]{}, 5227 (2001), gr-qc/0102069.
M. Bojowald, [*Class. Quant. Grav.*]{} [**18**]{}, 1055 (2001), gr-qc/0008052.
M. Bojowald, [*Class. Quant. Grav.*]{} [**18**]{}, 1071 (2001), gr-qc/0008053.
M. Bojowald, [*Phys. Rev. Lett.*]{} [**87**]{}, 121301 (2001), gr-qc/0104072.
M. Bojowald, [*Class. Quant. Grav.*]{} [**18**]{}, L109 (2001), gr-qc/0105113.
For more details of factor ordering etc, see\
M. Bojowald, [*Class. Quant. Grav.*]{} [**19**]{}, 2717 (2002), M. Bojowald, gr-qc/0105113.
[^1]: e-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
bibliography:
- 'references\_syed\_husain.bib'
title: 'A new dynamical core of the [G]{}lobal [E]{}nvironmental [M]{}ultiscale ([GEM]{}) model with a height-based terrain-following vertical coordinate'
---
Introduction {#s_intro}
============
The dynamical core of the Global Environmental Multiscale (GEM) model, used operationally by Environment and Climate Change Canada (ECCC) for numerical weather prediction (NWP), employs a log-hydrostatic-pressure-type terrain-following vertical coordinate. The system of nonlinear model equations is linearized around a basic state and is then reduced to an elliptic boundary value (EBV) problem through numerical discretization and elimination of variables [@gdm14]. The existing pressure-type coordinate system then permits the use of a direct solver for the discretized EBV problem to resolve the dynamical component of the flow. The direct solver starts by separating the EBV problem vertically in terms of the vertical eigenvectors of the part of the coefficient matrix that only includes the discretized difference and average operators in the vertical direction [@qle10]. For $N$ number of model vertical levels, the resulting $N$ vertically-decoupled two-dimensional Helmholtz problems are then separated along the longitude, leading to a system of tridiagonal problems depending only on the latitude. The tridiagonal problems are finally solved using LU decomposition without pivoting. Such an approach is computationally more efficient than most iterative methods, particularly for the spatial resolutions of the current operational NWP systems at ECCC.\
One of the principal incentives for the adoption of the existing pressure-type vertical coordinate in GEM is the computational advantage of the direct solution approach that is permissible with such a coordinate. However, numerical experiments carried out at ECCC have revealed that vertical separability, which is an imperative for the direct solution approach, can become quite restrictive for very high spatial resolutions, e.g., for sub-kilometer horizontal grid spacing. Furthermore, the reduction of the 2D Helmholtz problems into the 1D tridiagonal problems requires projection of right hand side (RHS) of the 2D problems along the pertinent eigenvectors which is based on Fourier transformation. This necessitates transposing the coefficient matrix that involves global communications, and therefore, becomes inefficient for very large number of processor cores. As a result, the direct solver is found to lose scalability with increasing number of processor cores. Initial research at ECCC as well as published research literature [@msc14] reveal that optimized three-dimensional iterative solvers may possess better scalability in these circumstances. The different limitations of the existing direct solver, particularly its potential lack of scalability for future generations of massively parallel supercomputers, therefore warrants the development of more scalable iterative solvers at ECCC. A height-based vertical coordinate is considered to be more amenable to such iterative solvers as the metric terms originating from the vertical coordinate transformation appear explicitly in the discretized EBV.\
Another, but more challenging, problem pertaining to the existing GEM dynamical core is the fact that the current model exhibits strong numerical instability over steep orography. Tests conducted at ECCC have determined the maximum permissible terrain slope for maintaining stability to be approximately 45$^\circ$ [@vbg15]. This is generally considered to be a limitation inherent to the terrain-following coordinate (TFC) systems [@zan12]. With a growing demand for very high-resolution operational NWP systems, steep orographic slopes are expected to become more prevalent in the near future. Improving model stability over steep mountains is therefore of critical importance for the future development of sub-kilometer NWP systems. A number of approaches have been investigated to improve numerical stability with the existing pressure-type vertical coordinate in GEM. These include increased off-centering in the discretized vertical momentum equation, a vertically-variable basic state temperature profile, and modifications to the nonhydrostatic contributions in the linear system arising from the discretized GEM formulation. However, none of these approaches has been found to lead to any meaningful stability improvement for steep orography.\
Although the instability induced by steep orography is often characterized as a limitation of the terrain-following nature of the vertical coordinate itself, [@ssw07] were successful in resolving flow around the Pentagon with a model involving height-based TFC where the maximum slope was well above the widely acknowledged 45$^\circ$ threshold. Previous experience with the Mesoscale Compressible Community model at ECCC also suggests that a dynamical core with height-based TFC does not suffer from similar severe orography-induced instability [@gbd05]. Apparently, a dynamical core with a height-based TFC can lead to improved numerical stability through better implicit treatment of the metric terms arising from the vertical coordinate transformation through iterative solvers. More importantly, conventional numerical approximation of the horizontal gradients in a TFC becomes less accurate with increasing terrain slope as well as with increasing vertical resolution close to the model surface [@mah84]. In this context, [@zan12] argues that the pressure gradient term, in particular, becomes susceptible to triggering numerical instability when the height difference between two adjacent grid points along a terrain-following vertical level is much larger than the vertical grid resolution adjacent to the level. Numerical approximation of the horizontal gradient terms in the TFC, however, can be significantly improved following the corrections proposed by [@mah84]. These corrections require determination of the modified horizontal differencing stencils associated with each grid-point location that minimize the error in the metric corrections for the terrain-following nature of the coordinate. The existing pressure-type TFC varies with time and, therefore, would require determination of the pertinent grid-point locations in the vertical for the modified differencing stencils at every time step. On the contrary, the height-based TFC is time-invariant and thus would require the determination of these grid-point locations only once at the beginning of the time integration. Therefore, from a computational efficiency standpoint, a height-based TFC is more suitable for implementing improved numerical approximation of horizontal gradients to address instabilities induced by steep orography.\
The aforementioned challenges associated with the existing log-hydrostatic-pressure-type TFC motivated the development of a new dynamical core for the GEM model that utilizes a height-based TFC. The primary objective of the present study is to demonstrate that, for the model configurations where orography-induced numerical instability is not relevant—i.e., for horizontal grid resolutions within the hydrostatic regime—the new dynamical core developed at ECCC with height-based TFC makes predictions that are equivalent to those from the existing model. The present study also explores the appropriate strategy for coupling the operational Physics Parameterization Package (PPP) of RPN (Recherche en prévision numérique) with the new height-based dynamical core. Different setups for numerical experiments are utilized to compare the newly-developed dynamical core with the existing one covering both hydrostatic and nonhydrostatic scenarios. The experiments include two-dimensional theoretical cases [@rob93; @slf02] as well as three-dimensional global forecasts over a Yin-Yang grid [@qle11].\
Relevant background information on the GEM dynamical core with the proposed height-based TFC—from the spatial and temporal discretizations to the derivation of the elliptic boundary value problem—is presented in section \[s\_mod\_descri\]. The different solution methods utilized for the discretized elliptic problem is discussed in section \[s\_solver\]. The issue of coupling between the dynamical core and the parameterized physics forcings is presented in section \[s\_coupling\]. Section \[s\_evaluation\] contains the comparisons between the existing and the proposed dynamical cores in the context of two-dimensional theoretical benchmark cases as well as three-dimensional deterministic global predictions. The conclusions are then summarized in section \[s\_summary\].\
Model Description {#s_mod_descri}
=================
Governing equations
-------------------
The GEM model equations originate from the Euler equations. With the traditional shallow atmosphere approximation [@phi66], the system of equations in a spherical coordinate $(\lambda,\phi,r)$ can be expressed as follows:
$$\dv{u}{t}-\left(f+\frac{\tan\phi}{a}u\right)v+\frac{1}{\rho}\pdv{p}{x} = \left(\dv{u}{t}\right)_{phys},
\label{e_u_mom}$$
$$\dv{v}{t}+\left(f+\frac{\tan\phi}{a}u\right)u+\frac{1}{\rho}\pdv{p}{y} = \left(\dv{v}{t}\right)_{phys},
\label{e_v_mom}$$
$$\dv{w}{t}+\frac{1}{\rho}\pdv{p}{z} +g = \left(\dv{w}{t}\right)_{phys},
\label{e_w_mom}$$
$$\dv{\ln \rho}{t}+\pdv{u}{x}+\pdv{v}{y}+\pdv{w}{z}-\frac{\tan\phi}{a}v=\left(\dv{\ln \rho}{t}\right)_{phys},
\label{e_conti}$$
$$\dv{\ln T}{t}-\kappa\dv{\ln p}{t}=\left(\dv{\ln T}{t}\right)_{phys},
\label{e_thermo}$$
where Eqs. (\[e\_u\_mom\])–(\[e\_thermo\]) govern the evolutions of the $u$, $v$, and $w$ components of velocity, mass and energy, respectively. The spatial coordinates in the above equations are denoted by $(x,y,z)$ which are related to the spherical coordinate $(\lambda,\phi,r)$ through the differential relations given by $$dx=a\cos\phi d\lambda, dy=ad\phi, dz=dr,
\label{e_spherical}$$ such that $u$, $v$ and $w$ are the physical wind components. In Eq. (\[e\_spherical\]), $a$ denotes Earth’s radius. The Lagrangian derivative in this case can be expressed as $$\dv{}{t}=\pdv{}{t}+u\pdv{}{x}+v\pdv{}{y}+w\pdv{}{z}.
\label{e_lagrange}$$ In addition to the four independent variables $(x,y,z,t)$, the system of five equations (\[e\_u\_mom\])–(\[e\_thermo\]) involves six dependent variables, namely, the velocity components $u$, $v$, and $w$, the temperature $T$, the pressure $p$, and the density $\rho$. Also, in the above equations, $f$ is the Coriolis parameter and $\kappa=R/c_p$ where $R$ is the gas constant and $c_p$ is the specific heat of air at constant pressure. The terms on the RHS of Eqs. (\[e\_u\_mom\])–(\[e\_thermo\]) with subscript “$phys$” denote the various physical forcings. Depending on the equation, these physical forcings may arise from different sources that include friction, diabatic heating and frictional dissipation of kinetic energy. A sixth equation is required to close the system described by the six dependent variables and is provided by the equation of state, given by $$p=\rho RT.
\label{e_state}$$
It is important to note that the atmospheric substance does not only contain dry air but also water vapor and different types of hydrometeors. The displacement and evolution of water vapor and hydrometeors in the atmosphere are governed by their own evolution equations. However, they will also affect the RHS terms through fluxes of water vapor and precipitation which constitute sources of mass. The total air density in the presence of water in its different forms can be expressed as $$\rho=\rho_d+\rho_w=\rho_d(1+r_w),
\label{e_rho}$$ where $\rho_d$ is the dry air density, $\rho_w=\frac{\rho_w}{\rho_d}$ is the density of water vapor and hydrometeors, and $r_w$ is the mixing ratio for the total water content of the atmosphere. The equation of state in such a scenario is strictly given by $$p=(\rho_dR_d+\rho_vR_v)T,
\label{e_state_2}$$ where $\rho_v$ and $R_v$ are the density and gas constant of water vapor. Eq. (\[e\_state\_2\]) can then be further rearranged in terms of the total air density $\rho$ as $$p=\rho R_d T_v,
\label{e_state_3}$$ where $T_v$ is the virtual temperature of moist air which is given by $$T_v=\frac{1+\frac{R_v}{R_d}r_v}{1+r_w}T,
\label{e_tv}$$ where $r_v=\frac{\rho_v}{\rho_d}$ is the water vapor mixing ratio. Rewriting the dynamical equations (\[e\_u\_mom\])–(\[e\_thermo\]) in terms of $T_v$ is helpful as the equations can then be expressed using only the dry gas constant that does not vary due to the atmospheric water content. It also allows to account for the effects of water vapor buoyancy and condensed water loading implicitly. Furthermore, from the adiabatic point of view, the introduction of $T_v$ has no consequence since the water content is conserved during dynamical transport.\
Vertical coordinate
-------------------
The log-hydrostatic-pressure-type terrain-following vertical coordinate of the operational GEM model [@gdm14] has the form $$\ln\pi=\xi+Bs,
\label{e_vgd_p}$$ where $\xi$ defines the terrain-following vertical coordinate, $\pi$ denotes the hydrostatic pressure, $B$ is a metric term prescribing the rate of flattening of the vertical coordinate with elevation, and $s=\ln(\pi_s/p_{ref})$ with $\pi_s$ being the hydrostatic pressure at the surface and $p_{ref}=10^3$ hPa is a reference pressure. The definition of this vertical coordinate follows the concept of the generalized hydrostatic-pressure-type hybrid coordinate proposed by [@lap92]. Further details regarding the log-hydrostatic-pressure-type vertical coordinate are provided by @gdm14 and @hgi17.\
The present study proposes to develop a GEM dynamical core where the vertical coordinate, given by Eq. (\[e\_vgd\_p\]), in the existing dynamical core is replaced by a height-based TFC. The traditional formulation for height-based TFC can be expressed as $$\zeta(z)=H\frac{z-z_S}{z_T-z_S},
\label{e_vgd_z1}$$ where $z$ is the the true geometric height, $z_S$ and $z_T$ are the surface and the model top level heights, and $H$ is a scaling constant. A more general form of Eq. (\[e\_vgd\_z1\]) can be devised as $$z=A+Bz_S,
\label{e_vgd_z2}$$ where $A=(z_T/H)\zeta$ and $B=(1-\zeta/H)$. Assigning $H=z_T$ implies $z_T=\zeta_T$ and, as a result, Eq. (\[e\_vgd\_z2\]) becomes $$z=\zeta+Bz_S
\label{e_vgd_z3},$$ which is similar to Eq. (\[e\_vgd\_p\]) in form. The vertical coordinate for the proposed dynamical core in the present study, however, is further generalized as $$z=\zeta+B_1z_{SL}+B_2(z_S-z_{SL}),
\label{e_vgd_zs1}$$ which follows the concept of SLEVE (Smooth LEvel VErtical)-like coordinate system proposed by [@slf02], where $z_{SL}$ denotes the large-scale components of the orography. The vertical coordinate defined by Eq. (\[e\_vgd\_zs1\]) permits separate rates of flattening for the large and small-scale contributions of the orography on the terrain-following vertical coordinate with changing elevation through the metric terms $B_1$ and $B_2$ that are defined as $$B_n=\bigg(\frac{\zeta_T-\zeta}{\zeta_T-\zeta_S}\bigg)^{r_n},
\label{e_vgd_zbn}$$ where $r_n=[r_{n,max}-(r_{n,max}-r_{n,min})\lambda_k]$ and $\lambda_k=(\zeta_1-\zeta_k)/(\zeta_1-\zeta_S)$. The values of $r_{n,min}$ and $r_{n,max}$ together determine the rate of flattening of the vertical levels with increasing height. The subscript $k$ of $\lambda$ indicates the model vertical level number. Furthermore, the value of $k$ decreases with increasing height above the surface such that $k$=1 indicates the top-most model level.\
Henceforth, in this paper, the two dynamical cores with vertical coordinates based on log-hydrostatic-pressure and height are referred to as GEM-P and GEM-H, respectively. Different aspects of the GEM-P dynamical core, including the model formulation, discretization and numerical solution of the discretized problem along with the various modifications to the formulation over the past years, have been discussed in detail in the existing literature [@ycg02; @qle11; @gdm14; @hgi17]. The following subsections therefore only present the relevant details of the proposed GEM-H dynamical core.\
GEM-H formulation
-----------------
The development of the GEM-H formulation requires further modifications to the system of equations (\[e\_u\_mom\])–(\[e\_thermo\]). First, virtual temperature, given by Eq. (\[e\_tv\]), is introduced in the system of equations along with an isothermal basic state temperature $T_*$ such that $T_v=T'_v+T_*$, where $T'_v$ is the temperature deviation. The corresponding basic state pressure $p_*$ is defined hydrostatically as $\partial(\ln p_*)=-g\partial{z}/(R_dT_*)$. The equation of state, given by Eq. (\[e\_state\_3\]), is then used to eliminate density $\rho$ as a prognostic variable followed by a transformation of the resulting equations from the geometric height coordinate to the terrain-following $\zeta$-coordinate. The vertical coordinate transformation leads to the replacements of the independent variables $(x,y,z)$ associated with the $z$-coordinate by $(X,Y,\zeta)$ that are defined in the $\zeta$-coordinate. As a result, the system of equations (\[e\_u\_mom\])–(\[e\_thermo\]) is modified as follows: $$\dv{u}{t}-\bigg(f+\frac{\tan\phi}{a}u\bigg)v+\frac{T_v}{T_*}\bigg(\pdv{q}{X} -\frac{J_X}{J_{\zeta}}\pdv{q}{\zeta}\bigg)= \left(\dv{u}{t}\right)_{phys},
\label{e_u_1}$$ $$\dv{v}{t}+\bigg(f+\frac{\tan\phi}{a}u\bigg)u+\frac{T_v}{T_*}\bigg(\pdv{q}{Y} -\frac{J_Y}{J_{\zeta}}\pdv{q}{\zeta}\bigg)= \left(\dv{v}{t}\right)_{phys},
\label{e_v_1}$$ $$\dv{w}{t}+\frac{T_v}{T_*}\bigg(\frac{1}{J_{\zeta}}\pdv{q}{\zeta}-g\frac{{T_v}^{'}}{T_v}\bigg)= \left(\dv{w}{t}\right)_{phys},
\label{e_w_1}$$ $$\dv{}{t}\bigg(\frac{q}{c_*^2}+\ln{J_\zeta}\bigg)+\pdv{u}{X}+\frac{1}{\cos\phi}\pdv{(\cos\phi{v})}{Y}+\pdv{\dot{\zeta}}{\zeta}-\frac{g}{c^{2}_{*}} w =\left(\dv{\ln \rho T_v}{t}\right)_{phys}
\label{e_c_1},$$ $$\dv{}{t}\bigg[\ln\bigg(\frac{T_v}{T_*}\bigg)-\frac{q}{c_{pd}T_*}\bigg]+\frac{N^{2}_{*}}{g}w=\left(\dv{\ln T_v}{t}\right)_{phys},
\label{e_t_1}$$ where $q=R_dT_*\ln(p/p_*)$ is the pressure deviation from the basic-state pressure $p_*$, $\dot{\zeta}=\dv{\zeta}{t}$ is the vertical motion with respect to the transformed $\zeta$-coordinate, $N^2_*=g^2/(c_{pd}T_*)$ is the square of the basic-state Brunt-Väisälä frequency and $c^2_*=R_dT_*/(1-\kappa_d)$ is the square of the speed of sound. In Eq. (\[e\_c\_1\]), $\kappa$ is replaced by $\kappa_d=R_d/c_{pd}$ as an approximation. It is also important to note that, the physical forcings associated with the modified continuity equation, given by Eq. (\[e\_c\_1\]), now includes the same diabatic heating term that appears in the thermodynamic equation, given by Eq. (\[e\_t\_1\]).
In the above equations, the terms $J_X$, $J_Y$ and $J_\zeta$ appears due to the vertical coordinate transformation where $J_X=\pdv{z}{X}$, $J_Y=\pdv{z}{Y}$ and $J_\zeta=\pdv{z}{\zeta}$. It is however important to note that the coordinate transformation used to derive the Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]) is incomplete and only the following derivative operators have been transformed: $$\pdv{}{x}=\pdv{}{X}-\frac{J_X}{J_\zeta}\pdv{}{\zeta},
\label{e_trans_x}$$ $$\pdv{}{y}=\pdv{}{Y}-\frac{J_Y}{J_\zeta}\pdv{}{\zeta},
\label{e_trans_y}$$ $$\pdv{}{z}=\frac{1}{J_\zeta}\pdv{}{\zeta},
\label{e_trans_z}$$ $$\dv{}{t}=\pdv{}{t}+u\pdv{}{X}+v\pdv{}{Y}+\dot{\zeta}\pdv{}{\zeta}.
\label{e_lagrange_zeta}$$ As a result, the original vertical velocity $w$ has not been completely eliminated from the system. The system of equations in the $\zeta$-coordinate has its own vertical velocity in the form of $\dot{\zeta}=\dv{\zeta}{t}$, whereas $w=\dv{z}{t}$ remains in the system as a kinematic relation that needs to be dealt with explicitly. [@czg04] have demonstrated that such an approach is perfectly equivalent to a full coordinate transformation. As the treatment of advection in GEM is based on the semi-Lagrangian approach [@hgi17], the kinematic relation defining $w$ is also solved semi-Lagrangially. However, for convenience, the kinematic relation is modified as $$\frac{d}{dt}(z-\zeta)+\dot{\zeta}-w=0,
\label{e_k_1}$$ where Eq. (\[e\_k\_1\]) along with the Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]) constitute the complete system of equations for the GEM-H formulation.
One important aspect of any NWP model is how the effects of the physics forcings, as presented in the RHS of the Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]), are accounted for as the model equations are integrated at each time step. One way is to resolve the dynamical equations in the absence of any physics forcing and then modify the solution with the parameterized forcing as adjustments outside the dynamics step. Another possible approach is to compute the physics forcing and combine their effects with the nonlinear terms in a semi-implicit way within the dynamics step. This important aspect of the GEM model with particular focus on its impact pertaining to the GEM-H dynamical core is discussed in further details in section \[s\_coupling\].\
Spatial grid and discretization
-------------------------------
The objective of this project has been to implement the option of a dynamical core based on height-type vertical coordinate in addition to the existing pressure-type coordinate in the GEM model. The strategy has been to add the new coordinate option with minimal changes to the dynamical core and the rest of the GEM model source code. Therefore, the spatial grid structures in GEM-H, both in the horizontal and the vertical, are kept the same as those in GEM-P, which implies a staggered Arakawa C grid [@ara88] in the horizontal and a staggered Charney-Phillips grid [@cph53] in the vertical. The horizontal and vertical grid structures are presented in Fig. \[f\_grid\].
In addition to being similar to GEM-P for the limited-area model (LAM) grids, the global grid system is also kept unchanged in GEM-H, and is therefore, based on a Yin-Yang grid system (Qaddouri and Lee 2011). The Yin-Yang system combines two overlapping latitude-longitude LAM grids to form a global grid following the Schwarz method for non-matching domain decomposition [@qll08] and thus avoids pole-related singularity and convergence issues associated with a conventional global lat-lon grid. Further details on the Yin-Yang grid are provided by [@qle11].\
Discretization in time
----------------------
The general form of an individual equation in the system comprised of Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]) and (\[e\_k\_1\]) can be expressed as $$\frac{dF_i}{dt}-G_i=P_i
\label{e_gen_1}$$ where $F_i$ denotes the advected quantity for an individual equation $i$ within the system, $G_i$ is the associated dynamics source term with linear and nonlinear components, and $P_i$ denotes the corresponding physics forcing. Similarly to GEM-P, treating the advection terms in a semi-Lagrangian way and applying a two-time-level Crank-Nicholson temporal discretization leads to $$\frac{F_i^A-F_i^D}{\Delta{t}}-\frac{1+b_i}{2}G_i^A-\frac{1-b_i}{2}G_i^D = \mathbf{s_c}\overline{P_i}
\label{e_gen_2}$$ where $\Delta t$ indicates the time-step length, and the superscripts $A$ and $D$ denote the arrival and departure positions of the air parcels at the current time $t$ and the previous time $(t-\Delta t)$, respectively. The integrals of the source terms $G_i$ for the different dynamical equations are approximated by trajectory averages. The parameter $b_i$ denotes the off-centering weight factor for the averaging of the dynamics source terms. When $b_i=0$, the averaging of the source term is fully centered, whereas $b_i>0$ implies additional weight placed on the implicit component of the source term.
Historically, off-centering was implemented in GEM-P primarily to address spurious resonance originating from stationary orographic forcing [@rsr94]. However, it also suppresses computational noise and improves numerical stability. These other beneficial impacts have been found to be equally important in the current and previous implementations of GEM-P for the different operational NWP systems at ECCC. As the principal objective of this study is to have a GEM-H dynamical core that is equivalent to GEM-P for the different operational GEM-based NWP systems, off-centering has been retained in GEM-H. Also, following the latest implementation of GEM-P [@hgi17], a differential approach for off-centering has been adopted for GEM-H where $b_i$ varies depending on the dynamical equation denoted by the subscript $i$. At present, the system of equations, given by Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]) and (\[e\_k\_1\]), are separated into three groups with an associated off-centering parameter for each group as follows:
(i) [$b_m$ for the horizontal momentum equations \[Eqs. (\[e\_u\_1\])–(\[e\_v\_1\])\],]{}
(ii) [$b_h$ for the continuity and thermodynamic equations \[Eqs. (\[e\_c\_1\])–(\[e\_t\_1\])\], and]{}
(iii) [$b_{nh}$ for the vertical momentum and kinematic equations \[Eqs. (\[e\_w\_1\]) and (\[e\_k\_1\])\].]{}
On the RHS of Eq. (\[e\_gen\_2\]), the term $\overline{P_i}$ denotes the parameterized physics source term and the parameter $\mathbf{s_c}$ indicates the mode of coupling between physics and dynamics. Depending on the chosen method for dynamics-physics coupling, the value of $\mathbf{s_c}$ can be either 0 or 1. Also, the approach for dynamics-physics coupling determines how the contribution of $\overline{P_i}$ is accounted for in the model. Further discussions regarding the coupling of the parameterized physics forcing with the dynamical core is presented in section \[s\_coupling\].
Trajectory calculations
-----------------------
Semi-Lagrangian treatment of advection requires the solution of kinematic displacement equations of the form
$$\begin{aligned}
\dv{X}{t}=u, \dv{Y}{t}=v, \\
\dv{\zeta}{t}=\dot{\zeta},
\end{aligned}$$
\[e\_traj\_1\]
to determine the departure positions of the air parcels. In the context of GEM-P, [@hgi17] have shown that the consistency in the numerical discretizations between the dynamical and trajectory equations is fundamentally important for accurate solution of the advection problem. In order to be numerically consistent, similarly to the treatment of the dynamics source term in Eq. (\[e\_gen\_1\]), the averaging of the velocities in Eq. (\[e\_traj\_1\]) needs to be done using the trapezoidal rule. Furthermore, the interpolation scheme employed to determine the wind field at the departure positions for the trajectory calculations need to be the same as the one applied to determine the source terms in the dynamical equations at the departure positions. In the case of GEM-P, cubic interpolation is used for both the wind field and the dynamical source terms to achieve numerical consistency. Following the conclusions of [@hgi17], similar consistent trajectory calculation approach is adopted in GEM-H, i.e., trapezoidal rule for evaluating the integral of the source term in Eq. (\[e\_traj\_1\]) along with cubic interpolation to determine the wind field at the departure positions.
The elliptic problem
--------------------
In order to solve the system of equations associated with the GEM-H formulation, each equation of the form (\[e\_gen\_2\]) is rearranged to separate the linear and nonlinear components of the implicit part and is expressed as $$L_i=R_i-N_i,
\label{e_lnr}$$ where $L_i=(F^A_i/\tau_i-G^A_i)_{linear}$, $N_i=F^A_i/\tau-G^A_i-L_i$, and $R_i=F^D_i/\tau_i+\beta_iG^D_i+\mathbf{s_c}\overline{P_i}$ with $\beta_i=(1-b_i)/(1+b_i)$ and $\tau_i=\Delta t(1+b_i)/2$. Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]) and (\[e\_k\_1\]) are rearranged as in Eq. (\[e\_lnr\]), giving $$L_u = \frac{u}{\tau_m} + \delta_Xq - \mathbf{s_i } J_X\overline{J^{-1}_\zeta\delta_\zeta q}^{X\zeta},
\label{e_l_u}$$ $$L_v = \frac{v}{\tau_m} +\delta_Yq - \mathbf{s_i } J_Y\overline{J^{-1}_\zeta\delta_\zeta q}^{Y\zeta},
\label{e_l_v}$$ $$L_w = \frac{w}{\tau_{nh}} + (\mathbf{s_i}J^{-1}_\zeta + \mathbf{s_d})\delta_\zeta q - g\frac{T^{'}_v}{T_v},
\label{e_l_w}$$ $$L_c=\frac{1}{\tau_h}\bigg(\frac{q}{c_*^2}+\ln{J_\zeta}\bigg)+\delta_Xu+\frac{1}{\cos_\phi}\delta_Y(\cos\phi_v)+\delta_\zeta\dot{\zeta}-\varepsilon\overline{w}^{\zeta},
\label{e_l_c}$$ $$L_T=\frac{1}{\tau_h}\bigg(\frac{T^{'}_v}{T_v}-\frac{\overline{q}^\zeta}{c_{pd}T_*}\bigg)+\mu w,
\label{e_l_t}$$ $$L_z=\frac{z-\zeta}{\tau_{nh}}+\dot{\zeta}-w,
\label{e_l_k}$$ where the symbol $\delta_i$ denotes the finite difference operator along the $i$-direction, and the overline operator $\overline{( )}^j$ implies spatial averaging in the $j$-coordinate. For convenience of notation the terms $\frac{g}{c^{2}_{*}}$ and $\frac{N^{2}_{*}}{g}$ have been replaced by $\varepsilon$ and $\mu$, respectively. The corresponding nonlinear components $N_i$ associated with the discretized forms of the Eqs. (\[e\_u\_1\])–(\[e\_t\_1\]) and (\[e\_k\_1\]) as well as the associated RHS terms $R_i$ are provided in Appendix A. The parameters $\mathbf{s_i}$ and $\mathbf{s_d}$ in the above equations denote the choice of the solver for the dynamical core, where the subscripts $i$ and $d$ stand for $iterative$ and $direct$ respectively. Based on the selected solution approach, these parameters can be either 1 or 0, and are mutually exclusive such that $\mathbf{s_i}=1-\mathbf{s_d}$. Further discussion on the solution approaches is presented in Section \[s\_solver\]. As shown by [@cst88], the solution of the system of equations of the type (\[e\_lnr\]) requires nonlinear iterations for convergence, where the nonlinear terms $N_i$ are re-evaluated during each iteration using the latest values of the prognostic variables. Furthermore, Crank-Nicholson iterations are required, where the $R_i$ terms are re-evaluated during each iteration at the departure positions calculated using the latest velocity estimates. At present, the GEM-based operational NWP systems at ECCC utilize two Crank-Nicholson iterations and within each Crank-Nicholson step two nonlinear iterations are carried out. As a result, irrespective of the solution approach, the solver is called four times during each dynamical time step.
The discretized equations with the left hand sides (LHSs) given by Eqs. (\[e\_l\_u\])–(\[e\_l\_k\]) are then reduced into a single elliptic boundary value (EBV) problem through elimination of variables, where the LHS of the final elliptic problem has the form $$L^{'''}_c=\delta_XA+\frac{1}{\cos\phi}\delta_Y[\cos\phi B]+\delta_\zeta C-\varepsilon\overline{C}^\zeta-\gamma q,
\label{e_l_ebv}$$ where $A=\bigg(\delta_Xq - \mathbf{s_i } J_X\overline{J^{-1}_\zeta\delta_\zeta q}^{X\zeta}\bigg)$, $B=\bigg(\delta_Yq - \mathbf{s_i } J_Y\overline{J^{-1}_\zeta\delta_\zeta q}^{Y\zeta}\bigg)$ and $C=\Gamma\bigg[(\mathbf{s_i}J^{-1}_\zeta+\mathbf{s_d})\delta_\zeta q-\mu\overline{q}^\zeta\bigg]$. Also, in Eq. (\[e\_l\_ebv\]), $\gamma=1/(c^2_*\tau_h\tau_m)$ and $\Gamma=1/(\tau_m/\tau_{nh}+N^2_*\tau_h\tau_m)$. It is important to note that, with $\mathbf{s_i}=1$, the terms $A$, $B$ and $C$ (with $\mu=0$) are simply the components of the gradients in the terrain-following coordinate system. The sequence of steps involved in deriving the EBV problem, i.e., the final form of $L^{'''}_c$, is provided in Appendix B.
Initial and boundary conditions
-------------------------------
As with the case of any initial value problem, in order to initiate integration in time, the GEM-H dynamical core requires initial values of all the prognostic variables. At present, ECCC’s operational data assimilation system provides analyzed initial values for the horizontal wind components $u$ and $v$, virtual temperature $T_v$ and surface pressure $p_s$. The remaining prognostic variables $w$, $\dot{\zeta}$ and $q$ are computed in a diagnostic manner at time $t=0$. The initial value of $q$ is obtained from the analyzed surface pressure $p_s$ with a hydrostatic approximation. Substituting $\dv{w}{t}=0$ in Eq. (\[e\_w\_1\]) gives $$\frac{1}{J_{\zeta}}\pdv{q}{\zeta}=g\frac{{T_v}^{'}}{T_v}
\label{e_hydro}$$ as a hydrostatic approximation. The value of $q$ at the different model levels are then obtained by integrating Eq. (\[e\_hydro\]) where at the surface, due to the hydrostatic approximation, $q_s=R_dT_*\ln (p_s/p_*)$. The initial value of $\dot{\zeta}$ is computed by assuming $\pdv{\rho}{t}=0$ at time $t=0$ in the continuity equation. In the $\zeta$-coordinate, this takes the form $$\pdv{}{X}\bigg(\rho u \pdv{z}{\zeta}\bigg)+\frac{1}{\cos\phi}\pdv{}{Y}\bigg(\cos\phi\rho v \pdv{z}{\zeta}\bigg)+\pdv{}{\zeta}\bigg(\rho \dot{\zeta} \pdv{z}{\zeta}\bigg)=0,
\label{e_zdot_ini}$$ which is then discretized to compute the initial value of $\dot{\zeta}$. Once $\dot{\zeta}$ is known, the initial value of $w$ is obtained from its definition in the $\zeta$-coordinate. i.e., $w\equiv\dv{z}{t}=uJ_X+vJ_Y+\dot{\zeta}J_\zeta$.
The boundary conditions for the upper and lower boundaries are given by $\dot{\zeta}_T=\dot{\zeta}_S=0$. This implies that the vertical motion in the $\zeta$-coordinate vanishes at the surface and the model top which is flat. For LAM problems, GEM-H also requires lateral boundary conditions which are obtained from the driving fields. As the global Yin-Yang system is based on two interacting geometrically identical LAM domains, it therefore similarly requires lateral boundary conditions. In this case, the boundary conditions for one sub-domain (Yin or Yang) depend on the solution in the other. Thus the solution of the global problem is obtained by iteratively solving the two sub-problems separately and updating the values in the overlapping region until a certain convergence criteria is satisfied [@qle11].
Solution of the EBV problem {#s_solver}
===========================
The EBV problem to be resolved by GEM-H at each model time-step can be expressed as $$\nabla_\zeta^2q + \mathbf{M}q=\mathbf{R},
\label{e_ebv_gf}$$ by replacing $L^{'''}_c$ in Eq. (\[e\_l\_ebv\]) with $(\nabla^2_\zeta+\mathbf{M})q$, where $\nabla_\zeta^2=(\delta_{XX} + \frac{1}{\cos \phi}\delta_Y (\cos\phi \delta_Y))$ is a discretized two-dimensional horizontal operator in the $\zeta$-coordinate, $\mathbf{M}$ contains all the remaining terms of $L^{'''}_c$ that include the discretized difference and averaging operators in the horizontal and vertical dimensions, $q$ is the unknown and $\mathbf{R}$ includes the explicit RHS terms as well as the implicit nonlinear terms. It is important to note that the nonlinear terms in $\mathbf{R}$ require iterations for convergence, irrespective of the solution approach. At present, the GEM dynamical core uses two iterations for the sufficient convergence of the nonlinear terms and two iterations for trajectories. As a result, the solver is called into action four times during each dynamical step.
Once Eq. (\[e\_ebv\_gf\]) is solved to obtain the unknown $q$, the other prognostic variables are obtained through back substitution as presented in Appendix C. As has been mentioned earlier, two general approaches are available for solving the elliptic problem – direct and iterative. The selection of these approaches depends on which terms are included in $\mathbf{M}$, and is determined by the values of the terms $\mathbf{s_i}$ and $\mathbf{s_d}$ in Eqs. (\[e\_l\_u\])–(\[e\_l\_t\]).
The direct solver
-----------------
The direct solver works by first decoupling Eq. (\[e\_ebv\_gf\]) in the vertical. It is achieved through the expansion of the unknown $q$ and the RHS $\mathbf{R}$ in terms of the eigenvectors that diagonalizes the operator $\mathbf{M}$ [@qle11]. This is only possible when the operator $\mathbf{M}$ does not include contributions from the metric terms arising from the vertical coordinate transformation that involves horizontally variable coefficients imparting horizontal coupling. Therefore, the implementation of direct solver in GEM-H requires a ‘simplified approach’ where all the metric terms of the relevant discretized equations are treated as nonlinear terms. This is achieved by setting $\mathbf{s_d}=1$. For $N_k$ number of vertical levels used in the model, vertical separation reduces Eq. (\[e\_ebv\_gf\]) to a set of $N_k$ independent horizontal Helmholtz problems of the form $$\nabla_\zeta^2\tilde{q} + m\tilde{q}=\tilde{\mathbf{R}},
\label{e_helmh}$$ where $\tilde{q}$ and $\tilde{\mathbf{R}}$ are the vertical projections of $q$ and $\mathbf{R}$, respectively, and $m$ is the eigenvalue of the operator $\mathbf{M}$. The horizontal solution of the algorithm then proceeds by expanding $\tilde{q}$ and $\tilde{\mathbf{R}}$ in terms of the eigenvectors that diagonalize the $X$-component of the two-dimensional operator $\nabla_\zeta^2$. For $N_i$ number of grid points along the longitude $X$, this leads to $N_i$ independent tridiagonal problems of $N_j$ dimension for each model vertical level, where $N_j$ denotes the number of grid points along the latitude $Y$. The total number of tridiagonal problems to solve is therefore $N_k\times N_i$. Solution to the tridiagonal problems are computed by Gaussian elimination without pivoting, and afterwards, the final three-dimensional solution $q$ is reconstituted [@qle10].
The ‘simplified approach’ has been primarily implemented in GEM-H to take advantage of the computational performance of the direct solver. The direct solver uses Fast Fourier Transform to compute the horizontal solution $\tilde{q}$ and outperforms any iterative approach implemented at the ECCC by a substantial margin for the configurations of the various GEM-based NWP systems running operationally. The ‘simplified approach’ thus makes the application of GEM-H for such configurations feasible and keeps the option of a future replacement of the GEM-P core with GEM-H. The simplified approach, however, works as long as the vertical-horizontal coupling, imparted through the metric terms, is not too significant so that these terms can be treated efficiently through nonlinear iterations. This approach works unless the maximum terrain slope is not sufficiently steep (less than 30$^\circ$) which is generally the case for most of the operational GEM-based NWP systems. However, with increasing spatial resolution, the slopes in grid-scale orography also increase, particularly over complex terrain, which leads to increased vertical-horizontal coupling and, at one point, makes the ‘simplified approach’ inapplicable.
The iterative solver
--------------------
When all the metric terms in $A$ and $B$ (see Eq. \[e\_l\_ebv\]) are included in $\mathbf{M}$, the resulting vertical-horizontal coupling makes the problem non-separable. This is done by setting $\mathbf{s_i}=1$. In such a scenario, the three-dimensional equation of the form (\[e\_ebv\_gf\]) can only be solved at each time step by using an iterative solver. The current iterative solver for the EBV problem in GEM-H is based on the flexible generalized minimal residual (FGMRES) method [@saa93; @qle10].
The fully discretized system of equations of the form (\[e\_ebv\_gf\]) can be further generalized as $$\mathbf{A}q=\mathbf{R}
\label{e_axb}$$ where coefficient matrix $\mathbf{A}$ contains the discretized operator $(\nabla_\zeta^2 + \mathbf{M})$. The FGMRES method approximates the solution in a Krylov sub-space of small dimension by minimizing the Euclidean norm of its residual. A major advantage of such a method is that instead of explicitly generating the coefficient matrix $\mathbf{A}$, one only needs to compute the vector resulting from the action of the underlying operator $(\nabla_\zeta^2 + \mathbf{M})$ on the vector $q$. Efficient functioning of such an iterative solver, however, requires a pre-conditioner. At present, the pre-conditioner is based on the block Jacobi iteration for the EBV in Eq. (\[e\_ebv\_gf\]), where all the metric terms in $\mathbf{M}$ are absent. This pre-conditioner improves the convergence rate of the FGMRES solver. However, the time of execution is still high compared to the fast direct solver. Significant research is currently underway at ECCC to devise iterative solvers that are competitive with the direct approach and will work for both GEM-P and GEM-H. At present, the current implementation of the iterative solver in GEM-H—although not as efficient—provides the necessary reference for the direct solver approach.
Dynamics-physics coupling {#s_coupling}
=========================
Along with the dynamical core, parameterization of the subgrid-scale physical processes constitutes the other fundamental component of any NWP model. Coupling between the dynamical core and the parameterized subgrid-scale physical processes is of critical importance. How to devise the most appropriate coupling strategy is still an unsettled question [@bbb18]. This issue is being actively studied at ECCC. However, it is not the objective of this study to delve deep into the fundamental questions around dynamics-physics coupling. Rather, in this section, the issue of coupling the RPN Physics Package with the GEM dynamical core is discussed in order to determine which approach is the most feasible for GEM-H among the options that are available for GEM-P.
It is important to note that, during every model time step in GEM, the RPN Physics Package is executed after the dynamical equations have been resolved by the dynamical core and thus the physics schemes utilize the outputs of the dynamics step as inputs. However, irrespective of the vertical coordinate used in the dynamical core, the physics schemes use a traditional $\sigma$-coordinate in the vertical, defined as $$\sigma=\frac{\pi}{\pi_s},
\label{e_sigma}$$ where $\pi$ is the hydrostatic pressure. Also, the physics schemes work within a one-dimensional configuration where each processor core only has access to the vertical structure of the meteorological fields associated with a single horizontal grid point. The various physical processes are parameterized sequentially where the tendencies estimated by one parameterization scheme affects the ones that follow. The parameterization sequence during each physics step initiates with the radiation scheme and is followed by the parameterizations of the surface processes, gravity wave drag, boundary layer turbulence, convection and grid-scale condensation. At the end of the physics step, the tendencies from the different physical parameterization schemes are aggregated to compute the grid-scale tendencies for the wind components, temperature, water vapor and the hydrometeors.
Different coupling approaches in GEM
------------------------------------
Particularly in the context of GEM-P, two approaches are presently available to couple the RPN Physics Package with the GEM dynamical core. A brief discussion on these methods will be helpful in establishing the rationale behind the approach selected for the GEM-H dynamical core.
(i) [Split method: As has been mentioned earlier, $\mathbf{s_c}$ determines the mode of coupling between dynamics and physics. If $\mathbf{s_c}=0$ then the dynamical equations are resolved in the absence of any physical forcing and at the end of the dynamics step their contributions are incorporated as adjustments in the so called ‘split mode’. In the absence of physics forcing, Eq. (\[e\_gen\_2\]) becomes $$\frac{F_i^{A*}-F_i^D}{\Delta{t}}-\frac{1+b_i}{2}G_i^{A*}-\frac{1-b_i}{2}G_i^D = 0
\label{e_gen_3}$$ where $F_i^{A*}$ is the interim solution of the dynamics step. Once Eq. (\[e\_gen\_3\]) is resolved, the physics source term is then applied as grid point adjustments as follows $$(\delta F)_{phys}=F_i^{A} - F_i^{A*} = \Delta t P_i.
\label{e_gen_4}$$ Thus, in the split method, the dynamics step predicts an inviscid and adiabatic solution that is modified through adjustments attributable to the parameterized physics forcings in order to obtain the complete solution at the end of each model time step. ]{}
(ii) [Explicit method: The second option for dynamics-physics coupling treats the physics source terms explicitly by setting $\mathbf{s_c}=0$ and replacing $\overline{P_i}$ by $P_i^D$. This method is referred to as the ‘explicit method’ and moves by solving equations of the following form at each model time step $$\frac{F_i^A-F_i^D}{\Delta{t}}-\frac{1+b_i}{2}G_i^A-\frac{1-b_i}{2}G_i^D = P_i^D.
\label{e_gen_7}$$ In this approach, physics tendency $P_i$ from the previous time step is combined with the RHS term $R_i$, followed by the determination of $R_i^D$ at the departure positions in a semi-Lagrangian way. In other words, the explicit method works by directly incorporating the impact of physics forcings as tendencies into the discretized dynamical equations.]{}
Coupling in GEM-H
-----------------
Although both of the aforementioned coupling methods are available for GEM-P, it is important to note that all the operational NWP systems based on GEM-P at present utilize the split method for dynamics-physics coupling. Nevertheless, there exists strong concern about the split method in general, and a brief discussion highlighting the pertinent issues will be helpful.
In the context of model formulations for both GEM-P and GEM-H, the term $F_i$ does not necessarily coincide with the prognostic model variables. For example, in the case of GEM-P in its hydrostatic mode, as presented by [@gdm14]:
$$\begin{aligned}
F=\left\{u,v,Bs+\ln(1+\pdv{B}{\zeta}),\ln\frac{T_v}{T_*}-\kappa_dBs\right\}, \\
P=\left\{\dv{}{t}\bigg(u,v,\ln\rho,\ln T_v\bigg)\right\}_{phys},
\end{aligned}$$
\[e\_couple\_gemp\]
whereas the prognostic variables are $u$, $v$, $s$, $\dot{\zeta}$, and $T_v$. Therefore, in the actual model implementation of the split method, the prediction from the dynamics step is utilized to compute the interim state of the prognostic variables and adjustments are then applied to these variables at the end of the physics step. It is important to note that in this case, the only adjustments that have been found to not result in any issue of major concern are $(\delta u)_{phys}$, $(\delta v)_{phys}$ and $(\delta T_v)_{phys}$. Also in GEM-P, adjustments are required for density, as in effect $$\left(\dv{\ln \rho}{t}\right)_{phys}\equiv\dv{}{t}\ln(1+r_w).
\label{e_split_density}$$ Although water vapor and hydrometeors are updated through physical parameterizations, the only variable that could be updated in the split mode appears to be $s$ through a surface pressure adjustment $\delta\pi_s$ that may be computed as $$\delta\pi_s=\int{\delta \ln(1+r_w)d\pi},
\label{e_delta_pis}$$ which takes into account the net inflow/outflow of mass through the Earth’s surface at every model grid point. Unfortunately, the vertical distribution of this change in mass through water vapor and precipitation fluxes cannot be correctly accounted for in the split mode, and is found to produce considerable noise in the wind forecast.
Furthermore, in the context of three-time-level discretization, [@clz98] have shown that the split method can lead to erroneous results for long time steps that are permissible with the semi-implicit semi-Lagrangian models. Similar conclusions were drawn from a theoretical analysis for two-time-level schemes by [@swc02]. Figure \[f\_split\]a shows the geopotential height contours at 400 hPa from a 72-h global forecast with ECCC’s 25-km resolution Global Deterministic Prediction System (GDPS) using the GEM-P dynamical core. The results correspond to split method for dynamics-physics coupling. Although the distribution of geopotential height for the meso and large scales does not reveal any issue of immediate concern, when one looks at the smallest scales, i.e., scales of about a few grid lengths, some spurious computational noise is visible (see Fig. \[f\_split\]b). Historically, the operational NWP systems at ECCC have been utilizing spatial filters over the model-predicted meteorological fields of interest, like the geopotential height, to smooth out any computational noise in the model outputs. As a result, the kind of computational noise shown in Fig. (\[f\_split\]b) has not been troublesome for the meteorologists using ECCC’s operational NWP outputs. Nevertheless, the computational noise associated with the split method remains as a concern. However, as shown in Fig. \[f\_split\]c, when the split method for coupling is replaced by the explicit method, the noise in the geopotential height disappears. It should be noted that, although explicit coupling can impose stability limitations in terms of the acceptable length of time steps, ECCC’s operational NWP system configurations are found to function with explicit coupling without requiring any adjustments to the time-step lengths.
Similarly to GEM-P, the $F_i$ terms do not coincide with the prognostic variables for all of the GEM-H model equations, where
$$\begin{aligned}
F=\left\{u,v,w,\frac{q}{c^2_*}+\ln J_\zeta,\ln\frac{T_v}{T_*}-\frac{q}{c_{pd}T_*}\right\}, \\
P=\left\{\dv{}{t}\bigg(u,v,w,\ln\rho T_v,\ln T_v\bigg)\right\}_{phys}.
\end{aligned}$$
\[e\_couple\_gemh\]
Particularly, the presence of the physics forcing term $\left(\dv{\ln \rho T_v}{t}\right)_{phys}$ in the modified continuity equation (\[e\_c\_1\]) poses an additional challenge for the GEM-H formulation, as far as the split method is concerned. If one attempts to apply this tendency as a hydrostatic adjustment to pressure in the split mode then of course its effect is not limited to the continuity equation alone. Rather, applying the adjustments to pressure, i.e. changing $q$ in the case of GEM-H, also affects the thermodynamic equation. Such an adjustment leads to over compensation and is found to result in spurious bias in the temperature in upper troposphere and stratosphere, which is unacceptable (not shown). Also, the jet-level wind is found to be adversely affected. As a result, in GEM-H the physics contributions are accommodated through the explicit method by including them as explicit grid-scale tendencies in the RHS of the discretized dynamical equations. For a fair comparison between the new and the existing dynamical core, the results for both the GEM-P and GEM-H cores presented in the rest of this paper are obtained with explicit dynamics-physics coupling. It is also important to mention that, even though parameterization of physical processes like boundary-layer turbulence can modify vertical motion $w$, at present the impact of physics on $w$ is neglected. This is the case for both GEM-P and GEM-H.
A hybrid ‘split-explicit method’ has also been tested with GEM-H, where the physics contributions to the thermodynamic and horizontal momentum equations are accounted for through the ‘split method’ while the contributions to the continuity equation is accommodated using the ‘explicit method’. Such a hybrid approach with GEM-H produces results that are equivalent to those obtained with the ‘split method’ for GEM-P. However, questions remain about the numerical consistency of the hybrid split-explicit method.
It is important to mention that the current implementation of the RPN Physics Package, when coupled with the GEM-P (GEM-H) dynamical core through the explicit method, leads to some deterioration in temperature bias in the upper troposphere compared to the split (split-explicit) method. This implies that further research is necessary to have a more consistent dynamics-physics coupling with the explicit method. Particularly, the grid-scale condensation scheme in GEM, for 10 km or coarser horizontal resolutions, has been found to exhibit large sensitivity with the explicit method which leads to under-prediction of clouds (R. McTaggart-Cowan, ECCC, personal communication). This implies that some process-specific adjustments in the computation of the relevant physics tendencies may be required to improve the overall dynamics-physics coupling. The challenges imposed by the process-specific issues are also being explored by other operational NWP centers [@bbb18]. Currently, work is underway at ECCC to explore the various issues within the coupling interface as well as the parameterizations of the different physical processes to improve the dynamics-physics coupling in general. Further discussion on this issue, however, is beyond the scope of this paper.
Evaluation of GEM-H {#s_evaluation}
===================
One of the most important objectives of this study has been to develop a dynamical core for the GEM model that utilizes a height-based TFC and is capable of producing predictions that are equivalent to the results obtained by the pressure-based dynamical core. In order to evaluate the consistency and performance of the new dynamical core, a number of numerical experiments covering a wide range of scales—ranging from microscales to the meso and synoptic scales—have been carried out. These include two-dimensional theoretical test cases involving bubble convection [@rob93] and nonhydrostatic mountain waves [@slf02] as well as three-dimensional global NWP. The two-dimensional test cases selected in this paper have become ubiquitous tools in testing the consistency and performance of nonhydrostatic dynamical cores. Also, the availability of the GEM-P core provides the opportunity to have reference solutions for all these cases.
Robert’s bubble convection case
-------------------------------
[@rob93] presented a two-dimensional theoretical case involving the evolution of a warm bubble within a dry isentropic atmosphere. Initially, the bubble has a diameter of 500 m and is placed 10 m above a flat surface within a 1 km $\times$ 1 km computational domain, and has a uniform potential temperature of 30.5$^\circ$C. Also, the basic-state atmosphere is at rest under a hydrostatic equilibrium with an isentropic basic-state temperature of 30$^\circ$C. As the bubble has a potential temperature excess of 0.5$^\circ$C compared to the surrounding atmosphere, it rises due to the action of the buoyancy force. The absence of any orographic variation makes this experiment an excellent benchmark to test the functioning of advection and buoyancy within a dynamical core during the early stages of its development.
The numerical experiment for bubble convection is carried out with a spatial grid resolution of 10 m and a time step of 5 s. No explicit numerical diffusion is used. The resulting evolution of the bubble, in terms of its potential temperature distribution at two different times (7 min and 10 min), is presented in Fig. \[f\_bubble\] for both GEM-P and GEM-H. The bubbles predicted by the two cores initially deform into a somewhat mushroom-like shape (see at $t=7$ min) and then are deformed further (at $t=10$ min). Overall, the predictions from GEM-P and GEM-H are equivalent for the entire range of scales - from the large to the smallest scales. Such a good resemblance between the two predictions imply negligible impact of the choice of the vertical coordinate and the other modifications in model formulations in the absence of any orographic variation at the model surface. It also indicates that the representation of the advection and buoyancy effects are comparable between the two GEM cores. It is important to note that, due to considerable differences in the model formulations and spatiotemporal discretizations, it is difficult to compare the evolution of the bubbles between two completely separate models in a quantitative manner. Only qualitative comparisons are feasible. Therefore, the lesser resemblance between the results from [@rob93] and the GEM dynamical cores are not unusual. Although the predictions from the two GEM cores have some large-scale resemblance to the results presented by [@rob93], significant differences appear at the upper half of the bubble - particularly at $t=10$ min. However, the upper structure of the bubble compares better with the predictions by [@spu92]. Also, because of the implicit dissipation associated with the semi-Lagrangian approach, the GEM solution does not suffer from computational noise like models based on Eulerian advection [@jua00]. Overall, as GEM-P is being used operationally at ECCC, the resemblance between GEM-H and GEM-P is of more significant importance, as it confirms consistency of the GEM-H formulation and a neutral impact of the vertical coordinate modification on buoyancy and advection.
Schär’s mountain case
---------------------
[@slf02] presented a linear two-dimensional theoretical test case of mountain waves which is an excellent benchmark for verifying nonhydrostatic dynamical cores, particularly in determining the presence of possible inconsistencies in the numerical details [@hgi17; @mdw10; @gbd05; @ksf03]. The bottom boundary profile of the idealized mountain for this case is defined by $$z_s=z_0 e^{{-(x/a)}^2}\cos^2(\pi x/l_x),
\label{e_schaer_mtn}$$ where $z_0$=250 m, $l_x$=4 km, $a$=5 km, and $\pi$ is the conventional mathematical constant. The upstream flow conditions are given by uniform upstream wind $U$=10 m s$^{-1}$, upstream surface temperature $T_{surf}$=288 K, upstream surface pressure $p_0$=1000 hPa, and a constant Brunt-Väisälä frequency $N_*$=0.01 s$^{-1}$. All other conditions for the simulations of this test case is similar to those presented by [@hgi17].
One major advantage of Schär’s mountain case is the availability of a steady-state analytical solution of the corresponding to the linearized problem as a reference [@slf02]. The simulated quasi-steady vertical velocity obtained after 4 hours of integration with both the GEM-P and GEM-H dynamical cores are presented in Fig. \[f\_schaer\_1\]. The analytical solution of the problem, as presented by [@slf02], is a combination of rapidly decaying small-scale nonhydrostatic mountain waves close to the surface and large-scale hydrostatic waves extending to much higher altitudes. As can be seen in Fig. \[f\_schaer\_1\], the solutions with both types of vertical coordinates generate these two regimes of the mountain waves and are very similar. Results shown in this figure represent simulations that have been carried out without any off-centering in the discretized dynamical equations and with consistent trajectory calculations, i.e., integration of the wind field in the trajectory equations is based on the trapezoidal rule while the wind field at the departure positions is obtained with cubic interpolation. Inconsistent trajectory calculations were found to produce similar distortion in the large-scale hydrostatic waves with GEM-H (not shown) as has been found for GEM-P earlier by [@hgi17]. Although the GEM-H results presented here corresponds to the direct solver based on the simplified approach, results with the iterative solver is found to be almost identical (not shown). Figure \[f\_schaer\_2\] reveals that off-centering in the discretized dynamical equations leads to distortions in the vertical velocity distribution, irrespective of the type of the vertical coordinate. The solutions presented here are obtained with uniform off-centering involving $b_m=b_h=b_{nh}=0.2$, which are the standard values used in the current GEM-based operational NWP systems. The results correspond to $\Delta t$=32 s, for which the maximum Courant number is approximately 0.76. Reducing the time step to even 4 s is unable to remove these distortions. Also, reducing the level of off-centering is found to reduce the level of distortion in the mountain waves, but the distortions are only completely eliminated when $b_m=b_h=b_{nh}=0$ (not shown). This conforms to the conclusions drawn by [@hgi17] in the context of GEM-P. Furthermore, as has been shown by [@hgi17], consistent trajectory calculations in the presence of off-centering necessitates off-centered averaging applied to the integrals of the source term on the RHS of Eq. (\[e\_traj\_1\]). With uniform off-centering applied to the discretized dynamical equations, the discretized trajectory equations also require uniform off-centering of the same degree. Figure \[f\_schaer\_3\] reveals that in the presence of consistent off-centering in the trajectory calculations, the distortions in the vertical velocity distribution are eliminated for the both dynamical cores. In the presence of differential off-centering, i.e., with different values of $b_h$ and $b_{nh}$ for hydrostatic and nonhydrostatic contributions in the system of dynamical equations, a similar differential approach is required for off-centering in the discretized trajectory equations. As has been shown for GEM-P by [@hgi17], in order to achieve numerical consistency, the off-centering in the discretized source terms in Eq. (\[e\_traj\_1\]a) and (\[e\_traj\_1\]b) need to be equal to the values of $b_h$ and $b_{nh}$ used in the discretized dynamical equations, respectively.
Global deterministic prediction
-------------------------------
A series of 5-day global forecasts, with 25 km horizontal grid spacing, has been carried out covering winter and summer periods for the Northern Hemisphere to compare the predictions by GEM-H and GEM-P from NWP standpoint. Each seasonal period includes 44 cases where the initialization between two consecutive cases are apart by 36 hours. The first summer and winter cases start at 0000 UTC of 25 June 2014 and 19 December 2014, respectively. The global predictions have been obtained with uniform off-centering in the discretized dynamical equations ($b_m=b_h=b_{nh}=0.2$). [@hgi17] have shown that inconsistent off-centering has negligible effects for three-dimensional NWP applications. Therefore, the global forecasts are carried out without any off-centering in the discretized trajectory equations. Furthermore, as has been mentioned in section \[s\_coupling\], physics is coupled with dynamics through the explicit method for both GEM-H and GEM-P for all the tests presented in this section.
First, comparisons are made in the spectral space by comparing the variance spectra associated with the different meteorologically important fields. In order to compute the spectral variance of any meteorological field, it is first interpolated from the Yin-Yang grid to a global Gaussian grid. Afterwards, variance spectra of the field are calculated by decomposing the field at a given pressure level using the spherical harmonics. Figure \[f\_spectra\_gz\] shows the spectral variance of the geopotential height and temperature fields for 120-h forecasts at three different pressure levels for an average over 10 cases for the winter period. The results show spectral similarity between GEM-P and GEM-H for the entire range of scales resolved by the global model - from synoptic to mesoscales. The spectra of kinetic energy and vertical velocity are presented in Fig. \[f\_spectra\_ke\]. The spectral slope of kinetic energy is critical for accurate representation of atmospheric dynamics, and as can be seen in this figure, both GEM-H and GEM-P have the same spectral slope at the synoptic and mesoscales. The vertical motion is also important, particularly for physical processes like convection. Fig. \[f\_spectra\_ke\] shows close spectral similarity between the vertical motions from the two dynamical cores. This implies that changing the vertical coordinate has negligible sensitivity to the extremely important physical process like convection and convection-driven precipitation. Also, the comparisons in the spectral space confirms that the height-based TFC does not lead to any spurious noise or damping in the meteorological fields for any model-resolved length scale.
Objective forecast scores are computed by comparing the model predictions against radiosonde observations at different pressure levels. The evaluation is based on the bias and standard deviation of error (SDE) in model predictions for the individual cases as well as for the average of the 44 cases covering each seasonal period. Figure \[f\_arcad\_1\] presents the vertical profile of error in the predictions from GEM-P (blue) and GEM-H (red) for the winter period. These figures represent global average scores of 120-hour forecast from 44 winter cases for zonal wind (UU), wind speed (UV), geopotential height (GZ), and temperature (TT). An important thing to note while reading this figure is the presence of the statistical confidence scores at the different pressure levels along the left and right vertical axes of the individual subplots for bias and SDE, respectively. A confidence value (in %) shaded in blue (red) color implies statistically significant improvement obtained with the GEM-P (GEM-H) core with respect to the other. The confidence score for the average of SDE and bias are estimated by applying the $F$-test and $t$-test, respectively. Figure \[f\_arcad\_1\] reveals that although there are small differences in the bias for the average of the 44 winter cases, there is no statistically significant difference in the SDE. When tested in the absence of physics forcings, no statistically significant difference is found between the two dynamical cores in either bias or SDE (not shown). The meteorological fields are interpolated from the TFC of the dynamics to the $\sigma$-coordinate for physics through vertical interpolation which can lead to small differences in the vertical for the different definitions of the TFC. Physical parameterizations can be sensitive to the position of the vertical levels and is apparently responsible for the small bias differences shown in Fig. \[f\_arcad\_1\]. Even though small differences in bias are present, the objective scores from GEM-P and GEM-H can be safely assumed to be equivalent as a whole. During the summer period, the two dynamical cores have also been found to be similarly equivalent (not shown).
The geopotential height at 1000 hPa in Fig. \[f\_arcad\_1\] shows a negative bias for both dynamical cores. This indicates a loss of mass conservation, which is a consequence of the non-conservative nature of semi-Lagrangian advection. This can be improved by introducing a simple global mass fixer that works by conserving the global mean surface pressure after each dynamics step in the model. The scores in the presence of a global mass fixer is shown in Fig. \[f\_arcad\_2\], where the bias at the lowest model level is eliminated for both dynamical cores. The overall scores for the two cores are again, as expected, found to be equivalent in the presence of a global mass fixer.
Overall, the results presented in the Figures \[f\_bubble\]-\[f\_arcad\_2\] clearly demonstrate that the implementation of GEM-H as presented in this paper produces results that are equivalent to the existing GEM-P dynamical core.
Summary {#s_summary}
=======
A newly-developed dynamical core for ECCC’s GEM model with a height-based terrain-following vertical coordinate has been presented. With increasing focus on three-dimensional iterative solvers at ECCC driven by the limitations of the operational direct solver as well as the strong numerical instability induced by steep-orography for sub-kilometer resolution NWP, a dynamical core with height-based TCF is expected to be better placed to address the future NWP challenges at ECCC. The principal objective of this paper is to provide information pertaining to the different aspects of the new height-based dynamical core including changes to the model formulation, discretizations, solvers for the discretized problem and the strategy for coupling the new core with the RPN physics package. Another important objective is to demonstrate that the new GEM-H core is capable of making meteorological predictions that are equivalent to the existing GEM-P dynamical core, which is based on a log-hydrostatic-pressure-type vertical coordinate.
Numerical experiments have been conducted throughout the different stages of GEM-H development. Initially, the bubble convection test revealed that the advection and the buoyancy effects in GEM-H are accurately represented and are producing results that are equivalent to GEM-P. When tested for the idealized Schär’s mountain case, the nonhydrostatic and hydrostatic components of the mountain waves predicted by GEM-H are found to be very close to the GEM-P predictions as well as the analytical solution. The dynamics source terms in GEM are averaged over the air parcel trajectories using the trapezoidal method and the calculation of the RHS terms in the dynamical equations are carried out using cubic interpolation. Although it has not been shown explicitly, similarly to GEM-P, in the absence of any off-centering in the discretized dynamical equations, numerical consistency in GEM-H requires a trapezoidal averaging of the source terms in the discretized trajectory equations along with a cubic interpolation for the wind fields at the departure positions. Furthermore, in the presence of off-centering, the Schär’s mountain case shows that the discretized sources terms in the trajectory equations also necessitate off-centered averaging for the sake of consistent numerics in both GEM-H and GEM-P. In general, the knowledge acquired over years regarding the different numerical aspects of the GEM-P dynamical core is proven to be equally applicable to the case of GEM-H. Comparisons between GEM-H and GEM-P for global deterministic predictions are also presented. The results are found to be equivalent in the spectral space confirming that GEM-H does not produce any spurious noise or damping over the model-resolved scales. When compared against upper-air radiosonde observations, except for small differences in bias, GEM-H and GEM-P are found to produce equivalent results.
The rationale behind the choice of the solution approach for the discretized EBV resulting from the GEM-H formulation as well as the method for dynamics-physics coupling is discussed in detail. Although the general structure of the EBV originating form the GEM-H discretized system of equations require an iterative approach, a simplified approach has been devised where the horizontally-variable metric terms—attributable to the vertical-coordinate transformation—are coupled with the nonlinear terms and are treated with nonlinear iterations. This makes the EBV vertically separable and allows the use of a direct solver which is computationally very efficient for the currently operational NWP system configurations. A three-dimensional iterative solver based on FGMRES is also developed for situations when the simplified approach is not feasible. The fact that the GEM-H core can utilize the direct solver approach for global and regional scale model resolutions, eliminates the concern of computational efficiency as far as its implementation in the current and near-future plans for operational NWP systems at ECCC is concerned.
Improving any NWP model is a continuous process. As a stable GEM-H dynamical core is now developed, a number of other relevant issues are currently being studied. The objective is to improve the GEM model in general and the GEM-H dynamical core in particular. One of the most important short-term goal in this regard is to devise a more numerically consistent and accurate coupling between RPN Physics and GEM dynamics which will benefit both the dynamical cores. Also, extensive research is being conducted to develop highly optimized three-dimensional iterative solvers that can be competitive against the operationally-used direct solver while scaling better for very large number of processor cores for the future generations of massively parallel supercomputers. Currently, the Yin-Yang system uses the Schwarz method where the global solution is produced by iteratively solving two elliptic sub-problems for the two sub-domains (Yin and Yang) separately and updating the solutions in the overlapping regions until a certain convergence criteria is satisfied. One promising solution to reduce the execution time of the iterative solver for the Yin-Yang system is to remove the Schwarz iterations and to solve the two elliptic sub-problems as one by using FGMRES [@zal12]. Also, pre-conditioners based on other types of methods, e.g., incomplete factorization, block Gauss-Siedel or multigrid method, could be used in order to improve the convergence rate of the FGMRES solver. Finally, on the GEM-H front, another important short-term objective is to improve its numerical stability over steep orography by implementing more accurate numerical approximation of the horizontal gradients in the discretized dynamical equations.\
The authors would like to sincerely thank the members of the Numerical Methods Research Group at RPN for all their thoughtful comments and suggestions during the course of GEM-H development. The authors would like to particularly thank Michel Desgagné, Stéphane Gaudreault, Rabah Aider and Vivian Lee for their contributions during the implementation of the GEM-H source code. Also, comments from the internal reviewer, Dr. Christopher Subich, have helped to considerably improve the overall presentation of the paper.
The nonlinear components of (\[e\_u\_1\])–(\[e\_t\_1\]) and (\[e\_k\_1\]), associated with linear components (\[e\_l\_u\])–(\[e\_l\_k\]), are given by $$N_u = -\bigg(f+\frac{tan\phi}{a}u\bigg)\overline{v}^{XY}+ \bigg(\frac{\overline{T_v}^{X\zeta}}{T_*}-1\bigg)\delta_Xq - \bigg(\frac{\overline{T_v}^{X\zeta}}{T_*}-\mathbf{s_i}\bigg)J_X\overline{J^{-1}_\zeta\delta_\zeta q}^{X\zeta},
\label{e_n_u}$$ $$N_v = \bigg(f+\frac{tan\phi}{a}\overline{u}^{XY}\bigg)\overline{u}^{XY}+ \bigg(\frac{\overline{T_v}^{Y\zeta}}{T_*}-1\bigg)\delta_Yq - \bigg(\frac{\overline{T_v}^{Y\zeta}}{T_*}-\mathbf{s_i}\bigg)J_Y\overline{J^{-1}_\zeta\delta_\zeta q}^{Y\zeta},
\label{e_n_v}$$ $$N_w = \bigg[\bigg(\frac{T_v}{T_*} - \mathbf{s_i}\bigg)J^{-1}_\zeta-\mathbf{s_d}\bigg]\delta_\zeta q - \bigg(\frac{T_v}{T_*}-1\bigg) g\frac{T^{'}_v}{T_v},
\label{e_n_w}$$ $$N_c=0,
\label{e_n_c}$$ $$N_T=\frac{1}{\tau_h}\bigg[\ln\bigg(\frac{T_v}{T_*}\bigg)-\frac{T^{'}_v}{T_v}\bigg],
\label{e_n_t}$$ $$N_z=0
\label{e_n_k}.$$
The corresponding RHS terms, in the absence physics forcing, take the following forms $$R_u = \frac{u}{\tau_m}-\beta_m\bigg[-\bigg(f+\frac{tan\phi}{a}u\bigg)\overline{v}^{XY}+ \frac{\overline{T_v}^{X\zeta}}{T_*}\bigg(\delta_Xq - J_X\overline{J^{-1}_\zeta\delta_\zeta q}^{X\zeta}\bigg)\bigg],
\label{e_r_u}$$ $$R_v = \frac{v}{\tau_m}-\beta_m\bigg[\bigg(f+\frac{tan\phi}{a}\overline{u}^{XY}\bigg)\overline{u}^{XY}+ \frac{\overline{T_v}^{Y\zeta}}{T_*}\bigg(\delta_Yq - J_Y\overline{J^{-1}_\zeta\delta_\zeta q}^{Y\zeta}\bigg)\bigg],
\label{e_r_v}$$ $$R_w = \frac{w}{\tau_{nh}} - \beta_{nh}\bigg[\frac{T_v}{T_*} \bigg(J^{-1}_\zeta\delta_\zeta q - g\frac{T^{'}_v}{T_v}\bigg)\bigg],
\label{e_r_w}$$ $$R_c=\frac{1}{\tau_h}\bigg(\frac{q}{c_*^2}+\ln{J_\zeta}\bigg)-\beta\bigg[\delta_Xu+\frac{1}{\cos\phi}\delta_Y(\cos\phi v)+\delta_\zeta\dot{\zeta}-\varepsilon\overline{w}^{\zeta}\bigg],
\label{e_r_c}$$ $$R_T=\frac{1}{\tau_h}\bigg[\ln\bigg(\frac{T_v}{T_*}\bigg)-\frac{\overline{q}^\zeta}{c_{pd}T_*}\bigg] -\beta[\mu w],
\label{e_r_t}$$ $$R_z=\frac{z-\zeta}{\tau_{nh}}-\beta_{nh}[\dot{\zeta}-w]
\label{e_r_k}.$$
The LHS of the EBV problem to be solved at every time step in GEM-H is derived by manipulating the discretized system of equations of the form (\[e\_lnr\]). The sequence of operations to derive the EBV are provided below in terms of its linear components: $$L^{'}_c=\delta_XL_u+\frac{1}{\cos\phi}\delta_Y(\cos\phi L_v) - \frac{1}{\tau_m}\bigg(L_c-\frac{\ln J_\zeta}{\tau_h}\bigg),$$ $$L^{'}_z=L_z-\frac{z-\zeta}{\tau_{nh}},$$ $$L^{'}_w=L_w+L^{'}_z/\tau_{nh},$$ $$L^{'}_T=L_T+\mu L^{'}_z,$$ $$L^{''}_T=\Gamma (g\tau_h L^{'}_T+L^{'}_w),$$ $$L^{''}_c=L^{'}_c+\frac{\varepsilon}{\tau_m}\overline{L^{'}_z}^\zeta,$$ $$L^{'''}_c=L^{''}_c+\delta_\zeta L^{''}_T-\varepsilon\overline{L^{''}_T}^\zeta.$$
Similar manipulations are also applied to the nonlinear and RHS components of the discretized equations to obtain $R^{'}_c$, $N^{'}_c$, $R^{'}_z$, $N^{'}_z$, $R^{'}_w$, $N^{'}_w$, $R^{'}_T$, $N^{'}_T$, $R^{''}_T$, $N^{''}_T$, $R^{''}_c$, $N^{''}_c$, $R^{'''}_c$ and $N^{'''}_c$.
The solution of the EBV computes the unkown $q$. The rest of the variables are then calculated from the discretized dynamical equations through back-substitution in the following sequence: $$u=\tau_m\bigg[R_u-N_u-\bigg(\delta_Xq - \mathbf{s_i} J_X\overline{J^{-1}_\zeta\delta_\zeta q}^{X\zeta}\bigg)\bigg]
\label{e_back_u},$$ $$v=\tau_m\bigg[R_v-N_v-\bigg(\delta_Yq - \mathbf{s_i} J_Y\overline{J^{-1}_\zeta\delta_\zeta q}^{Y\zeta}\bigg)\bigg]
\label{e_back_v},$$ $$\dot{\zeta} = \tau_m \bigg[R^{''}_T - N^{''}_T - \Gamma\bigg((\mathbf{s_i}J^{-1}_\zeta + \mathbf{s_d})\delta_\zeta q - \mu\overline{q}^\zeta\bigg)\bigg],
\label{e_back_zd}$$ $$w=\dot{\zeta}-L^{'}_z,
\label{e_back_w}$$ $$T_v=gT_*\bigg[g- \frac{\dot{\zeta}}{\tau_{nh}} + (\mathbf{s_i}J^{-1}_\zeta + \mathbf{s_d})\delta_\zeta q - R^{'}_w -N^{'}_w\bigg]^{-1}.
\label{e_back_T}$$
![a) Vertical Charney-Phillips grid. b) Horizontal Arakawa C grid.[]{data-label="f_grid"}](figure_grid.png "fig:"){width="45pc"}\
![a) Geopotential height (dam) at 400 hPa after 72 hours for a 25-km global forecast initiated at 1200 UTC of 25 January 2015 obtained with GEM-P using the ‘split method’ for dynamics-physics coupling (contour intervals of 6 dam). b) The enlarged view over a small section of the global domain (contour intervals of 0.5 dam). c) Same as in Fig. \[f\_split\]b, but with ‘explicit method’ for the dynamics-physics coupling.[]{data-label="f_split"}](figure_split.png "fig:"){width="45pc"}\
![Potential temperature ($^\circ$C) distribution for the rising bubble with: a) GEM-P at $t=7$ min, b) GEM-H at $t=7$ min, c) GEM-P at $t=10$ min, and d) GEM-H at $t=10$ min.[]{data-label="f_bubble"}](figure_bubble.png "fig:"){width="45pc"}\
![Steady-state vertical velocity contours (at 0.1 m s$^{-1}$ intervals) for the Schär mountain case predicted by : a) GEM-P and b) GEM-H. No off-centering is used in the discretized dynamical and trajectory equations.[]{data-label="f_schaer_1"}](figure_schaer_one.png "fig:"){width="43pc"}\
![Same as in Fig. \[f\_schaer\_1\], but with uniform off-centering ($b_m=b_h=b_{nh}=0.2$) used in the discretized dynamical equations. The absence of off-centering in the discretized trajectory equations leads to inconsistent trajectory calculations and distortions in the mountain waves.[]{data-label="f_schaer_2"}](figure_schaer_two.png "fig:"){width="43pc"}\
![Same as in Fig. \[f\_schaer\_1\], but with consistent off-centering, i.e., off-centering is applied to both the discretized dynamical and trajectory equations.[]{data-label="f_schaer_3"}](figure_schaer_three.png "fig:"){width="43pc"}\
![Spectral variance of (left column) geopotential height (GZ) and (right column) temperature (TT) for 120-h global forecasts with 25 km horizontal grid spacing. Results are presented for three different pressure levels (top: 100 hPa, middle: 500 hPa, bottom: 850 hPa) that are obtained by averaging the spectra for 10 Northern Hemisphere winter forecasts.[]{data-label="f_spectra_gz"}](figure_spectra_gz_tt.png "fig:"){width="33pc"}\
![Spectral variance of (left column) kinetic energy (KE) and (right column) vertical velocity (WW) for 120-h global forecasts. All other conditions as in Fig. \[f\_spectra\_gz\].[]{data-label="f_spectra_ke"}](figure_spectra_ke_ww.png "fig:"){width="33pc"}\
![Comparison of 120-h 25-km GDPS forecasts obtained with GEM-P (blue) and GEM-H (red) dynamical cores against radiosonde observations for zonal wind (UU), wind speed (UV), temperature (TT), and geopotential height (GZ). The dashed and solid lines, respectively, indicate bias and standard deviation of error (SDE). The scores are obtained by averaging over 44 Northern Hemisphere winter cases. The red and blue shaded numbers along the left (right) vertical axes within each panel indicate the confidence in percentage in the statistically significant improvements in bias (SDE) for the dynamical core associated with each color. Significance for bias and SDE are computed using $t$- and $F$-test, respectively.[]{data-label="f_arcad_1"}](figure_arcad_120h_wint_psadj0.png "fig:"){width="45pc"}\
![Same as in Fig. \[f\_arcad\_1\], but with a global mass fixer in the simulations for both GEM-P and GEM-H to improve mass conservation.[]{data-label="f_arcad_2"}](figure_arcad_120h_wint_psadj1.png "fig:"){width="45pc"}\
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'E. Alecian'
- 'O. Kochukhov'
- 'C. Neiner'
- 'G.A. Wade'
- 'B. de Batz'
- 'H. Henrichs'
- 'J.H. Grunhut'
- |
\
J.-C. Bouret
- 'M. Briquet'
- 'M. Gagne'
- 'Y. Naze'
- 'M.E. Oksala'
- 'T. Rivinius'
- |
\
R.H.D. Townsend
- 'N.R. Walborn'
- 'W. Weiss'
- the MiMeS collaboration
bibliography:
- 'harps\_letter.bib'
date: 'Received September 15, 1996; accepted March 16, 1997'
title: 'First HARPSpol discoveries of magnetic fields in massive stars [^1] '
---
Introduction
============
MiMeS[^2] (Magnetism in Massive Stars) is a large collaboration that aims to address many issues concerning the magnetism of massive stars. One goal in particular is to determine the global magnetic properties of massive stars with the help of Large Programs (LP) that have been allocated on the high-efficiency high-resolution spectropolarimeters ESPaDOnS (Canada France Hawaii Telescope, Hawaii) and Narval (Telescope Bernard Lyot, France). These programs aim to observe about 200 massive OB field stars (the Survey Component or SC), in order to search for magnetic fields, [confirm those previously suspected,]{} and derive statistical properties [[@wade09; @grunhut11b]]{}. They also aim to observe intensely about 30 already known magnetic massive stars (the Targeted Component [or]{} TC) in order to map in detail their surface magnetic fields. This smaller sample of stars is dedicated to the study of the interplay of magnetic fields with the stellar structure, environment and evolution at high mass [e.g. @grunhut09; @oksala11].
In 2010, the polarimeter HARPSpol was commissioned at the 3.6m-ESO telescope (La Silla, Chile). For the first time, we can access the Southern hemisphere with data quality similar to ESPaDOnS and Narval. Therefore, a Large Program was established to complete the ESPaDOnS/Narval field sample, and to take the first steps toward observing massive stars in various open clusters [and associations]{} of different ages, to investigate the magnetic field evolution, and the impact of magnetic fields on stellar evolution.
The HARPSpol sample is divided in two components (SC and TC) to follow the same strategy as the Narval and ESPaDOnS LPs. The HARPSpol SC sample contains about 180 stars including $\sim$110 stars in 7 clusters, and $\sim$70 stars in the field of the Galaxy. The former have been selected from the Catalogue of Open Cluster Data [@kharchenko05], while the latter have been chosen from the International Ultraviolet Explorer (IUE) data archive but also from other catalogues or publications containing highly probable magnetic stars, in accordance with the ESPaDOnS/Narval target selection [@wade09].
This HARPSpol Large Program [was]{} allocated four separate runs over two years. During the first run in May 2011, 57 stars were observed including one magnetic calibrator and one TC target for which the results will be presented in a forthcoming paper. In this letter, we report on the first discoveries of magnetic fields in massive stars with HARPSpol in HD 130807 and HD 122451, and we confirm the magnetic field in HD 105382 previously detected with the low-resolution spectropolarimeter FORS 1 [@kochukhov06; @hubrig06]. Among the 55 SC stars observed during this run, those three stars are the only ones in which a magnetic field was detected. In Section 2, we present the observations and reduction techniques. In Section 3, we detail the HARPSpol results on each star, and discuss them in Section 4.
Observations
============
[@l@c@c@c@r@r@r@[$\pm$]{}l@c@l@]{} Date (d/m) & HJD & $t_{\rm exp}$ & \# & S/N & S/N & & $P_{\rm det}$ &\
UT & (2 455 000+) & (s) & & & (LSD) & & &\
\
23/05 05:23 & 704.724 & 4000 & 1 & 500 & 4600 & 292 & 26 & 1.00000 & DD\
27/05 06:06 & 708.754 & 1200 & 1 & 520 & 4800 & -94 & 26 & 1.00000 & DD\
28/05 05:58 & 709.748 & 6000 & 2 & 630 & 5700 & 677 & 21 & 1.00000 & DD\
\
\
23/05 04:16 & 704.677 & 1340 & 5 & 1680 & 33200 & -43 & 20 & 1.00000 & DD\
27/05 05:21 & 708.723 & 1200 & 20 & 3000 & 61800 & -83 & 14 & 1.00000 & DD\
28/05 04:34 & 709.690 & 1680 & 14 & 1260 & 26000 & -66 & 29 & 0.98173 & ND\
\
\
25/05 01:19 & 706.555 & 3200 & 1 & 990 & 9650 & -622 & 26 & 1.00000 & DD\
26/05 01:59 & 707.582 & 4000 & 1 & 900 & 8840 & -298 & 32 & 1.00000 & DD\
28/05 00:10 & 709.506 & 4800 & 1 & 920 & 8900 & -406 & 32 & 1.00000 & DD\
We used the HARPSpol polarimeter [@piskunov11], combined with the HARPS spectrograph [@mayor03], installed at the 3.6m ESO telescope at La Silla Observatory (Chile), yielding spectra with resolving power $\lambda / \Delta \lambda$ of about $105\,000$, and covering the 380–690 nm wavelength region. All spectra were recorded as sequences of 4 individual sub-exposures taken in different configurations of the polarimeter, in order to yield a full circular polarisation analysis, as described by @donati97. The data were reduced using the package “REDUCE” described by @piskunov02. After reduction, we obtained the intensity Stokes $I$ and the circular polarisation Stokes $V$ spectra of the stars, both normalised to the continuum. A null spectrum ($N$) was also computed in order to diagnose spurious polarisation signatures, and to help to verify that the signatures in the Stokes $V$ spectrum are of stellar origin. The log of the observations is presented in Table 1.
To increase the effective signal to noise ratio (S/N) of our data, we applied the Least Squares Deconvolution [LSD; @donati97] procedure using tailored line masks of appropriate temperature and gravity for each star. The masks were first computed using Kurucz ATLAS 9 models of solar abundance [@kurucz93], with intrinsic line depths larger than 0.1. We then excluded from these masks [hydrogen]{} Balmer lines, and lines whose Landé factor is unknown. Finally we have modified the line depths [to]{} take into account the relative depth of the lines of the observed spectra. The resulting masks contain 394, 592, and 394 lines for HD 1030807, HD 122451 and HD 105382, respectively. The S/N of the LSD Stokes $V$ profiles is about 10 times larger than the S/N in the original spectra (Table \[tab:log\]).
In order to perform a reliable magnetic field diagnosis, we have computed the detection probability inside the LSD $V$ profiles [as described in @donati97]. We consider that an observation displays a “definite detection" (DD) [of Stokes $V$ Zeeman signature]{} if the probability is larger than 0.99999, a “marginal detection" (MD) if it falls between 0.999 and 0.99999, and a “null detection" (ND) otherwise (see Table 1). All observations of HD 130807 and HD 105382 display DD while two DD and one ND have been obtained for HD 122451. The LSD $I$, $V$, and $N$ profiles are plotted in Fig. \[fig:lsd\]. In almost all of our observations Zeeman signatures, as broad as the $I$ profiles, are clearly detected in the $V$ profiles, while the $N$ profiles are consistent with the noise. These results allow us to confidently affirm that magnetic fields are present at the surface of these stars.
We measured for each observation the line-of-sight component of the magnetic field averaged over the visible stellar surface (the so-called longitudinal magnetic field or $B_{\ell}$), by integrating the $I$ and $V$ profiles over the ranges $[-50,60]$, $[-80,100]$, and $[-70,105]$ [km.s$^{-1}$]{}for HD 130807, HD 122451, and HD 105382, respectively [as described by @alecian09]. The values are reported in Table 1.
Results
=======
HD 130807
---------
{width="6cm"} {width="6cm"} {width="6cm"}
[HD 130807 ($o$ Lup) is member of the Sco-Cen association [@kharchenko05]. A companion was detected at an angular distance varying from 0.07 to 0.14 arcsec [@perryman97; @mcalister90]. According to the angular separation, the light of both components entered the HARPSpol fibers during our observations of this target.]{}
[From a visual inspection]{} we find that most of the spectrum of HD 130807 is consistent with a synthetic spectrum of a single star of [effective temperature]{} $T_{\rm eff}=18000$ K, [surface gravity]{} $\log g=4.25$ (cgs), broadened by $v\sin i=25$ [km.s$^{-1}$]{}, calculated using TLUSTY non-LTE atmosphere models and the SYNSPEC code [@hubeny88; @hubeny92]. However, we observe that all He [i]{} lines are substantially weaker than the synthetic ones calculated with solar abundance (Fig. \[fig:sphd130807\]), while the Si [ii]{} lines are considerably stronger. The Si, N and Fe lines show variability in depth and shape on a timescale of 1 d. These characteristics suggest that HD 130807 is an He-weak star with abundance spots on its surface [@jaschek74].
Magnetic signatures are detected in almost all the lines of the spectrum, similar to the LSD one (Fig. \[fig:lsd\] [*left*]{}). Many additional lines are observed in the spectrum that could be due to Fe [ii]{}, Fe [iii]{}, or Ti [ii]{} enhancements. All these lines show Zeeman signatures similar to the others, with the same variations from one night to the other. They can therefore be attributed to the same star, rather than a companion, and they are probably the result of the chemical peculiarities at the surface of the star.
A significant shift in radial velocity ($\sim 6$ km/s) is detected in the strongest spectral lines including Balmer lines, between May 22 and May 26-27. The maximum reported angular separation between both visual components implies a distance $\ge17$ AU, and therefore a period $\ge27$ years. This radial velocity shift cannot therefore be due to the reported visual companion. A third companion very close to the primary could explain these variations, but more observations are required to fully understand all the peculiarities observed in the spectrum.
The variations observed in the $V$ profiles over 6 days (Fig. \[fig:lsd\] [*left*]{}) can be understood in terms of the oblique rotator (OR) model that consists of an inclined dipole placed inside a rotating star [@stibbs50]. The rotational modulation of the shape of the $V$ profiles and of the $B_{\ell}$ values that vary from [-94 to 677 G]{} (Table 1) suggest that the rotation period of the star should be between 1 and 6 days. More observations, well sampled over the rotation period, are required in order to fully characterise the magnetic field and better constrain the period of HD 130807.
HD 122451
---------
[HD 122451 ($\beta$ Cen) is a double-lined spectroscopic binary with components of similar effective temperatures (25000 K) and gravities ($\log g=3.5$, cgs), and a $\beta$ Cep-type pulsating primary. The system is highly eccentric ($e = 0.835$) and orbits with a period of 357 days [@ausseloos02; @ausseloos06; @davis05].]{} We obtained [three]{} observations of HD 122451 during 3 different nights. In order to avoid potential false magnetic detections due to pulsations, we have split each observation into many sequences of four spectra (Table \[tab:log\]), so that the exposure time for one sequence is much shorter than the pulsation period.
The spectra of HD 122451 clearly show two components of similar temperatures but different broadening, confirming the SB2 nature of the system. We adopt the same definition as @ausseloos06 for the primary and secondary, i.e. as the broad-line and narrow-line components, respectively. In order to measure the [$v\sin i\;$]{}of both stars, we performed a least-squares fit to few individual spectral lines with the sum of two functions calculated as the convolution of a Gaussian of instrumental width and a rotation function as described by @gray92 [see details of the fitting procedure [used by]{} @alecian08]. We find a [$v\sin i\;$]{}of $190\pm20$ [km.s$^{-1}\;$]{}and $75\pm15$ [km.s$^{-1}\;$]{}for the primary and secondary, respectively. When compared with TLUSTY/SYNSPEC synthetic spectra our observations are consistent with $T_{\rm eff}=25000$ K and $\log g=3.5$ (cgs), in agreement with the work of @ausseloos06. The spectral lines appear distorted and show rapid variations very likely due to $\beta$ Cep-type pulsations. No obvious abundance peculiarity, nor manifestation of circumstellar matter is observed within the spectra.
In Fig. \[fig:lsd\] ([*middle*]{}), we superimposed the LSD $I$, $V$, and $N$ profiles of our observations. According to the ephemeris of @ausseloos06, the 3 observations are roughly at the same orbital phase ($\sim$0.5), and both components have [similar]{} radial velocities ($\sim$14 and $\sim$4 [km.s$^{-1}\;$]{}for the primary and secondary respectively), which explains why it is difficult to distinguish both components in the profiles. The shape of the LSD $I$ profile shows variations during the run that can be understood in terms of radial pulsations in the primary, which would occasionally broaden the profile. As a result, both components can be clearly distinguished in the profile of May 27 (full black line in Fig. \[fig:lsd\] [*middle*]{}), while it is less obvious in the other observations.
![Spectrum (full black) of HD 130807 plotted around He [i]{} 4471 [Å]{} and Mg [ii]{} 4481 [Å]{}. Synthetic spectra of 18000 K (dashed red) and 17000 K (dot-dot-dot-dashed blue) are overplotted.[]{data-label="fig:sphd130807"}](obssyn_hd130807_psp01.ps){width="9cm"}
Zeeman signatures are detected in many individual spectral lines, [as in]{} the LSD $V$ profiles. The signatures are as broad as the secondary profile, meaning that the magnetic field is detected only in the secondary component of the system. However, considering the faint Zeeman signatures in the secondary, and the broad line shape of the primary, a magnetic field of the same strength as the secondary’s could exist in the primary, without being detected in our observations. [In order to estimate the $B_{\ell}$ values of the secondary, we need to extract the $V$ and $I$ profiles of the secondary only, from the profiles of the binary. Without any evidence of a magnetic field in the primary, we neglect its contribution to the $V$ profile, which we consider to be entirely from the secondary. On the contrary, the $I$ profile needs to be corrected. With]{} this aim, we have first fitted the $I$ profile of the binary with the method described above for the individual spectral lines. Then we have subtracted from the observed $I$ profile the fit of the primary. Finally we have measured $B_{\ell}$ using the corrected $I$ and the original $V$ profiles. The values are reported in Table 1.
HD 105382
---------
HD 105382 (= HR 4618) is member of the Sco-Cen association [@kharchenko05]. @briquet04 classified it as He-weak with He patches enhanced where Si is depleted, and derived a $T_{\rm eff}$ of $17400\pm800$ K, a $\log g=4.18\pm0.20$ (cgs), a rotation period of $1.295 \pm 0.001$ d and an inclination angle [of the rotation axis to the line of sight]{} $i=50\pm10$[$^{\circ}$]{}.
In our [three]{} spectra, we observe strong variations in the spectral lines, mainly in He [i]{}, Si [ii]{} and Fe [iii]{}, that are due to abundance spots on the stellar surface described by @briquet04. Clear Zeeman signatures are detected in the metallic and Balmer lines, as well as in the LSD $V$ profiles (Fig. \[fig:lsd\] [*right*]{}). The rotation phases of our observations, calculated with a rotation period of 1.295 d, are very different (0.35, 0.14, and 0.63), and yet the $V$ profiles are all similarly negative (Table 1). According to the OR model, this implies that the magnetic obliquity angle (with respect to the rotation axis) cannot be very high ($|\beta| < 40^{\circ}$ if $i = 50^{\circ}$), otherwise the positive magnetic pole would sometimes appear on the visible stellar hemisphere, creating a positive profile at least once during the run.
HD 105382 was independently discovered as magnetic by @kochukhov06 and @hubrig06. @briquet07 derived the longitudinal field from FORS 1 observations and found values ranging from $-923$ G to $840$ G. Among their four values, the May 2004 [observation]{} ($840\pm58$ G) is clearly inconsistent with our data as positive values are not expected. @bagnulo11 very carefully re-reduced the same FORS 1 data and [found values consistent with those of Briquet et al. (2007)]{} except for that observation ([for which they derived]{} $B_{\ell}=-29\pm69$ G).
We performed a least-square sinusoidal fit to our $B_{\ell}$ values simultaneously with the Briquet et al. (2007) data and could find a solution only by removing the May 2004 datapoint. We also performed an independent fit using the re-reduced data of @bagnulo11 and found a similar result. In both cases, the derived period is consistent with that of Briquet et al. (2004). The fitted $B_{\ell}$ values are very similar in both cases, [varying]{} from $-670$ G to $-20$ G, [and]{} implying a magnetic obliquity of $\sim38$[$^{\circ}\;$]{}and a polar field strength of $\sim2.3$ kG, assuming a dipole field [@borra80].
Discussion
==========
We report direct detections of magnetic fields in three hot B-type stars (18000 K - 25000 K), among a sample of 55 stars in which we were searching for magnetic fields with HARPSpol. Two of them (HD 122451 and HD 130807) are completely new detections. For the other one - HD 105382 [- this]{} is the first direct detection of a Zeeman signature. One of the main MiMeS results is the systematic detection of chemical peculiarities at the surface of magnetic hot stars [and conversely, e.g. @grunhut11a]. Among the three stars discussed in this paper two are unambiguously He-weak. The third one (HD 122451) belongs to a binary system with a $\beta$ Cep primary, that makes the interpretation of the spectrum and the detection of peculiarities inside spectral lines very difficult. More observations well sampled over the orbital period of the system are required in order to first confirm a magnetic detection in only the secondary, and then disentangle the pulsation and chemical peculiarity effects.
The interplay between radiative and magnetic forces [@hunger99] is usually assumed to be at the origin of the over- or under-abundant He spots at the surface of hot magnetic B stars. These spots are very often correlated with the stellar magnetic fields [e.g. @veto90]. In the case of HD 105382, @briquet04 found a large He spot at a latitude of 60[$^{\circ}$]{}. If our estimate of its magnetic obliquity is confirmed, this spot would be situated close to the South magnetic pole, demonstrating once more the importance of magnetic fields in the formation of chemical spots.
More observations of these magnetic stars are planned within the HARPSpol large program in order to perform detailed mapping of their magnetic fields, and better confront the models and theories of magnetic massive stars.
We wish to thank the Programme National de Physique Stellaire (PNPS) for their support. JHG and GAW acknowledges support from NSERC. M.B. acknowledges the Fund for Scientific Research – Flanders for a grant for a long stay abroad, and she is a F.R.S.-FNRS Postdoctoral Researcher. RHDT acknowledges support from NSF grant AST-0908688. This research has made use of the SIMBAD database and the VizieR catalogue access tool, operated at CDS, Strasbourg (France), of INES data from the IUE satellite and of NASAs Astrophysics Data System.
[^1]: Based on observations collected at the European Southern Observatory, Chile (Program ID 187.D-0917)
[^2]: http://www.physics.queensu.ca/$\sim$wade/mimes
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'If ultraluminous X-ray sources (ULXs) are powered by accretion onto stellar remnant black holes, then many must be accreting at super-Eddington rates. It is predicted that such high accretion rates should give rise to massive, radiatively-driven winds. However, observational evidence of a wind, in the form of absorption or emission features, has remained elusive. As such, the reported detection of X-ray spectral residuals in [*XMM-Newton*]{} spectra of NGC 5408 X-1, which could be related to absorption in a wind is potentially very exciting. However, it has previously been assumed by several authors that these features simply originate from background diffuse plasma emission related to star-formation in the ULX’s host galaxy. In this work we utilise the spatial resolving power of [*Chandra*]{} to test whether we can rule out this latter interpretation. We demonstrate that the majority of the luminosity in these spectral features is emitted from a highly localised region close to the ULX, and appears point-like even with [*Chandra*]{}. It is therefore highly likely that the spectral features are associated with the ULX itself, and little of the flux in this spectral component originates from spatially extended emission in the host galaxy. This may be consistent with the suggestion of absorption in an optically thin phase of a super-Eddington wind. Alternatively, we could be seeing emission from collisionally ionised material close to the black hole, but critically this would be difficult to reconcile with models where the source inclination largely determines the observed X-ray spectral and timing properties.'
author:
- Andrew D Sutton
- 'Timothy P. Roberts'
- Matthew J Middleton
bibliography:
- 'refs.bib'
title: 'X-ray spectral residuals in NGC 5408 X-1: diffuse emission from star formation, or the signature of a super-Eddington wind?'
---
Introduction
============
Whilst it was previously suggested that ULXs could contain intermediate mass black holes [@colbert_and_mushotzky_1999], several well-developed arguments have been put forward to dispute this (e.g. @grimm_etal_2003 [@king_2004]). Instead, many ULXs may actually be powered by accretion onto stellar-remnant black holes, which has been confirmed in a few cases (e.g. @liu_etal_2013 [@middleton_etal_2013; @motch_etal_2014]; although at least one ULX contains a neutron star, see @bachetti_etal_2014). Some of these may be more massive than the typical $\sim 10 \Msun$ stellar-mass black holes that we see throughout our Galaxy, as stellar collapse in regions of low metallicity could leave remnants with masses of up to $\sim 80 \Msun$ [@zampieri_and_roberts_2009; @mapelli_etal_2010]. However, the Eddington luminosity of even an $80 \Msun$ black hole is only $\sim 10^{40}~\ergs$, and ULXs are observed to exceed this. Hence, if the majority of ULXs do contain stellar-remnant black holes, then many must be accreting matter at close to or in excess of the Eddington limit. It is therefore important to note that high quality ULX X-ray spectra are observed to differ from the standard sub-Eddington states, with many displaying both a soft excess and high energy curvature (e.g. @stobbart_etal_2006 [@gladstone_etal_2009; @bachetti_etal_2013]).
Whilst the X-ray spectra from ULXs below $\sim 3 \times 10^{39}~\ergs$ are typically broad, disc-like continua, those from more luminous sources have a characteristic two component shape. These peak in either a hard or soft component, in what are termed the hard and soft ultraluminous regimes respectively [@sutton_etal_2013b]. The hard component may originate in a cool, optically thick corona [@gladstone_etal_2009], or the hot inner disc itself, with a large colour correction [@kajava_etal_2012]. The soft emission may originate from the base of a radiatively driven wind, which is predicted to arise at super-Eddington accretion rates (@shakura_and_sunyaev_1973 [@poutanen_etal_2007; @kajava_and_poutanen_2009]; but see @miller_etal_2013). Although it has previously been mooted that the distinction between these two regimes is largely driven by accretion rate, recent work suggests that the inclination of the ULX is also key [@sutton_etal_2013b; @middleton_etal_2015]. One potential physical mechanism for introducing the inclination dependence is the super-Eddington outflow, which is expected to be funnel-shaped [@shakura_and_sunyaev_1973; @kawashima_etal_2012]. Whilst the soft emission is relatively isotropic, the hard emission from the centre of the accretion disc can be geometrically beamed by the wind funnel [@king_2009]. Hence, when viewed close to face-on the hard emission is focussed towards the observer’s line-of-sight and the ULX appears with a hard ultraluminous spectrum. However, at higher inclinations the hard emission is scattered away from the line-of-sight and the source appears with a soft ultraluminous spectrum. This interpretation is supported by X-ray timing evidence [@sutton_etal_2013b], if the edge of the wind is clumpy at its base [@middleton_etal_2011a], as is the case in radiation-magnetohydrodynamic simulations [@takeuchi_etal_2013].
One potential test of the funnel-shaped wind model is a search for the signatures of the outflow in ULX X-ray spectra, in the form of absorption and emission lines. In this vein, [@walton_etal_2012; @walton_etal_2013] examined the Fe K band in NGC 1313 X-1 and Ho IX X-1, finding that any narrow atomic absorption or emission features must be intrinsically weak or absent. However, both NGC 1313 X-1 and Ho IX X-1 typically have hard ultraluminous X-ray spectra.[^1] In the funnel-shaped wind model the absorbing material would predominantly intercept an observer’s line-of-sight in soft ultraluminous sources, and we would not expect as strong absorption features to be detected in hard ultraluminous sources. Instead, to test the model we ideally want to search for atomic features in soft ultraluminous ULXs, but the requirement for a high count rate makes such an experiment infeasible in the Fe K band in these sources. However, it is potentially very interesting that some soft ultraluminous ULXs do show spectral residuals to their best-fitting continuum models (e.g. @stobbart_etal_2006), including the object that we scrutinise in this paper: NGC 5408 X-1 (2XMM J140319.6$-$412258; e.g. @strohmayer_and_mushotzky_2009 [@miller_etal_2013; @middleton_etal_2014]).
At a distance of 4.8 Mpc [@karachentsev_etal_2002], NGC 5408 X-1 is one of the best studied ULXs. It has been observed by several of the current generation of X-ray satellites, on a multitude of occasions. Observations include: an [ *XMM-Newton*]{} large programme (e.g. @pasham_and_strohmayer_2012); [ *Swift*]{} XRT monitoring [@kaaret_and_feng_2009; @grise_etal_2013]; and 8 [ *Chandra*]{} exposures, which we re-analyse here. The flux variability of the source rules out an X-ray supernova remnant and confirms that it is powered by accretion onto a compact object [@kaaret_etal_2003; @soria_etal_2004]. It persistently displays a distinct soft ultraluminous two component X-ray spectrum in [*XMM-Newton*]{} data [@sutton_etal_2013b] at an average 0.3–10 keV unabsorbed luminosity of $\sim 1.1 \times
10^{40} \ergs$ (@strohmayer_2009; although we note that they fit the high energy spectrum with a soft power-law, so may over-estimate the intrinsic luminosity, cf. @middleton_etal_2014). Additional soft residuals have been detected in the [*XMM-Newton*]{} spectra which can be well modelled as thermal plasma emission [@strohmayer_and_mushotzky_2009; @miller_etal_2013; @middleton_etal_2014]. It has previously been assumed that these were the result of diffuse star-formation related emission in the host galaxy [@strohmayer_etal_2007; @strohmayer_and_mushotzky_2009; @miller_etal_2013]. However, we know from observational studies of galaxies that the X-ray luminosity of such emission is correlated with star-formation rate ($L_{0.5-2 {\rm keV}} / SFR \approx
8.3 \times 10^{38} \ergs~(M_{\odot}~{\rm yr}^{-1})^{-1}$; @mineo_etal_2012), and [@middleton_etal_2014] contend that the luminosity of the putative thermal plasma emission ($\sim 2.5 \times 10^{38}~{\rm erg~s^{-1}}$, calculated from @miller_etal_2013) greatly exceeds that inferred from star-formation, even over the entirety of NGC 5408 ($\approx 3 \times
10^{37} \ergs$, calculated from a 24 $\rm \mu m$ flux density of $0.42 \pm 0.04~{\rm Jy}$, @dale_etal_2005). Instead, [@middleton_etal_2014] show that the putative plasma emission features could actually be commensurate with broadened, blueshifted absorption in a partially ionised, optically thin medium. Such a medium would be expected to occur in a super-Eddington wind, as it becomes optically thin at large distances ($> 10^3 R_{\rm g}$) from the central black hole.
In this paper, we utilise [*Chandra*]{} archival data to determine the spatial origin of the soft plasma emission-like features in the X-ray spectrum of NGC 5408 X-1. The high angular resolution of [*Chandra*]{} is absolutely key to this analysis. Having constrained the spatial origin of these spectral features, we are able to make important inferences as to their nature. In section \[results\] we examine the spatial and spectral properties of the ULX and surrounding regions in [*Chandra*]{} ACIS data, and interpret these results in section \[discussion\].
Analysis and results {#results}
====================
Before examining the [*Chandra*]{} X-ray observations of NGC 5408 X-1, we briefly consider the optical/UV data. NGC 5408 X-1 is located around $\sim 12$ arcsec from the major regions of star-formation in NGC 5408 (Figure \[hst\]; @kaaret_etal_2003). This would fall well within a typical [*XMM-Newton*]{} source extraction region centred on the ULX, thus star-formation-related emission could feasibly contaminate the X-ray spectrum. However, NGC 5408 X-1 is sufficiently displaced from these star-formation regions that [*Chandra*]{} ACIS can spatially resolve them from the ULX. There is a much smaller stellar association containing $\sim 20$ OB stars, located around 4 arcsec North East of NGC 5408 X-1 [@grise_etal_2012], and this may still slightly contaminate the [*Chandra*]{} spectrum of the source.
{width="12cm"}
-------- -------------- --------------- -- --
Obs ID Date Exposure time
(yyyy-mm-dd) (ks)
4555 2003-12-20 5.2
4556 2004-02-09 4.9
4557 2004-12-20 5.1
4558 2005-01-29 5.2
11032 2010-05-02 12.2
11033 2010-05-15 12.2
11034 2010-05-28 12.0
13018 2010-09-12 12.0
-------- -------------- --------------- -- --
: [*Chandra*]{} ACIS-S observations of NGC 5408 X-1
\[obs\]
Details of the [*Chandra*]{} ACIS-S observations of NGC 5408 X-1 taken with the 1/8 sub-array.
![Exposure map of the merged [*Chandra*]{} observations. The approximate location of NGC 5408 X-1 is indicated by the red cross. As the ULX is close to the nominal ACIS-S aim-point in all of the observations, their individual footprints all overlap at the position of the target.[]{data-label="expmap"}](chan_expmap.ps){width="8cm"}
The region around NGC 5408 has been observed with [*Chandra*]{} on a total of 9 occasions, 8 of which are suitable for studying X-1. In this work, we analyse these 8 observations, which were taken using the 1/8 sub-array of ACIS-S, with X-1 being positioned close to on-axis. In the case of NGC 5408 X-1, the use of the 1/8 sub-array is critical to mitigate the effects of pileup. We estimate residual pileup fractions of $\sim 5$ per cent based on [PIMMS]{} (v4.7b: with ACIS Pile up and Background Count Estimation)[^2] simulations of an absorbed power-law with parameters from [@gladstone_etal_2009] and an unabsorbed 0.3–10 keV luminosity of $1.1 \times 10^{40}~\ergs$ [@strohmayer_2009]. Details of the observations included in this work are given in Table \[obs\], and a combined exposure map is shown in Figure \[expmap\]. We downloaded each dataset from the HEASARC archive,[^3] and analysed them using tools in [ciao]{} 4.6,[^4] with calibration database 4.6.3. We note that we would not expect the analysis to change significantly if [ciao]{} 4.7 was used instead.
The first stage of the analysis was to confirm that the ULX appeared point-like at the spatial resolution of [*Chandra*]{}. To do this, the individual observations were reprocessed using the [chandra\_repro]{} script, before being stacked using [merge\_obs]{}. The stacked image is shown in Figure \[chan\_img\], where it has been divided by the total exposure map of the [*Chandra*]{} observations (with $8 \times 8$ pixel binning). A number of faint point sources are evident, which would be confused with the ULX in typical 30 arcsec [*XMM-Newton*]{} source extraction regions.
![Three colour [*Chandra*]{} ACIS-S image of NGC 5408 X-1 and the surrounding regions, centred on the ULX. The RGB colours correspond to 0.2-1.5 keV, 1.5-2.5 keV and 2.5-8.0 keV respectively. To create the image, all eight observations reported in Table \[obs\] were stacked, and the resulting image was divided by the total exposure map (with 8$\times$8 pixel binning), before being convolved with a 2.4 arcsec FWHM Gaussian to smooth the final image. The green annulus shows the 6–30 arcsec region in which we estimated the extended emission, and the red box corresponds to the footprint of Figure \[arestore\]. The smaller, arbitrarily sized red circles show the location of other sources detected by [wavdetect]{} within 30 arcsec of X-1.[]{data-label="chan_img"}](chan_merged.ps){width="8cm"}
We applied the [src\_extent]{} script to the individual observations to determine whether NGC 5408 X-1 is extended. In preparation for this, we simulated the [*Chandra*]{} ACIS-S point spread function (PSF) for each observation using [ chart]{} ([*Chandra*]{} ray tracer)[^5] and the [marx]{} (model of [*AXAF*]{} response to X-rays version 5.0.0; @davis_etal_2012) software suite. As inputs to [chart]{}, we used the off-axis angles, azimuthal angles and exposure times from each observation, along with absorbed power-law model spectra. These model spectra were fitted to the point source spectra (see below for details of the spectral extraction) in the 0.3–10 keV energy range using [ sherpa]{}, and extracted in an appropriate format using the [ save\_chart\_spectrum]{} tool from the [chart\_spectrum]{} contributed [ sherpa]{} package. The simulated PSFs and the reprocessed event files from each observation were used as inputs to [src\_extent]{}. This script estimates the size of a source using the Mexican Hat Optimization algorithm, and compares this with a PSF. In all but one case, this analysis indicated that the source was consistent with having zero intrinsic size. The sole exception to this was observation 11034, which had an estimated intrinsic size of $0.14^{+0.05}_{-0.04}$ arcsec (which is not flagged as extended by [srcextent]{} at the default threshold criteria, and we note that the PSF and observed source sizes are consistent at the 90 per cent significance level).
In order to further examine the potential extended emission in observation 11034 on finer scales, we used [arestore]{} in [ciao]{} to deconvolve the observed data with a simulated PSF. For this purpose, we created a PSF using [chart]{} and [marx]{} with 10 times the exposure time of the real observation. The [ chart]{} raytraces are deterministic rather than statistical, so increasing the number of photons improves our sampling of the PSF. Images of both the observed data and the simulated PSF were created with 0.5 pixel binning. These were used as inputs to [arestore]{}, which restores images which have been degraded by a blurring function using the Lucy-Richardson deconvolution algorithm [@richardson_1972; @lucy_1974]. We ran the deconvolution algorithm for 100 iterations, after which some additional structure was evident around $\sim$ 0.7 arcsec to the West of the ULX (Figure \[arestore\]). However, this region corresponds to a known asymmetry fixed in spacecraft coordinates,[^6] and the excess of counts is highly likely to be this artefact. To test for any other significant resolved sources we ran [wavdetect]{} on the restored image, with wavelet radii of 1, 2 and 4 image pixels. Apart from the ULX itself and the known artefact, no other sources with greater than 10 counts were detected in the region shown in Figure \[arestore\]. As such, we find no convincing evidence that the ULX can be further spatially resolved by [*Chandra*]{} ACIS.
In addition to the spatial analysis, we also looked for an excess of spectral counts in the regions surrounding the [*Chandra*]{} point source, to further test whether the putative plasma emission component could originate there. We fit this emission with a variety of models, a summary of which is given in Table \[extended\_fits\]. We note that the nature of the features in the [*XMM-Newton*]{} spectrum is not clear, and they may originate from either emission or absorption. Throughout the rest of this work we model them as plasma emission, but do not intend to imply that this is the correct physical interpretation. Since the response of [*Chandra*]{} ACIS changed significantly during the 7 year time span over which the observations were taken, especially at low energies, it is not acceptable to simply extract a merged spectrum. As such, we used [ specextract]{} in [ciao]{} to extract energy spectra from each observation in a 6–30 arcsec annular region centred on NGC 5408 X-1 (Figure \[chan\_img\]), with appropriate parameters for an extended source ([correctpsf]{}=no and [weight]{}=yes). It is clear from Figure 3 that several point sources were resolved by [*Chandra*]{} within $\sim 30$ arcsec of NGC 5408 X-1. As such, we used [wavdetect]{} with [scales]{}=‘1, 2, 4, 8, 16’ and [sigthresh]{}=$1 \times 10^{-5}$ (i.e. $\sim$ 1 false detection over the number of pixels in the image) to identify sources near to X-1. Sources within 30 arcsec of the ULX are highlighted in Figure 3, and were masked out when defining the extended source region. The [ specextract]{} script also produces appropriate response matrix files (RMFs), ancillary response files (ARFs) and background spectra, which were estimated from a large circular region, far from the ULX.
Model$^a$ E range$^b$ Mask$^c$ Goodness-of-fit$^d$
------------------------------------------------------------------------------------ ------------- ---------- ---------------------
$\textsc{tbabs} \times \textsc{mekal}$ 0.3–2 keV y 94%
$\textsc{tbabs} \times \textsc{mekal}^e$ 0.3–2 keV y 76%
$\textsc{tbabs} \times \textsc{mekal}$ 0.3–2 keV n 99%
$\textsc{tbabs} \times (\textsc{mekal} + \textsc{tbabs} \times \textsc{powerlaw})$ 0.3–10 keV n 97%
Summary of the goodness-of-fit statistics for spectral models simultaneously fitted to the extended emission from all of the [*Chandra*]{} observations. All model parameters were assumed to be invariable between observations. We note that from a comparison of the first and second models it seems that fixing a parameter of the model to a non-optimal value actually improves the goodness of fit. This is because different fit and test statistics were used. Notes: $^a$[xspec]{} model fitted to the extended spectrum; $^b$energy range in which the model was fitted; $^c$flag for whether a point source mask was applied when extracting the extended spectrum, to remove contaminating emission from resolved point sources; $^d$goodness of fit of the model in terms of the percentage of 10000 simulated realisations based on the model which had improved Anderson-Darling statistics compared to the real data; $^e$unlike the other models, here the [mekal]{} temperature is fixed to the value given in [[@miller_etal_2013]]{}.
The unbinned extended source spectra, RMFs, ARFs and background spectra were read in to [xspec]{} (version 12.8.1), and simultaneously fitted with a plasma emission model in the 0.3–2 keV energy range. When fitting the model there were insufficient data in each observation to allow us to use binning and the $\chi^2$ statistic, so we used the modified Cash statistic in [xspec]{}, which allows for a background spectrum to be subtracted.[^7] Specifically, the spectral model that we used was $\textsc{tbabs} \times \textsc{mekal}$. The absorption component was fixed to the Galactic value ($5.93 \times 10^{20}~{\rm
cm}^{-2}$; @dickey_and_lockman_1990) and we used the solar abundance table of [@wilms_etal_2000]. As any extended emission would not be expected to vary significantly between observations, the [mekal]{} temperature and normalisation were fixed between the spectra. To assess the goodness-of-fit, we used the [goodness]{} command in [xspec]{} to simulate 10000 realisations from the best fitting model, and calculated the Anderson-Darling statistic to compare these with the real data. Of the 10000 realisations, 94 per cent had an improved fit-statistic compared to the real data, indicating that the model cannot be rejected at 95 per cent significance (equivalent to $2 \sigma$ for Gaussian distributed data). This model had a plasma temperature of $0.64 \pm 0.03~{\rm keV}$, which is in disagreement with the value of $0.87 \pm 0.02~{\rm keV}$ reported by [@miller_etal_2013] for a plasma emission spectrum fitted to the residual features in [*XMM-Newton*]{} EPIC data. However, when we repeated the above goodness-of-fit test with the [mekal]{} temperature fixed at 0.87 keV, we found that this variant of the model cannot be rejected at high significance either (76 per cent of 10000 simulated data sets had a lower Anderson-Darling test statistic; note that the test statistic differs from the fit statistic). We used the [cflux]{} convolution model in [xspec]{} to estimate the observed 0.3–2 keV flux of the extended emission, integrated over the annular region as $(2.5 \pm 0.2) \times 10^{-14}~{\rm erg~cm^{-2}~s^{-1}}$. At the distance of NGC 5408, this is equivalent to a 0.3–2 keV luminosity of $(7.0 \pm 0.5) \times
10^{37}~{\rm erg~s^{-1}}$. If we renormalise this to account for the difference in region size between our annular region with point source masks and a 30 arcsec circle, then this is increased by $\sim 6$ per cent to $(7.4 \pm 0.5)
\times 10^{37}~{\rm erg~s^{-1}}$. This is $\sim 1/3$ of the [mekal]{} flux seen in the [*XMM-Newton*]{} data.
The [*XMM-Newton*]{} source extraction region does not exclude the faint point sources resolved by [*Chandra*]{}, so we test whether these could be contributing any of the missing [mekal]{} flux. To do this, we repeated the above analysis in the same annular region without masking out the point sources. An absorbed [mekal]{} model could not be rejected at the equivalent of $3 \sigma$ significance in the 0.3–2 keV range (99 per cent of 10000 simulated spectra had an improved fit statistic). We estimated the observed 0.3–2 keV flux using the [cflux]{} model in [xspec]{} to be $(3.2 \pm 0.2) \times 10^{-14}~{\rm erg~cm^{-2}~s^{-1}}$ integrated over the annulus. After correcting for the difference in region size, this is around 20 per cent higher than the previous estimate with point sources excluded, but it still can only account for $\sim$ 37 per cent of the [*XMM-Newton*]{} [mekal]{} luminosity. However, when we tried adding a power-law to the spectral model (${\textsc{tbabs} \times (\textsc{mekal} + \textsc{tbabs} \times \textsc{powerlaw}})$ in [xspec]{}, with the model fitted in the 0.3–10 keV range, and having a rejection probability of 97 per cent) this resulted in a larger increase in the estimated 0.3–2 keV observed flux. Most of the difference in observed flux between the two models is in the 1–2 keV energy band, and it is likely due to fitting the models over different energy ranges. For the latter model the total observed flux was $(4.7 \pm 0.3) \times 10^{-14}~{\rm erg~cm^{-2}~s^{-1}}$, but only $(1.1 \pm 0.4) \times 10^{-14}~{\rm erg~cm^{-2}~s^{-1}}$ originated in the [mekal]{} component. As such, much of the faint flux would have been confused with the continuum spectrum in [*XMM-Newton*]{}, not the putative plasma emission features. Since the [mekal]{} flux here is lower than our estimate with the point sources excluded, this may suggest that there is still some unresolved faint point source emission. We therefore take $(7.4 \pm 0.5) \times 10^{37}~{\rm erg~s^{-1}}$ as a conservative estimate of the extended contribution to the [*XMM-Newton*]{} [mekal]{} component luminosity, but caution that it should be considered as an upper-limit.
We also examined point source spectra from the ULX to test whether there was evidence of [mekal]{}-like features, which could indicate that the spectral features have a localised origin close to the ULX. A summary of models fitted to the ULX spectra is given in Table \[ulx\_fits\]. Again, we used [ specextract]{} in [ciao]{} to extract the source and background spectra, RMFs and ARFs from each of the [*Chandra*]{} observations. Source spectra were extracted from 5 arcsec circular regions centred on the ULX, and background spectra from large, circular regions located on the same chip, away from the ULX and other point sources. As is appropriate for a point source, we used parameters [correctpsf]{} = yes and [weight]{} = no. The spectra and associated files were grouped and binned to a minimum of 20 counts per energy bin by the [specextract]{} script, allowing us to use the $\chi^2$ statistic.
Model$^a$ $\chi^2.{\rm dof}$
-------------------------------------------------------------------------------------------------------- --------------------
$\textsc{tbabs} \times \textsc{tbabs} \times \textsc{powerlaw}$ 716.7/623
$\textsc{tbabs} \times (\textsc{tbabs} \times \textsc{powerlaw} + \textsc{mekal})$ 622.4/621
$\textsc{tbabs} \times \textsc{tbabs} \times \textsc{simpl} \times \textsc{diskbb}$ 628.2/607
$\textsc{tbabs} \times (\textsc{tbabs} \times \textsc{simpl} \times \textsc{diskbb} + \textsc{mekal})$ 615.5/605
$\textsc{tbabs} \times (\textsc{tbabs} \times \textsc{simpl} \times \textsc{diskbb} + \textsc{mekal})$ 617.7/606
$\textsc{tbabs} \times \textsc{tbabs} \times (\textsc{simpl} \times \textsc{diskbb} + \textsc{mekal})$ 613.5/612
: Spectral models used to fit the ULX[]{data-label="ulx_fits"}
Summary of the goodness-of-fit statistics for a variety of spectral models simultaneously fitted to all of the [*Chandra*]{} ULX spectra. In all of the models the first absorption component was set equal to the Galactic neutral hydrogen column density in the direction of the source. The second absorption component was left free to model intrinsic absorption in the source/absorption in the host galaxy. This was free to vary between observations, except in the final model where it was assumed to be invariable. Similarly, ULX components in all of the models ([powerlaw]{}, [simpl]{} and [diskbb]{}) were allowed to vary between observations, whilst the [mekal]{} parameters was assumed to be invariable. Notes: $^a$[xspec]{} model fitted to the extended spectrum.
We read all of the grouped [*Chandra*]{} point source spectra into [xspec]{} and simultaneously fitted them with a simple phenomenological model of emission from a ULX, a doubly absorbed power-law ($\textsc{tbabs} \times \textsc{tbabs} \times
\textsc{powerlaw}$ in [xspec]{}). The first absorption component was fixed to the Galactic value, and the second was left free to model intrinsic absorption in the host galaxy and/or the ULX itself. The power-law spectral index and normalisation were allowed to vary between observations, to model any spectral variability in the ULX. Although it was not strictly necessary to simultaneously fit the spectra, we chose to do this to later allow us to assess the significance of an additional invariable [mekal]{} component. This model has a goodness-of-fit statistic of $\chi^2/{\rm dof} = 716.7/623$, which cannot formally be rejected at $3 \sigma$ significance. Next, we repeat the fitting process with an additional [mekal]{} component, which is kept constant between observations ($\textsc{tbabs} \times (\textsc{tbabs} \times \textsc{powerlaw} + \textsc{mekal})$ in [ xspec]{}). Keeping the [mekal]{} component fixed is reasonable if it originates in diffuse emission, although that is not necessarily the case here where we are looking at the point source. However, the data are such that we have insufficient spectral counts to constrain a variable [mekal]{} component, so we must make the assumption that it does not vary strongly. This additional component results in a large improvement in the goodness-of-fit over an absorbed power-law alone ($\Delta \chi^2 = 94.3$ for 2 degrees of freedom). However, the 0.3–2 keV observed luminosity in the [mekal]{} component ($(4.1 \pm 0.4) \times 10^{38}~{\rm erg~s^{-1}}$) is around double that seen in the [*XMM-Newton*]{} EPIC data. Clearly, this is not physically realistic, as the putative [mekal]{} contribution to the [*XMM-Newton*]{} spectrum is here integrated over a much smaller area. Instead, we suggest that as the [*XMM-Newton*]{} spectrum of NGC 5408 X-1 has a two component form (e.g. @gladstone_etal_2009 [@middleton_etal_2015]), perhaps we are fitting the plasma emission to the soft emission component from the ULX itself, not the more subtle spectral features. Instead, a two component model of emission from a ULX may be more appropriate.
{width="18cm"}
[*XMM-Newton*]{} ULX spectra have previously been fitted with models of emission from a multi-colour-disc plus Comptonisation in an optically thick medium ($\textsc{diskbb} + \textsc{comptt}$). However, the energy response of [*Chandra*]{} ACIS is such that we could not constrain the high energy curvature, even if it was present. So, we instead opted to use a multi-colour-disc with an empirical power-law approximation of Comptonisation ($\textsc{tbabs} \times \textsc{tbabs} \times \textsc{simpl} \times
\textsc{diskbb}$ in [xspec]{}). One key difference between [simpl]{} and a power-law is that it does not diverge at low energies. This was potentially advantageous here, as we were trying to test for the presence of a soft, faint diffuse emission component. As previously, we simultaneously fitted the model to all eight [ *Chandra*]{} spectra, with parameters free to vary between observations. Again, we did this, to allow for a comparison with an additional invariable [mekal]{} component. This results in a goodness-of-fit statistic of $\chi^2/{\rm dof} = 628.2/607$, which cannot be rejected at 2 $\sigma$ significance. From the model, we estimated inner disc temperatures of 0.17–0.22 keV and power-law spectral indices of typically 2.2–2.4, with the exception of observation 13018, which required a softer spectral index of $2.88^{+0.10}_{-0.08}$.
The addition of the [mekal]{} component to the two component ULX model (i.e. $\textsc{tbabs} \times (\textsc{tbabs} \times \textsc{simpl} \times \textsc{diskbb} + \textsc{mekal})$, Figure \[spec\]) resulted in a reduction to the fit statistic ($\chi^2/{\rm dof} = 615.5/605$; i.e. $\Delta \chi^2 = 12.7$ for 2 degrees of freedom), with a plasma temperature of $0.27^{+0.03}_{-0.02}$ keV. We note that this additional spectral component is not strongly required by the [*Chandra*]{} data, however we are not testing for its significance here, we are simply constraining how much flux it could contain. Our estimate of the [mekal]{} temperature is in disagreement with the value of $0.87 \pm 0.02$ keV reported by [@miller_etal_2013]. However, we repeated the spectral fitting with the [mekal]{} temperature fixed at 0.87 keV, and compared the fits using an F-test. Setting the [mekal]{} plasma temperature in this way resulted in an increase in $\chi^2$ of 2.2 for 1 degree of freedom. This corresponds to an F-test probability of 0.14, thus the improvement in the fit from allowing the plasma temperature to vary is only very marginal ($<2 \sigma$ significance). We estimated the observed 0.3–2 keV luminosity of the [mekal]{} component using the [cflux]{} convolution model in [ xspec]{} to be $L_{\rm X} = 1.3^{+0.7}_{-0.4} \times 10^{38}~\ergs$ (or $1.3^{+0.4}_{-0.3} \times 10^{38}~\ergs$ when the [mekal]{} temperature was fixed to 0.87 keV). This luminosity is consistent with the point source providing the observed [*XMM-Newton*]{} putative [mekal]{} component that remains after subtracting the spatially resolved diffuse emission. In this model, the physical interpretation of the second [tbabs]{} component is not clear, as it accounts for both absorption in the host galaxy and intrinsic to the ULX. Therefore, at least the galactic component of this absorption should be applied to the putative plasma emission features, and arguably the intrinsic component too, depending on their physical origin. As such, we also tried $\textsc{tbabs} \times \textsc{tbabs} \times (\textsc{simpl} \times \textsc{diskbb} + \textsc{mekal})$, with the first absorption component set to the Galactic value and the second free to vary, but fixed between observations. In this iteration the mekal temperature was $kT = 0.25 \pm 0.02~{\rm keV}$ and the 0.3–2 keV observed component luminosity was slightly higher at $L_{\rm X} = 2.2^{+0.4}_{-0.6} \times 10^{38}~{\rm erg~cm^{-2}~s^{-1}}$, but is still consistent with the previous estimate. This plasma temperature is still in disagreement with the value reported by [@miller_etal_2013], but the luminosity is consistent with that of the entire [mekal]{} component seen in [*XMM-Newton*]{} data.
Discussion and conclusions {#discussion}
==========================
Previous [*XMM-Newton*]{} analyses of NGC 5408 X-1 have identified spectral residuals that can be modelled as emission from a thermal plasma. As such, it has previously been assumed that these residuals originated in star-formation related diffuse emission in the host galaxy, and they were not associated with the ULX itself (e.g. @strohmayer_and_mushotzky_2009 [@miller_etal_2013]). However, [@middleton_etal_2014] noted that this would require the diffuse emission within the [*XMM-Newton*]{} source extraction region to be in excess of that expected for the entire galaxy based on its star-formation rate. Instead, [@middleton_etal_2014] suggested that the residuals may originate in a radiation-pressure-dominated wind coming from the ULX. In this work, we have exploited the spatial resolution of [*Chandra*]{} to place the best possible limits on the origin of the putative diffuse plasma emission.
From our analysis, we find no strong evidence that the ULX deviates from remaining essentially point-like in [*Chandra*]{} ACIS data. Whilst a hand-full of faint point sources not evident in the [*XMM-Newton*]{} observations are resolved, the ULX does not appear to sit within a strongly peaked distribution of diffuse emission. This is not particularly unexpected, given that the ULX is largely displaced from the major star-formation regions in NGC 5408. This result is also largely confirmed by a spectral analysis of the regions surrounding X-1. Although we do find an excess of counts within 30 arcsec of the ULX, these are only sufficient to account for a 0.3–2 keV luminosity of $\lesssim 7.4 \times 10^{37}~{\rm erg~s^{-1}}$. Admittedly, this limit is a factor $\sim 2$ greater than the predicted star-formation related luminosity integrated over the entirety of NGC 5408 [@middleton_etal_2014], but it may contain some degree of unresolved point source emission, and is still only sufficient to contribute $\sim 1/3$ of the 0.3–2 keV [mekal]{} flux seen in the [*XMM-Newton*]{} data ($\sim 2.5 \times10^{38}~{\rm erg~s^{-1}}$; calculated from @miller_etal_2013).
To test whether the [*Chandra*]{} ULX point source spectra could be harbouring the remainder of the putative plasma flux observed in the [*XMM-Newton*]{} data, we fitted them with a model for the continuum spectrum from a ULX, plus an additional [mekal]{} component. The [mekal]{} component had a 0.3–2 keV luminosity of $1.3^{+0.7}_{-0.4} \times
10^{38}~\ergs$ (or $1.3^{+0.4}_{-0.3} \times 10^{38}~\ergs$ assuming a plasma temperature of 0.87 keV), thus is consistent with contributing the missing flux. We know from the [*HST*]{} data that a small optical association falls within the 5 arcsec source extraction region [@grise_etal_2012], but [*Chandra*]{} sees a point source and it seems very unlikely that a minor stellar association could produce $\sim 4$ times the diffuse X-ray emission than is expected over the entire galaxy. Indeed, we very crudely estimate the X-ray luminosity of this association as $\sim 2 \times 10^{33}$–$2 \times 10^{35}~{\rm erg~s^{-1}}$ by assuming that all $\sim 20$ of the stars are colliding wind binaries [@mauerhan_etal_2010]. We note that the additional [mekal]{} component is not strongly required by the data, although this is hardly surprising given that it contributes only $\sim
4.6$ per cent of the flux [@strohmayer_etal_2007], and we are much more photon-limited here than in the [*XMM-Newton*]{} observations. Not only does [*XMM-Newton*]{} EPIC have greater effective area, but in the case of NGC 5408 X-1 the observations also have much greater exposure times: 6 [*XMM-Newton*]{} observations have between 28.6 and 88.2 ks of good time [@middleton_etal_2014], compared to at best 12.2 ks with [*Chandra*]{}. However, it is reasonable to include the [mekal]{} component here, as we have prior knowledge that two component phenomenological ULX models excluding it are ruled out by the [*XMM-Newton*]{} data (e.g. @strohmayer_etal_2007).
The majority of the putative plasma flux remains spatially unresolved, and is not displaced from the [*Chandra*]{} point source. Therefore, the putative [mekal]{} features in [*XMM-Newton*]{} data cannot be dominated by star-formation related diffuse emission. Rather, we favour scenarios where they are associated with the ULX. Similar spectral features have been reported in a number of other ULXs, including: Ho II X-1 [@miyaji_etal_2001; @dewangan_etal_2004], NGC 4395 X-1 [@stobbart_etal_2006], NGC 4559 X-1 [@roberts_etal_2004], NGC 6946 X-1 [@middleton_etal_2014] and NGC 7424 ULX2 [@soria_etal_2006b]. Notably these sources all have soft X-ray spectra, at least in the observations where residuals are reported [@stobbart_etal_2006; @sutton_etal_2013b; @middleton_etal_2014]. Furthermore, [@soria_etal_2006b] reported detections of NGC 7424 ULX2 with both hard and soft spectra ($\Gamma = 1.8^{+0.2}_{-0.1}$ and $2.2 \pm 0.2$ respectively when fitted with absorbed power-laws), with the plasma-like component only being detected when the source had a soft spectrum, thus ruling out the possibility that it originates from underlying emission. The fact that we typically see these plasma emission-like features in ULXs with soft X-ray spectra strongly suggests that they could be associated with a super-critical wind. However, similar features have been identified in ULXs with hard X-ray spectra, e.g. NGC 1313 X-1 [@bachetti_etal_2013] and Ho IX X-1 [@walton_etal_2014]. This does not necessarily rule out an association with an outflow. The wind would be expected to become physically diffuse as it moves away from the accretion disc, thus the optically thin phase could intercept an observers line-of-sight at low inclinations, even if the optically thick material does not. Additionally, the wind opening angle is expected to vary with accretion rate [@king_2008], so ULXs could shift between hard and soft ultraluminous states. Indeed, such shifts have been reported in NGC 1313 X-1 and NGC 5204 X-1 [@sutton_etal_2013b]. Furthermore, Middleton et al. (submitted) show that the strength of the the putative absorption features in NGC 1313 X-1 are anti-correlated with spectral hardness, which supports the interpretation that they are associated with an outflow.
One possibility, as suggested by [@middleton_etal_2014], is that putative mekal features do not originate from plasma emission, rather they are misdiagnosed absorption features from a blue-shifted, partially ionised, optically thin absorber. Such a material could originate from a radiation-pressure dominated wind driven by super-critical accretion, at large distances from the black hole. Alternatively, the features may truly be associated with thermal plasma emission. This could potentially occur if ejecta in the ULX wind collisionally ionise a nearby cloud of material, such as the outer-layers of a highly evolved massive stellar companion [@roberts_etal_2004]. However, observational data [@middleton_etal_2011a; @sutton_etal_2013b; @middleton_etal_2015] and simulations [@takeuchi_etal_2013] suggest that the distinction between hard and soft ultraluminous sources is determined at least in part by the observation angle. As there is no [*a priori*]{} reason that such plasma emission would be anisotropic, the stronger detections in soft ultralumious ULXs is troubling for the collisionally ionised plasma emission scenario. For this reason, we tend towards favouring absorption as the more likely explanation. Future missions with high spectral resolution, such as [*Astro-H*]{} and [*Athena*]{}, will be able to provide definitive tests of the nature of the features we discuss here. However, in the meantime further diagnostics can come from examining their evolution with changing X-ray continuum spectra [@middleton_etal_2015b].
[*Facilities*]{}:
[^1]: One of the [*XMM-Newton*]{} observations of NGC 1313 X-1 analysed by [@walton_etal_2013] was classified as soft ultraluminous and a second as ambiguous (either hard of soft ultraluminous) by [@sutton_etal_2013b]. However, out of a total of 164 ks of good time in the EPIC pn detector analysed by [@walton_etal_2013], these observations only contributed 9 and 10 ks respectively. A further observation with 3 ks of good EPIC pn time was not classified by [@sutton_etal_2013b]. As such, the vast majority of the NGC 1313 X-1 data studied by [@walton_etal_2013] comes from epochs when the source had a hard ultraluminous spectrum.
[^2]: <http://cxc.harvard.edu/toolkit/pimms.jsp>
[^3]: <http://heasarc.gsfc.nasa.gov>
[^4]: <http://cxc.harvard.edu/ciao/>
[^5]: <http://cxc.harvard.edu/chart/>
[^6]: [http://cxc.harvard.edu/ciao/caveats/psf\_artifact.html
](http://cxc.harvard.edu/ciao/caveats/psf_artifact.html
)
[^7]: [
http://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/X\\allowbreak S\\allowbreak a\\allowbreak p\\allowbreak p\\allowbreak e\\allowbreak n\\allowbreak d\\allowbreak i\\allowbreak x\\allowbreak Statistics.html](
http://heasarc.gsfc.nasa.gov/xanadu/xspec/manual/X\allowbreak S\allowbreak a\allowbreak p\allowbreak p\allowbreak e\allowbreak n\allowbreak d\allowbreak i\allowbreak x\allowbreak Statistics.html)
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Compressing or cooling a fluid typically enhances its static interparticle correlations. However, there are notable exceptions. Isothermal compression can reduce the translational order of fluids that exhibit anomalous waterlike trends in their thermodynamic and transport properties, while isochoric cooling (or strengthening of attractive interactions) can have a similar effect on fluids of particles with short-range attractions. Recent simulation studies by Yan [*et al.*]{} \[Phys. Rev. E [**76**]{}, 051201 (2007)\] on the former type of system and Krekelberg [*et al.*]{} \[J. Chem. Phys. [**127**]{}, 044502 (2007)\] on the latter provide examples where such structural anomalies can be related to specific changes in second and more distant coordination shells of the radial distribution function. Here, we confirm the generality of this microscopic picture through analysis, via molecular simulation and integral equation theory, of coordination shell contributions to the two-body excess entropy for several related model fluids which incorporate different levels of molecular resolution. The results suggest that integral equation theory can be an effective and computationally inexpensive first-pass tool for assessing, based on the pair potential alone, whether new model systems are good candidates for exhibiting structural (and hence thermodynamic and transport) anomalies.'
author:
- 'William P. Krekelberg'
- Jeetain Mittal
- Venkat Ganesan
- 'Thomas M. Truskett'
title: |
Structural anomalies of fluids:\
Origins in second and higher coordination shells
---
[^1]
Introduction {#sec:intro}
============
A bulk equilibrium fluid is translationally invariant; i.e., its one-particle density, $\rho^{(1)}({\bf r})=\rho$, is constant. Nonetheless, assuming spherically-symmetric interactions, the local density $\rho g(r)$ surrounding a reference particle is a function of distance $r$ from its center, where $g(r)$ is the radial distribution function (RDF) of the fluid.[@Hansen2006Theory-of-Simpl] Although the RDF depends on both the form of the interparticle interactions and the thermodynamic state, some features of its shape are fairly general. For example, the RDF vanishes for $r$ less than the effective exclusion diameter of the particles. For larger $r$, it shows an oscillatory decay toward unity with peaks loosely corresponding to coordination “shells”. Away from the critical point, the structure of the RDF typically persists for distances comparable to a few particle diameters, reflecting the short range of the interparticle correlations.
Studies of the liquid state have primarily focused on the particles in the first coordination shell. This is due in part to the important role that nearest neighbors are expected to play in determining many physicochemical properties. For example, both the non-ideal contribution to the equation of state of the hard-sphere fluid (see, e.g., Ref. ) and the collision frequency in Enskog theories for transport processes[@Chapman1970The-Mathematica] scale with the “contact” density $\rho g(\sigma)$, where $\sigma$ is the particle diameter. The hard-sphere equation of state is the standard reference system for perturbation theories.[@Hansen2006Theory-of-Simpl] It also accurately predicts how the thermodynamics of “hard-sphere” colloidal suspensions relate to their structure, as has recently been experimentally verified by confocal microscopy.[@Dullens2006Direct-measurem] Furthermore, analysis of first-shell contributions to hydration structure and thermodynamics helps to understand and make predictions about a wide variety of aqueous solution properties.[@Nemethy1962Structure-of-Wa; @Pratt1985Theory-of-Hydro; @Beglov1994Finite-represen; @Matubayasi1994Thermodynamics-; @Silverstein1999Molecular-model; @Lockwood1999Evaluation-of-F; @Lockwood2000Functional-Grou; @Hummer2000New-perspective; @Paulaitis2002Hydration-theor; @Asthagiri2004Hydration-Struc; @Dill2005MODELING-WATER-]
Although the second shell of the RDF has received comparatively less attention, there is evidence that it also contains structural information relevant for understanding nontrivial behaviors of liquids. One notable feature is the shoulder[@Truskett1998Structural-prec] that it develops near the freezing transition, which in turn becomes a pronounced split peak[@Finney1970Random-Packings; @Bennett1972Serially-Deposi; @Rahman1976Molecular-dynam; @Stillinger1984Packing-Structu; @Hiwatari1984Structural-char; @Clarke1987Numerical-simul; @Tsumuraya1990Local-structure; @Snook1991Structure-of-co; @Donev2005Pair-correlatio; @OMalley2005Structure-of-ha] in supercooled liquid and glassy states. Analysis of the configurations that give rise to this structural motif indicate that it reflects frustration of icosahedral[@Tsumuraya1990Local-structure] and emerging crystalline[@Truskett1998Structural-prec; @OMalley2005Structure-of-ha] order in the fluid. Understanding how these these types of structural features connect to relaxation processes of supercooled liquids is an active area of research (see, e.g., Refs. ).
In this work, however, we focus on the second and higher coordination shells of the RDF for a different reason: to understand their role in the [*structural anomalies*]{} of fluids. Interparticle correlations of most fluids are enhanced upon (i) compression or (ii) cooling (alternatively, strengthening of interparticle attractions). Nonetheless, there are a few systems of scientific interest that exhibit notably different behaviors. For example, compression induced disordering occurs in water and other fluids with anomalous waterlike trends in their thermodynamic and transport properties.[@Errington2001Relationship-be; @Shell2002Molecular-struc; @Truskett2002A-Simple-Statis; @Yan2005Structural-Orde; @Yan2006Family-of-tunab; @Oliveira2006Structural-anom; @Mittal2006Quantitative-Li; @Mittal2006Relationship-be; @Sharma2006Entropy-diffusi; @Oliveira2007Interplay-betwe; @Agarwal2007Ionic-melts-wit; @Agarwal2007Waterlike-Struc; @Yan2007Structure-of-th; @Oliveira2007in-press] Cooling (or attraction) induced disordering, on the other hand, can occur in fluids of particles with short-range attractive (SRA) interactions,[@Mittal2006Quantitative-Li; @Krekelberg2007How-short-range] such as concentrated suspensions of colloids.
These anomalies do not appear to be first-shell effects. Rather, they reflect how structuring in second and more distant coordination shells responds to changes in thermodynamic or system parameters. For example, Yan [*[et al.]{}*]{}[@Yan2007Structure-of-th] recently demonstrated in an insightful paper how the structural anomaly of the five-site transferable interaction potential (TIP5P) model[@Mahoney2000A-five-site-mod] for water is quantitatively related to compression induced translational disordering of molecules in the second coordination shell. Similarly, Krekelberg [*[et al.]{}*]{}[@Krekelberg2007How-short-range] have shown that the cooling (or attraction) induced structural anomaly of a square-well SRA fluid is due to weakening of second- and higher-shell pair correlations.
The goal here is to study the generality of the above findings. It is known that a number of models, with varying levels of molecular resolution, can qualitatively predict the structural anomalies of the aforementioned systems. [@Errington2001Relationship-be; @Oliveira2006Structural-anom; @Yan2007Structure-of-th; @Mittal2006Quantitative-Li; @Mittal2006Relationship-be; @Krekelberg2007How-short-range; @Shell2002Molecular-struc; @Truskett2002A-Simple-Statis; @Oliveira2007Interplay-betwe; @Agarwal2007Ionic-melts-wit; @Sharma2006Entropy-diffusi; @Agarwal2007Waterlike-Struc; @Oliveira2007in-press; @Yan2005Structural-Orde; @Yan2006Family-of-tunab] But do the anomalies exhibited by lower resolution models have the same microscopic origins as those of more detailed models? Moreover, can the behavior of the lower resolution models be predicted, at least qualitatively, by integral equation theory? If so, it would suggest that integral equation theory might serve as a valuable first-pass tool in assessing, based on the pair potential alone, whether new model systems might be good candidates for exhibiting structural anomalies.
Furthermore, although the structurally anomalous trends analyzed here are interesting in their own right, there is a more compelling reason to try to understand their origins. In short, they appear to be closely linked to other distinctive dynamic and thermodynamic behaviors. For example, in addition to being “structurally anomalous”, cold liquid water is also “dynamically anomalous” in that its self-diffusivity increases upon isothermal compression and “thermodynamically anomalous” in that its volume increases upon isobaric cooling. Errington and Debenedetti[@Errington2001Relationship-be] first noticed that these particular anomalies form a cascade in the temperature-density plane for the extended simple point charge (SPC/E) model[@Berendsen1987The-Missing-Ter] of water.[@Errington2001Relationship-be] Specifically, the thermodynamic anomaly occurs only for state points that also exhibit the dynamic anomaly. The dynamic anomaly, in turn, is only present for states that also exhibit the structural anomaly. Strong correlations between these three basic types of anomalies have since been documented for a wide variety of model systems with waterlike properties.[@Oliveira2006Structural-anom; @Yan2007Structure-of-th; @Shell2002Molecular-struc; @Truskett2002A-Simple-Statis; @Oliveira2007Interplay-betwe; @Sharma2006Entropy-diffusi; @Oliveira2007in-press; @Kumar2005Static-and-dyna; @Esposito2006Entropy-based-m; @Netz2006Thermodynamic-a; @Xu2006Thermodynamics-; @Szortyka2007Diffusion-anoma]
A similar connection between structural and dynamic anomalies has now also been identified for model SRA fluids.[@Mittal2006Quantitative-Li; @Krekelberg2007How-short-range] In those systems, the most commonly studied dynamic anomaly is an [ *increase*]{} in self-diffusivity upon cooling (or strengthening of interparticle attractions), which can occur at sufficiently high particle concentrations.[@Eckert2002Re-entrant-Glas; @Pham2002Multiple-Glassy; @Bergenholtz1999Nonergodicity-t; @Fabbian1999Ideal-glass-gla; @Dawson2001Higher-order-gl; @Zaccarelli2002Confirmation-of; @Sciortino2002One-liquid-two-] Krekelberg [*et al.*]{}[@Krekelberg2007How-short-range] discovered that the self-diffusivity anomaly for a square-well SRA fluid occurs only for state points that also exhibit the cooling (or attraction) induced structural anomaly discussed above. In other words, it appears that SRA fluids can also display a cascade of anomalies qualitatively similar to those of waterlike fluids.
Although structural and dynamic properties of these systems show unusual dependencies on quantities like temperature or density, the correlations between structure and dynamics are often similar to those found in simpler liquids (e.g., the hard-sphere fluid).[@Mittal2006Quantitative-Li; @Mittal2006Relationship-be; @Errington2006Excess-entropy-; @Sharma2006Entropy-diffusi; @Krekelberg2007How-short-range] In fact, it was recently demonstrated[@Errington2006Excess-entropy-] that the cascade of anomalies of one waterlike model system can be semi-quantitatively predicted based only on knowledge of the state dependencies of excess entropy, which measures structural order,[@Truskett2000Towards-a-quant] and quasi-universal excess entropy scalings[@Rosenfeld1999A-quasi-univers; @Rosenfeld1977Relation-betwee; @Dzugutov1996A-univeral-scal] for the transport coefficients. All of this suggests that investigations like the present one, which probe the physics of structural anomalies, might also provide insights into dynamic and thermodynamic anomalies as well.
Methods {#sec:sim_methonds}
=======
We used molecular dynamics simulation and integral equation theory to examine various models from two classes of fluids known to exhibit structural anomalies: those with waterlike properties and those comprising particles with SRA interactions. For the integral equation theory analysis, we numerically solved the Ornstein-Zernike equation[@Ornstein1914integral-equati] together with an approximate closure relation using the method of Labik *et al.*[@Labik1985A-RAPIDLY-CONVE] In the discussion of the models below, we mention the specific closures employed and provide further details about the molecular simulations.
We did not perform a systematic study here to determine which of many possible closure relations[@Hansen2006Theory-of-Simpl] provides the most quantitatively accurate description for each model. Rather, our focus was to explore whether integral equation theory solved with standard closure relations, such as Percus-Yevick (PY)[@Percus1958Analysis-of-Cla] or hypernetted-chain (HNC)[@van-Leeuwen1959New-method-for-], can in fact qualitatively predict both the structural anomalies and their microscopic origins in the RDF. Molecular simulations of the model systems provide the data necessary to make that basic determination.
Waterlike fluid models
----------------------
We investigated two waterlike models: (1) the SPC/E [@Berendsen1987The-Missing-Ter] model and (2) a lower resolution “core-softened” [@Oliveira2006Thermodynamic-a; @Oliveira2006Structural-anom] model. We chose the SPC/E model because it represents one of the most commonly studied effective pair potentials for water, and it is known to qualitatively reproduce many of water’s distinctive thermodynamic, dynamic, and structural properties.[@Errington2001Relationship-be] As a result, it provides a reasonable baseline against which to compare simpler, lower resolution models. Details of our molecular dynamics simulations of the SPC/E model are the same as reported in Ref. .
The core-softened model[@Oliveira2006Thermodynamic-a; @Oliveira2006Structural-anom] that we studied is more schematic. It is defined by the effective pair potential $U_{\RM{CS}}(r)$ \[see Figure \[fig:CS\_potential\](a)\], $$\label{eq:core_softened_potential}
U_{\RM{CS}}(r)=4\epsilon\Biggl[\biggl(\frac{\sigma}{r}\biggr)^{12}-\biggl(\frac{\sigma}{r}\biggr)^6\Biggr]
+5\epsilon\RM{exp}\Biggl[-\biggl(\left[\frac{r}{\sigma}\right]-0.7\biggr)^2\Biggr],$$ where $\epsilon$ is the characteristic energy scale. The main idea behind this potential is that it has two different kinds of repulsions that act at different length scales. The harsh $(\sigma/r)^{12}$ repulsion defines the effective hard-core diameter ($\sigma$), while the softer Gaussian repulsion extends to considerably larger distances. The end result is that the average interparticle separation, and hence the density, of this fluid can depend sensitively on both temperature and pressure. The model is similar to cold water in that it favors locally open (low-density) structures at moderate pressure and low temperature, but can collapse to denser structures when compressed or heated enough to overcome the soft Gaussian repulsion. Although this low resolution model does not provide an accurate molecular-level description of water, it does qualitatively reproduce many of its peculiar thermodynamic, structural, and kinetic behaviors.[@Oliveira2006Thermodynamic-a; @Oliveira2006Structural-anom; @Mittal2006Relationship-be; @Oliveira2007Interplay-betwe]
To compute the properties of the core-softened model, we performed molecular dynamics simulations in the microcanonical ensemble using $N=1000$ identical particles of mass $m$. We used the velocity-Verlet technique for integrating the equations of motion with a time step of $\Delta t=0.002\sigma\sqrt{m/\epsilon}$. For the integral equation theory analysis, we employed the HNC closure. We chose the HNC approximation because of its ability to describe the structure of another fluid with a soft Gaussian repulsion, the Gaussian-core model.[@Louis2000Mean-field-flui] We investigated both the SPC/E and core-softened models over a wide range of density and temperature, where they are known to exhibit structural anomalies.[@Errington2001Relationship-be; @Oliveira2006Thermodynamic-a; @Oliveira2006Structural-anom; @Mittal2006Relationship-be; @Oliveira2007Interplay-betwe]
![(a) Pair potential of the core-softened model $U_{\RM{CS}}(r/\sigma)/\epsilon$ \[see Eq. \]. (b) Pair potential of the model SRA fluid $U_{\RM{SRA}}(r/a)/\kB T$ discussed in the text for various values of polymer concentration $\Pp$. Further details on this SRA model are provided in Refs. and .[]{data-label="fig:CS_potential"}](CS_SRA_pot-scaled.eps)
SRA fluid models {#sra_models}
----------------
The first SRA fluid model that we considered qualitatively describes a solution of (explicit) colloidal particles attracted to one another by depletion interactions due to the presence of (implicit) non-adsorbing polymers. The details of the colloidal pair potential are provided in Refs. and , but we discuss some of its main features below. The colloids are spherical and their effective interactions consist of three parts. The first is a steeply repulsive, essentially hard-sphere (HS), contribution $U_{\mathrm{HS}}
(r) = \kB T (2 a/r)^{36}$, where $2a$ is the colloid diameter, $\kB$ is Boltzmann’s constant, and $T$ is temperature. The second term represents the aforementioned polymer-induced depletion attraction $U_{\RM{AO}} (r)$, approximated by the Asakura-Oosawa[@Asakura1958depletion] potential. The strength of this attraction is proportional to the volume fraction of polymers in solution $\Pp$, while the range is controlled by the radius of gyration of the polymers $R_{\mathrm g}$, set in this case to $a/5$. A soft repulsion $U_{\RM{R}}$ is also added to the effective interparticle potential[@Puertas2002Comparative-Sim] to prevent fluid-fluid phase separation. Figure \[fig:CS\_potential\](b) displays the total colloidal potential $U_{\RM{SRA}}=U_{\RM{HS}}+U_{\RM{AO}}+U_{\RM{R}}$ for three different polymer concentrations $\Pp$. The details of the molecular dynamics simulations that we performed for this fluid are the same as those reported in Ref. , with one exception. In the original study, a weakly polydisperse system was investigated. Here, all particles considered had identical radius $a$ and mass $m$. The advantage of focusing on a monodisperse system is that the pair correlations are unambiguously described by a single RDF, which facilitates the analysis discussed in the next section. For the integral equation theory of this SRA fluid, we employed the PY closure. The PY approximation is a natural choice here due to its simplicity and its ability to describe the structure of liquids with harshly repulsive, short-range potentials,[@Hansen2006Theory-of-Simpl] (in particular, other SRA fluids[@Dawson2001Higher-order-gl]).
We also considered a simpler model SRA fluid: a system of identical square-well particles with attractive well depth -$\epsilon$ and width 0.03$\sigma$, where $\sigma$ represents the hard-core diameter. This model is similar to others known to exhibit structural[@Krekelberg2007How-short-range] and dynamic [@Krekelberg2007How-short-range; @Zaccarelli2002Confirmation-of] anomalies. We also use the PY closure in our integral equation theory analysis of this fluid for the reasons mentioned above.
Quantification of structural order
----------------------------------
For each of the model fluids, we calculated the state dependencies of $-s_2/\kB$, $$\label{eq:s2}
-\frac{s_2}{\kB}=2\pi\rho \int_0^\infty r^2 \{ g(r) \RM{ln}\, g(r) -[g(r)-1]\},$$ where $s_2$ is the translational pair-correlation contribution[@Nettleton1958Expression-in-T; @Baranyai1989Direct-entropy-] to the excess entropy and $\rho$ is the number density. We used the orientationally averaged oxygen-oxygen RDF in Eq. \[eq:s2\] for the analysis of SPC/E water. It has been shown that $-s_2/\kB$ not only quantifies the translational order exhibited by a fluid (the tendency of pairs of particles to adapt preferential separations),[@Truskett2000Towards-a-quant] but it also strongly correlates with the transport coefficients (see, e.g., Refs. ). Other translational order parameters have also been introduced to study the structure of molecular and colloidal fluids,[@Truskett2000Towards-a-quant; @Torquato2000Is-Random-Close; @Errington2001Relationship-be] but these measures are known to correlate strongly with $s_2$,[@Truskett2000Towards-a-quant; @Errington2001Relationship-be] and thus we exclusively use the latter in our analysis.
To understand how the various coordination shells of the RDF contribute to $-s_2/\kB$, we also investigated the cumulative order integral $\Is2$, defined as[@Krekelberg2007How-short-range] $$\label{eq:Is2}
I_{s_2}(r)=2\pi\rho\int_0^r r'^2\{g(r')\RM{ln}g(r')-[g(r')-1]\}dr'.$$ Note that $I_{s_2}(r)\rightarrow
-s_2/\kB$ as $r\rightarrow \infty$.
Finally, we adopted the following criteria to identify structurally anomalous behavior.
\[eq:anomalies\] $$\begin{aligned}
\left(\pd{[-s_2]}{\rho}\right)_T&<0,&\text{$\rho-$structural
anomalies} \label{eq:anomaliesDens} \\
\left(\pd{[-s_2]}{[\kB
T/\epsilon]}\right)_\rho&>0,&\text{$T-$structural
anomalies}\label{eq:anomaliesT}\end{aligned}$$
As indicated in the Introduction, waterlike fluids exhibit $\rho-$structural anomalies[@Mittal2006Quantitative-Li; @Errington2006Excess-entropy-] and SRA fluids display $T-$structural anomalies.[@Krekelberg2007How-short-range]
Structural anomalies
====================
Waterlike fluids {#sec:struct-anom-waterl}
----------------
### SPC/E water
![Structural data for the SPC/E water model obtained from molecular dynamics simulations. (a) Structural order parameter $-s_2/\kB$ as a function of density $\rho$ at $T=220\kv,240\kv,260\kv,280\kv$ and $300\kv$. Vertical dotted lines are at $\rho=0.9 \gcm$ and $\rho=1.15 \gcm$, the approximate boundaries for the region of anomalous structural behavior. (Lower panel) Orientationally averaged oxygen-oxygen radial distribution function $g(r)$ and cumulative order integral $I_{s_2}(r)$ along the $T=220\kv$ isotherm \[black circles, dashed curve in (a)\] for three different density regions: (b,c) $\rho
\leq 0.9\gcm$ \[up to maximum in $-s_2(\rho)/\kB$\], (d,e) $0.9\gcm
\leq \rho \leq 1.15\gcm$ \[between maximum and minimum in $-s_2(\rho)/\kB$\], (f,g) $\rho \geq 1.15\gcm$ \[beyond minimum in $-s_2(\rho)/\kB$\]. The regions are indicated by circled numbers along top of (a) and lower panel. In the lower panel, arrows indicate direction of increasing density; dashed vertical line is at $r=0.31\,\RM{nm}$ and dotted vertical line is at $r=0.57\,\RM{nm}$, the approximate locations of the first and second minima in $g(r)$, respectively.](spce_s2figNum)
. \[fig:spce\_s2fig\]
First, we discuss the simulation results for how $-s_2/\kB$ (i.e., translational order) of the SPC/E water model changes with density $\rho$. As can be seen in Figure \[fig:spce\_s2fig\](a), SPC/E water displays the $\rho-$structural anomalies of Eq. over the density range $0.9\gcm \leq \rho
\leq 1.15\gcm$ and $T<280\kv$. To gain insights into the origins of this behavior, we examine the orientationally averaged oxygen-oxygen RDF and $I_{s_2}$ as a function of $\rho$ along the $T=220\kv$ isotherm for three different density regions: (1) the initial increase of ${{-s_2({\rho})/\kB}}$ at low densities \[$\rho\leq 0.9 \gcm$, Figs. \[fig:spce\_s2fig\](b,c)\], (2) the anomalous decrease of ${{-s_2({\rho})/\kB}}$ at intermediate densities \[$0.9 \gcm \leq \rho \leq 1.15
\gcm$, Figs. \[fig:spce\_s2fig\](d,e)\], and (3) the increase of ${{-s_2({\rho})/\kB}}$ at high densities \[$\rho>1.15 \gcm$, Figs. \[fig:spce\_s2fig\](f,g)\].
Compressing the fluid in the lower-density region (1) $(\rho\leq
0.9\gcm)$ has relatively little effect on the RDF \[Fig. \[fig:spce\_s2fig\](b)\], but it does lead to a small net increase in translational order. As can be seen from the behavior of $\Is2$ in Fig. \[fig:spce\_s2fig\](c), the changes come primarily from the second shell. The reason is that the coordination number of water (approximately four, reflecting local tetrahedral hydrogen-bonding to nearest neighbors) is insensitive to changes in density over this range.[@Sciortino1991Effect-of-defec] As a result, the increase of $\rho$ is compensated by a slight decrease in the first peak of the RDF, and thus the first-shell contribution to the structural order remains largely unchanged. In the second shell, however, the change in density does not affect the RDF (i.e., the strength of the correlations with the central molecule). This means that compression induced hydrogen-bond bending has allowed more total water molecules into the second shell, which in turn leads to an overall increase in translational order.
On the other hand, further increases in density \[region (2), $0.9\gcm \leq \rho \leq
1.15\gcm]$ result in a pronounced decrease in $-s_2(\rho)/\kB$, i.e. the $\rho$-structural anomaly. As can be seen in Fig. \[fig:spce\_s2fig\](d), the main implications of compression for the interparticle correlations are a dramatic flattening of the second coordination shell and an associated shifting inward of these molecules into the interstitial space between the first and second shells. These structural changes are consistent with the earlier simulation observation[@Sciortino1991Effect-of-defec] that high local density can force a fifth molecule from the second shell into the periphery of the otherwise four-coordinated first shell. Inspection of $\Is2$ \[Fig. \[fig:spce\_s2fig\](e)\] confirms that the decrease in structural ordering is almost entirely due to reduced correlations between the central and second-shell molecules. In fact, Yan [*et al.*]{} convincingly demonstrated that a similar structural anomaly in TIP5P water can also be attributed to compression induced translational disordering of the second shell.[@Yan2007Structure-of-th]
Can these structural changes explain water’s self-diffusity anomaly? Sciortino [*et al.*]{}[@Sciortino1991Effect-of-defec] argued, based on molecular simulation results, that the presence of a fifth molecule in the first coordination shell significantly lowers the barriers for translational and rotational motions of the central water molecule. This suggests that second-shell waters play a central role in water’s increased mobility under compression. Interestingly, since the self-diffusivity of SPC/E water is strongly correlated to $s_2$ over these conditions,[@Mittal2006Quantitative-Li] one can independently draw the same conclusion from the data in Fig. \[fig:spce\_s2fig\].
Finally, we observe that, at sufficiently high densities \[region (3), $\rho>1.15\gcm$\], translational order again increases upon compression. This is “normal” behavior for dense liquids, and it simply reflects the fact that smaller volumes force particles to adopt locally ordered (i.e., efficient) packing structures.[@Truskett2000Towards-a-quant; @Errington2001Relationship-be]
### Core-softened model
![Structural data obtained from molecular dynamics simulations of the core-softened potential discussed in the text. (a) Structural order parameter $-s_2/\kB$ as a function of reduced density $\rho^*=\rho \sigma^3$ at $T^*=\kB
T/\epsilon=0.2,0.3,0.4,0.5$ and $0.6$, where $\sigma$ is the particle diameter, and $\epsilon$ is the energy scale of the potential (see Eq. ). Arrow indicates direction of increasing $T^*$, and vertical dotted lines are at $\rho^*=0.08$ and $\rho^*=0.175$, the approximate boundaries of the region of anomalous structural behavior. (Lower panel) Radial distribution function $g(r)$ and cumulative order integral $I_{s_2}(r)$ along the $T^*=0.3$ isotherm \[red squares, dashed curve in (a)\] for three density regions: (b,c) $\rho^* \leq
0.08$ \[up to $-s_2(\rho^*)/\kB$ maximum\], (d,e) $0.08 \leq \rho^*
\leq 0.175$ \[between maximum and minimum in $-s_2(\rho^*)/\kB$\], (f,g) $\rho^*\geq 0.175$ \[beyond minimum in $-s_2(\rho^*)/\kB$\]. The regions are indicated by circled numbers along top of (a) and lower panel. In lower panels, arrows indicate direction of increasing density; numbers in legends indicate values of $\rho^*$; vertical dashed line is at $r=1.5\sigma$ and vertical dotted line is at $r=3.5\sigma$, the approximate locations of the first and second minima in $g(r)$, respectively.[]{data-label="fig:ramp_sim_s2fig"}](ramp_sim_s2figNum)
In this section, we investigate how the translational order of the lower resolution, core-softened model of Eq. \[eq:core\_softened\_potential\] responds to changes in density. First, we consider the results from the molecular dynamics simulations. One striking feature of the data is that the behavior of $-s_2/\kB$ as a function of reduced density $\rho^*=\rho\sigma^3$, displayed in Fig. \[fig:ramp\_sim\_s2fig\](a), is qualitatively similar to that of SPC/E water \[see Fig. \[fig:spce\_s2fig\](a)\]. Specifically, the core-softened model also displays $\rho$-structural anomalies over the density range $0.08 \leq \rho^* \leq 0.175$ that become more pronounced at lower temperature.
Clearly, the core-softened model is very different from the SPC/E model in that the former does not provide a molecular description of water, and thus it does not favor the formation of tetrahedrally coordinated hydrogen-bond networks, etc. Nonetheless, as we explain below, the main “microscopic” origins of its density-dependent trends in structural order are basically the same as those for the SPC/E model.
In order to appreciate the similarity between these two models, it is helpful to first notice one difference. In the SPC/E model, the attractive “hydrogen-bond” interactions promote the formation of a first coordination shell, even at relatively low density. In contrast, since there are no attractions in the core-softened model, the Gaussian repulsion prevents the “first” coordination shell (near the hard-core diameter, $1.0\leq r/\sigma\leq 1.5$) from forming until sufficiently high density ($\rho^*\gtrsim 0.1$). On the other hand, the “second” coordination shell (against the Gaussian repulsion, $1.5\leq r/\sigma \leq 3.5$) is present even at low density.
From a qualitative perspective, one might consider each core-softened particle as effectively representing a cluster of water molecules (e.g., a central water molecule and its four nearest neighbors).[@Yan2007Correspondence-] In fact, Yan [*et al.*]{} have recently presented evidence that a mapping of this sort has quantitative merit when one compares, in appropriately reduced form, the behaviors of a core-softened ramp model to TIP5P water.[@Yan2007Correspondence-] When viewed from this perspective, the formation of the first shell in the core-softened model at high density qualitatively corresponds, in the molecular picture, to [ *additional*]{} water molecules (5, 6, etc.) penetrating the first shell of an otherwise four-coordinated central water molecule. Once this physical relationship between the two models is recognized, the similarities between their structural properties are easy to understand. To illustrate this, we carried out a structural analysis of the core-softened model identical to that presented above for the SPC/E model.
In particular, we examined the behavior of the RDF and $I_{s_2}$ \[see Fig. \[fig:ramp\_sim\_s2fig\](b-g)\] for the core-softened model as a function of density along the $T^*=\kB T/\epsilon=0.3$ isotherm for three different density regions: (1) the initial increase of ${{-s_2({\rho^*})/\kB}}$ at low densities \[$\rho^*\leq 0.08$, Figs. \[fig:ramp\_sim\_s2fig\](b,c)\], (2) the anomalous decrease of ${{-s_2({\rho^*})/\kB}}$ at intermediate densities \[$0.08 \leq \rho^* \leq
0.175$, Figs. \[fig:ramp\_sim\_s2fig\](d,e)\], and (3) the increase of ${{-s_2({\rho^*})/\kB}}$ at high densities \[$\rho^*>0.175$, Figs. \[fig:ramp\_sim\_s2fig\](f,g)\].
As discussed above, the “first” shell of the RDF is not populated in this model at low density because the core-softened particles themselves loosely represent a central water and its four nearest neighbors. In this view, the initial compression of the core-softened fluid \[region (1), $\rho^*\leq 0.08$\] has an effect that is similar to that seen for SPC/E water. The modest increase in $-s_2/\kB$ that is observed is due to the increase in density and a minor enhancement of structuring in the second shell ($1.5\leq r/\sigma \leq 3.5$).
Further compression of the core-softened model \[region (2), $0.08\leq\rho^*\leq 0.175$\] leads to an anomalous decrease in structural order \[Fig. \[fig:ramp\_sim\_s2fig\](a)\]. Figs. \[fig:ramp\_sim\_s2fig\](d) and (e) indicate that the disordering is again due to a flattening and shifting inward of the second shell. Moreover, the “first” shell of the core-softened particles begins to emerge, which schematically represents, in the approximate molecular view discussed above, that additional water molecules are effectively penetrating into the four-coordinated first shell.
Similar to SPC/E water, it is known that there is a strong correlation between excess entropy and self-diffusivity for the core-softened model.[@Mittal2006Relationship-be] This information, together with the results shown here, support the view that the self-diffusivity anomaly of the core-softened model is also linked to its density-dependent second-shell structure.
As expected, at higher density \[region (3), $\rho^*\geq 0.175$\], compression leads to an increase in structural order due to simple-liquid-like structuring of particles in the first coordination shell \[Figs. \[fig:ramp\_sim\_s2fig\](f,g)\]. In short, the qualitative response to changes in density of the structural order and its coordination-shell contributions for the core-softened model are very similar to those of the more detailed SPC/E water model. This finding is consistent with the recent demonstration that one can approximately map the anomalies of TIP5P water onto those of a similar two-scale ramp potential.[@Yan2007Correspondence-]
As a final point, we show in Fig. \[fig:ramp\_HNC\_s2fig\] that the integral equation theory of the core-softened model can qualitatively predict all of the trends shown in Fig. \[fig:ramp\_sim\_s2fig\]. The ability of this approach to reproduce the structural features seen in simulations, together with the quasi-universal connection between structure and transport coefficients of liquids,[@Rosenfeld1999A-quasi-univers; @Rosenfeld1977Relation-betwee; @Dzugutov1996A-univeral-scal] suggests that integral equation theory might serve as a valuable first-pass tool in assessing whether other model systems represent good candidates for exhibiting static and dynamic anomalies. However, if the intention is to ultimately use it as a quantitatively accurate predictive tool, then more comprehensive investigations of alternative closure relationships, in the spirit of Ref. , will be necessary.
, where $\sigma$ is the particle diameter, and $\epsilon$ is the energy scale of the potential (see Eq. ). Arrow indicates direction of increasing $T^*$, and vertical dotted lines are at $\rho^*=0.075$ and $\rho^*=0.165$, the approximate boundaries of the region of anomalous structural behavior. (Lower panel) Radial distribution function $g(r)$ and cumulative order integral $I_{s_2}(r)$ along the $T^*=0.3$ isotherm \[red dashed curve in (a)\] for three different density regions: (b,c) $\rho^* \leq 0.075$ \[up to $-s_2(\rho^*)/\kB$ maximum\], (d,e) $0.075 \leq \rho^* \leq 0.165$ \[between maximum and minimum in $-s_2(\rho^*)/\kB$\], (f,g) $\rho^*\geq 0.165$ \[beyond minimum in $-s_2(\rho^*)/\kB$\]. The regions are indicated by circled numbers along top of (a) and lower panel. In lower panels, arrows indicate direction of increasing density; numbers in legends indicate values of $\rho^*$; vertical dashed line is at $r=1.5\sigma$ and vertical dotted line is at $r=3.5\sigma$, the approximate locations of the first and second minima in $g(r)$, respectively.[]{data-label="fig:ramp_HNC_s2fig"}](ramp_HNC_s2figNum)
SRA fluids {#sec:struct-anom-short}
----------
One of the key aspects of short-range attractive (SRA) fluids is that their structurally anomalous behavior occurs as a function of the reduced interparticle attractive strength $\epsilon/\kB T$ at constant particle packing fraction $\Pc$, where $-\epsilon$ represents the well depth of the interparticle attraction. In most typical atomic or molecular fluids, one finds that structural order ($-s_2/\kB$) increases with $\epsilon/\kB T$. SRA fluids are anomalous in that, at sufficiently high values of $\Pc$, the opposite trend can be observed;[@Krekelberg2007How-short-range; @Mittal2006Quantitative-Li] i.e., attractions counterintuitively decrease the amount of structural order.
In this section, we briefly discuss how we used molecular simulation and integral equation theory to gain insights into this trend. We accomplished this by exploring the various coordination-shell contributions to $-s_2/\kB$ for the two model SRA fluids discussed in Section \[sra\_models\].
### Colloid-polymer mixture
![Structural data obtained from molecular dynamics simulations of the model colloid-polymer SRA fluid discussed in the text. (a) Structural order parameter $-s_2/\kB$ as a function of polymer volume fraction $\Pp$ (i.e., strength of colloid attractions) at colloid packing fraction $\Pc=0.4$. Vertical dotted line at $\Pp=0.1$, the location of the minimum in ${{-s_2({\Pp})/\kB}}$. (Lower panel) Radial distribution function $g(r)$ and cumulative order integral $I_{s_2}(r)$ along the isochore $\Pc=0.4$ \[black circles in (a)\] for two polymer concentration ranges: (b,c) $\Pp\leq 0.1$ \[below minimum in $-s_2(\Pp)/\kB$\], (d,e) $\Pp\geq 0.1$ \[above $-s_2(\Pp)/\kB$ minimum\]. The regions are indicated by circled numbers along top of (a) and lower panel. In lower panels, arrows indicate direction of increasing $\Pp$; the parameter $a$ indicates colloidal particle radius; vertical dashed line is at $r=3a$ and vertical dotted line is at $r=5a$, the approximate locations of the first and second minima in $g(r)$, respectively.[]{data-label="fig:cfp_sim_s2fig"}](CFP_sim_s2figNum)
We begin by investigating the behavior of the model colloid-polymer system[@Puertas2002Comparative-Sim; @Puertas2003Simulation-stud] by molecular simulation. The effective colloid-colloid pair potential for this model was presented earlier in Fig. \[fig:CS\_potential\](b) for several values of polymer packing fraction $\Pp$. Since the reduced well-depth of this potential, $\epsilon/\kB T$, scales as $\Pp$, we analyze structural order below as a function of the latter.
In particular, Fig. \[fig:cfp\_sim\_s2fig\](a) illustrates how $-s_2/\kB$ varies as a function of $\Pp$ at a particle packing fraction of $\Pc=0.4$. As expected for SRA fluids, ${{-s_2({\Pp})/\kB}}$ exhibits a minimum at $\Pp\approx 0.1$. In other words, this fluid displays the structural anomaly of Eq. for $\Pp\leq 0.1$. To understand the origins of this trend, we study the RDF and the cumulative order integral $\Is2$ as a function of $\Pp$ in two qualitatively different regions: (1) the anomalous decrease in ${{-s_2({\Pp})/\kB}}$ at low polymer concentrations (low interparticle attractions), and (2) the “normal” increase of ${{-s_2({\Pp})/\kB}}$ at higher polymer concentrations (higher interparticle attractions).
What specific changes to coordination shell structure explain the attraction induced disordering that occurs at small $\Pp$ \[region (1), $\Pp \leq 0.1$\]? First, note that strengthening interparticle attractions considerably increases but narrows the first peak of the RDF \[Fig. \[fig:cfp\_sim\_s2fig\](b)\]. These two effects essentially cancel so that the first-shell contributions to $\Is2$ are insensitive to $\Pp$ over this range \[see Fig. \[fig:cfp\_sim\_s2fig\](c)\]. However, attractions also slightly shift the higher coordination shells of the RDF inward and diminish their overall correlation with the central particle. These latter modifications to the structure of the second and higher coordination shells give rise to the anomalous decrease in the structural order of this system. They are also consistent with behavior observed in the recent Krekelberg [*et al.*]{}[@Krekelberg2007How-short-range] simulations of the square-well SRA fluid discussed in the Introduction. A microscopic interpretation of this trend is that SRA interactions drive weak particle clustering at low $\Pp$ (explaining the sharpening and narrowing of the first peak). This clustering, in turn, opens up channels of free volume in the fluid and disrupts the uniform hard-sphere-like packing order in the second and higher coordination shells.[@Krekelberg2006Model-for-the-f; @Krekelberg2006Free-Volumes-an; @Puertas2004Dynamical-heter; @Sciortino2002One-liquid-two-]
Under conditions where the aforementioned structural anomaly occurs, increases in $\Pp$ also increase the mobility of the fluid.[@Krekelberg2007How-short-range] Very similar to the waterlike fluids discussed above, it is known that $s_2$ and self-diffusivity are strongly correlated for the model colloid-particle mixture.[@Mittal2006Quantitative-Li] As a result, the self-diffusvity anomaly appears to also derive from subtle structuring effects in the second and higher coordination shells.
As one would expect, however, increasing $\Pp$ ultimately increases structural order, if the interactions are sufficiently attractive \[region (2), $\Pp \geq 0.1$\]. The attractions lead to the formation of strongly bonded particle clusters,[@Krekelberg2006Model-for-the-f; @Krekelberg2006Free-Volumes-an; @Puertas2004Dynamical-heter; @Sciortino2002One-liquid-two-] which is reflected by the increased height of the first peak of the RDF \[Fig. \[fig:cfp\_sim\_s2fig\](d)\] and the associated rise in the first-shell contribution to $\Is2$ \[Fig. \[fig:cfp\_sim\_s2fig\](e)\].
![Structural data for the model colloid-polymer SRA fluid discussed in the text from integral equation theory. (a) Structural order parameter $-s_2/\kB$ as a function of polymer volume fraction $\Pp$ at colloid packing fractions $\Pc=0.3,0.325,0.35,0.375,0.4,0.425,0.45,0.475$ and $0.5$. Arrow indicates direction of increasing $\Pp$, and vertical dotted line is at $\Pp=0.1$, the approximate boundary of the region of anomalous structural behavior. (Lower panel) Radial distribution function $g(r)$ and cumulative order integral $I_{s_2}(r)$ along the isochore $\Pc=0.475$ \[dashed violet curve in (a)\] for two polymer concentration ranges: (b,c) $\Pp\leq 0.1$ \[below minimum in $-s_2(\Pp)/\kB$\], (d,e) $\Pp\geq 0.1$ \[above $-s_2(\Pp)/\kB$ minimum\]. The regions are indicated by circled numbers along top of (a) and lower panel. The parameter $a$ indicates colloid radius. In lower panels, arrows indicate direction of increasing $\Pp$, vertical dashed line is at $r=3a$, and vertical dotted line is at $r=5a$, the approximate locations of the first and second minima in $g(r)$, respectively.[]{data-label="fig:cfp_py_s2fig"}](CFP_PY_phic0475_s2figNum)
In closing, we test in Figs. \[fig:cfp\_py\_s2fig\] and \[fig:sw\_py\_s2fig\] whether integral equation theory is able to qualitatively capture these attraction induced structural changes for both model systems introduced in Section \[sra\_models\]: the colloid-polymer fluid and the square-well fluid, respectively. Comparison of Fig. \[fig:cfp\_py\_s2fig\] with Fig. \[fig:cfp\_sim\_s2fig\] and Fig. \[fig:sw\_py\_s2fig\] with Figure 4 of Ref. demonstrate that this is indeed the case. The success of these predictions strengthens the case that integral equation theory will be a useful tool in assessing whether future model systems of interest might display structural anomalies.
![Structural data for the square-well fluid discussed in the text obtained from integral equation theory. (a) Structural order parameter $-s_2/\kB$ as a function of reduced attractive strength $\epsilon/\kB T$ at particle packing fractions $\Pc=0.4,0.45,0.5,0.525,0.55,0.56$ and $0.57$. Arrow indicates direction of increasing $\Pc$, and vertical dotted line is at $\ekt=0.9$, the approximate boundary of the region of anomalous structural behavior. (Lower panel) Radial distribution function $g(r)$ and cumulative order integral $I_{s_2}(r)$ along the isochore $\Pc=0.55$ \[dashed blue curve in (a)\] for two attractive strength ranges: (b,c) $\ekt \leq 0.9$ \[below minimum in $-s_2(\epsilon/\kB T)/\kB$\], (d,e) $\ekt \geq 0.9$ \[above $-s_2(\epsilon/\kB T)/\kB$ minimum\]. The regions are indicated by circled numbers along top of (a) and lower panel. The parameter $\sigma$ indicates colloid diameter. In lower panels, arrows indicate direction of increasing $\ekt$; vertical dashed line is at $r=1.4\sigma$ and vertical dotted line is at $r=2.3\sigma$, the approximate locations of the first and second minima in $g(r)$, respectively.[]{data-label="fig:sw_py_s2fig"}](SW_pyinteq_phic055_s2figNum)
Conclusions
-----------
Although the structural order of a fluid is usually enhanced by isothermal compression or isochoric cooling, a few notable systems show the opposite behaviors. Specifically, increasing density can disrupt the structure of waterlike fluids, while lowering temperature (or strengthening of attractive interactions) can weaken the correlations of fluids with short-range attractions. The two-body translational contribution to the excess entropy provides a quantitative measure of these changes in structural order. It is a particularly insightful quantity to study because (i) its contributions from the various coordination shells of the radial distribution function can be readily determined, and (ii) it correlates strongly with self-diffusivity, which allows it to provide insights into the dynamic anomalies of these fluids.
Here, we have presented a comprehensive study, by both molecular simulation and integral equation theory, of the coordination shell contributions to the two-body excess entropy for several model systems. These models incorporate different levels of molecular resolution, but all exhibit the aforementioned structural anomalies. The results of this study support the emerging view that the structural anomalies of these fluids can generally be attributed to quantifiable changes in the second and higher coordination shells of the radial distribution function. They also demonstrate that integral equation theory can serve as an effective first-pass tool for assessing, based on the pair potential alone, whether new model systems are good candidates for exhibiting static and dynamic anomalies.
We thank Anatol Malijevsky for providing an efficient code for numerical solution of the integral equation theories analyzed in this work. W.P.K. acknowledges financial support of the National Science Foundation for a Graduate Research Fellowship. T.M.T. acknowledges financial support of the National Science Foundation (CTS 0448721), the David and Lucile Packard Foundation, and the Alfred P. Sloan Foundation. V.G. acknowledges financial support of the National Science Foundation (CTS-0347381) and the Robert A. Welch Foundation. Computer simulations for this study were performed at the Texas Advanced Computing Center (TACC).
[81]{} natexlab\#1[\#1]{}bibnamefont \#1[\#1]{}bibfnamefont \#1[\#1]{}citenamefont \#1[\#1]{}url \#1[`#1`]{}urlprefix\[2\][\#2]{} \[2\]\[\][[\#2](#2)]{}
, ** (, , ), ed.
, in **, edited by (, , ), vol. of **, p. .
, ** (, ), ed.
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, , , , ****, ().
, , , ****, ().
, , , , ****, ().
, , , , ****, ().
, , , , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , , ****, ().
, , , , , ****, ().
, , , , ****, ().
, , , ****, ().
, , , ****, ().
, , , ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , , , , ****, ().
, , , , (), .
, , , , ****, ().
, ****, ().
, , , ****, ().
, , , , , ****, ().
, , , , ****, ().
, , , , ****, ().
, , , , ****, ().
, ****, ().
, ****, ().
, , , , , , , , , , ****, ().
, ****, ().
, , , , , ****, ().
, , , , , , , , , ****, ().
, , , , , , ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, , , , ****, ().
, , , ****, ().
, , , ****, ().
, , , ****, ().
, ****, ().
, , , ****, ().
, ****, ().
, ****, ().
, , , ****, ().
, , , ****, ().
, , , , , .
, , , ****, ().
, , , ****, ().
[^1]: Corresponding Author
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Pfaffians of matrices with entries $z[i,j]/(x_i+x_j)$, or determinants of matrices with entries $z[i,j]/(x_i-x_j)$, where the antisymmetrical indeterminates $z[i,j]$ satisfy the Plücker relations, can be identified with a trace in an irreducible representation of a product of two symmetric groups. Using Young’s orthogonal bases, one can write explicit expressions of such Pfaffians and determinants, and recover in particular the evaluation of Pfaffians which appeared in the recent literature.'
---
[**Pfaffians and Representations of the Symmetric Group**]{}
[Alain Lascoux]{}\
CNRS, IGM Université de Marne-la-Vallée\
77454 Marne-la-Vallée Cedex, France\
Email: [email protected]\
[*Key words:*]{} Pfaffians; Symmetric Group; Representations
[*AMS classifications:*]{} 05E05; 15A15
Introduction
============
Determinants or Pfaffians of order $n$ can be written in terms of the symmetric group $\mfS_n$. Determinants can be considered as generators of a 1-dimensional alternating representations. But in the case of a determinant or a Pfaffian $$\left| \frac{a_i-a_{j+n}}{x_i-x_{j+n}} \right|_{1\leq i,j\leq n}
\qquad ,\qquad
\Pfaff\left(\, \frac{a_i-a_j}{x_i+x_j}\, \right)_{1\leq i<j\leq n} \, ,$$ three symmetric groups occur : $\mfS_n^a$ acts on the indeterminates $a_i$, $\mfS_n^x$ acts on the $x_j$, and $\mfS_n^{ax}$ acts on the indices of all indeterminates simultaneoulsy. This “diagonal action” satisfy a Cauchy-type property, each irreducible representation of $\mfS_n^a$ occuring in the expansion of the determinant, or of the Pfaffian, being tensored with a representation of $\mfS_n^x$ of conjugate type.
When $n=2m$ is even, since the space generated by the orbit of the polynomial $(a_1 \moins a_2)(a_3\moins a_4)\cdots (a_{n-1}\moins a_n)$ under $\mfS_n^a$ is a copy of the irreducible representation $V_{[m,m]}^a$ of type $[m,m]$, this forces the Pfaffian to lie in the space $V_{[m,m]}^a \otimes V_{[2,\ldots,2]}^x$.
An easy analysis shows that moreover the Pfaffian $\Pfaff\left(\frac{a_i-a_j}{x_i+x_j} \right)$ is diagonal in Young’s orthogonal basis (and thus, can be considered as a trace). In fact, the same analysis remains valid (this is our main theorem, Th.\[th:Pfaffzg\]) in the more general case $$\Pfaff\left( z[i,j]\, g[i,j] \right)_{1\leq i<j\leq n} \, ,$$ when taking antisymmetric indeterminates $z[i,j]=\moins z[j,i]$ satisfying the Plücker relations (we say *Plücker indeterminates*), instead of $(a_i-a_j)$, and symmetric indeterminates $g[i,j]=g[j,i]$ instead of $(x_i+x_j)^{-1}$. For specific $z[i,j]$ and $g[i,j]$, one may be able to write another element belonging to the same representation. Checking that two elements in the same irreducible representation coincide is very easy, and reduces to compute some specializations.
The most general case that we consider is $\Pfaff\left( a[i,j]b[i,j] z[i,j]^{-1} \right)$, with three families of Plücker indeterminates. In that case the Pfaffian factorizes in two factors separating the $a[i,j]$’s and $b[i,j]$’s (Th.\[th:Pfaffabz\]).
A connection with the theory of symmetric functions is provided by specializing the Plücker indeterminates into $S_\l(A+x_i +x_j)(x_i-x_j)$, $S_\l(A)$ being a fixed Schur function, to the alphabet of which one adds the letters $x_i,x_j$ ([@Cbms]). Thus Th.\[th:Pfaffabz\] gives the factorization of $$\Pfaff\left( \frac{ S_\l(A+a_i+a_j)
S_\mu(B+b_i+b_j)}{S_\nu(Z+z_i+z_j)}
\frac{(a_i\moins a_j)(b_i\moins b_j)}{(z_i-z_j)} \right) \, ,$$ for three Schur functions, and three families of indeterminates.
In the case of $\Pfaff\left(\frac{a_i-a_j}{x_i+x_j} \right)$ first considered by Sundquist [@Sundquist], and that we have taken as our generic case, it is easy to write a determinant which also lies in the space $V_{[m,m]}^a \otimes V_{[2,\ldots,2]}^x$. Specializing half of the $a_i$’s to $1$, the others to $0$, one recovers the determinantal expression of Sundquist for this Pfaffian (Th.\[th:Sundquist\]).
Ishikawa [@Ishikawa], Okada [@Okada], M. Ishikawa, S. Okada, H. Tagawa and J. Zeng [@IshiOkada] have given different generalizations of Sundquist’s Pfaffian. We show how to connect their results to Th.\[th:Pfaffzg\] and Th.\[th:Pfaffabz\].
In section 8, we go back to determinants, and show how to relate $$\left| \frac{z[i,j]}{x_i^2-x_j^2} \right|_{1\leq i\leq m<j\leq n}
\quad \text{and}\quad \Pfaff\left(\frac{z[i,j]}{x_i+x_j} \right) \, ,$$ the indeterminates $z[i,j]$ still satisfying the Plücker relations. A corollary of this analysis is that the above determinant is, up to straightforward factor, symmetrical in $x_1,\ldots, x_n$, and not only symmetrical in $x_1,\ldots, x_m$ and $x_{m+1},\ldots, x_{2m}$ separately. Some determinants $\det\Bigl( S_\l(A+x_i+x_j)\\
S_\mu(B+x_i+x_j)^{-1} \Bigr)$ present a special interest in the theory of orthogonal polynomials, or of the six-vertex model.
To be self-contained, and for lack of a reference appropriate to our needs, we first recall some properties of representations. In the last section, we give more details about the polynomial bases that one deduces from Young’s orthogonal idempotents.
Representations of the symmetric group
======================================
Young’s idempotents
-------------------
The group algebra $\cH$ of the symmetric group $\mfS_n$ has by definition a linear basis consisting of all the permutations of $1,2,\ldots,n$.
Young described another basis $e_{tu}$, indexed by pairs of standard tableaux of the same shape with $n$ boxes. These elements are *matrix units*, in the sense that they satisfy the relations $$\begin{aligned}
e_{t,u} e_{u,v} &=& e_{t,v} \, , \\
e_{t,u} e_{w,v} &=& 0 \quad \text{if}\ w\neq u\, . \end{aligned}$$
In particular, the $e_{t,t}$ are idempotents: $e_{t,t} e_{t,t}= e_{t,t}$, and the identity decomposes as $$1 = \sum_t e_{t,t} \, ,$$ where the sum is over all standard tableaux of $n$ boxes. The subsum $$\label{IdempCentral}
e_\l=\sum_{t\in Tab(\l)} e_{t,t}$$ over standard tableaux of a given shape $\l $ is the *central idempotent* of index $\lambda$.
Specht representations
----------------------
Given any $t$, the right module $e_{t,t}\, \cH$ is an irreducible representation of the symmetric group, with basis $\{ e_{t,u}:\, u $ has the same shape as $t\}$.
There are simpler models of irreducible representations, in particular spaces of polynomials which are called *Specht representations*, though they have been defined by Young[^1].
Bases are still indexed by standard tableaux of a given shape, but now tableaux are interpreted as polynomials as follows.
A column tableau $\smallyoung{\s k\cr\s j\cr\s i\cr }$ is interpreted as the Vandermonde determinant in the variables $x_i,x_j,\ldots, x_k$ : $$\young{k\cr j\cr i\cr} = (x_i-x_j)\, (x_i-x_k)\, (x_j-x_k)
:= \Delta^x(i,j,k) \, ,$$ and a tableau stands for the product of its columns : $$\young{ 5\cr 3 &6\cr 1&2&4\cr} =
\Delta^x(1,3,5)\, \Delta^x(2,6) \, \Delta^x(4)\, .$$
We shall denote this polynomial $\Delta_t^x$ and call it *Specht polynomial*. The orbit of any $\Delta_t^x$ under the symmetric group (permuting the variables $x_i$) has $n!$ elements, whose linear span is of dimension the number of standard tableaux of the same shape as $t$.
More precisely, Young obtained, in the case of zero characteristic :
\[th:Specht1\] Given a partition $\l$, the linear span of the polynomials $\Delta_t^x$, $t$ varying over the set $\Tab(\lambda)$ of standard tableaux of shape $\l$, is an irreducible representation of the symmetric group.
Fixing a shape $\lambda$, there are two “extreme tableaux”: the one such that its columns are filled with consecutive letters, and that we shall call *top tableau* and denote $\zeta$. For shape $[2,3,4]$, the top tableau is $$\zeta = \young{3 &6\cr 2 & 5 & 8\cr 1&4 &7 &9\cr}$$ and gives the Specht polynomial $$\Delta_\zeta^x =
\Delta^x(1,2,3)\, \Delta^x(4,5,6)\, \Delta^x(7,8)\, \Delta^x(9)\, .$$
Similarly, the *bottom tableau* has its rows filled with consecutive letters. We denote it by $\aleph$: $$\aleph = \young{8 &9\cr 5&6&7\cr 1&2&3&4 \cr} \,$$ with Specht polynomial $$\Delta_\aleph^x =
\Delta^x(1,5,8)\, \Delta^x(2,6,9)\, \Delta^x(3,7)\, \Delta^x(4)\, .$$
The standard tableaux of a given shape may be generated by using simple transpositions, starting with $\zeta$. By the notation $\Tab(\l)$ we mean this ranked poset, with top element $\zeta$ and bottom one, $\aleph$. The *distance* $\ell(t,u)$ of two tableaux is the distance in $\Tab(\l)$.
The decomposition of any element of the Specht representation in the basis $\Delta_t^x$ is given by a so-called *straightening algorithm* (cf. [@DKR; @Desarmenien; @Turbo]).
We shall need only one coefficient in such an expansion, the coefficient of $\Delta_\aleph^x$. Given a tableau $t$ with $n$ boxes, and a function $f(x_1,\ldots, x_n)$, let us write $f(t)$ for the specialization where each $x_i$ is specialized to $r$ if $i$ lies on row $r$ (rows are numbered from the bottom, starting with $0$).
\[th:Specht2\] Given a partition $\lambda$, and a linear combination $f(x_1,\ldots, x_n)= \sum_t c_t\, \Delta_t^x $, with coefficients $c_t$ independent of $x_1,\ldots, x_n$, then the coefficient $c_\aleph$ is equal to $$f(\aleph)/ \Delta_\aleph^x(\aleph) \ .$$
All other tableaux than $\aleph$ have in some column two entries which lie in the same row of $\aleph$.
The Specht representation can occur in many disguises. Let us call *Plücker indeterminates* anti-symmetric indeterminates $z[i,j]=-z[j,i]$, satisfying Plücker relations[@BL] for all quadruples of different integers : $$z[i,j] z[k,l] -z[i,k]z[j,l] + z[j,k]z[i,l]= 0 \ .$$
A typical example is obtained by taking a $2\times \infty$ generic matrix $M$, and defining $z[i,j]$ to be the minor on columns $i,j$ of $M$. More generally, one takes an $N\times\infty$ generic matrix $M$, one chooses $N-2$ columns of index $\a,\b,\ldots$ and define $z[i,j]$ to be the minor of maximal order of $M$ on columns $i,j,\a,\b,\ldots$, with $i,j\neq \a,\b,\ldots$.
The following proposition gives another description of Specht representations for shape $[m,m]$.
\[th:Specht3\] Given an even positive number $n=2m$, let $z[i,j]$, $1\leq i, j \leq n$ be Plücker indeterminates. Given any numbering $u$ of the boxes of the diagram $[m,m]$, let $z[u]$ be the product of all $z[i,j]$, where $[i,j]$ is a column of $u$.Let the symmetric group $\mfS_n$ act on the variables $z[i,j]$ by permutation of $1,2,\ldots,n$.
Then the correspondence $z[u]=\prod z[i,j] \to \prod (x_i-x_j) $ induces an isomorphism of representations of $\mfS_n$. In particular, $\{ z[t]\}$, where $t$ runs over all standard tableaux of shape $[m,m]$, is a linear basis of the span of all $z[u]$.
In short, when one has $$z[1,2]z[3,4] -z[1,3]z[2,4] +z[1,4]z[2,3] =0 \,$$ one can as well read $$(a_1-a_2)(a_3-a_4) - (a_1-a_3)(a_2-a_4) + (a_1-a_4)(a_2-a_3) =0$$ or $$\young{ 2 &4\cr 1&3\cr} - \young{ 3 &4\cr 1&2\cr}
+ \young{ 4 &3\cr 1&2\cr} = 0$$ without loss of generality.
We shall use, for a rectangular shape with two columns of length $m=n/2$, three models of representations. The first one is the usual Specht representation, generated by the action of $\mfS_n$ on $$\Delta^x(1,2,\ldots,m)\, \Delta^x(m\plus1,\ldots,n)\, .$$
The second model is the image of the first one under the correspondence $(x_i-x_j) \to z[i,j]$. The Specht polynomial corresponding to the top tableau of shape $2^m$ will now be $$\Y^{z[\,]}(\zeta) :=
\Delta^{z[\,]}(1,2,\ldots,m)\, \Delta^{z[\,]}(m\plus1,\ldots,n)
= \prod_{1\leq i<j\leq m} z[i,j]\, \prod_{m+1\leq i<j\leq n} z[i,j]
\, .$$
For the third one, one starts with a symmetric matrix $G$, with entries $g[i,j]=g[j,i]$. Let us denote the minor consisting of rows $i_1,\ldots, i_m$ and columns $i_{m+1},\ldots, i_n$ by $$g[ i_1,\ldots, i_m\, |\, i_{m+1},\ldots, i_n] \, .$$ The symmetric group $\mfS_n$ acts formally by permuting the indices of such minors. Now, Kronecker [@Kronecker; @Muir] has shown that such minors satisfy the Plücker relations[^2] $$\sum_{i=0}^m (-)^i\,
g[1,\ldots,m\moins1, m\plus i\, |\, m\plus1,\ldots,
\widehat{m\plus i},\ldots, n] =0 \, .$$
This implies the following proposition :
Let $G$ be a symmetric matrix of order $n=2m$. Then the linear span of the minors in the orbit of $$g[1,\ldots,m\, |\, m\plus1,\ldots,n]$$ under permutation of indices, is an irreducible representation of $\mfS_n$ of shape $2^m$.
Using the correspondence $$g[ i_1,\ldots, i_m\, |\, i_{m+1},\ldots, i_n]
\leftrightarrow \Delta^x(i_1,\ldots, i_m)\, \Delta^x(i_{m+1},\ldots, i_n)$$ one can still speak of a *Specht basis* for these minors of a symmetric matrix.
For example, for $m=3$, the space has basis $$g[123\, |\, 456],\, g[124\, |\, 356],\,
g[125\, |\, 346],\, g[134\, |\, 256],\,
g[135\, |\, 246],\, ,$$ and one can directly check the relation $$g[123\, |\, 456] -g[124\, |\, 356] +g[125\, |\, 346]
- g[126\, |\, 345] =0 \, .$$
In detail, for $m=2$, the sum $ g[12|34]-g[13|24]+g[14|23] $ expands into $$g[1, 3] g[2, 4] - g[1, 4] g[2, 3] - g[1, 2] g[3, 4] + g[1, 4] g[3, 2]
+ g[1, 2] g[4, 3] - g[1, 3] g[4, 2]$$ which is indeed zero, because $g[i,j]=g[j,i]$.
Plücker relations, hence Specht representations, also occur in the theory of symmetric functions.
Indeed, given two alphabets[@Cbms] $A=\{a\}$, $B=\{b\}$, the *complete functions* $S_k(A-B)$ are defined by the generating function $$\sum_k z^k S_k(A-B) = \prod_{b\in B} (1-zb)
\prod_{a\in A} (1-za)^{-1} \, ,$$ putting $S_{k}=0$ for $k<0$. The *Schur function* $S_v(A-B)$, $v\in \Z^r$, has the determinantal expression $\det\left( S_{v_j+j-i}(A-B) \right)$. When $B$ is the two-letters alphabet $B=\{x,y\}$, then $(x-y)S_v(A-B)$, denoted $(x\moins y) S_v(A -x-y)$, is equal to the maximal minor on columns $v_1,v_2\plus 1, \ldots, v_r \plus r\moins 1, x,y$ of $$\begin{array}{r} \s columns \\ \phantom{S_0(A)}\\
\phantom{S_0(A)}\\ \phantom{\vdots}\\ \phantom{S_0(A)} \end{array}
\begin{bmatrix}
\s 0 & \s 1 &\s 2 & \s 3 & \cdots & & \s x & \s y \\
\noalign{\kern 3pt \hrule \kern 3pt}
S_0(A) & S_1(A) & S_2(A) &S_3(A) &\cdots & & x^{r+1} & y^{r+1} \\
S_{-1}(A) & S_0(A) & S_1(A) & S_2(A) &\cdots & & x^r & y^r \\
\vdots & \vdots &\vdots &\vdots & & &\vdots &\vdots \\
S_{-r-1}(A) & S_{-r}(A) & S_{-r+1}(A) & S_{-r+2}(A) &\cdots & & 1 & 1
\end{bmatrix} \, .$$
Therefore, for a given $v\in\Z^r$, and a given alphabet $A$, the $z[i,j] = (x\moins y) S_v(A -x-y)$ (resp. $z[i,j] = (x\moins y) S_v(A +x+y)$) satisfy the Plücker relations.
Orthogonal representations
--------------------------
Given a shape $\lambda$, then $ \Delta_\zeta^x\, e_{\zeta,\zeta} = \Delta_\zeta^x$, and the polynomials $\Delta_\zeta^x\, e_{\zeta,t}$, for $t\in \Tab(\l)$, constitute another basis of the Specht representation. Taking a linear order compatible with the poset structure, then the matrix of change of basis is lower triangular. A more precise information is given by using the Yang-Baxter relations (see the last section).
Let us call *Young’s basis* the basis proportional to $\Delta_\zeta^x\, e_{\zeta,t}$, such that the leading term (with respect to the poset $\Tab(\l)$) of each $ \Y(t)$ be $\Delta_t^x$, and denote it $\{ \Y(t) :\, t\in \Tab(\l)\}$.
For example, for shape $[3,3]$, the poset of tableaux is $$\begin{array}{rcl}
& \young{2&4&6\cr 1&3&5\cr} \\[4pt]
\raise 3pt \hbox{$\s s_2$} \diagup &
& \hskip -6pt\diagdown\hskip -6pt\diagdown
\hskip -6pt\diagdown \raise 3pt \hbox{$\s s_4$} \\{}
\young{3&4&6\cr 1&2&5\cr} & & \young{2&5&6\cr 1&3&4\cr}\\[10pt]
{\s s_4} \diagdown\hskip -6pt\diagdown\hskip -6pt\diagdown &&
\diagup {\s s_2} \\
& \young{3&5&6\cr 1&2&4\cr} & \\[13pt]
& \phantom{\s s_3} \big\| {\s s_3} &\\[2pt]
& \young{4&5&6\cr 1&2&3\cr}
\end{array}$$ and the matrix expressing Young’s basis in terms of the Specht basis (reading successive rows) is $$\left [\begin {array}{ccccc} 1&0&0&0&0\\\noalign{\medskip}-1/2&1&0&0
&0\\\noalign{\medskip}-1/2&0&1&0&0\\\noalign{\medskip}1/4&-1/2&-1/2&
1&0\\\noalign{\medskip}2/3&-1/3&-1/3&-1/3&1\end {array}\right ]\,
=\,
\left [\begin {array}{ccccc} 1&0&0&0&0\\\noalign{\medskip}1/2&1&0&0&0
\\\noalign{\medskip}1/2&0&1&0&0\\\noalign{\medskip}1/4&1/2&1/2&1&0
\\\noalign{\medskip}-1/4&1/2&1/2&1/3&1\end {array}\right ]^{-1}
\, .$$
In that special case, the two allowable linear orders on the graph give the same matrices, we did not need numbering the tableaux.
To handle other models of irreducible representations, we first need to characterize the elements corresponding to $\Delta_\zeta^x$ in these models. In fact, $\Delta_\zeta^x$ is the only polynomial in the Specht representation, such that $\left( \Delta_\zeta^x\right)^{s_i} =- \Delta_\zeta^x$ for all $i$ such that $i,i+1$ are in the same column of $\zeta$. This property still characterizes a unique element in a different copy of the Specht representation, that we shall still denote $ \Y(\zeta)$.
We shall still call *Young basis* the basis proportional to $\{ \Y(\zeta)\, e_{\zeta,t}\}$, with the same factors of proportionality as in the case where we start with $\Delta_\zeta^x$. More details about how to compute such a basis are given in the last section.
Cauchy Formula
--------------
Let us now take two symmetric groups $\mfS_n^x$, $\mfS_n^a$ acting respectively on the variables $x_1,\ldots, x_n$, and $a_1,\ldots, a_n$, with generators $s_i^x$, $s_i^a$.
We also use the group $\mfS_n= \mfS_n^{ax}$, which permutes simultaneously the variables $a_i$ and $x_i$. We write its generators $s_i$, instead of $s_i^{ax}$. These are such that $$s_i = s_i^a s_i^x = s_i^x s_i^a \, .$$
We shall give a decomposition of the two 1-dimensional idempotents\
$\square:= (n!)^{-1}\sum \sigma$ and $\nabla:=(n!)^{-1} \sum (\moins 1)^{\ell(\sigma)} \sigma$, with respect to the groups $\mfS_n^x$ and $\mfS_n^a$.
Indeed, by definition $$\square^{ax} = \frac{1}{n!}\, \sum_\sigma \sigma^x \sigma^a\ ,$$ with pairs of permutations in $\mfS_n^a$, $\mfS_n^x$ commuting with each other.
Taking the basis of Young idempotents, instead of the basis of permutations, one obtains a Cauchy-type formula : $$\label{CauchyCarre}
\square^{ax} = \sum_{t,u} \frac{1}{d(t)} e_{t,u}^x\, e_{t,u}^a \ ,$$ where the sum is over all pairs of standard tableaux of the same shape, and where $d(t)$ is the number of tableaux of the same shape as $t$.
The element $\nabla^{ax}$ is obtained from $\square^{ax}$ under the isomorphism induced by : $$s_i^x \to \widehat{s_i^x}:= - s_i^x \ , \ s_i^a \to s_i^a$$
From the expressions of $\square$ and $\nabla$ as sums of products of permutations, one sees that for any simple transposition, $$\begin{aligned}
\square^{ax} \, s_i^x = \square^{ax} \, s_i^a \ &\& &
s_i^x\, \square^{ax} = s_i^a\, \square^{ax} \\
\nabla^{ax} \, s_i^x = -\nabla^{ax} \, s_i^a \ &\& &
s_i^x\, \nabla^{ax} = -s_i^a\, \nabla^{ax} \ .\end{aligned}$$
By taking products, one gets $$\square^{ax} \, s_i^x s_j^x \cdots s_k^x =
\square^{ax} \, s_i^a s_j^x \cdots s_k^x
= \square^{ax} \, s_j^x \cdots s_k^x s_i^a =\cdots \ ,$$ and therefore, for any permutation $\sigma$, $$\begin{aligned}
\square^{ax} \, \sigma^x = \square^{ax} \, (\sigma^a)^{-1}
\ &\& & \ \sigma^x\, \square^{ax} = (\sigma^a)^{-1} \, \square^{ax} \\
\nabla^{ax} \, \sigma^x = (\moins 1)^{\ell(\sigma)}
\nabla^{ax} \, (\sigma^a)^{-1}
\ &\& & \ \sigma^x\, \nabla^{ax} = (\moins 1)^{\ell(\sigma)}
(\sigma^a)^{-1}\, \nabla^{ax} \ . \end{aligned}$$
Restricting $\square^{ax}$ or $\nabla^{ax}$ to a representation of $\mfS_n^a$ of type is achieved by multiplying $\square^{ax}$ or $\nabla^{ax}$ by $e_\l^a$.
Since an idempotent $e_{t,t}^x$ is sent under $s_i^x\to -s_i^x$ to $e_{t^\sim,t^\sim}^x$, where $t\to t^\sim$ denotes the transposition of tableaux, using expression (\[IdempCentral\]), we obtain the two equivalent expansions : $$\begin{aligned}
\square^{ax}\, e_\l^a &=& \sum_{t,u\in \Tab(\l)}
\frac{1}{d(t)}\, e_{t,u}^a\, e_{t,u}^x \ , \\
\label{DecompoNabla}
\nabla\, e_\l^a &=& \sum_{t,u\in \Tab(\l)}
(-1)^{\ell(t,u)} \frac{1}{d(t)} e_{t,u}^a\, e_{t^\sim,u^\sim}^x \ .\end{aligned}$$
For example, there are $4$ standard tableaux with three boxes : $$\a= \smallyoung{\s1& \s2&\s3\cr}\ ,\
\b=\smallyoung{\s2\cr \s1&\s3\cr}\ ,\
\g=\smallyoung{\s3\cr \s1&\s2\cr}\ ,\
\d=\smallyoung{\s3\cr \s 2\cr \s 1\cr} \ ,$$ and the elements $\square^{ax}$ and $\nabla^{ax}$ decompose as : $$\begin{aligned}
\square^{ax} &=& e_{\a,\a}^a e_{\a,\a}^x + \frac{1}{2}\left(
e_{\b,\b}^a e_{\b,\b}^x + e_{\b,\g}^a e_{\b,\g}^x
+ e_{\g,\b}^a e_{\g,\b}^x +e_{\g,\g}^a e_{\g,\g}^x \right)
+ e_{\d,\d}^a e_{\d,\d}^x \\
\nabla^{ax} &=& e_{\a,\a}^a e_{\d,\d}^x + \frac{1}{2}\left(
e_{\b,\b}^a e_{\g,\g}^x - e_{\b,\g}^a e_{\g,\b}^x
- e_{\g,\b}^a e_{\b,\g}^x +e_{\g,\g}^a e_{\b,\b}^x \right)
+ e_{\d,\d}^a e_{\a,\a}^x\, ,\end{aligned}$$ the middle part being the component of type $[1,2]$.
Pfaffians
=========
Let $Z=\bigl[ z[i,j] \bigr]$ be an anti-symmetric matrix of even order $n$. Its determinant is the square of a function of the $z[i,j]$, which is called the *Pfaffian* of $Z$. The Pfaffian is a certain sum, with coefficients $\pm$, of products $z[i,j]\cdots z[k,l]$. We refer to [@Knuth] for an historical and complete presentation.
Deciding to write each monomial in $z[i,j]$ according to some lexicographic order on the variables, one can erase $z$, brackets and commas, and use permutations : $$z[i,j]\, z[k,l]\, z[p,q] \to [i\, j\,\, k\, l\, p\, q]$$
The Pfaffian of $Z$ has become an alternating sum of permutations which can be defined recursively as follows [@Knuth]. For any vector $v\in \N^n$ of even length $n$, let $$\Pf(v) = \sum_{i=1}^{n-1} (-1)^i\, \Pf(v\setminus \{v_i,v_n\})\cdot
[v_i,v_n] \ ,$$ where the product is the concatenation product, and where $v\setminus \{v_i,v_n\}$ means suppressing the components $v_i,v_n$ inside $v$.
The initial case is $\Pf([\, ]) = [\, ]$.\
$\Pf([1,2])=[1,2] \quad ; \quad
\Pf([1,2,3,4])=[1,2,3,4]-[1,3,2,4]+[2,3,1,4] \ ,$\
$ \Pf([1,2,3,4,5,6])=
[1,2,3,4,5,6]-[1,2,3,5,4,6] +[1,2,4,5,3,6]\\
\hspace*{10pt} -[1,3,2,4,5,6]+[1,3,2,5,4,6]-[1,3,4,5,2,6] -[1,4,2,5,3,6]\\
\hspace*{10pt} +[1,4,3,5,2,6]+[2,3,1,4,5,6] -[2,3,1,5,4,6]+[2,3,4,5,1,6]\\
\hspace*{10pt} +[2,4,1,5,3,6] -[2,4,3,5,1,6]-[3,4,1,5,2,6]+[3,4,2,5,1,6]\, . $
Let $\Pf_n := \Pf([1,\ldots,n])$ be the above sum of permutations (we use the notation $\Pfaff(Z)$ for the Pfaffian of an antisymmetric matrix, and $\Pf_n$ for the formal sum of permutations). Our data, the $z[i,j]$, were such that $z[i,j]=-z[j,i]$, and that $z[i,j] z[k,l]=z[k,l]z[i,j]$. We shall see in the next proposition,whose proof is immediate, that the permutations appearing in $\Pf_n$ are cosets representatives, modulo the symmetries possessed by the $z[i,j]$.
Indeed, let $\mfS_{n/2}$ be the symmetric group which permutes the blocks $[1,2],[3,4],\ldots, [n\moins 1,n]$, and let $\Theta_n$ be the sum of its elements. Simple transpositions $s_1,s_3,\ldots, s_{n-1}$ commute with $\Theta_n$.
\[Pfaff2Sg\] For even $n$, the alternating sum of all permutations can be factorized as follows : $$\begin{aligned}
n!\, \nabla &=& (1-s_1)(1-s_3)\cdots (1-s_{n-1})\, \Theta_n\, \Pf_n \\
&=& \Theta_n\, (1-s_1)1-s_3)\cdots (1-s_{n-1})\, \Pf_n\, . \end{aligned}$$
As a consequence, we may write the Pfaffian of $Z$ as $$\Pfaff(Z) = z[1,2]\, z[3,4]\cdots z[n\moins 1,n]\, \nabla
\, \frac{n!}{2^n (n/2)!} \ ,$$ since $s_1,s_3,\dots$ and the permutations in $\Theta_n$ act trivially on the product $z[1,2]z[3,4]\cdots$.
For example, for $n=6$, $$\begin{gathered}
\sum _{\sigma\in\mfS_6} (\moins 1)^{\ell(\sigma)}\sigma =
(1-s_1)(1-s_3)(1-s_5) \bigl([123456]+[125634]+[341256] \\
+[345612]+[561234]+[563412] \bigr)\,
\Pf_6\, .\end{gathered}$$
An interesting approach, due to Luque and Thibon [@LuqueThibon], to combinatorial properties of Pfaffians is through shuffle algebras. As a matter of fact, the same methods give also the *Hafnian*, i.e. the image of the Pfaffian (as an element of the group algebra) under the involution $s_i\to -s_i$, $i=1,\ldots, n\moins 1$. We shall not need this approach, having written the Pfaffian in terms of the alternating sum of all permutations.
In [@LLT] one finds Pfaffians and determinants associated to a family of formal series, which are needed in geometry.
Pfaffian, with the help of two symmetric groups
===============================================
The main case that we want to treat now is the case of an antisymmetric matrix with entries $$z[i,j]\, g[i,j] \, ,$$ the $z[i,j]$ satisfying the Plücker relations, and the $g[i,j]$ being symmetrical: $g[i,j]=g[j,i]$.
Of course, any antisymmetric matrix $N= \Bigl[ n[i,j]\Bigr]$ can be written in this way, introducing extra variables $a_i$, and writing $$n[i,j] = (a_i-a_j)\, \frac{n[i,j]}{a_i-a_j} \ .$$
The Pfaffian, being a sum of products of $z[i,j]$, belongs to the irreducible representation of shape $[m,m]$ (with respect to the symmetric group $\mfS_n^z$ acting on the indices of the indeterminates $z[i,j]$, $i=1,\ldots,n$, $m=n/2$). Hence, it can be expressed as a linear combination of Specht elements : $$c_\zeta(g) z[\zeta] +\cdots +c_{\aleph}(g) z[\aleph] \ .$$
Thanks to Prop.\[th:Specht3\], the coefficients $c_t(g)$ are the same as for the specialization $z[i,j]= (a_i-a_j)$.
The case of the Pfaffian of $ (a_i-a_j)\, (x_i+x_j)^{-1}$ has been treated by Sundquist [@Sundquist], but we need a more extensive description than his.
The coefficient $c_{\aleph}(g)$ is obtained by specializing $a_1=1 = \cdots = a_m$, $a_{m+1} =0=\cdots = a_n$. In that case, the sum over all permutations in $\mfS_n^{a,g}$ $$\sum (-)^{\ell(\sigma)} \left( (a_1\moins a_2) g[1,2]\,
(a_3\moins a_4) g[3,4]\, \cdots \right)^\sigma$$ reduces to $$\sum (-)^{\ell(\sigma)}
\bigl( g[1,m\plus 1]\, g[2,m\plus 2]\, \cdots
g[m\moins 1, n] \bigr)^\sigma \ ,$$ where the sum is now only over the subgroup $ \mfS_{m} \times \mfS_{m}$ which permutes $1,\ldots, m$, and $m\plus 1,\ldots, 2m$ separately.
This sum is equal to $$\label{Detg}
m!\, g[1\ldots m\, \,|\, m\plus 1\ldots n] \, .$$ Therefore the Pfaffian of the matrix $\Bigl[(a_i-a_j)g[i,j] \Bigr]$ is, up to a normalization constant, equal to the image of $$\Omega_{a,g}:= (a_1-a_{m+1})\cdots (a_m-a_n)\,
g[1\ldots m\, \,|\, m\plus 1\ldots n]$$ under the anti-symmetrization $\nabla^{ag}$.
This last element belongs to the Specht representation of $\mfS_n^a$ of shape $[m,m]$, and therefore, thanks to (\[DecompoNabla\]), belongs to the space[^3] $$V^a_{[m,m]} \otimes V^g_{[2^m]} \, .$$
The action of $\nabla^{ag} = \sum \pm e_{t,t}\, e_{t^\sim,t^\sim}$ restricts to the tableaux $t$ of shape $[m,m]$, and, consequently, the Pfaffian is proportional to $$\sum_{t\in \Tab([m,m])} (-)^{\ell(\aleph,t)}\,
\Omega_{a,g}\, e_{\aleph,t}^a\, e_{\aleph^\sim,t^\sim}^g \, .$$
Eventually, taking into account normalizations, one can expand the Pfaffian in the Young basis : $$\Pfaff\left((a_i-a_j)g[i,j]\right) =
\sum_{t\in \Tab([m,m]} (-1)^{\ell(\zeta,t)}\, \Y^a(t)\, \Y^g(t^\sim) \, .$$
In short, the Pfaffian may be considered as a trace in the space $ V^a_{[m,m]} \otimes V^g_{[2^m]}$. We summarize the preceding considerations in the following theorem.
\[th:Pfaffzg\] Let $n=2m$ be an even positive integer. Let $z[i,j]$ be Plücker indeterminates, let $g[i,j]$ be indeterminates symmetrical in $i,j$, for $i,j=1\ldots n$. Then $$\begin{aligned}
\Pfaff\left( z[i,j]g[i,j]\right) &=&
d(\aleph)\, \Y^z(\aleph)\, \Y^g(\aleph^\sim)\, \nabla^{z,g} \\
&=& \sum_{t\in \Tab([m,m])} (-1)^{\ell(\zeta,t)}\, \Y^z(t)
\Y^g(t^\sim) \, ,\end{aligned}$$ where $d(\aleph)$ is the number of Young tableaux of shape $[m,m]$.
This gives two ways of computing a Pfaffian. Either by a summation over all tableaux, or by antisymmetrization of the element $\Y^z(\aleph)\, \Y^g(\aleph^\sim)$ (one could take any other tableau than $\aleph$, or take Specht polynomials instead of Young polynomials).
For example, for $n=6$, $z[i,j]=a_i-a_j$, $g[i,j]=(x_i^4-x_j^4)(x_i-x_j)^{-1}$, one sees that $$g[123\, |\, 456]= \Delta^x(1,2,3)\Delta^x(4,5,6)
S_{1,1,1}(x_1,\ldots, x_6) \, .$$ In particular, $g[123\, |\, 456]$ is the product of a Specht polynomial by a symmetric function in $x_1,\ldots,x_6$. Therefore, the Pfaffian is equal, up to a numerical factor, to $$(a_1\moins a_4)(a_2\moins a_5)(a_3\moins a_6)\Delta^x(1,2,3)\Delta^x(4,5,6)
\nabla^{ax}\, S_{1,1,1}(x_1,\ldots, x_6) \, .$$
Three symmetric groups
======================
One can use $k$ families of Plücker indeterminates $z^1[i,j], z^2[i,j],\ldots, z^k[i,j]$, together with a last family of symmetric, or antisymmetric, indeterminates $g[i,j]$, according to the parity of $k$. A Pfaffian $$\Pfaff\left( z^1[i,j]\cdots z^k[i,j]\, g[i,j] \right)$$ still belongs to the irreducible representation $$V_{[m,m]}^{z^1} \otimes \cdots \otimes V_{[m,m]}^{z^k}$$ of $\mfS_n^{z^1} \times \cdots \times \mfS_n^{z^k}$.
Therefore, it can be expanded into the Young basis : $$\Pfaff\left( z^1[i,j]\cdots z^k[i,j]\, g[i,j] \right) =
\sum_{t_1,\ldots,t_k} \Y^{z^1}(t_1)\cdots \Y^{z^k}(t_k)\,
f(t_1,\ldots,t_k; g) \, ,$$ sum over $k$-tuples of standard tableaux of shape $[m,m]$. To find the coefficients $ f(t_1,\ldots,t_k; g) $, one may take indeterminates $a_i^r:\, i=1\ldots n,\, r=1\ldots k$, and put $z^1[i,j]=a_i^1-a_j^1,\ldots, z^k[i,j]=a_i^k-a_j^k$.
The main difference with the case $k=1$ is that a single specialization of the $a_i^r$ is not enough to determine the Pfaffian. However, since any element of $V^a_{[m,m]}$ is characterized by the set of all its specializations $$a_{\sigma_1}=1=\cdots=a_{\sigma_m} \, ;\,
a_{\sigma_{m+1}}=0=\cdots=a_{\sigma_n} \, , \sigma\in\mfS_n \, .$$ It is in fact sufficient to take the specializations corresponding to the permutations obtained by reading the standard tableaux as permutations (reading rows from bottom to top). In that way, the number of specializations is equal to the number of indeterminate coefficients, and representation theory tells us that this system is solvable.
I do not see anything more to say for general $k$, but shall restrict to $k=2$.
\[th:Pfaffabz\] Let $a[i,j],b[i,j],z[i,j]$, $1\leq i<j\leq n=2m$ be three families of Plücker indeterminates. Then $$\Pfaff\left(\frac{a[i,j] b[i,j]}{z[i,j]} \right) =
\bigl( \Y^a(\aleph) \Y^z(\zeta)\, \nabla^{az} \bigr)
\bigl( \Y^b(\aleph) \Y^z(\zeta)\, \nabla^{bz} \bigr)
\frac{d(\aleph)^2}{\prod z[i,j]} \, ,$$
where $\aleph$ is the bottom tableau of shape $[m,m]$, and $\zeta$, the top tableau of shape $2^m$, $d(\aleph)$ still being the number of tableaux of the same shape as $\aleph$.
As we already used, we take $a[i,j]=a_i-a_j$, $b[i,j]=b_i-b_j$. For $z[i,j]$, we can assume that we are given a generic $N\times (N\plus n-1)$ matrix, $N$ sufficiently big. $$M= \begin{bmatrix}
x_1 & x_2 &\cdots &x_n & \cdots \\
y_1 & y_2 &\cdots &y_n & \cdots \\
\cdots & \cdots & \cdots & \cdots & \fbox{$ K$}
\end{bmatrix} \, ,$$ with $K$ a submatrix of order $N\moins 2$. One then takes $z[i,j]$ to be the maximal minor containing $x_i,y_j$ and $X_i$ (resp $Y_j$) to be the minor of order $N\moins 1$ containing $K$ and $x_i$ (resp. $y_j$). Sylvester’s relation [@BL] states that $\det(K) \, z[i,j]= X_iY_j -X_j Y_i$. In all, the value of the Pfaffian is determined by the case $$\Pfaff\left( \frac{(a_i-a_j)(b_i-b_j)}{X_iY_j -X_j Y_i } \right)\, .$$
The first step is still to specialize $a_1=1=\cdots=a_m$, $a_{m+1}=0=\cdots= a_n$, as for Th.\[th:Pfaffzg\]. The Pfaffian becomes $$F:=\det\left( (b_i-b_j)(X_iY_j -X_j Y_i)^{-1}
\right)_{1\leq i\leq m <j\leq n} \, .$$ One has now to specialize $b_1,\ldots,b_n$. Instead of taking all permutations $\sigma\in\mfS_n$, by reordering rows and columns, one can suppose that there exist $\a,\b,\g$: $\sigma= [1,\ldots,\a, \a\plus \b,\ldots, \a\plus \g,
\a\plus 1,\ldots, \a\plus b\moins 1, \a\plus \g\plus 1,\ldots, n]$. In that case, the specialization $b_{\sigma_1}=1, \ldots, b_{\sigma_m}=1$, $b_{\sigma_{m+1}}=0, \ldots, b_{\sigma_n}=0$ of $F$ factorizes into two blocks, each of which is of Cauchy type $\det( (X_iY_j-Y_iX_j)^{-1} )_{i,j=1\ldots k}$. For example, for $m=4$, the specialization $b_1=b_2=b_5=b_6=1$, $b_3=b_4=b_7=b_8=0$ is $$\begin{vmatrix}
0 & 0 & (X_1Y_7-X_7Y_1)^{-1} & (X_1Y_8-X_8Y_1)^{-1} \\
0 & 0 & (X_2Y_7-X_7Y_2)^{-1} & (X_2Y_8-X_8Y_2)^{-1} \\
-(X_3Y_5-X_5Y_3)^{-1} & -(X_3Y_6-X_6Y_3)^{-1} & 0 & 0 \\
-(X_4Y_5-X_5Y_4)^{-1} & -(X_4Y_6-X_6Y_4)^{-1} & 0 & 0 \\
\end{vmatrix}\, .$$ The evaluation of a Cauchy determinant is, of course, immediate (since 1812), and in final, for any $\sigma$, the specialization $b_{\sigma_1}=1, \ldots, b_{\sigma_m}=1$, $b_{\sigma_{m+1}}=0, \ldots, b_{\sigma_n}=0$ of $F$ is equal to $$(\moins 1)^{\ell(\sigma)} \Delta^z(\sigma_1,\ldots,\sigma_m)
\Delta^z(\sigma_{m+1},\ldots,\sigma_n)\,
\prod_{1\leq i\leq m <j\leq n} (z_i-z_j)^{-1} \, .$$ Therefore, $F$ coincides, up to a numerical factor, with $$\begin{gathered}
b_1\cdots b_m \Delta^z(1,\ldots,m)\Delta^z(m\plus1,\ldots,n) \nabla^{b,z}
\prod_{1\leq i\leq m <j\leq n} (z_i-z_j)^{-1} \\
= (b_1-b_{m+1})\cdots (b_m-b_n) \Delta^z(m\plus1,\ldots,n) \nabla^{b,z}
\prod_{1\leq i\leq m <j\leq n} (z_i-z_j)^{-1} \, .\end{gathered}$$ From this specialization, one writes the Pfaffian as $$\Y^b(\aleph) \Y^z(\zeta)\, \nabla^{bz}
\frac{\Delta^z(1,\ldots,m) \Delta^z(m\plus 1,\ldots,n)}
{\Delta^z(1,\ldots,n)} \, \nabla^{abz} \, .$$ Since any permutation in $\mfS^{abz}$ commutes with $\nabla^{bz}\Delta^z(1,\ldots,n)^{-1}$, the theorem follows
Okada [@Okada Th.3.4] (see also [@IshiOkada Formula 1.8]) has already computed $\Pfaff\bigl( (a_i\moins a_j)(b_i\moins b_j) (x_i\moins x_j)^{-1} \bigr)$. His formula can be written $$\begin{gathered}
\label{Pfaffabx}
\Pfaff\left(\frac{(a_i-a_j)(b_i-b_j)}{x_i^2-x_j^2}\right) \\ =
\prod_{i<j}\frac{x_i+x_j}{x_i-x_j}\,
\Pfaff\left(\frac{a_i-a_j}{x_i+x_j} \right)
\, \Pfaff\left(\frac{b_i-b_j}{x_i+x_j} \right) \, .\end{gathered}$$
In [@Okada] and [@IshiOkada], one finds many evaluations of Pfaffians and determinants, with entries which are specializations of Plücker indeterminates. For example, Okada [@Okada Th.3.4] takes $$a[i,j]= \begin{vmatrix} 1+ a_i x_i & x_i +a_i \\
1+ a_j x_j & x_j +a_j\end{vmatrix}
\ \text{or}\
a[i,j] =
\begin{vmatrix}
1+ a_i x_i^2 & x_i +a_i x_i & x_i^2 +a_i\\
1+ a_j x_j^2 & x_j +a_j x_j & x_j^2 +a_j\\
1+cz^2 & z+cz & z^2 +c
\end{vmatrix} \, .$$
All these families satisfy the Plücker relations, and one could take determinants of higher order of the type below, as indeterminates $z[i,j]$. On the other hand, Pfaffians involving elliptic functions as in [@Okada2] do not fall in this category, the Riemann relations replacing in that case Plücker relations.
Special Pfaffians
=================
There are cases where $\Y^z(\aleph)\, \Y^g(\aleph^\sim)\, \nabla^{z,g}$ can be written as a determinant. Indeed, let us take $z[i,j]=(a_i-a_j)$, and $g[i,j]=(x_i+x_j)^{-1}$ as Sundquist[@Sundquist]. For any integer $k$, let $U(a,x^k)$ be the determinant of oder $n$ $$\label{UU}
U(a,x^k) :=\Bigl|a_i x_i^0,\, a_i x_i^k,\dots,
a_i x_i^{k(m-1)},\, x_i^0,\, x_i^k,\ldots, x_i^{k(m-1)}
\Bigr|_{i=1\ldots n}\ .$$
The Laplace expansion of $U(a,x)$ along its first $m$ columns shows that it belongs to the Specht representation of $\mfS_n^x$ of shape $[2^m]$, and therefore, that $U(a,x)$ is an element of $V_{[m,m]}^a\otimes V_{[2^m]}^x$, which is identified by the specialization $a_1=1 = \cdots = a_m$, $a_{m+1} =0=\cdots = a_n$. Needless to add that this specialization is $$\Delta(x_1,\ldots,x_m)\, \Delta(x_{m+1},\ldots, x_n) \, .$$
On the other hand, the Pfaffian $\Pfaff\left( (a_i\moins a_j)(x_i\plus x_j)^{-1} \right)$ belongs to the same space, and the same specialization sends it, according to (\[Detg\]), and thanks to the Cauchy identity relative to the determinant $\Bigl|(x_i+y_j)^{-1} \Bigr|_{i,j=1\ldots m}$, to $$c_{\aleph}(x) = \frac{\Delta^x(1,\ldots,m)\,
\Delta^x(m\plus 1,\ldots, n) }
{\prod_{i\leq m <j} x_i+x_j }\, .$$
Taking a symmetrical denominator, one rather writes $$c_{\aleph}(x) = \frac{\Delta^{xx}(1,\ldots,m)\,
\Delta^{xx}(m\plus 1,\ldots, n) }
{\prod_{1\leq i<j\leq n} x_i+x_j }\, .$$
Therefore, this specialization coincides with the one of $U(a,x^2)$, and one recovers the following theorem.
\[th:Sundquist\] Let $n=2m$, $a_1,\ldots,a_n$, $x_1,\ldots,x_n$ be indeterminates. Let $U(a,x)$ be the determinant *.*
Then $$\begin{aligned}
\Pfaff\left( \frac{a_i-a_j}{x_i+x_j} \right)
&=& \frac{d}{n!} \sum_\sigma (\moins 1)^{\ell(\sigma)}
\Bigl( (a_1\moins a_{m+1})\cdots (a_{m-1}\moins a_n)\, \times \\
& & \hspace{30pt}
\Delta^{xx}(1,\ldots,m) \Delta^{xx}(m\plus 1,\ldots, n) )\Bigr)^\sigma
\, \prod_{1\leq i<j\leq n} (x_i+x_j)^{-1} \\
&=& \det\bigl(U(a,x^2) \bigr)\,
\prod_{1\leq i<j\leq n} (x_i+x_j)^{-1} \, ,\end{aligned}$$ where the sum is over all permutations $\sigma\in\mfS_n^{a,x}$, and $d$ is the number of Young tableaux of shape $[m,m]$.
We have given another expression in Th\[th:Pfaffzg\], using the Young basis. To stay nearer the expansion of $U(a,x)$, we can also use the Specht basis, but there will be extra terms corresponding to pairs of non-orthogonal tableaux (as soon as $n=6$).
Indeed, for $n=4$, we have $$\begin{gathered}
\prod_{i<j}(x_i+x_j)\,
\Pfaff\left( \frac{a_i-a_j}{x_i+x_j} \right) =
(a_1-a_2)(a_3-a_4)(x_1^2 -x_3^2)(x_2^2-x_4^2) \\
-(a_1-a_3)(a_2-a_4)(x_1^2 -x_2^2)(x_3^2-x_4^2) \, .\end{gathered}$$
For $n=6$, the expansion in the Specht basis is $$\begin{gathered}
\prod_{i<j}(x_i+x_j)\,
\Pfaff\left( \frac{a_i-a_j}{x_i+x_j} \right) =
(a_1-a_2)(a_3-a_4)(a_5-a_6) \times \\
\Delta(x_1^2,x_3^2,x_5^2) \Delta(x_2^2, x_4^2, x_6^2)
\bigl(1 -s_2^{ax} -s_4^{ax} +s_2^{ax}s_4^{ax} -
s_2^{ax}s_4^{ax}s_3^{ax} \bigr) \\
- (a_1-a_2)(a_3-a_4)(a_5-a_6) \, \Delta(x_1^2,x_2^2,x_3^2)
\Delta(x_4^2, x_5^2, x_6^2) \\
=\begin{vmatrix}
\, a_1 & a_1 x_1^2 &a_1 x_1^4 & 1 & x_1^2 & x_1^4 \, \\
\, a_2 & a_2 x_2^2 &a_2 x_2^4 & 1 & x_2^2 & x_2^4 \, \\
\, a_3 & a_3 x_3^2 &a_3 x_3^4 & 1 & x_3^2 & x_3^4 \, \\
\, a_4 & a_4 x_4^2 &a_4 x_4^4 & 1 & x_4^2 & x_4^4 \, \\
\, a_5 & a_5 x_5^2 &a_5 x_5^4 & 1 & x_5^2 & x_5^4 \, \\
\, a_6 & a_6 x_6^2 &a_6 x_6^4 & 1 & x_6^2 & x_6^4
\end{vmatrix} \, .\end{gathered}$$ The first five terms, written as images of the first one, are the Specht polynomials for pairs of tableaux transposed of each other, but there is a sixth term corresponding to the only non-zero entry outside the diagonal in the matrix of scalar products.
In the case where the $a_i's$ are fixed powers of the indeterminates $x_i$’s, then the determinant $U(a,x^2)$ is a determinant of powers of $x_i$’s, proportional to a Schur function. Thus, Th.\[th:Sundquist\] implies
\[cor:PfafPowers\] Let $r,k$ be positive integers, $n=2m$ be an even integer. Let $q=2(k-1)$, and $\lambda$ be the (increasing) partition $[0,r,q, q\plus r, \ldots, (m\moins 2)q, (m\moins 2)q+r ,
(m\moins 1)q, m\moins 1)q+r]$ Then $$\Pfaff\left( \frac{x_i^{r+1}-x_j^{r+1}}{x_i^k+x_j^k} \right) =
\frac{\Delta^x(1,\ldots,n)}{\prod_{1\leq i<j\leq n} x_i^k+x_j^k}\
S_\l(x_1,\ldots, x_n) \, .$$
We can give more interesting examples of Pfaffians $\Pfaff\left( z[i,j]g[i,j]\right)$, with $g[i,j]$ a symmetric function in $x_i,x_j$, which admits a symmetric function in $x_1,\ldots, x_n$ as a factor.
For example, take a partition $\lambda$, an alphabet $B$, and variables $x_1,\ldots,x_n$.
Chosing a positive $k$, we want to evaluate $$\Pfaff\left((a_i-a_j) S_\l(B+ x_i\plus x_j)\frac{x_i-x_j}{x_i^k-x_j^k}
\right) \, .$$ According to Th.\[th:Pfaffzg\], we need only compute the specialization $a_1=1,\ldots,a_m=1$, $a_{m+1}=0,\ldots, a_n=0$, which is equal to $$\det\left({S_\lambda(B+x_i\plus x_j)}\frac{x_i-x_j}{x_i^k -x_j^k}
\right)_{ 1\leq i\leq m <j\leq n} \, .$$
To proceed further, one needs to evaluate such determinants. We shall do that in the next section. For the moment, let us only use the fact that the determinant in question is *the product of a symmetric function* $f(x_1,\ldots, x_n)$ by $\Delta^x(1\ldots m)\Delta^x(m\plus 1\ldots n)
\prod_{1\leq i\leq m <j\leq n} (x_i\moins x_j) (x_i^k \moins x_j^k)^{-1} $.
The symmetric function can then be factored out, so that the Pfaffian $\Pfaff\left((a_i-a_j) S_\l(B+ x_i\plus x_j)
(x_i\moins x_j) (x_i^k \moins x_j^k)^{-1} \right)$ is finally equal to $$\label{PfaffSf}
f(x_1,\ldots, x_n)\, \Pfaff\left( (a_i-a_j)
\frac{x_i-x_j}{x_i^k -x_j^k}\right) \, .$$ Thus, the evaluation of the Pfaffian of order $2m$ has been reduced to the evaluation of a determinant of order $m$.
Ishikawa [@Ishikawa], Okada [@Okada], and Ishikawa-Okada-Tagawa-Zeng [@IshiOkada] have given many generalizations of Sundquist’s Pfaffian. Instead of using Plücker coordinates, they use specific determinants (which, of course, satisfy built-in Plücker relations).
Determinants and two symmetric groups
=====================================
The fundamental surveys of Krattenthaler [@Kratt; @Kratt2] describe many methods to evaluate determinants. We would like to add to them one more method, using two symmetric groups.
In the course of proving Th.\[th:Pfaffabz\], we have met a determinant which happened to possess an unsuspected global symmetry, and that we record now (notice that $\nabla^{az}$ is given by a summation on the full symmetric group $\mfS_n$).
\[cor:Detaz\] Let $\{ a[i,j]\}$, $\{ z[i,j]\}$, $1\leq i <j\leq n=2m$ be two families of Plücker indeterminates. Then $$\det\left( \frac{ a[i,j]}{z[i,j]}
\right)_{1=1\ldots m,\, j=m\!+\!1\ldots n}
= d(\aleph)\, \Y^a(\aleph) \Y^z(\zeta) \, \nabla^{az} \, \prod_{1\leq i <j\leq n}
z[i,j]^{-1} \, ,$$ where $\aleph$ is the bottom tableau of shape $[m,m]$, $\zeta$, the top tableau of shape $2^m$.
The special case where $z[i,j]=x_i-x_j$, or $z[i,j]=x_i^k-x_j^k$ is worth commenting, since it reveals a symmetry in $x_1,\ldots, x_n$ that we emphasize in the next theorem theorem.
To evaluate $\det\bigl( a[i,j] (x_i-x_j)^{-1}\bigr)_{1=1\ldots m,\, j=m\!+\!1\ldots n}$, we already used that we need only take $a[i,j]=a_i-a_j$, and specialize $[a_1,\ldots, a_n]$ in all permutations of $[1^m,0^m]$. However, writing $$R^x(1\ldots m\, |\, m\plus 1\ldots n) := \prod\nolimits_{1\leq i\leq m}
\prod\nolimits_{m+1\leq j\leq n} (x_i-x_j) \, ,$$ it is clear that $$R^x(1\ldots m\, |\, m\plus 1\ldots n)\,
\Biggl|\frac{a_i-a_j}{x_i-x_j} \Biggr|_{1\leq i\leq m<j\leq 2m}$$ has the same specializations as $U(a,x)$.
Thanks to Th.\[th:Pfaffzg\] and Th.\[th:Sundquist\], going back to the variables $z[i,j]$, taking variables $x_i^2$ instead of $x_i$, we have just obtained :
\[th:DetSf\] Let $z[i,j]$, $1\leq i<j \leq n=2m$ be Plücker indeterminates, and let $x_1,\ldots, x_n$ be indeterminates. Then $$\begin{aligned}
\notag
\det\left| \frac{z[i,j]}{x_i^2-x_j^2} \right|_{1\leq i\leq m<j\leq n}
&=& \frac{1}{R^{xx}(1\ldots m\, |\, m+1\ldots n) } \\
& & \hspace{15pt} \times \
\sum_{t\in \Tab([m,m]} (-1)^{\ell(\zeta,t)}\, \Y^z(t) \Y^{xx}(t^\sim) \\
&=& \Y^z(\aleph) \Y^{xx}(\aleph^\sim)\, \nabla^{xz}
\frac{d(\aleph)}{R^{xx}(1\ldots m | m\plus1\ldots n) } \\
&=& \frac{\prod_{1\leq i<j\leq n}{x_i+x_j}}{
R^{xx}(1\ldots m\, |\, m+1\ldots n) } \,
\Pfaff\left(\frac{z[i,j]}{x_i+x_j} \right) \, . \end{aligned}$$ the Pfaffian being of order $n$, and the superscript $xx$ meaning using the indeterminates $x_i^2$ instead of $x_i$.
In particular, $R^x(1\ldots m|m\plus 1\ldots n)\,
\left| \frac{z[i,j]}{x_i-x_j} \right|$ is symmetrical in $x_1,\ldots, x_n$.
Taking $z[i,j]= S_\lambda(B+x_i+x_j)(x_i-x_j)$ and changing the powers of the variables $x_i$ in denominator, we get the following corollary.
\[th:DetSf2\] Let $\lambda$ be a partition, $B$ be an alphabet, $m,k$ be two positive integers. Then $$\frac{1}{\Delta^x(1\ldots m)\Delta^x(m\plus 1\ldots 2m)}
\left(\prod_{1\leq i\leq m<j \leq 2m} \frac{x_i^k-x_j^k}{x_i-x_j}\right)\,
\left| \frac{S_\lambda(B \plus x_i\plus x_j)}{
S_{k-1}(x_i+x_j)} \right|_{1\leq i\leq m<j\leq 2m}$$ is a function symmetrical in $x_1,\ldots, x_{2m}$.
Notice that $\Bigl|p_2(B+x_i+x_j) \Bigr|_{1\leq i\leq 2<j\leq 4}$, where $p_2$ is the second power sum, does not furnish a symmetric function in $x_1,\ldots, x_4$ (this does not contradict the corollary, because, fortunately, $p_2$ is not a Schur function).
For $k=1$, and $\lambda=r^p$, a rectangular partition, one gets an identity which is useful in the theory of orthogonal polynomials [@Cbms Prop. 8.4.3] : $$\begin{gathered}
\frac{1}{ \Delta^x(1\ldots m)\Delta^x(m\plus 1\ldots 2m)}
\left| S_{r^p}(B+x_i+x_j)\right|_{1\leq i\leq m<j\leq n} \\
= \left(S_{(r+1)^{p-1}}(B) \right)^{m-1}\,
S_{(r-m+1)^{p+m-1}}(B+x_1+\cdots +x_n) \, .\end{gathered}$$
More precisely, given *moments* $\mu_k=(\moins 1)^k S_{1^k}(B)$, supposed to be sufficiently generic, then $\{S_{n^n}(B+x) \}$ is a family of orthogonal polynomials, with respect to this moments [@Cbms Ch.8]. The Christoffel-Darboux kernel of order $r$, $K_r(x,y)$, is proportional to $S_{r^{r+1}}(B+x+y)$, and Cor.\[th:DetSf2\] states that the determinant with entries $K_r(x_i,y_j)$, $1\leq i,j\leq n$ is a symmetric function of $x_1,\ldots,x_n,y_1,\ldots,y_n$ and gives its precise value. This property can be directly proved, using Bazin relation on minors [@L-Shi] (see also the article of Rosengren about the relations between Pfaffians and kernels [@Rosengren]).
The case $k=2$, and $\lambda=\rho :=[1,\ldots,r]$, has been settled by [@IshiOkada; @Okada2], and reads $$\begin{gathered}
\left| \frac{S_\rho(B+x_i+x_j)}{x_i+x_j}\right|_{1\leq i\leq m<j\leq n}
\\ =
\frac{ \Delta^x(1\ldots m)\Delta^x(m\plus 1\ldots 2m) }{
\prod_{1\leq i\leq m<j \leq 2m} x_i+x_j}
\left(S_\rho(B) \right)^{m-1}\,
S_\rho(B+x_1+\cdots +x_n) \, .\end{gathered}$$
In these two cases, the symmetric function has been further factorized, compared to the case of a general partition $\lambda$. This induces a factorization of Pfaffians : $$\begin{gathered}
\Pfaff\left(a[i,j]S_{r^p}(B+x_i+x_j) \right) \\
= \left(S_{(r+1)^{p-1}}(B) \right)^{m-1}\,
S_{(r-m+1)^{p+m-1}}(B+x_1+\cdots +x_n) \hspace*{45pt} \\
\Y^a(\aleph) \Delta^x(1\ldots m)
\Delta^x(m\plus 1\ldots 2m) \nabla^{ax} \, ,\end{gathered}$$ $$\begin{gathered}
\Pfaff\left( \frac{ a[i,j]}{x_i+x_j}\, S_\rho(B+x_i+x_j) \right) \\
= \left(S_\rho(B) \right)^{m-1}\,
S_\rho(B+x_1+\cdots +x_n) \,
\Pfaff\left( \frac{a[i,j]}{x_i+x_j}\right) \, .\end{gathered}$$
Our last example will be related to the six-vertex model in physics. Stroganov [@Stroganov] found that the determinant with entries $$\sin( x_i-y_j +\eta)^{-1} \sin( x_i-y_j -\eta)^{-1} \, ,$$ with $\eta=exp(\pi\sqrt{-1}/3)$, $i,j=1\ldots n$, is the product of a symmetric function in $x_1,\ldots,x_n,y_1,\ldots,y_n$ by $\Delta^x(1,\ldots,n) \Delta^y(1,\ldots,n)$.
Since $$-4\sin( x_i-y_j +\eta) \sin( x_i-y_j -\eta)=
1+ 2 \cos(2(x\moins y)) = 1 +a/b + b/a ,$$ with $a= exp(2\pi x\sqrt{-1})$, $b=exp(2\pi y\sqrt{-1})$, Stroganov’s case is the evaluation of the determinant $\Bigl| \bigl( x_i/y_j-y_j/x_i \bigr)\bigl(
(x_i/y_j)^3 -(y_j/x_i)^3 \bigr)^{-1} \Bigr|$.
The following lemma gives a more general case, as a corollary of Th. \[th:DetSf\].
Let $k,r$ be two positive integers. Then $$\det\Bigl( (x_i^r-x_j^r)(x_i^k-x_j^k)^{-1}
\Bigr)_{i=1\ldots m,\, j=m+1\ldots 2m}$$ is equal to the product of the Schur function in $x_1,\ldots, x_{2m}$ of index $$[0,\g,\b, \b\plus \g,2\b, 2\b\plus \g,\ldots, (n\moins 1)\b,
(n\moins 1)\b\plus \g]$$ times $\Delta^x(1\ldots m) \Delta^x(m\plus 1\ldots 2m)$, where $\g=r-1$, $\b=k-2$.
For example, for $m=3$, $r=2$, $k=5$, then the determinant is equal to $$\Delta^x(1,2,3)\Delta^x(4,5,6)\, s_{[0,1,3,4,6,7]}(x_1,\ldots, x_6)\, .$$
Note: Young’s basis
===================
Young first defined *natural idempotents*, giving rise to what we have called the *Specht basis*. He then obtained orthogonal idempotents by an orthogonalization process which was later clarified by Thrall (see Rutherford [@Rutherford]).
The easiest way of obtaining Young’s orthogonal idempotents is to characterize them as simultaneous eigenvectors for the *Jucys-Murphy elements* $$\xi_j := \sum_{i<j} (i,j) \ , \ j=0,\ldots,n \, ,$$ where the sum is over transpositions (cf. Okounkov-Vershik [@OV]).
However, this approach does not provide the relations between the different idempotents for the same shape, and is inappropriate for our decomposition of Pfaffians.
We need to reinterpret Young’s orthogonalization in terms of the *Yang-Baxter relations* : $$\begin{gathered}
\label{eq:Yang1}
\left( s_i+\frac{1}{\a}\right)
\left(s_{i+1}+\frac{1}{\a \plus\b}\right) \left(s_i+\frac{1}{\b}\right) =
\left( s_{i+1}+\frac{1}{\b}\right) \left(s_{i}+\frac{1}{\a\plus \b}\right)
\left(s_{i+1}+\frac{1}{\a}\right) \\ \label{eq:Yang2}
\left( s_i+\frac{1}{\a}\right)\, \left(s_j+ \frac{1}{\b}\right) =
\left(s_j+ \frac{1}{\b}\right)\,
\left( s_i+\frac{1}{\a}\right) \ ,\ |i-j|\neq 1\end{gathered}$$ The graphical representation of these relations is easy to remember (taking $i=1$) : $$\hspace{-1cm} \begin{array}{rcl}
& [1\,2\,3] &\\[2pt]
\hbox{${\hskip -4mm\raise 3mm
\hbox{$ s_1+\frac{1}{\a} $}\, \diagup\hskip -2mm\diagup }$} &
& \diagdown \raise 3mm \hbox{$\, s_2+\frac{1}{\b} $} \\[5pt]
[2\,1\,3] & & [1\,3\,2] \\[5pt]
\hbox{$ s_2+\frac{1}{\a +\b} \, \bigg|$\hskip 3.7mm} &
&\hbox{\hskip 3.4mm $\bigg\| \, s_1+\frac{1}{\a+\b} $} \\[2pt]
[2\,3\,1] & &[3\,1\,2] \\[5pt]
\hbox{$ s_1+\frac{1}{\b} \ \diagdown\hskip -2mm\diagdown$} &
&\diagup \ s_2+\frac{1}{\a} \\[5pt]
&[3\,2\,1] &
\end{array} \hskip 15mm
\begin{array}{rcl}
& [1\,2\,3\,4] &\\[2pt]
\hbox{$\hskip -4mm\raise 3mm \hbox{$ s_1+\frac{1}{\a} $}\, \diagup\hskip
-2mm\diagup $} &
& \diagdown \raise 3mm\hbox{$\, s_3+\frac{1}{\b}$} \\[5pt]
[2\,1\,3\,4] & &[1\,2\,4\,3] \\[5pt]
s_3+\frac{1}{\b} \ \diagdown & & \diagup\hskip -2mm\diagup \ s_1+\frac{1}{\a} \\[5pt]
&[2\,1\,4\,3] &
\end{array}$$ The standard Young tableaux of a given shape are the vertices of a graph obtained by generating them with simple transpositions, starting from the top $\zeta$. We keep the same directed graph, but label an edge by $s_i+1/\rho$ : $$t \ \stackrel{s_i+1/\rho}{\hbox to 33pt{\rightarrowfill}}
\ t\, s_i \, ,$$ where $\rho$ is the diagonal distance (the difference of *contents*) between the letters $i$ and $i\plus 1$ in $t$.
A path in such a graph is interpreted as the product, in the group algebra, of the edges composing it, and the Yang-Baxter relations insure that two paths having the same end points evaluate to the same element in the group algebra.
We replace now $\zeta$ by the Specht polynomial $\Delta_\zeta^a$, and define, for any other standard tableau of the same shape, the *Young polynomial* $\Y^a(t)$ by : $$\Y^a(t) = \Delta_\zeta^x\, \left(s_i +\frac{1}{\rho}\right)
\cdots \left( s_j+\frac{1}{\rho'}\right) \, ,$$ whenever $$\zeta \ \stackrel{s_i+1/\rho}{\hbox to 33pt{\rightarrowfill}}
\ \cdots \stackrel{s_j+1/\rho'}{\hbox to 33pt{\rightarrowfill}}
\ t$$ is a path from $\zeta$ to $t$.
$$\begin{array}{rcl}
& \smallyoung{\s 2&\s4&\s6\cr \s1&\s3&\s5\cr} \\[4pt]
\raise 5pt \hbox{$\s s_2 -\frac{1}{2}$} \swarrow &
& \searrow \raise 5pt \hbox{$\s s_4-\frac{1}{2}$} \\{}
\smallyoung{\s3&\s4&\s6\cr \s1&\s2&\s5\cr} & &
\smallyoung{\s2&\s5&\s6\cr \s1&\s3&\s4\cr}\\[10pt]
{\s s_4-\frac{1}{2}} \searrow &&
\swarrow {\s s_2 -\frac{1}{2}} \\
& \smallyoung{\s3&\s5&\s6\cr \s1&\s2&\s4\cr} & \\[13pt]
& \phantom{\s s_3} \big\downarrow
\raise 3pt \hbox{$\s s_3-\frac{1}{3} $} &\\[2pt]
& \smallyoung{\s4&\s5&\s6\cr \s1&\s2&\s3\cr}
\end{array}\quad\quad
\begin{array}{rcl}
& \Y^a(\zeta) \\[4pt]
\raise 5pt \hbox{$\s s_2 -\frac{1}{2}$} \swarrow &
& \searrow \raise 5pt \hbox{$\s s_4-\frac{1}{2}$} \\[5pt]
\Y^a(t_2) & & \Y^a(t_3)\\[10pt]
{\s s_4-\frac{1}{2}} \searrow &&
\swarrow {\s s_2 -\frac{1}{2}} \\
& \Y^a(t_4) & \\[6pt]
& \phantom{\s s_3} \big\downarrow
\raise 3pt \hbox{$\s s_3-\frac{1}{3} $} &\\[6pt]
& \Y^a(\aleph)
\end{array}$$
The graph on the right side describes the generation of Young’s basis, starting with $\Y\left( \begin{smallmatrix}
2&4&6\\ 1&3&5 \end{smallmatrix}\right)$, and applying those $s_i +1/\rho$ which label the edges.
Our graph is directed, but since $-\rho$ is the distance between $i,i\plus1$ in $ts_i$, if $\rho$ is the distance in $t$, and since $$(s_i+1/\rho)\, (s_i -1/\rho) = 1-1/\rho^2 \, , \rho \neq 1\, ,$$ one could use a double orientation by normalizing the edges, taking\
$(s_i+1/\rho)/\sqrt{ 1-1/\rho^2}$ instead of $(s_i+1/\rho)$.
Pfaffians are obtained by taking a space $V_{[m,m]}^a\otimes V_{[2^m]}^x$, and using a pair of orthonormal bases that we write below in terms of the two Young bases. The graph on the left describes the orthonormal basis for shape $[3,3]$, generated downwards, and the graph on the right, the basis for shape $[2,2,2]$, generated upwards. The Pfaffian is obtained by taking the sum of products of the corresponding vertices of the two graphs : $$\begin{array}{rcl}
& \Y^a(\zeta) \\[4pt]
\raise 5pt \hbox{$\s c\,(s_2 -\frac{1}{2})$} \swarrow &
& \searrow \raise 5pt \hbox{$\s c(s_4-\frac{1}{2})$} \\[5pt]
c\, \Y^a(t_2) & & c\, \Y^a(t_3)\\[10pt]
{\s c\, (s_4-\frac{1}{2})} \searrow &&
\swarrow {\s c\, (s_2 -\frac{1}{2})} \\
& c^2\Y^a(t_4) & \\[6pt]
& \phantom{\s s_3} \big\downarrow{\s c'\, (s_3-\frac{1}{3})}&\\[6pt]
&c^2c'\, \Y^a(\aleph)
\end{array}
\quad\quad
\begin{array}{rcl}
& c^2c'\, \Y^x(\zeta^\sim) \\[4pt]
\raise 5pt \hbox{$\s c\, (s_2 -\frac{1}{2})$} \nearrow &
& \nwarrow \raise 5pt \hbox{$\s c\, (s_4-\frac{1}{2})$} \\[5pt]
cc'\, \Y^x(t_2^\sim) & & cc'\, \Y^x(t_3^\sim)\\[10pt]
{\s c(s_4-\frac{1}{2})} \nwarrow &&
\nearrow {\s c\, (s_2 -\frac{1}{2})} \\
& c'\, \Y^x(t_4^\sim) & \\[6pt]
& \phantom{\s s_3} \big\uparrow {\s c'\, (s_3-\frac{1}{3})} &\\[6pt]
& \Y^x(\aleph^\sim)
\end{array} \ .$$ The normalization constants $\frac{1}{\sqrt{1-1/\rho^2}}$ are $c= \frac{1}{\sqrt{1-1/4}}$ and $c'=\frac{1}{\sqrt{1-1/9}}$, because the diagonal distances involved are $\pm 2$ and $\pm 3$.
In conclusion, the sum $\sum_{t\in\Tab([3,3])} \pm \Y^a(t) \Y^x(t^\sim)$ is equal, up to the global factor $c^2c'$, to the same sum when using the orthonormal basis rather than the Young polynomials (which are, as we chose to define them, only an orthogonal basis).
Notice that the sum of products of Young polynomials can be written $$\begin{gathered}
\Y^a(\aleph)\Y^x(\aleph^\sim)\,
\Biggl( 1-(s_3^a+{\s\frac{1}{3}})(s_3^x-{\s\frac{1}{3}})
\frac{1}{1-\frac{1}{9}} \times \\
\left( 1-(s_2^a+{\s\frac{1}{2}})(s_2^x-{\s\frac{1}{2}})
\frac{1}{1-\frac{1}{4}} \right)
\left( 1-(s_4^a+{\s\frac{1}{2}})(s_4^x-{\s\frac{1}{2}})
\frac{1}{1-\frac{1}{4}} \right)
\Biggr)\\
= \quad \Y^a\left(\begin{smallmatrix} 4&5&6\\ 1&2&3 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 3&6\\ 2&5\\ 1&4\end{smallmatrix}\right)
- \Y^a\left(\begin{smallmatrix} 3&5&6\\ 1&2&4 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 4&6\\ 2&5\\ 1&3\end{smallmatrix}\right)
\times \\
\left( 1-(s_2^a+{\s\frac{1}{2}})(s_2^x-{\s\frac{1}{2}})
\frac{1}{1-\frac{1}{4}} \right)
\left( 1-(s_4^a+{\s\frac{1}{2}})(s_4^x-{\s\frac{1}{2}})
\frac{1}{1-\frac{1}{4}} \right) \\
= \cdots = \quad
\Y^a\left(\begin{smallmatrix} 4&5&6\\ 1&2&3 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 3&6\\ 2&5\\ 1&4\end{smallmatrix}\right)
- \Y^a\left(\begin{smallmatrix} 3&5&6\\ 1&2&4 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 4&6\\ 2&5\\ 1&3\end{smallmatrix}\right) \\
+ \Y^a\left(\begin{smallmatrix} 3&4&6\\ 1&2&5 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 5&6\\ 2&4\\ 1&3\end{smallmatrix}\right)
+\Y^a\left(\begin{smallmatrix} 2&5&6\\ 1&3&4 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 4&6\\ 3&5\\ 1&2\end{smallmatrix}\right)
-\Y^a\left(\begin{smallmatrix} 2&4&6\\ 1&3&5 \end{smallmatrix}\right)
\Y^x\left(\begin{smallmatrix} 5&6\\ 3&4\\ 1&2\end{smallmatrix}\right)\end{gathered}$$
In the preceding sections, we did not have recourse to normalization constants, but used the Young basis and checked the overall factor by computing a specialization of the Pfaffian.
*Acknowledgment. The author benefits from the ANR project BLAN06-2\_134516. This work was partly done in January-February 2005, during a Combinatorial Semester at the Mittag-Leffler Institute. I thank the Institute for its warm hospitality.*
[1]{}
C. Carré, A. Lascoux, B. Leclerc. *Turbo straightening*, Int J. of Algebra and Comp. (1992) 275–290.
J. Désarmenien. *An algorithm for the Rota straightening formula*, Discrete M. [**30**]{} (1980) 51–68.
J. Désarmenien, J. King, G.C. Rota. *Invariant Theory, Young bitableaux and combinatorics*, Adv. in M. [**27**]{} (1978) 63–92.
M. Ishikawa. *Minor summation formula and a proof of Stanley’s open problem*, arXiv:math.CO/0408204.
M. Ishikawa, S. Okada, H. Tagawa and J. Zeng. *Generalizations of Cauchy’s determinant and Schur’s Pfaffian*, arXiv:math.CO/0411280.
G. James, A. Kerber. *The representation theory of the symmetric group*, Encyclopedia of Maths, vol [**16**]{} Addison-Wesley (1981).
D. Knuth. *Overlapping Pfaffians*, Electron. J. Combin. 3 (2) (The Foata Festschrift) (1996), 151163.
C. Krattenthaler. *Advanced determinant calculus*, Sem. Lothar. Combin. 42 (1999), B42q.
C. Krattenthaler. *Advanced determinant calculus: A Complement*, Linear Alg.
L. Kronecker. *Die Subdeterminanten symmetrischer Systeme*, Sitzungsbericht Akad. Wissenschaften (1882) 824
D. Laksov, A. Lascoux and A. Thorup, *On Giambelli’s theorem on complete correlations*, Acta Math. [**162**]{} (1989), 143–199.
A. Lascoux. *Symmetric functions & Combinatorial operators on polynomials*, CBMS/AMS Lectures Notes [**99**]{}, (2003).
A. Lascoux, Shi He. *Généralisation de la Formule de Darboux-Christoffel pour les polynômes orthogonaux*, Comptes Rendus [**300**]{} (1985) 681–683.
B. Leclerc. *On Identities satisfied by minors of a matrix*, Adv. in Math. [**100**]{} (1993) 101–132.
J.-G. Luque and J.-Y. Thibon. *Pfaffian and Hafnian identities in shuffle algebras*, Adv. Appl. Math. 29 (2002), 620646.
. *Symmetric functions and [Hall]{} polynomials*. Oxford University Press, Oxford, 2nd edition (1995).
T. Muir. *On certain aggregates of determinant minors*, Proc. R.Soc. Edinburgh 23 (1900) 142-154.
S. Okada. *Enumeration of symmetry classes of alternating sign matrices and characters of classical groups*, arXiv:math./0408234, J. Alg. Comb. [**23**]{} (2006) 43-69.
S. Okada. Title: An elliptic generalization of Schur’s Pfaffian identity arXiv:math./0412038.
A. Okounkov, A. Vershik. *A new approach to representation theory of symmetric groups*, [Selecta Math.]{}, [**2**]{} (1996) 581–605.
H. Rosengren. *Pfaffians, determinants, and multivariable Christoffel-Darboux kernels*, arXiv:math./0606391.
D.E. Rutherford. *Substitutional Analysis*, Edinburgh, at the University Press, 1948.
J. R. Stembridge. *Non-intersecting paths, Pfaffians and plane partitions*, Adv. Math. 83 (1990), 96–131.
Yu. G. Stroganov. *Izergin-Korepin determinant reloaded* arXiv:math. /0409072
T. Sundquist. *Two variable Pfaffian identities and symmetric functions*, J. Algebraic Combin. 5 (1996), 135–148.
A. Young. , University of Toronto Press (1977).
[^1]: Young [@Young Theorem IV, p.591] uses the picturesque terminology “has the same substitutional qualities”, to state that the space $e_{t,t}\, \cH$ is isomorphic to the space generated by some products of Vandermonde determinants.
[^2]: and, of course, all the relations obtained by permuting the rows and the columns of the original matrix, in such a way as to obtain another symmetric matrix. We could reprove directly Kronecker’s relations by introducing extra variables $a_1,a_2,\ldots$ and evaluating the Pfaffian of the antisymmetric matrix $\bigl[(a_i-a_j)g[i,j] \bigr]_{i,j=1,\ldots,n}$, as will become clear later.
[^3]: We do not need to know Kronecker’s relations. Having only a component of type $[m,m]$ for $\mfS_n^a$ forces a component of type $[2^m]$ for $\mfS_n^g$.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'In this article we study reaction-diffusion systems containing self and cross-diffusion using a free energy approach. This framework naturally leads to the formulation of an energy law, and to a numerical method respecting a discrete version of the latter (a linear second order finite-element method with adaptive mesh refinement). Applying this numerical scheme to a Gray-Scott system augmented with self and cross-diffusion terms, we verify numerically the energy law, and unveil patterns distinct from those obtained with linear diffusion.'
address: 'MathNeuro Team, Inria Sophia Antipolis Méditerranée, Sophia Antipolis Cedex, France'
author:
- Benjamin Aymard
bibliography:
- 'RDarxiv.bib'
title: 'On simulation of reaction-diffusion systems containing self and cross-diffusion'
---
reaction-diffusion ,self-diffusion ,cross-diffusion ,energy law ,finite-element method ,Gray-Scott ,Turing pattern. 00-01,99-00
NOTE FOR REVIEWERS: if the manuscript is accepted, the source of the code used in simulations will be available online.
Introduction {#introduction .unnumbered}
============
Since the seminal work of Alan Turing in 1952 [@turing], it is known that the introduction of linear diffusion in dynamical systems may lead, under certain conditions, to the formation of instabilities known as Turing patterns. Turing’s predictions have been confirmed numerous times, both in experiments, like the CIMA reaction [@CIMA], and in modelling: genetics [@fisher], chemotaxis [@KS], chemical reactions [@BZ], patterns formation in animal and plant biology [@GM], and in neuroscience [@patternFN].
Reaction-diffusion systems can be generalized by including nonlinear diffusion terms: self-diffusion, corresponding to a diffusion proportional to the density, and cross-diffusion, a tendency to diffuse along the gradient of another co-existing density. Such generalization is particularly relevant in biology, for instance in predator-prey systems (notably the so-called SKT system [@SKT]) or in epidemiology models. These systems have been widely studied [@mimura], [@martinez], [@boris], [@trescases], and pattern formation has been reported [@gambino], [@turingSKT], [@turingSIR], [@turingSelf].
In this work we study reaction-diffusion systems including self and cross-diffusion from an energetic point of view, using an approach inspired by diffuse-interface theory. The most prominent example is the Cahn-Hilliard system [@CH] proposed in 1958: one equation describes density evolution, with a diffusion along the gradient of a so-called chemical potential, and one equation defines the chemical potential as a functional derivative of a free energy. This formulation gives a natural framework for writing a free energy law, and in designing a finite-element method based on a mixed formulation [@JCP2013]. Using these tools, we perform numerical simulations on a generalization of a Gray Scott system [@GS], known as a good example for generating a wide variety of patterns [@pearson]. We study how Turing patterns differ when they are generated with different combinations of self and cross-diffusion, compared to linear diffusion.
In the first section, we present the general reaction-diffusion model that we study under its free energy formulation (the name can be misleading, as the model does not necessarily derive from a free energy, but can still be written in a similar form). We establish the corresponding energy law verified by solutions of the system. The second section is devoted to the design of the numerical method approaching solutions to this model. We demonstrate a discrete version of a mass and energy law. The last section presents numerical simulations of pattern formation in presence of different combinations of linear, self and cross-diffusion.
Energy law
==========
In this section we introduce the general reaction-diffusion model we will study throughout the article, under the form of a free energy-like system, and we derive an energy law from it. This system is classified as a nonlinear parabolic partial differential equation system. Interested readers may find an introduction to the topic of parabolic partial differential equations for biology, and in particular reaction-diffusion equations, in [@perthame].
Model
-----
We consider the dynamics of M interacting species in a bounded domain $\Omega \subset \mathbb{R}^d$ with $d$ the spatial dimension. Let us note $\phi_i(\textbf{X},t)$ the density of species $i$ at position $\textbf{X}$ and time $t$. The evolution of the system is described by:
$$\begin{aligned}
\forall i \in \{1,...,M\}, \notag \\
\partial_t \phi_i - \nabla^2 \mu_i = R_i,\notag \\
R_i = R_i(\phi_1,...,\phi_M),\notag \\
\mu_i = [d_{i} + d_{ii} \phi_i^{\alpha_i} + \sum_{j \not= i}d_{ij} \phi_j^{\beta_{ij}}]\phi_i,
\label{PDE}\end{aligned}$$
with $R_i$ the reaction terms, and $\mu_i$ may be called chemical potentials in the framework of thermodynamics, if they derive from a free energy. This system involves several levels of diffusion: a linear diffusion for each species, tuned by parameters $d_{i} \geq 0$, describing the tendency to fill up the space, a self-diffusion for each species, tuned by parameters $d_{ii} \geq 0$ and positive $\alpha_i \geq 0$, describing the tendency of each species to avoid crowding, and cross-diffusion terms between species, tuned by parameters $d_{ij} \geq 0$ and powers $\beta_{ij} \geq 0$, describing the tendency of each species to avoid another. In order to close mathematically this problem, we also have to add initial conditions, and boundary conditions, under the form of confinement of populations in the domain $\Omega$:
$$\begin{aligned}
\forall i \in \{1,...,M\}, \notag\\
\phi_i(\textbf{X},t=0) = \phi_i^0(\textbf{X}),\notag \\
\nabla \mu_i \cdot n = 0 \mbox{ on }\partial \Omega,
\label{BC}\end{aligned}$$
with $n$ the outward unit normal vector.
Multiplying by test functions in $H^1(\Omega)$, integrating over the physical domain $\Omega$, and using integration by parts, we get the weak form of the system:
$$\begin{aligned}
\forall i \in \{1,...,M\}, \forall \theta_i, \nu_i \in H^1(\Omega),\notag \\
(\partial_t \phi_i,\theta_i) + (\nabla\mu_i , \nabla \theta_i)
- (R_i,\theta_i) = 0,\notag\\
(\mu_i,\nu_i) = ([d_{i} + d_{ii} \phi_i^{\alpha_i} + \sum_{j\not=i}d_{ij} \phi_j^{\beta_{ij}}]\phi_i,\nu_i).
\label{weakForm}\end{aligned}$$
where $(f,g) = \int_{\Omega}fg$ denotes the scalar product. The mass of each species $i$ is defined as the integral of its density over the physical domain: $$M_i(t) = \int_{\Omega}\phi_i(\textbf{X},t).
\label{M}$$
The mass variation of the solutions of \[weakForm\] verify the mass law: $$\frac{d}{dt} M_i(t) = \int_{\Omega}R_i.
\label{Mlaw}$$
Consider $\theta_i = 1$ in weak formulation \[weakForm\]. Let us remark that in the case $R_i=0$, the system is purely diffusive, and therefore is mass conservative.
Energy law
----------
Decomposing potentials $\mu_i$ as a sum of a gradient and a non gradient part we may define a function $F$ from the equality:
$$\mu_i = (\nabla F)_i + g_i,$$
and a function $G$ from an integral:
$$G_i(\phi_1(t),...,\phi_M(t)) = \int_0^t g_i \partial_t \phi_i.$$
Solutions of system \[weakForm\], if they exist, admit an energy law of the form:
$$\frac{d}{dt}
\left(E(\phi_1(t),...,\phi_M(t))\right)
=
- \sum_{i=1}^M \| \nabla \mu_i \|^2_{L^2(\Omega)}
+ \sum_{i=1}^M (R_i,\mu_i),
\label{Elaw}$$
with:
$$E(\phi_1(t),...,\phi_M(t))
=
\int_{\Omega}
F(\phi_1(t),...,\phi_M(t))
+ \sum_{i=1}^M G_i(\phi_1(t),...,\phi_M(t)),
\label{E}$$
Let us consider $\theta_i =\mu_i, \nu_i = \frac{\partial \phi_i}{\partial t}$ in weak formulation \[weakForm\]: $$\begin{aligned}
\forall i \in \{1,...,M\},\\
(\frac{\partial \phi_i}{\partial t},\mu_i) + (\nabla\mu_i , \nabla \mu_i) - (R_i),\mu_i) = 0,\\
(\mu_i,\frac{\partial \phi_i}{\partial t}) = ([d_{i} + d_{ii} \phi_i^{\alpha_i} + \sum_{j\not=i}d_{ij} \phi_j^{\beta_{ij}}]\phi_i,\frac{\partial \phi_i}{\partial t}).\end{aligned}$$
By summing all equations and cancelling cross terms we get:
$$\sum_{i=1}^M \left(\mu_i,\frac{\partial \phi_i}{\partial t}\right)
=
- \sum_{i=1}^M (\nabla \mu_i , \nabla \mu_i)
+ \sum_{i=1}^M(R_i,\mu_i).$$
We conclude using the decomposition of $\mu_i$: $$\sum_{i=1}^M\left(\mu_i,\frac{\partial \phi_i}{\partial t}\right)
= \sum_{i=1}^M \left((\nabla F)_i + g_i,\frac{\partial \phi_i}{\partial t}\right)
= \frac{d}{dt}\left( F + \sum_{i=1}^M G_i,1 \right).$$
In the purely diffusive case, the free energy of the system naturally decreases until reaching a state with null gradient of potential $\mu_i$.
In the symmetric case $d_{ij}=d_{ji}$ and $\beta_{ij} = \beta_{ji} = 2$, the energy writes: $$E(\phi_1,...,\phi_M) =
\int_{\Omega}\
\left(\sum_{i=1}^M
d_{ii}\frac{\phi_i^{\alpha_i+1}}{\alpha_i+1}
+ d_i\frac{\phi_i^2}{2}
+ \sum_{j>i} d_{ij}\frac{\phi_i^2 \phi_j^2}{2}
\right).
\label{symLaw}$$
For two populations, in the triangular case $d_{12} > 0$ and $d_{21} = 0$, the energy writes: $$E(\phi_1,...,\phi_M) =
\int_{\Omega}\left(
d_{11}\frac{\phi_1^{\alpha_1+1}}{\alpha_1+1}
+ d_{22}\frac{\phi_2^{\alpha_2+1}}{\alpha_2+1}
+ d_1\frac{\phi_1^2}{2}
+ d_2\frac{\phi_2^2}{2}
+ d_{12}G_1
\right),
\label{unSymLaw}$$ with: $$G_1 = \int_0^t \phi_1 \phi_2 \partial_t \phi_1.$$
Numerical Method
================
In this section we derive a numerical method in order to simulate the system introduced in the first section. The strategy follows the methodology developed in [@JCP2013] and [@JCP2019]. It relies on a finite-element method and an adaptive mesh refinement. After introducing the method, we show that numerical solutions respect a discrete version of the energy law. Interested readers may find an introduction to finite-elements and meshing strategy in [@Allaire] and [@Pironneau]. Numerical simulations were done using the free open source software FreeFem++ [@freefem].
Scheme
------
Let us denote by $\Delta t$ the time step, and $\phi_i^n$ the density of species $i$ at time $t^n = n \Delta t$. We propose a numerical method based on a splitting strategy, solving sequentially the reaction part of the equation, then the diffusion part. Using a Strang splitting [@strang] ensures a second order accuracy in time, as long as the two components are second order in time.
For the reaction part, we use a classical second order method, namely the Heun method. Let us denote by $A_{\Delta t}$ the operator defined by:
$$\begin{aligned}
\forall i \in \{1,...,M\},\notag \\
\phi_i^{*} = \phi_i^n + \Delta t R_i(\phi_1^n,...,\phi_M^n),\notag \\
A_{\Delta t}(\phi_i^{n}) = \phi_i^n + \frac{\Delta t}{2}(R_i(\phi_1^n,...,\phi_M^n) + R_i(\phi_1^*,...,\phi_M^*).
\label{operatorA}\end{aligned}$$
Let us denote by $B_{\Delta t}$ the operator approximating the diffusive part, defined in a weak sense by: $$\begin{aligned}
\forall i \in \{1,...,M\}, \forall \theta_i, \nu_i \in H^1(\Omega),\notag\\
\left( \displaystyle\frac{B_{\Delta t}(\phi_i^{n}) - \phi_i^n}{\Delta t},\theta_i \right) + (\nabla\mu_i^{n+1/2} , \nabla \theta_i) = 0,\notag\\
(\mu_i^{n+1/2},\nu_i) = (\tilde{F}_i(\phi_1^n,...,\phi_M^n,\phi_1^{n+1},...,\phi_M^{n+1}), \nu_i).
\label{operatorB}\end{aligned}$$
with $\tilde{F}_i$ defined by:
$$\tilde{F}_i(\phi_1^n,...,\phi_M^n,\phi_1^{n+1},...,\phi_M^{n+1}) =
\mu_i(\phi_1^n,...,\phi_M^n)
+ \frac{1}{2}(\phi_i^{n+1} - \phi_i^n)\frac{\partial \mu_i}{\partial \phi_i}(\phi_1^n,...,\phi_M^n).$$
Using operators $A_{\Delta t}$ and $B_{\Delta t}$ the Strang splitting writes:
$$\phi^{n+1} = A_{\Delta t/2} B_{\Delta t} A_{\Delta t/2} (\phi^n).
\label{scheme}$$
If reaction terms write $R_i = \phi_i S_i(\phi_1,...,\phi_M) + T_i$ with $T_i \geq 0$, and if the density is positive at time $n$, the numerical operator $A_{\Delta t}$ defined by \[operatorA\] admits a unique solution. Moreover, the solution remains positive for $\Delta t$ small enough.
Operator $A_{\Delta t}$ is explicit, therefore the solution is unique. Let us re-write operator $A_{\Delta t}$ as: $$A_{\Delta t}(\phi_i^n) = \phi_i^n (1 + \Delta t S_i) + T_i.$$ If $S_i$ is non-negative (creation term), then $A_{\Delta t}(\phi_i^n)$ is also non-negative. If $S_i$ is negative, then choosing $\Delta t = \min \limits_{i \in \{1,...,M \}} \frac{1}{\| S_i \|}$ ensures that $\phi_i^n (1 + \Delta t S_i)$ is non-negative, and therefore $A_{\Delta t}(\phi_i^n)$ is also non-negative.
If diffusion coefficients $d_{ij}$ are all non-negatives, if the density is non-negative at time $n$, then for any $\Delta t$ the numerical operator $B_{\Delta t}$ defined by \[operatorB\] admits at most one solution.
Let us consider two solutions to system \[weakForm\], that we denote $(\phi_1,...,\phi_M,\mu_1,...,\mu_M)$ and $(\tilde{\phi_1},...,\tilde{\phi_M},\tilde{\mu_1},...,\tilde{\mu_M})$. Let us also note $\Delta_i = \phi_i - \tilde{\phi_i}$ and $\delta_i = \mu_i - \tilde{\mu_i}$. In the numerical scheme \[scheme\] we consider $\theta_i = \delta_i$ and $\nu_i = \frac{1}{\Delta t}\Delta_i$, leading to: $$\| \nabla \delta_i \|^2 + \left( \frac{1}{2} \frac{\partial \mu_i}{\partial \phi_i }\Delta_i,\Delta_i \right) = 0.$$ Using the definition of $\mu_i$, and hypothesis of positivity of the solution at time $n$, we see that the second term of the sum is positive. Therefore, the two terms of the sum have to be null. In particular, the gradient of $\delta_i$ is null. Plugging this result in the first equation of the scheme shows that $\Delta_i$ is also null. Finally, using this nullity in the second equation shows that $\delta_i$ is also null. This shows that if a solution exists, it is necessary unique.
Conservation of positivity for operator $B_{\Delta t}$ has been observed numerically in all the simulations done for this article. However, the proof remains open.
Let us define the useful notations: $\forall i \in \{1,...,M\},$ $$\begin{aligned}
R_i^1 = R_i(\phi_i^n),\\
R_i^2 = R_i(\phi_i^n + \Delta t R_1),\\
R_i^3 = R_i(B_{\Delta t} (A_{\Delta t/2} (\phi^n),\\
R_i^4 = R_i(B_{\Delta t} (A_{\Delta t/2} (\phi^n) + \Delta t R_3).\\\end{aligned}$$
The numerical scheme verifies a discrete mass law similar to \[Mlaw\] under the form: $$\forall n \geq 1, \int_{\Omega} \phi_i^{n+1}
= \int_{\Omega} \phi_i^n
+ \frac{\Delta t}{4}(R_i^1 + R_i^2 + R_i^3 + R_i^4).$$ In particular, in the purely diffusive case $R_i=0$, the mass is conserved.
Consider $\theta_i = 1$ to get: $$\begin{aligned}
\int_{\Omega}\phi_i^{n+1}
& = &\int_{\Omega}\phi_i^{n+1/2} + \frac{\Delta t}{4}(R_i^1 + R_i^2) \\
& = &\int_{\Omega}\phi_i^{n} + \frac{\Delta t}{4}(R_i^1 + R_i^2 + R_i^3 + R_i^4).\end{aligned}$$
The numerical operator $B_{\Delta t}$, definde by \[operatorB\], respects a discrete energy law of the form: $$\frac{E(\phi_1^{n+1},...,\phi_M^{n+1})-E(\phi_1^{n},...,\phi_M^{n})}{\Delta t}
= - \sum_i \| \nabla \mu_i^{n+1/2}\|^2 + \sum_i \epsilon_i,$$ with $\epsilon_i$ error terms defined by: $$\epsilon_i = \int_{\Omega}
\frac{1}{\Delta t}(\phi_i^{n+1}-\phi_i^n)F_i(\phi_1^n,\phi_2^n,\phi_1^{n+1},\phi_2^{n+1})
-\frac{1}{\Delta t}((F + G_i)(\phi^{n+1})-(F + G_i)(\phi^n)).$$ The error terms are second order accurate: $$\epsilon_i = O (\Delta t^2).$$
Let us consider $\theta_i =\mu_i^{n+1/2}, \nu_i = \frac{\phi_i^{n+1} - \phi_i^n}{\Delta t}$ in the numerical scheme. Using a Taylor expansion we get: $$(F+G_i)(\phi^{n+1})
\approx (F+G_i)(\phi^n)
+ (\phi^{n+1}-\phi^n)f_i(\phi_1^n,...,\phi_M^n)$$ $$+\frac{1}{2}(\phi_i^{n+1}-\phi_i^n)^2\frac{\partial f_i}{\partial \phi_i}(\phi_1^n,...,\phi_M^n)
+\frac{1}{6}(\phi_i^{n+1}-\phi_i^n)^3\frac{\partial^2 f_i}{\partial \phi_i^2}(\phi_1^n,...,\phi_M^n).$$ Dividing this expression by $\Delta t$ and using the definition of $F_i$, we get: $$\epsilon_i
= \frac{1}{6}\int_{\Omega}\left(\frac{(\phi_i^{n+1}-\phi_i^n)}{\Delta t}\right)^3\frac{\partial^2 f_i}{\partial \phi_i^2}\Delta t^2.$$ This expression is controlled as: $$| \epsilon_i |
\leq \Delta t^2 \| \partial_t \phi_i\|_{C([0,T],L^3(\Omega)}
\left\| \frac{\partial^2 f_i}{\partial \phi_i^2} \right\|_{C([0,T],L^{\infty}(\Omega))}.$$
The numerical scheme \[scheme\] respects a discrete energy law analogous to (\[Elaw\]) of the form: $$\frac{E^{n+1} - E^n}{\Delta t}
=
- \sum_{i=1}^M \| \nabla \mu_i^{n+1/2}\|^2
+ \left(\frac{R_i^1 + R_i^2 + R_i^3 + R_i^4}{4},\mu_i\right)
+ \xi
,
\label{discreteElaw}$$ with: $\xi = \frac{E^{n+1} - E^n}{\Delta t} -
\sum_{i=1}^M \left( \tilde{F}_i \left(A_{\Delta t}(\phi_i^n),B_{\Delta t} (A_{\Delta t}(\phi_i^n))\right),
\frac{\phi_i^{n+1} - \phi_i^n}{\Delta t} \right)$. Error term $\xi$ tends to $0$ as $\Delta t$ tends to $0$.
Let us write numerical scheme \[scheme\] in an alternative form: $$\begin{aligned}
\forall i \in \{1,...,M\}, \\
(\phi_i^{n+1},\theta_i)
&= (\phi_i^n,\theta_i) +
\Delta t \left(\frac{R_i^1 + R_i^2 + R_i^3 + R_i^4}{4},\theta_i\right)
- \Delta t(\nabla \mu_i^{n+1/2},\nabla \theta_i),\\
(\mu_i^{n+1/2},\nu_i) &= \left( \tilde{F}_i \left(A_{\Delta t}(\phi_i^n),B_{\Delta t} (A_{\Delta t}(\phi_i^n))\right),\nu_i\right),\\\end{aligned}$$ Then, considering $\theta_i = \mu_i^{n+1/2}$, $\nu_i = \frac{\phi_i^{n+1} - \phi_i^n}{\Delta t}$ and summing all the equations we get: $$\sum_{i=1}^M\left(\tilde{F}_i,\frac{\phi_i^{n+1} - \phi_i^n}{\Delta t}\right)
=
\sum_{i=1}^M \left(\frac{R_i^1 + R_i^2 + R_i^3 + R_i^4}{4},\mu_i\right)
- \sum_{i=1}^M \|\nabla \mu_i^{n+1/2} \|^2$$ Using the definition of $\xi$, we obtain formula \[discreteElaw\]. Let us remark that, for all $i \in \{1,...,M\}$ we get: $$\begin{aligned}
\lim\limits_{\Delta t \rightarrow 0} &
\sum_{i=1}^M\left( \tilde{F}_i \left(A_{\Delta t}(\phi_i^n),B_{\Delta t} (A_{\Delta t}(\phi_i^n))\right),\frac{\phi_i^{n+1} - \phi_i^n}{\Delta t}\right)\\
& = \sum_{i=1}^M\left(\mu_i(t^n),\frac{d\phi_i}{dt}(t^n)\right)
= \frac{dE}{dt}(t^n),\end{aligned}$$ Therefore we can conclude, using the definition of $\xi$, that: $\lim \limits_{\Delta t \rightarrow 0} \xi = 0$.
In order to compute the free energy in non symmetric cases, when there are terms $G_i$ in formula \[Elaw\], we evaluate them with the second order approximation: $$G_1^n \approx \frac{\Delta t}{2}
\left(
\phi_1^{n+1} \phi_2^{n+1}\frac{(\phi_1^{n+1} - \phi_1^n)}{\Delta t}
+ \phi_1^n \phi_2^n \frac{(\phi_1^n - \phi_1^n)}{\Delta t}
\right).$$
In order to achieve space discretization, we define a mesh $T_h$ of the domain $\Omega$ composed by $N_t$ triangles: $$\Omega \approx \bigcup_{t=1}^{N_t} T_h.$$
From this space discretization, we define an approximation of the functional spaces. Let us note: $$V_h = \{v \in C(\Omega), v_{|Th} \in T\}.$$
Replacing $H^1(\Omega)$ by discrete functional space $V_h$ in weak form \[weakForm\], the variational formulation is reduced to a linear system that we solve at each time step. Depending on the context, for the sake of computational speed, we might choose $T=P1$ and adapt the mesh simultaneously according to both densities $\phi_i$ and potentials $\mu_i$, or we might keep the mesh constant in time with $T=P2$.
Numerical test
--------------
As a first test, we consider a purely diffusive situation, with no reactive term. It permits to check qualitatively the numerical operator $B_{\Delta t}$ \[operatorB\], and energy evolution \[Elaw\]. We consider a square domain $\Omega = [0,1]^2$, and we start with two densities localized as disks: $$\begin{aligned}
\phi_1(t=0) = \mathbbm{1}_{D_1((0.4,0.5),0.2)} \notag,\\
\phi_2(t=0) = \mathbbm{1}_{D_2((0.6,0.5),0.2)}\label{IC11}.\end{aligned}$$ Diffusion parameters are set to: $d_i = 1e^{-5}, d_{ii} = 0, d_{ij}=0$ for simple diffusion, $d_i = 1e^{-5}, d_{ii} = 2e^{-5}, d_{ij}=0$ for self-diffusion, $d_i = 1e^{-5}, d_{ii} = 0, d_{ij}=2e^{-5}$ for symmetric cross-diffusion, $d_1 = d_2 = 1e^{-5}, d_{11} = 0, d_{22} = 2e^{-5}, d_{12}=2e^{-5}, d_{21} = 0$ for asymmetric cross-diffusion. Powers parameters are set to: $\alpha_1 = \alpha_2 = 2$, $\beta_{12} = \beta_{21} = 1$. Numerical parameters are: $\Delta t = 0.2, h_{\min} = 0.001, h_{\max} = 0.07$.
Results are displayed in Figure \[figCrossDiffusion\]. The first vector field, representing linear diffusion, is used as a witness test. By comparison, we see the effect of self-diffusion, with a clear difference of intensity visible in the inner density circle. We also notice that asymmetric cross-diffusion creates an asymmetric density distribution, as opposed to symmetric cross-diffusion, preserving the axial symmetry of the initial condition. As expected, according to \[Elaw\], the free energy is decreasing, in all cases, with a similar shape. It has been computed with formula \[symLaw\] for cases 1, 2 and 3, and with formula \[unSymLaw\] for case 3. Mass of densities $\phi_1$ and $\phi_2$ is conserved up to machine precision.
![Comparison between linear, self and cross-diffusion (symmetric and asymmetric). Simulation of system \[PDE\] with no reaction term ($\forall i, R_i = 0$) and boundary conditions \[BC\]. Top: initial condition \[IC11\] for $\phi_1$ and time evolution of free energy \[Elaw\]. Middle: vector fields $\nabla \mu_1$ for linear diffusion and self-diffusion. Down: vector fields $\nabla \mu_1$ for symmetric and asymmetric cross-diffusion.[]{data-label="figCrossDiffusion"}](init.png "fig:"){width="0.45\linewidth"} ![Comparison between linear, self and cross-diffusion (symmetric and asymmetric). Simulation of system \[PDE\] with no reaction term ($\forall i, R_i = 0$) and boundary conditions \[BC\]. Top: initial condition \[IC11\] for $\phi_1$ and time evolution of free energy \[Elaw\]. Middle: vector fields $\nabla \mu_1$ for linear diffusion and self-diffusion. Down: vector fields $\nabla \mu_1$ for symmetric and asymmetric cross-diffusion.[]{data-label="figCrossDiffusion"}](Ecurve "fig:"){width="0.45\linewidth"} ![Comparison between linear, self and cross-diffusion (symmetric and asymmetric). Simulation of system \[PDE\] with no reaction term ($\forall i, R_i = 0$) and boundary conditions \[BC\]. Top: initial condition \[IC11\] for $\phi_1$ and time evolution of free energy \[Elaw\]. Middle: vector fields $\nabla \mu_1$ for linear diffusion and self-diffusion. Down: vector fields $\nabla \mu_1$ for symmetric and asymmetric cross-diffusion.[]{data-label="figCrossDiffusion"}](vectorDiffusion.png "fig:"){width="0.45\linewidth"} ![Comparison between linear, self and cross-diffusion (symmetric and asymmetric). Simulation of system \[PDE\] with no reaction term ($\forall i, R_i = 0$) and boundary conditions \[BC\]. Top: initial condition \[IC11\] for $\phi_1$ and time evolution of free energy \[Elaw\]. Middle: vector fields $\nabla \mu_1$ for linear diffusion and self-diffusion. Down: vector fields $\nabla \mu_1$ for symmetric and asymmetric cross-diffusion.[]{data-label="figCrossDiffusion"}](vectorSelfDiffusion.png "fig:"){width="0.45\linewidth"} ![Comparison between linear, self and cross-diffusion (symmetric and asymmetric). Simulation of system \[PDE\] with no reaction term ($\forall i, R_i = 0$) and boundary conditions \[BC\]. Top: initial condition \[IC11\] for $\phi_1$ and time evolution of free energy \[Elaw\]. Middle: vector fields $\nabla \mu_1$ for linear diffusion and self-diffusion. Down: vector fields $\nabla \mu_1$ for symmetric and asymmetric cross-diffusion.[]{data-label="figCrossDiffusion"}](vectorCrossDiffusion.png "fig:"){width="0.45\linewidth"} ![Comparison between linear, self and cross-diffusion (symmetric and asymmetric). Simulation of system \[PDE\] with no reaction term ($\forall i, R_i = 0$) and boundary conditions \[BC\]. Top: initial condition \[IC11\] for $\phi_1$ and time evolution of free energy \[Elaw\]. Middle: vector fields $\nabla \mu_1$ for linear diffusion and self-diffusion. Down: vector fields $\nabla \mu_1$ for symmetric and asymmetric cross-diffusion.[]{data-label="figCrossDiffusion"}](vectorUnsymCrossDiffusion.png "fig:"){width="0.45\linewidth"}
Pattern formation
=================
Modified Gray-Scott
-------------------
The Gray-Scott system [@GS] is a generic reaction-diffusion system known to contain rich dynamics, and able to generate a wide variety of patterns (see [@pearson]). A famous example of application is the chemical reaction of three components in a stirred tank: a reactant U, a catalyst V, and an inert product P verifying $U+V+P=1$. From the chemistry point of view, the dynamics writes: $$\begin{aligned}
U + 2V \rightarrow 3V, \\
V \rightarrow P.\end{aligned}$$ Let us denote by $\phi_1$ (resp. $\phi_2$) the concentration of $U$ (resp. $V$). The classical system writes: $$\begin{aligned}
\partial_t \phi_1 - \nabla^2 \phi_1 &= -\phi_1\phi_2^2 + F(1-\phi_1), \notag\\
\partial_t \phi_2 - \nabla^2 \phi_2 &= \phi_1\phi_2^2 - (F+k)\phi_2,\label{GrayScott}\end{aligned}$$ The term $F(1-\phi_1)$ is the replenishment term for reactant $U$, the term $\phi_1\phi_2^2$ is the reaction term, and the term $- (F+k)\phi_2$ is a vanishing term.
In this paragraph, we consider a modification of this system, by adding self and cross-diffusion terms: $$\begin{aligned}
\partial_t \phi_1 - \nabla^2 \mu_1 &= -\phi_1\phi_2^2 + F(1-\phi_1), \notag \\
\partial_t \phi_2 - \nabla^2 \mu_2 &= \phi_1\phi_2^2 - (F+k)\phi_2,\notag\\
\mu_1 = (d_{1} &+ d_{11} \phi_i^{\alpha_1} + d_{12} \phi_2^{\beta_{12}})\phi_1,\notag\\
\mu_2 = (d_{2} &+ d_{22} \phi_2^{\alpha_2} + d_{21} \phi_1^{\beta_{12}})\phi_2. \label{mGrayScott}\end{aligned}$$ Note that this system has the form of \[PDE\].
Simulations
-----------
We consider a 2D case, in a square domain $\Omega = [0,1]^2$. Following [@pearson], as initial conditions we consider a square, denoted by S, centered at $(0.5,0.5)$ with size $0.05$, and we define: $$\begin{aligned}
\phi_1(t=0) = (0.5 + 0.05 U)\mathbbm{1}_{S} + \mathbbm{1}_{\Omega \setminus S},\notag\\
\phi_2(t=0) = (0.25 + 0.05 V)\mathbbm{1}_{S},\label{ICGS}\end{aligned}$$ with $U,V$ random numbers uniformly distributed on $[0,1]$. Numerical parameters are set to: $\Delta t = 1$, $h_{\min} = 0.001$, $h_{\max} = 0.07$. Model parameters are set to: $F=0.037, k=0.06$, $\alpha_1 = \alpha_2 = 2$, $\beta_{12} = \beta_{21} = 1$. By default, diffusion parameters not mentioned in case description are set to $0$.
Results are displayed in Figure \[figRD\] and Figure \[figRDenergy\]. First case ($d_1 = 2e^{-5}, d_2 = 1e^{-5}$) is a witness test, and shows the classical Gray-Scott pattern for linear diffusion. Second case ($d_{11} = d_{22} = 1e^{-5}$) and fourth case ($d_2 = 1e^{-5}, d_{11}=1e^{-5}$) generate a ring-shaped travelling wave, that differ by their contours: fourth case generates a circular contour, whereas second case generates an almost circular ring containing inflexion points. Third case ($d_{11} = d_{22} = 1e^{-5}, d_{12}=1e^{-5}$) generates a complex pattern, made of concentric circular stripes, and containing non circular spots. Fifth case ($ d_1 = 2e^{-5}, d_{22} = 1e^{-5}$) and sixth case ($ d_1 = 3e^{-5}, d_2 = 1e^{-5}, d_{12} = 1e^{-5}$) generate a combination of spots and stripes, mainly in alternance. Classically, in system \[GrayScott\], different patterns such as spots, stripes and waves are obtained by varying the reactive term. In model \[mGrayScott\], we observe different kinds of patterns, even for the same reactive term. Interestingly, in all cases, the free energy \[E\] decreases with respect to time.
![Pattern formation driven by combinations of linear, self and cross-diffusion. Simulations of modified Gray-Scott reaction-diffusion system \[mGrayScott\] with boundary conditions \[BC\] starting from initial conditions \[ICGS\]. Plots of density $\phi_1$ (except on case 5 where we represent $\phi_2$). The same reaction term triggers different patterns depending on the combinations of nonlinear diffusion coefficients.[]{data-label="figRD"}](GStest1.png "fig:"){width="0.49\linewidth"} ![Pattern formation driven by combinations of linear, self and cross-diffusion. Simulations of modified Gray-Scott reaction-diffusion system \[mGrayScott\] with boundary conditions \[BC\] starting from initial conditions \[ICGS\]. Plots of density $\phi_1$ (except on case 5 where we represent $\phi_2$). The same reaction term triggers different patterns depending on the combinations of nonlinear diffusion coefficients.[]{data-label="figRD"}](GStest2.png "fig:"){width="0.49\linewidth"}\
![Pattern formation driven by combinations of linear, self and cross-diffusion. Simulations of modified Gray-Scott reaction-diffusion system \[mGrayScott\] with boundary conditions \[BC\] starting from initial conditions \[ICGS\]. Plots of density $\phi_1$ (except on case 5 where we represent $\phi_2$). The same reaction term triggers different patterns depending on the combinations of nonlinear diffusion coefficients.[]{data-label="figRD"}](GStest3.png "fig:"){width="0.49\linewidth"} ![Pattern formation driven by combinations of linear, self and cross-diffusion. Simulations of modified Gray-Scott reaction-diffusion system \[mGrayScott\] with boundary conditions \[BC\] starting from initial conditions \[ICGS\]. Plots of density $\phi_1$ (except on case 5 where we represent $\phi_2$). The same reaction term triggers different patterns depending on the combinations of nonlinear diffusion coefficients.[]{data-label="figRD"}](GStest4.png "fig:"){width="0.49\linewidth"}\
![Pattern formation driven by combinations of linear, self and cross-diffusion. Simulations of modified Gray-Scott reaction-diffusion system \[mGrayScott\] with boundary conditions \[BC\] starting from initial conditions \[ICGS\]. Plots of density $\phi_1$ (except on case 5 where we represent $\phi_2$). The same reaction term triggers different patterns depending on the combinations of nonlinear diffusion coefficients.[]{data-label="figRD"}](GStest5.png "fig:"){width="0.49\linewidth"} ![Pattern formation driven by combinations of linear, self and cross-diffusion. Simulations of modified Gray-Scott reaction-diffusion system \[mGrayScott\] with boundary conditions \[BC\] starting from initial conditions \[ICGS\]. Plots of density $\phi_1$ (except on case 5 where we represent $\phi_2$). The same reaction term triggers different patterns depending on the combinations of nonlinear diffusion coefficients.[]{data-label="figRD"}](GStest6.png "fig:"){width="0.49\linewidth"}
![Time evolution of free energy \[E\] corresponding to Figure \[figRD\] using the same order. In all cases, free energy is a decreasing function of time.[]{data-label="figRDenergy"}](Etest1.png "fig:"){width="0.45\linewidth"} ![Time evolution of free energy \[E\] corresponding to Figure \[figRD\] using the same order. In all cases, free energy is a decreasing function of time.[]{data-label="figRDenergy"}](Etest2.png "fig:"){width="0.45\linewidth"} ![Time evolution of free energy \[E\] corresponding to Figure \[figRD\] using the same order. In all cases, free energy is a decreasing function of time.[]{data-label="figRDenergy"}](Etest3.png "fig:"){width="0.45\linewidth"} ![Time evolution of free energy \[E\] corresponding to Figure \[figRD\] using the same order. In all cases, free energy is a decreasing function of time.[]{data-label="figRDenergy"}](Etest4.png "fig:"){width="0.45\linewidth"} ![Time evolution of free energy \[E\] corresponding to Figure \[figRD\] using the same order. In all cases, free energy is a decreasing function of time.[]{data-label="figRDenergy"}](Etest5.png "fig:"){width="0.45\linewidth"} ![Time evolution of free energy \[E\] corresponding to Figure \[figRD\] using the same order. In all cases, free energy is a decreasing function of time.[]{data-label="figRDenergy"}](Etest6.png "fig:"){width="0.45\linewidth"}
Conclusion {#conclusion .unnumbered}
==========
In this article we have studied reaction-diffusion systems containing self and cross-diffusion using a free energy framework. Using this alternative form, we have theoretically established an energy law \[Elaw\], that was later exemplified by numerical simulations \[figRDenergy\]. We have derived a numerical method for general reaction-diffusion systems with nonlinear diffusion terms \[scheme\]. We have shown key properties for this method. In order to exemplify this scheme, we have introduced a modified version of the classical Gray Scott system augmented with self and cross-diffusion terms \[mGrayScott\]. During numerical simulation of this model, we have observed patterns that are clearly distinct from those obtained with linear diffusion even for the same reaction term Figure \[figRD\].
As a further work, a full and systematic numerical exploration, in the spirit of Pearson [@pearson], might be done on system \[mGrayScott\], in order to identify all the possible patterns. This work may be extended to other classical systems such as Fitzhugh-Nagumo, SKT or SIR, considering not only a linear diffusion, but a combination of linear, self and cross-diffusion.
During all simulations, we have observed that free energy \[E\] was a decreasing function of time. Theoretically, we have shown that free energy was a decreasing function of time in the purely diffusive case \[Elaw\]. Could this proof be extended, under hypothesis, to the reactive case? Could we get qualitative information about the system by tailoring this general expression to specific reactive terms?
Acknowledgement {#acknowledgement .unnumbered}
===============
The author would like to thank Romain Veltz and Mathieu Desroches for fruitful discussion and feedback about this work.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'A diverse variety of processes — including recurrent disease episodes, neuron firing, and communication patterns among humans — can be described using inter-event time (IET) distributions. Many such processes are ongoing, although event sequences are only available during a finite observation window. Because the observation time window is more likely to begin or end during long IETs than during short ones, the analysis of such data is susceptible to a bias induced by the finite observation period. In this paper, we illustrate how this length bias is born and how it can be corrected without assuming any particular shape for the IET distribution. To do this, we model event sequences using stationary renewal processes, and we formulate simple heuristics for determining the severity of the bias. To illustrate our results, we focus on the example of empirical communication networks, which are temporal networks that are constructed from communication events. The IET distributions of such systems guide efforts to build models of human behavior, and the variance of IETs is very important for estimating the spreading rate of information in networks of temporal interactions. We analyze several well-known data sets from the literature, and we find that the resulting bias can lead to systematic underestimates of the variance in the IET distributions and that correcting for the bias can lead to *qualitatively* different results for the tails of the IET distributions.'
author:
- Mikko Kivelä
- 'Mason A. Porter'
bibliography:
- 'ietest.bib'
title: '**Estimating inter-event time distributions from finite observation periods in communication networks**'
---
Introduction
============
The newfound wealth of large data sets in the modern era of “Big Data” necessitates statistical analyses of such data. This has been prevalent in the study of human behavior, as the digital footprints left behind by electronic activities provide a deluge of data. One of the most important problems in the study of human dynamics, which benefits directly from such data, is to quantify temporal activity patterns in human behavior. For example, this problem has been approached via the characterization of time sequences of human activities [@saramaki2015; @Eckmann2004Entropy; @Barabasi2005Origin; @Vazquez2006Modeling; @Rybski2012Communication; @Candia2008Uncovering; @Malmgren2008Poissonian; @altmann2009; @Wu2010Evidence; @Karsai2011Small; @Jiang2013Calling; @Goh2008Burstiness; @Kivela2012Multiscale; @Duarte2007Traffic; @Radicchi2009Human; @jo2015] and the analysis of “temporal networks” [@Holme2012Temporal; @holme13] (i.e., networks that change in time). *Inter-event times* (IETs) give the times between each pair of events (e.g., sending an e-mail, making a phone call, or doing any other activity), and the way that they are distributed has received intense scrutiny because they can be used to characterize temporal processes.
Electronic records often have a huge number of data points. Such data often includes many subjects, but it may or may not also include a similar wealth of longitudinal points. For example, there exist data sets with thousands or even millions of people but with observation periods that only last a few months [@Eckmann2004Entropy; @Candia2008Uncovering; @Karsai2011Small; @Jiang2013Calling]. Moreover, even when the observation period is long, a given individual might rarely be active during that time. This is the case, for example, in recent studies of e-mail communication [@Eckmann2004Entropy; @Vazquez2006Modeling; @Malmgren2008Poissonian; @Goh2008Burstiness; @Barabasi2005Origin], mobile phone calling [@Karsai2011Small; @Kivela2012Multiscale; @Jiang2013Calling; @saramaki2015], website usage [@Duarte2007Traffic; @Radicchi2009Human; @Rybski2012Communication], and donations to charities [@wipprecht-thesis]. As we will illustrate in this article, data sets in which the observation windows are comparable in scale to the IETs are vulnerable to finite-size biases. This can arise due to short observation windows and/or sparse records of activity. This effect biases the tails of observed IET distributions, thereby creating a very serious issue, as the properties of distribution tails are often among the most important empirical features that one needs to consider [@Holme2012Temporal] and models of human dynamics have been validated or refuted based on their predictions of the shape of IET distributions [@Barabasi2005Origin; @Vazquez2006Modeling; @Malmgren2008Poissonian; @Min2009Waiting; @Oliveira2009Impact; @Wu2010Evidence; @Jiang2013Calling]. Furthermore, the variance of IET distributions can have a large effect on dynamical processes that occur on a system [@Vazquez2007Impact; @Karsai2011Small; @Miritello2011Dynamical; @Takaguchi2011Voter; @Hoffmann2012Generalized; @Kivela2012Multiscale; @Jo2014Analytically; @porter2014], and the IET-distribution variance has been used to classify the processes that produce these distributions [@Goh2008Burstiness; @Zhou2008Role].
Several approaches have been used to account for the bias introduced by a finite temporal-window size. In particular, it is common to disregard all of the boundary effects and use the observed IETs [@Duarte2007Traffic; @Rybski2012Communication; @Candia2008Uncovering; @Goh2008Burstiness; @Karsai2011Universal; @Jiang2013Calling]. Such biases are sometimes acknowledged: for example, the exponential tail of an IET distribution is sometimes construed as a finite-size effect [@Wu2010Evidence; @Jiang2013Calling]. One can try to ameliorate the bias by introducing temporal periodic boundary conditions [@Karsai2011Small; @Kivela2012Multiscale], but such a solution does not give an unbiased estimator for an IET distribution. Another approach to dealing with a finite observation period is to correct the probability of observing an IET value by dividing it by the probability that an IET of that length is not truncated by the observation window [@Holme2003Network]. As we discuss in Section \[sec-estimate\], for stationary renewal processes, the latter probability always decreases linearly with the growth of the observed IET length. This linear correction has also been observed in empirical data by resampling using different observation window sizes [@Vazquez2007Impact].
The error in an observed IET distribution is very small if the tail of the IET distribution is sufficiently short relative to the length of the time window. This is usually the case if one uses event sequences with a large number of events. Unfortunately, in practice, this tends to entail that one can use only a small subset of available data. For example, some studies on temporal communication patterns that were based on data sets of thousands or millions of people only used subsets of the most active people that ranged from a single person to about $10\%$ of the data [@Barabasi2005Origin; @Wu2010Evidence; @Malmgren2008Poissonian; @Jiang2013Calling]. This approach discards valuable data and biases the analysis towards the behavior of very active individuals.
The use of IET distributions by scholars has a long history, and the problem of inferring an IET distribution from a finite observation period arises in a diverse set of fields — such as engineering and medicine, where the problem has been studied using renewal processes [@Feller1971Introduction] and other models for recurrent events [@Nelson2003Recurrent; @Cook2007Statistical]. Due to the generic nature of the problem, several statistical tools have been developed for estimating IET distributions for renewal processes [@Vardi1982Nonparametric; @Denby1985Shortcut; @Mcclean1995Nonparametric; @Soon1996Nonparametric; @Pena2001Nonparametric; @Gill2010Product; @Zhu2014Parametric]. Additionally, some techniques based on survival analysis and event-history analysis have been used to analyze temporal network data [@Butts2008Relational; @Dubois2013Hierarchical; @Melo2015Universal]. Similar problems have also been encountered when analyzing geological data [@Laslett1982Censoring; @Pickering1995Sampling] and estimating inter-spike intervals of firing neurons [@Pawlas2011Distribution].
In the present paper, we concentrate on stationary renewal processes that produce $N$ event sequences observed in a finite time window of length $T$. See Fig. \[fig:model\_illustration\]a,b for an illustration. We focus on renewal processes because they are minimal models for producing event sequences with arbitrary IET distributions. However, real processes are often more complicated than stationary renewal processes. For example, communication patterns and many natural phenomena — such as earthquakes, neuronal spike trains, and disease epidemics—arise from processes that have memory [@Goh2008Burstiness; @Karsai2011Universal]. Other processes, such as inhomogeneous Poisson processes and processes in which cascades of activity can be triggered by prior events, also yield tractable models for human dynamics [@Malmgren2008Poissonian; @Malmgren2009Universality; @Zipkin2014Pointprocess].
![(a) A stationary renewal process generates an infinite sequence of events. We place a time window of length $T$ in an arbitrary place on the timeline. (b) We consider only the events that lie inside of the time window. The time from the beginning of the time window to the first event is the *residual waiting time* $\tau_R$, and one can derive its distribution from $p(\tau)$ [@Feller1971Introduction]. (c) The *observed inter-event times* are the IETs that lie completely inside of the time window. (d) The *censored IETs* are the ones that are cut by the time window. An IET that is cut by the end (respectively, beginning) of the time window is said to be *forward censored* (respectively, *backward censored*). An IET that is truncated must be longer than the forward (and backward) censoring time: $\tau_{fc} \leq \tau_3$ (and $\tau_{bc} \leq \tau_0$) [@modelfootnote]. []{data-label="fig:model_illustration"}](figs/model_illustration.pdf){width="1.0\linewidth"}
Estimation of inter-event time distributions {#sec-estimate}
============================================
We seek to estimate the IET distribution $p(\tau)$ of the underlying process when we are given only the time stamps of the events inside of the observation window. A naive method would be to use the distribution $p^\prime(\tau)$ for observed IETs to estimate the real distribution $p(\tau)$ (see Fig. \[fig:model\_illustration\]c). Unfortunately, in general, the observed IETs and the real IETs do not follow the same distribution.
In Fig. \[fig:model\_iets\], we illustrate the difference between $p(\tau)$ and $p^\prime(\tau)$ for stationary renewal processes with exponential and power-law IET distributions. This difference grows linearly when the IET length $\tau$ approaches the window size $T$, and $p^\prime(\tau)=0$ for $\tau > T$. The growth occurs because a longer IET makes it more likely that the observation window either starts or end between the two events that correspond to that IET. Observed IETs are always distributed so that there is a linear cutoff at the end time $T$ of the time window. In other words, $$p^\prime(\tau) \propto (T - \tau) p(\tau)
\label{eq:tauobs}$$ when the number $N$ of event sequences tends to infinity [@Soon1996Nonparametric]. To give intuition for Eq. , note for a stationary renewal process that the probability of observing an event is uniform for a whole observation window. This implies that $(T - \tau)/T$ is the probability that an IET of length $\tau$ following an event chosen uniformly at random in the interval is not cut short by the end of the observation window.
In the worst case, the linear bias in Eq. can lead to qualitatively incorrect conclusions about the shape of the tail of an IET distribution. It is therefore important to correct for this bias. Note that this bias is more severe than that from an upper truncation, in which data points that are larger than a certain threshold value are not observed [@Burroughs2001Upper; @Deluca2013Fitting].
![ We simulate $N=10^5$ event sequences using stationary renewal processes for which the real IET distribution satisfies (a) $p(\tau) \propto e^{-\tau}$ and (b) $p(\tau) \propto \tau^{-2.1}$. We consider window sizes $T$ (which we indicate with dashed vertical lines) of (a) 0.5, 1, 2, and 5 and (b) 5, 10, 20, and 40. We calculate IETs for each event sequence, pool them together, and plot cumulative IET distributions $P_{\ge}(\tau)$. The dots indicate the observed IET distribution, and the crosses indicate the estimates of the real IET distribution using the Kaplan–Meier (KM) estimator. The solid black line is the theoretical $p(\tau)$ distribution, and the dotted curves are the theoretical distributions $p^\prime(\tau)$ for IETs \[see Eq. \] that lie completely inside of each time window. A nonparametric maximum likelihood estimator (NPMLE) [@Soon1996Nonparametric] gives qualitatively similar results. See Fig. \[fig:model\_iets\_binned\] in Appendix \[sec:altplots\] for the same distributions plotted using probability densities instead of cumulative probabilities. []{data-label="fig:model_iets"}](figs/exp_plaw_taudists_single_column.pdf){width="1.0\linewidth"}
There exist both parametric [@Zhu2014Parametric] and nonparametric [@Vardi1982Nonparametric; @Denby1985Shortcut; @Mcclean1995Nonparametric; @Soon1996Nonparametric; @Pena2001Nonparametric; @Gill2010Product; @Cook2007Statistical] estimators for the real IET distribution $p(\tau)$. A straightforward nonparametric way to estimate IETs is to use the Kaplan–Meier (KM) estimator [@Kaplan1958Nonparametric] by considering the IETs inside of the time window as uncensored observations and the IETs that are truncated by the end of the time window as censored observations [@Gill1981Testing; @Denby1985Shortcut]. Additionally, because the stationary renewal process that generates the event sequences is symmetric in time, we can increase the accuracy of our estimate by repeating this estimation process backwards in time [@Denby1985Shortcut]. That is, each uncensored IET is counted twice, and the censored IETs at the boundaries of the time window are counted only once [^1]. One can estimate the variance of the KM estimator using Greenwood’s formula [@Kaplan1958Nonparametric; @Greenwood1926Natural], which has to be modified slightly to take into account double-counting of the uncensored IETs [@Denby1985Shortcut]. See Fig. \[fig:model\_illustration\]d for a schematic and Fig. \[fig:model\_iets\] for an example how the KM estimator corrects the bias introduced by the finite observation window for simulated data [@ietsoft]. See Appendix \[sec:km\_estimator\] for details on how to use the KM estimator to estimate IET distributions.
The derivation of the KM estimator for IETs is based on a partial likelihood approach for data produced with a stationary renewal process [@Denby1985Shortcut]. The KM estimator only assumes that the sampled IETs are produced from the IET distribution independently of the windows of observation (i.e., the times from events to the end of the observation period). That is, the KM estimator disregards some information on how the data were produced if it is used for data that is known to be produced by a stationary renewal process. Vardi [@Vardi1982Nonparametric] defined a nonparametric maximum likelihood estimator (NPMLE) method for data produced with a stationary renewal process. Soon and Woodroofe [@Soon1996Nonparametric] later generalized Vardi’s method for continuous-time situations as well as for situations in which there are event sequences in which no events are observed during the observation time window. Note, however, that methods based on the KM estimator and Vardi’s NPMLE can yield estimates that are very close to each other even though the KM estimator is more computationally efficient than Vardi’s NPMLE estimator [@Denby1985Shortcut]. One can also use a reduced-sample estimator, which ignores data points close to the boundaries of an observation window, although Pawlas et al. [@Pawlas2011Distribution] observed for several different generative models of event sequences that it gives less accurate estimates than a method based on the KM estimator.
When does one need to worry about finite window-size effects? {#subsec-worry}
-------------------------------------------------------------
The bias introduced by using the observed IET distribution as an estimate for the real IET distribution for a given process can be very small even if data are produced by sampling from a renewal process using a finite time window. This is the case if the time-window length is sufficiently long. In this case, one does not need to worry about finite-size effects or make any corrections to account for them. We will next give some guidelines for determining when this happy situation holds.
As we discussed at the beginning of Section \[sec-estimate\], the bias in an IET distribution grows linearly with IET length. It is thus useful to compare the bias in the smallest observed IET to the bias in the largest observed IET, as their ratio gives an estimate for the largest error in the distribution. If the smallest possible IET is $\tau_0$, then Eq. implies that $$\frac{p^{\prime}(\tau )}{p^{\prime}(\tau_0)} =\left(1-\frac{\tau}{T}\right) \frac{p(\tau )}{p(\tau_0)}\,.
\label{eq:rot}$$ Equation can be used as a rule of thumb for assessing if a finite time window distorts an observed IET distribution. For example, if the largest data point (i.e., the rightmost point in an observed IET distribution) is more than $100$ times smaller than the length of the observation window, then the error that results using the observed IETs for estimating the real IET distribution is less than $1\%$ for IET values that are smaller than the maximum observed IET value.
Equation (\[eq:rot\]) gives an estimate for the relative probabilities of observed IETs, but it does not indicate anything about the distribution’s tail, which is not observed. This can be an issue if there are very small amounts of data or if one wants to calculate summary statistics of an IET distribution that are very sensitive to the properties of the tail (e.g., moments of an IET distribution, measures of event burstiness [@Goh2008Burstiness], and so on). The moments $\mu_m^\prime$ of an observed IET distribution are lower than the moments $\mu_m$ of the real IET distribution. However, if we have an estimate $p^{\rm{est}}( \tau )$ for the real IET distribution $p( \tau )$ for $\tau \leq \tau_{\rm{max}}$, then we can obtain an estimate for the moments using $$\begin{aligned}
\mu_{m}^{\rm{est}}=\int_0^{\tau_{\rm{max}}} \tau^m p^{\rm{est}}( \tau )d\tau + \tau_{\rm{max}}^m P_{\ge}^{\rm{est}}(\tau_{\rm{max}})\,,
\label{eq:muest}\end{aligned}$$ where $P_{\ge}^{\rm{est}}$ is the estimator of cumulative distribution of the IETs. That is, in this estimator, we use $p^{\rm{est}}( \tau )$ for the IET distribution for $\tau \leq \tau_{\rm{max}}$ and replace the unobserved tail by adding all of the remaining probability mass, $P_{\ge}^{\rm{est}}(\tau_{\rm{max}})$, to the point $\tau_{\rm{max}}$. Assuming that the estimate for the IET distribution is perfect (i.e., $p^{\rm{est}}( \tau )=p( \tau )$ when $\tau \leq T$), we obtain a sharper lower bound for the moments using $\mu_{m}^{\rm{est}}$ than using $\mu_m^\prime$. That is, $\mu_m^\prime \leq \mu_{m}^{\rm{est}} \leq \mu_{m}$. We illustrate this issue in Section \[sec:data\] using empirical data. Note, in practice, that $\tau_{\rm{max}}$ is close to $T$ for these data sets.
Analysis of empirical data {#sec:data}
==========================
We now use the methods that we described in Section \[sec-estimate\] to reanalyze several public data sets that have been studied previously in the literature. For each data set, we concentrate on temporal sequences of messages that are sent by individuals.
The Eckmann et al. e-mail data set [@Eckmann2004Entropy] contains time stamps of about $3 \times 10^5$ e-mails between 3188 people during $83$ days. This data set has been examined by several authors, and the shape of the IET distributions of individuals with high e-mailing frequencies has received particularly close scrutiny (and has attracted controversy) [@Barabasi2005Origin; @Stouffer2005Comment; @Barabasi2005Reply; @Malmgren2008Poissonian; @Karsai2011Universal]. The [pussokram.com](pussokram.com) (POK) data set [@Holme2003Network; @Rybski2012Communication] is a communication record of an online community with about $3 \times 10^4$ people who sent $5 \times 10^5$ messages during the entire 492-day lifetime of the site. Because the data recording started from the birth of the POK website, it is not reasonable to construe message sequences in this data set as having been produced by a stationary process. However, it is still reasonable to consider the data as being forward censored (see Fig. \[fig:model\_illustration\]). Rybski et al. [@Rybski2012Communication] plotted the distribution of all IETs as well as distributions grouped according to the number of sent messages. Their plots contain noticeable dips at the end of the IET distributions, but it is not clear in their paper if this feature arises because of intrinsic human behavior or is instead due to the finite length of the data. A third data set that we examine was introduced by Wu et al. [@Wu2010Evidence], who studied IETs of short messages sent within three different companies during one month. We present our reanalysis of data from company 1, which includes about $5 \times 10^5$ messages sent by about $4 \times 10^4$ people. The results for the two other companies are similar. To obtain good statistics, Wu et al. concentrated on communication patterns between the few pairs of users who sent very large numbers of messages to each other. For each data set, we consider the observation window for each user to be the observation window of the whole system, although additional information of users leaving or joining the system could have been used to construct individual observation windows if such information were available.
![Results for the empirical data sets. We consider activation times for each node as a single event sequence. In panels (a)–(c), we show the IET distributions that we obtain from combining IET distributions of all node activation sequences. The dots indicate the observed IET distributions, and the crosses indicate the estimates of the IET distributions using the KM estimator. In panels (d)–(i), we bin the event sequences according to the number of events in them ([$\times$]{}: $n = 3$, [$+$]{}: $n = 6$, [$\circ$]{}: $n \in [8, 9]$, [$\bigtriangledown$]{}: $n \in [14, 25]$, [$\square$]{}: $n \in [51, 150]$). We skip every other bin to make the figure easier to read), and we normalize the IETs according to the bin’s mean IET. In panels (d)–(f), we show cumulative distributions of observed IETs normalized by the mean $\mu_1^\prime$ of the observed IETs. In panels (g)–(i), we show KM estimates for the cumulative IET distributions normalized by the mean $\mu_1^{\rm{KM}}$ calculated from the estimated IET distribution. The shaded regions are the 95% confidence intervals [@Denby1985Shortcut]. The data sets are (a, d, g) the Eckmann et al. e-mail data [@Eckmann2004Entropy], (b, e, h) POK messages [@Rybski2012Communication], and (c, f, i) the Wu et al. short-message data [@Wu2010Evidence]. See Fig. \[fig:data\_iets\_loglog\] in Appendix \[sec:altplots\] for the same distributions plotted using doubly logarithmic axes. []{data-label="fig:data_iets"}](figs/data_iets.pdf){width="1.0\linewidth"}
Each of the data sets includes a large number of IETs that are sufficiently close to the time-window length to affect the observed IET distribution. We illustrate this fact in panels (a)–(c) of Fig. \[fig:data\_iets\]. For each data set, we show both the observed IET distribution and the KM estimate of the IET distribution. It is clear that the shape of the tail of the observed IET distributions is qualitatively different from that of the KM estimate of the IET distribution. The dip that is often observed in the tail of an IET distribution that includes IETs that are close to the observation-window length [@Duarte2007Traffic; @Karsai2011Universal; @Rybski2012Communication; @Jiang2013Calling] can be explained by the finite observation window in each of the data sets that we study.
In Table \[table:summary\], we compare some summary statistics of the KM estimate of the IET distributions and the observed IET distributions to gain a better understanding of how much the two differ. The first two moments and residual waiting times calculated from the IET distribution given by the KM estimator are often more than $100\%$ larger than ones calculated from the observe IET distribution. These differences can have a huge impact on processes that act on top of temporal networks, and it is clear that the bias introduced by a finite observation-window size can be a major problem in these situations. For example, the mean residual waiting time $\tau_R$ — which is vastly smaller when calculated using the observed IET distributions than when calculated using the IET distributions obtained with the KM estimator — is related to the speed of spreading in networks [@Vazquez2007Impact; @Karsai2011Small; @Miritello2011Dynamical; @Kivela2012Multiscale; @Jo2014Analytically], because it is the expected time until the next event after a node is infected at a time chosen uniformly at random.
Data $\mu_1^\prime$ $\mu_1^{\rm{KM}}$ $\sqrt{\mu_2^\prime}$ $\sqrt{\mu_2^{\rm{KM}}}$ $\mu_1^\prime (\tau_R)$ $\mu_1^{\rm{KM}} (\tau_R)$ $\tau_{fc,bc}$
--------------- ---------------- ------------------- ----------------------- -------------------------- ------------------------- ---------------------------- ----------------
E-mail $0.908$ $1.51$ $3.20$ $6.88$ $5.62$ $15.6$ $17.5$
POK $5.13$ $28.4$ $23.1$ $106$ $51.9$ $198$ $240$
Short message $0.633$ $1.40$ $2.11$ $4.89$ $3.53$ $8.53$ $8.73$
: The first two moments of the IETs calculated from the observed IET distribution and using Eq. for the IET distribution produced by the KM estimator, estimates of the residual waiting times using the formula $\mu_1(\tau_R)=\frac{1}{2}\frac{\mu_2}{\mu_1}$ [@Kivela2012Multiscale], and the mean of forward and backward censoring times $\tau_{fc,bc}$. (For the POK data set we only calculate the mean of the forward censoring times.) Note that data produced by a stationary renewal process has forward-censoring and backward-censoring times that are are distributed as the residual waiting times for values that are smaller than the window size $T$. []{data-label="table:summary"}
In studies of empirical data, it is often assumed that each event sequence is produced by an IET distribution with the same characteristic shape but different underlying rate. Different event sequences would then arise using the same scaling function $f$ but with a different mean value $\tau_0$ [@Corral2003Local; @Saichev2006Universal; @Candia2008Uncovering; @Goh2008Burstiness; @Karsai2011Small; @Rybski2012Communication]. The IET distribution for a sequence with finite mean $\tau_0$ is defined as $p(\tau | \tau_0) = \frac{1}{\tau_0} f(\tau / \tau_0)$ (see Appendix \[sec:scaling\]). In panels (d)–(i) of Fig. \[fig:data\_iets\], we plot the IET distributions (for each data set) in which we group event sequences with similar numbers of events. We include event sequences that have fewer than 151 events because sequences with few events are the most susceptible to finite-size effects. Sequences with at most 150 events encompass 90%–99% of all sequences (depending on the data set). We observe that normalized IET distributions for event sequences with few events decrease much faster than the IET distributions for sequences with many events. This result is expected, and it results from the bias introduced by the finite observation window. There is a very good collapse of the tails of the KM estimates of the normalized IET distributions for the e-mail communication and short-message communication data. This is remarkable, given that collapse is not expected to be perfect even for data that perfectly follows the characteristic distribution model (see Appendix \[sec:scaling\]). The difference between the IET distributions of the POK data and the two other data sets may be due to users who leave the service permanently. This process would lead to the last IET being infinitely long, which would manifest as the tail of the cumulative distribution approaching a value that corresponds to the fraction of people in each group who have left the service. One would expect this fraction to be smaller for groups with a larger number of messages if the probability of leaving the service is lower for people who have sent more messages.
Conclusions and Discussion
==========================
We investigated the effects that a finite observation window can have on observed inter-event times (IETs). For a stationary renewal process, we illustrated that the finite time window introduces a linear cutoff to the observed IET distribution at the end of the time window (see Fig. \[fig:model\_iets\]). We showed how to correct this bias using nonparametric estimators, such as the KM estimator or an NPMLE, for a stationary renewal process. We also illustrated that these estimators work well even for event sequences with small numbers of events if these sequences can be grouped together. We then used these methods to reanalyze three data sets of human communication, and we found that using the observed IET distributions without correcting for the finite-size bias can seriously distort the shape and key summary statistics of IET distributions.
Human behavior is rather heterogeneous in many aspects, and in particular, the event sequences of different people contain widely disparate number of events. Many authors have argued that it is possible to represent such sequences using a scaling function that is independent of the underlying rate of events [@Corral2003Local; @Saichev2006Universal; @Candia2008Uncovering; @Goh2008Burstiness; @Karsai2011Small]. However, there is an additional bias if one infers the underlying rate from the observed number of events (see Appendix \[sec:scaling\]), and it is important to develop statistical methods that are able to assume an underlying model for a characteristic IET distribution. Moreover, methods for testing whether an IET distribution has some specific shape are also susceptible to finite-size effects, and parametric analogs of the methods that we have employed should be applied in such situations [@Zhu2014Parametric]. Further, in the present paper we are focusing on the IET distributions of multiple event sequences, but finite-size effects should also be taken into consideration when estimating summary statistics such as moments or burstiness [^2] of single event sequences.
The need for the wide dissemination and use of correction methods like KM estimators or NPMLEs for IET distributions is underscored by the rapidly growing analysis of temporal data streams. Nonparametric methods for correcting for biases that are introduced by a finite observation window have existed for several decades [@Vardi1982Nonparametric; @Denby1985Shortcut; @Soon1996Nonparametric]. Surprisingly, such methods (to our knowledge) do not seem to have been used when analyzing human communication patterns, although there have been some ad-hoc attempts to directly correct for the linear bias [@Holme2003Network; @Vazquez2007Impact]. Additionally, although we have focused on human communication patterns, the problem of correcting for these finite-size effects is a general one, and similar methods have been reinvented in multiple fields. For example, the KM estimator was used for window-censored data in the 1980s [@Denby1985Shortcut], and its use for such data was independently reinvented many years later in the context of estimating the inter-spike intervals of neurons [@Pawlas2011Distribution]. Appropriately taking into account finite-size effects makes it possible to obtain accurate estimates for the tail of an IET distribution and to optimally exploit data that consists of a large number of event sequences with only a small number of events (as opposed to high-frequency event sequences, which are largely free of such significant finite-size effects).
Both authors were supported by the European Commission FET-Proactive project PLEXMATH (Grant No. 317614). We thank Andrea Bertozzi, Carlos Gershenson, Adilson Motter, Se Wook Oh, and Jari Saramäki for helpful comments; and we thank Jean-Pierre Eckmann for providing us with the e-mail data set. We also thank several anonymous referees for helpful comments.
Kaplan–Meier estimator for inter-event times {#sec:km_estimator}
============================================
We now discuss how to use the Kaplan–Meier (KM) estimator [@Kaplan1958Nonparametric] to estimate the IET distribution of a stationary renewal process when one only observes events in a finite time window. Our approach is similar to the “shortcut method” of Denby and Vardi [@Denby1985Shortcut]. Unlike them, however, we do not add a point $\tau_M$ that is much larger than the observed IET values to the IET-distribution estimate.
The KM estimator is a nonparametric estimator for lifetimes (or times of death) in the presence of censored lifetimes (or losses) [@Kaplan1958Nonparametric]. Corresponding to each lifetime $\tau_i$, there is a censoring time $\tau_{c,i}$, and we observe the lifetime if it is shorter than or the same as the censoring time (i.e., if $\tau_i \leq \tau_{c,i}$) and censor it if it is longer than the censoring time (i.e., if $\tau_i > \tau_{c,i}$). That is, for each $i$ we observe a single time $t_i$ that is either a lifetime $t_i=\tau_i$ or a censoring time $t_i=\tau_{c,i}$. (If the lifetime $\tau_i$ is censored, we say that it is a “censored lifetime,” and we say that the time $\tau_{c,i}$ that it is censored is its “censoring time.”) The KM estimator $\hat{P}_{\ge}$ for the cumulative distribution of lifetimes is $$\begin{aligned}
\hat{P}_{\ge}(t)= \prod_{s \le t }\left(1- \frac{\delta_s}{n_s}\right)\,,\end{aligned}$$ where $n_s$ is the number of lifetimes that are known to be at least as long as $s$ (i.e., $n_s=\sum_{s^\prime \geq s}[\delta_{s^\prime}+c_{s^\prime}]$), the parameter $\delta_s$ is the number of lifetimes that are observed at time $s$, and $c_s$ is the number of lifetimes that are censored at time $s$.
One can estimate the variance of the KM estimator using Greenwood’s formula [@Kaplan1958Nonparametric; @Greenwood1926Natural]: $$\begin{aligned}
\mathrm{Var}\left(\hat{P}_{\ge}(t)\right)=\hat{P}_{\ge}^2(t) \sum_{s \le t }\frac{\delta_s}{n_s(n_s - \delta_s)}\,.
\label{eq:greenwood}\end{aligned}$$ One can then use the variance estimate to construct confidence intervals for the estimate of an IET distribution. For example, if the $\hat{P}_{\ge}(t)$ values are normally distributed, then the confidence intervals are $$\hat{P}_{\ge}(t) \pm z_{\alpha/2} \sqrt{\mathrm{Var}\left(\hat{P}_{\ge}(t)\right)}\,,$$ where $1-\alpha$ is the confidence level and $z_{\alpha}$ is the quantile function of the standard normal distribution. In general, however, the $\hat{P}_{\ge}(t)$ values are not normally distributed, which can lead to confidence intervals that are not restricted to lie in the interval $[0,1]$. One usually addresses this situation by applying a transformation $g$ to the $\hat{P}_{\ge}(t)$ values to obtain a set of values that better follow a normal-distribution approximation. One can then calculate the confidence interval for the transformed random variable so that $$g\left(\hat{P}_{\ge}(t)\right) \pm z_{\alpha/2} \sqrt{\mathrm{Var}\left(g\left(\hat{P}_{\ge}(t)\right)\right)}\,.$$ Choices for the transformation include $g(p)=\ln(p)$, $g(p)=\ln\left(-\ln(p)\right)$, and $g(p)=\arcsin(\sqrt{p})$. (See, e.g., Borgan and Knut [@Borgan1990Note] for a discussion about choosing the transformation.) In Fig. \[fig:data\_iets\] of the main text, we used the transformation $g(p)=\ln\left(p/(1-p)\right)$ to follow the choice in Ref. [@Denby1985Shortcut].
One can use the KM estimator to estimate IETs of a renewal process by considering the observed IETs as observed lifetimes and the IETs that are truncated by the end of a time window (i.e., the IETs that are forward censored) as censored lifetimes. If the renewal process is stationary, then one can also repeat this procedure by reversing the direction of time [@Denby1985Shortcut]. In other words, one can consider both backward-censoring and forward-censoring times as censored lifetimes, and the observed IETs are twice counted as observed lifetimes. This makes it possible to use the information in the backward-censoring times in the construction of the estimator for the IET distributions. Note that the variance estimator of Greenwood’s formula in Eq. needs to be multiplied by $2$ in order to account for the fact that uncensored data points are used twice [@Denby1985Shortcut].
Alternative illustrations of IET distributions {#sec:altplots}
==============================================
![As in Fig. \[fig:model\_iets\], we simulate $N=10^5$ event sequences using stationary renewal processes. Now, however, we plot probability densities instead of cumulative probabilities. We plot IET distributions $p(\tau)$ for $N=10^6$ event sequences that we simulate from a stationary renewal process for which (a) $p(\tau) \propto e^{-\tau}$ and (b) $p(\tau) \propto \tau^{-2.1}$. We consider window sizes $T$ (which we indicate with dashed vertical lines) of (a) 0.5, 1, 2, and 5 and (b) 5, 10, 20, and 40. The dots indicate the observed IET distribution, and the crosses indicate the estimates of the real IET distribution using the KM estimator. The solid black line is the theoretical $p(\tau)$ distribution, and the dotted curves are the theoretical distributions $p^\prime(\tau)$ for IETs \[see Eq. \] that lie completely inside of each time window. []{data-label="fig:model_iets_binned"}](figs/exp_plaw_taudists_binned.pdf){width="1.0\linewidth"}
Figure \[fig:model\_iets\_binned\] corresponds to Fig. \[fig:model\_iets\] in the main text, but we now show probability densities instead of cumulative probabilities. Figure \[fig:data\_iets\_loglog\] corresponds to Fig. \[fig:data\_iets\] in the main text, but we now plot the IET distributions using doubly logarithmic axes.
![Results for the empirical data sets (also see Fig. \[fig:data\_iets\]) plotted using doubly logarithmic axes. We consider activation times for each node as a single event sequence. In panels (a)–(c), we show IET distributions that we obtain by combining IET distributions of all node activation sequences. The dots indicate the observed IET distributions, and the crosses indicate the estimates of the IET distributions using the KM estimator. In panels (d)–(i), we bin the event sequences according to the number of events in them ([$\times$]{}: $n = 3$, [$+$]{}: $n = 6$, [$\circ$]{}: $n \in [8, 9]$, [$\bigtriangledown$]{}: $n \in [14, 25]$, [$\square$]{}: $n \in [51, 150]$). We skip every other bin to make the figure easier to read, and we normalize the IETs according to the bin’s mean IET. In panels (d)–(f), we show cumulative distributions of observed IETs normalized by the mean $\mu_1^\prime$ of observed IETs. In panels (g)–(i), we show KM estimates for the cumulative IET distributions normalized by the mean $\mu_1^{\rm{KM}}$ calculated from the estimated IET distribution. The shaded regions are the 95% confidence intervals [@Denby1985Shortcut]. The data sets are (a, d, g) the Eckmann et al. e-mail data [@Eckmann2004Entropy], (b, e, h) POK messages [@Rybski2012Communication], and (c, f, i) the Wu et al. short-message data [@Wu2010Evidence]. []{data-label="fig:data_iets_loglog"}](figs/data_iets_loglog.pdf){width="1.0\linewidth"}
Analyzing event sequences selected based on the number of events in them {#sec:scaling}
========================================================================
Distributions of number of events {#sec:nevents}
---------------------------------
One can quantify the activity of the people in the data sets discussed in the main text by counting the number of events that each person has in his/her event sequence. Most of the people in the data that we examine exhibit very little activity, although there are also people that are significantly more active (by several orders of magnitude). One would not expect such a distribution if all event sequences were produced by a single renewal process. To illustrate this point, we construct a renewal process whose IET distribution we infer using the KM estimator. (See Fig. \[fig:data\_iets\] in the main text.) Using this model process, we produce a new data set that has the same number of event sequences as the original data. In Fig. \[fig:nevents\], we plot the activity distribution for the original data and the data produced by the model processes. The distributions of events observed in our data sets and the ones observed for the model are significantly different: almost all of the event sequences produced by the renewal process that we construct contain between $10$ and $100$ events, and there are no sequences with a very small number or a very large number of events. It is clearly very unlikely that all of the event sequences in the data were produced by a single renewal process.
![Cumulative distributions for the numbers of events in several communication data sets. We indicate the distribution of the original data using green circles, and we use blue triangles to indicate the distribution of the process that assumes that the data were produced by a single IET distribution. (See the text for details.) (a) Eckmann et al. e-mail data [@Eckmann2004Entropy], (b) POK messages [@Rybski2012Communication], and (c) Wu et al. short-message data [@Wu2010Evidence]. []{data-label="fig:nevents"}](figs/eckmann1_user_send_nevents_cum.pdf "fig:"){width="0.49\linewidth"} ![Cumulative distributions for the numbers of events in several communication data sets. We indicate the distribution of the original data using green circles, and we use blue triangles to indicate the distribution of the process that assumes that the data were produced by a single IET distribution. (See the text for details.) (a) Eckmann et al. e-mail data [@Eckmann2004Entropy], (b) POK messages [@Rybski2012Communication], and (c) Wu et al. short-message data [@Wu2010Evidence]. []{data-label="fig:nevents"}](figs/pok_user_send_nevents_cum.pdf "fig:"){width="0.49\linewidth"} ![Cumulative distributions for the numbers of events in several communication data sets. We indicate the distribution of the original data using green circles, and we use blue triangles to indicate the distribution of the process that assumes that the data were produced by a single IET distribution. (See the text for details.) (a) Eckmann et al. e-mail data [@Eckmann2004Entropy], (b) POK messages [@Rybski2012Communication], and (c) Wu et al. short-message data [@Wu2010Evidence]. []{data-label="fig:nevents"}](figs/SD01_user_send_nevents_cum.pdf "fig:"){width="0.49\linewidth"}
Model with a scaling function {#app:universal}
-----------------------------
One way to relax the assumption that event sequences are produced by a single IET distribution is to suppose that each event sequence is produced by an IET distribution with the same characteristic shape, which given by a scaling function $f$ but with a different mean value $\tau_0$. The IET distribution for a model constructed using this scenario is $p(\tau | \tau_0) = \frac{1}{\tau_0} f(\tau / \tau_0)$, where $\tau_0$ is the mean IET of the sequence. Such a model has been fitted to several empirical data sets [@Corral2003Local; @Saichev2006Universal; @Candia2008Uncovering; @Goh2008Burstiness; @Karsai2011Small; @Rybski2012Communication].
Let’s consider a model in which we choose the distributions $f$ and $p_0(\tau_0)$ so that our model resembles a real set of event sequences but remains analytically tractable. The distribution for the number of events is often heavy-tailed in communication data [@Onnela2007Analysis] (e.g., see Fig. \[fig:nevents\]), and we choose to model the distribution for the number of events as $p(n) \propto n^{-\alpha}$ (where $n \geq 1$ and $\alpha=2.5$). To do this, we construct the distribution $p_0$ for the mean values $\tau_0$ so that the numbers of events in the sequences are distributed as the given power law. To ensure analytical tractability, we choose the function $f$ to be an exponential function. That is, our aggregate process is a combination of multiple Poisson processes.
For each event sequence, we draw an expected IET from the distribution $p_0(\tau_0)$. Event sequences are then produced by a renewal process with an IET distribution of $p(\tau)=f(\tau / \tau_0)/\tau_0$. The residual waiting-time distribution [@Feller1971Introduction] for the process is then $$\begin{aligned}
p_R(\tau_R) = \frac{1}{\tau_0} f_R(\tau_R/\tau_0)\,,
\label{puniversal}\end{aligned}$$ where $f_R$ is the residual waiting-time distribution for the process that is determined by the IET distribution $f$. By exploiting the expected relation $n=\frac{T}{\tau_0}$, we can approximate the IET distribution for the aggregate process: $$\begin{aligned}
p(\tau ) \propto \int_1^{\infty} n p_0(n) p\left(\tau | \tau_0=\frac{T}{n}\right) dn\,,\end{aligned}$$ which reduces to $$\begin{aligned}
p(\tau ) \propto E_{\alpha - 2}(\tau / T)\,,
\label{eq:iet_collapse_plaw_exp}\end{aligned}$$ where $E_\alpha(x)= \int_1^{\infty} e^{-t x}/t^\alpha dt $ is the exponential integral function [@DLMF].
In Fig. \[fig:collapse\_model\], we show numerical results for the model that we just described. In Fig. \[fig:collapse\_model\]a, we show both the distribution of observed IETs and a KM estimate that we compute when all of the event sequences are grouped together. It is clear that the observed IETs cannot be used to estimate the real IETs, but the KM estimator performs well in this task. One can also group event sequences with similar values for the parameter $\tau_0$. Plotting the IET distributions then causes the data to collapse onto a curve that follows the shape given by $f$ if the IET distributions are grouped according to the $\tau_0$ values that were used to generate them and rescaled using the mean of $\tau_0$ values. Each group — and especially the groups with large mean values of $\tau_0$ (i.e., with a small number of events) — is of course susceptible to finite-size window effects (see Fig \[fig:collapse\_model\]b), but one can correct for such effects using the same methods as one would use for data produced by a model with a single IET distribution. See the inset of Fig. \[fig:collapse\_model\]b.
There is often no way to access the underlying mean IET values $\tau_0$ even if the data is known to be produced by the model that we described above. Instead, one has to estimate $\tau_0$ values from data by calculating the mean IET for each sequence [@Candia2008Uncovering; @Goh2008Burstiness; @Rybski2012Communication; @Karsai2011Small]. This introduces another kind of bias, for which estimators that correct for finite observation windows are not designed. Our example with exponential $f$ illustrates this situation rather nicely. In Fig. \[fig:collapse\_model\]c, we show similar results as in Fig. \[fig:collapse\_model\]b, except that we group the event sequences using the observed number $n$ of events instead of using $\tau_0$ values of the underlying processes to calculate the expected number of events $\hat{n}=\frac{T}{\tau_0}$. The IET distributions of the event sequences with small numbers of events are not identified correctly as exponential distributions, but instead they follow the distribution defined in Eq. (see below) if one uses the observed number of events to group the event sequences. That is, when grouping event sequences with exactly $n$ events, we find that (1) their IET distributions are independent of the mean rates $\tau_0$ and (2) they cannot be rescaled to follow $f$ even after removing finite-size effects.
![Numerical calculations for a model in which we produce the event sequences using the IET distribution $p(\tau | \tau_0) = \frac{1}{\tau_0} f(\tau / \tau_0)$, where $f(\tau)=e^{-\tau}$ and the mean values $\tau_0$ are distributed such that the expected numbers of events satisfy the probability distribution $p(n) \propto n^{-2.5}$ (where $n \geq 1$). (a) Cumulative distribution of observed IETs (green dots) and a KM estimate for the cumulative distribution (blue crosses). The black curve is the theoretical estimate of Eq. for the real IET distribution $p(\tau ) \propto E_{\alpha - 2}(\tau / T)$, where $E_n$ is the exponential integral function [@DLMF]. (b) Cumulative distributions of observed IETs when we bin event sequences according to the expected number of observed events $\hat{n}=T/\tau_0$ ([$\times$]{}: $\hat{n} \in (2, 3]$, [$+$]{}: $\hat{n} \in (5, 6]$, [$\circ$]{}: $\hat{n} \in (7, 9]$, [$\bigtriangledown$]{}: $\hat{n} \in (13, 25]$, [$\square$]{}: $\hat{n} \in (50, 150]$). We skip every other bin to make the figure easier to read, and we divide the IETs in each bin by the mean $\tau_0$ value of the bin $\mu_1(\tau_0)$. In the inset, we show KM estimates for the cumulative distributions IETs of each bin. (c) Cumulative distributions of observed IETs when we bin event sequences according to the observed number $n$ of events. We divide the IETs in each bin by the mean observed IET value $\mu_1^\prime$ of the bin. The lines correspond to IET distributions predicted by Eq. (or to mixtures of them for bins that have event sequences with more than one $n$ value in them). In the inset, we show KM estimates for the cumulative distributions IETs of each bin and divide the IETs in each bin with $\mu_1^{\rm{KM}}$. []{data-label="fig:collapse_model"}](figs/collapse_model.pdf "fig:"){width="1.0\linewidth"} ![Numerical calculations for a model in which we produce the event sequences using the IET distribution $p(\tau | \tau_0) = \frac{1}{\tau_0} f(\tau / \tau_0)$, where $f(\tau)=e^{-\tau}$ and the mean values $\tau_0$ are distributed such that the expected numbers of events satisfy the probability distribution $p(n) \propto n^{-2.5}$ (where $n \geq 1$). (a) Cumulative distribution of observed IETs (green dots) and a KM estimate for the cumulative distribution (blue crosses). The black curve is the theoretical estimate of Eq. for the real IET distribution $p(\tau ) \propto E_{\alpha - 2}(\tau / T)$, where $E_n$ is the exponential integral function [@DLMF]. (b) Cumulative distributions of observed IETs when we bin event sequences according to the expected number of observed events $\hat{n}=T/\tau_0$ ([$\times$]{}: $\hat{n} \in (2, 3]$, [$+$]{}: $\hat{n} \in (5, 6]$, [$\circ$]{}: $\hat{n} \in (7, 9]$, [$\bigtriangledown$]{}: $\hat{n} \in (13, 25]$, [$\square$]{}: $\hat{n} \in (50, 150]$). We skip every other bin to make the figure easier to read, and we divide the IETs in each bin by the mean $\tau_0$ value of the bin $\mu_1(\tau_0)$. In the inset, we show KM estimates for the cumulative distributions IETs of each bin. (c) Cumulative distributions of observed IETs when we bin event sequences according to the observed number $n$ of events. We divide the IETs in each bin by the mean observed IET value $\mu_1^\prime$ of the bin. The lines correspond to IET distributions predicted by Eq. (or to mixtures of them for bins that have event sequences with more than one $n$ value in them). In the inset, we show KM estimates for the cumulative distributions IETs of each bin and divide the IETs in each bin with $\mu_1^{\rm{KM}}$. []{data-label="fig:collapse_model"}](figs/collapse_model2.pdf "fig:"){width="2.22"}
Deriving observed inter-event time distributions {#sec:iet_model}
------------------------------------------------
In this section, we derive a formula for the probability $p^{\prime}(\tau_i, n)$ of observing $\tau_i$ as the $i$th IET in a sequence with exactly $n$ events. We assume that the sequence is produced by a stationary renewal process with an IET distribution of $p(\tau)$ and that we observe it in a finite window that begins at time $0$ and ends at time $T$. We use $p^{\prime}(\tau_i, n)$ to approximate $p^{\prime}(\tau, n)$ when we observe a large number of independent sequences. See Ref. [@Feller1971Introduction] for an introduction to renewal processes.
The probability that the $n$th event after time $0$ takes place at time $t$ is $$\begin{aligned}
\label{thisresult}
p(t,n)=p_R {\ast}p^{{\ast}(n-1)}(t)\,,\end{aligned}$$ where $p_r(\tau_r)=\frac{1}{\mu_1} \int_{\tau_r}^{\infty} p(\tau) d\tau$ is the residual waiting-time distribution, ${\ast}$ is the convolution operator, $x^{{\ast}y}$ means that $x$ is convolved with itself $y$ times, and $\mu_1$ is the expected IET. We use Eq. to calculate the probability $p^\prime(n)$ of observing exactly $n$ events during a time window of length $T$. The probability $p^\prime(n)$ is equal to the probability that the $n$th event after time $0$ takes place at time $t \leq T$ and the subsequent IET $\tau_{n}$ is larger than $T-t$. That is, one can write the probability of observing exactly $n$ events as $$\begin{aligned}
p^\prime(n) &= \int_0^T p(t,n) \int_{T-t}^\infty p(\tau) d\tau dt \notag \\
&= \mu_1 p_R {\ast}p^{{\ast}(n-1)}{\ast}p_R(T)\,.
\label{eq:pn_obs}\end{aligned}$$
We now want to calculate the probability of observing $n$ events when we know the $i$th observed IET $\tau_i$ (where $i \in \{1,\dots,n-1\}$ and $n \geq 2$). We obtain this probability from Eq. by substituting $T$ with $T-\tau_i$ and $n$ with $n-1$ to yield $$\begin{aligned}
p^\prime(n|\tau_i)=\mu_1 p_R {\ast}p^{{\ast}(n-2)}{\ast}p_R(T-\tau_i)\,.\end{aligned}$$ The joint probability distribution of observing $n$ events with $\tau_i$ as the $i$th IET is thus $$\begin{aligned}
p^\prime(n,\tau_i)=p^\prime(n|\tau_i)p(\tau_i )\,.
\label{eq:pntau_obs}\end{aligned}$$ Observe that the probability distribution (\[eq:pntau\_obs\]) is independent of the index $i$ as long as $i \in \{1,\dots,n-1\}$. By contrast, for a single sequence, the quantities $\tau_i$ and $\tau_j$ (with $i \neq j$) are not independent. However, as long as there are sufficiently many event sequences, we can use Eq. to approximate the joint distribution of the IETs and the numbers of events.
For a Poisson process, the approximate observed IET distribution given the number of events is $$\begin{aligned}
p^\prime(\tau | n)=n\frac{(T-\tau)^{n-1}}{T^n}\,.
\label{eq:obs_iet_exp}\end{aligned}$$ The cumulative distribution is thus $$\begin{aligned}
P^\prime_{\ge }(\tau | n)=\frac{(T-\tau)^{n}}{T^n}\,.
\label{eq:obs_iet_exp_cum}\end{aligned}$$ Note that Eqs. (\[eq:obs\_iet\_exp\]) and (\[eq:obs\_iet\_exp\_cum\]) are independent of the rate of the Poisson process. We illustrate this independence in Fig. \[fig:collapse\_model\]c.
[^1]: We call this estimator “the KM estimator” in the rest of our article. It is also sometimes called a “a product-limit estimator”
[^2]: Note that the burstiness coefficient defined in Ref. [@Goh2008Burstiness] is based on the coefficient of variation, and naive estimates of the coefficient of variation are biased for small sample sizes [@Sokal1980Significance; @Breunig2001Almost].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
The $e^+e^-\to\pi^0\pi^0\gamma$ process was studied in the SND experiment at VEPP-2M $e^+e^-$ collider in the energy region 0.60–0.97 GeV. From the analysis of the energy dependence of measured cross section the branching ratios $B(\omega\to\pi^0\pi^0\gamma)= (6.6^{+1.4}_{-0.8}\pm0.6)\cdot10^{-5}$ and $B(\rho\to\pi^0\pi^0\gamma)=(4.1^{+1.0}_{-0.9}\pm0.3)\cdot10^{-5}$ were obtained.\
[**PACS: 13.65.+i,14.40.Cs**]{}
author:
- |
M.N.Achasov, K.I.Beloborodov, A.V.Berdyugin, A.G.Bogdanchikov,\
A.V.Bozhenok, D.A.Bukin, S.V.Burdin, A.V.Vasiljev, V.B.Golubev,\
T.V.Dimova, V.P.Druzhinin[^1], V.N.Ivanchenko, A.A.Korol, I.A.Koop,\
S.V.Koshuba, A.V.Otboev, E.V.Pakhtusova, A.A.Salnikov,\
S.I.Serednyakov, V.A.Sidorov, Z.K.Silagadze, A.N.Skrinsky,\
A.G.Skripkin, Yu.V.Usov, V.V.Shary, Yu.M.Shatunov.\
Budker Institute of Nuclear Physics,\
Novosibirsk State University,\
Novosibirsk 630090, Russia
title: '**Experimental Study of $\rho\to\pi^0\pi^0\gamma$ and $\omega\to\pi^0\pi^0\gamma$ Decays**'
---
Introduction
============
In 1998 and 2000 the experiments with Spherical Neutral Detector (SND)[@SND] at VEPP-2M $e^+e^-$ collider were carried out in the energy range $E=360-970$ MeV where cross section of $e^+e^-$ annihilation into hadrons is determined by the $\rho$ and $\omega$ meson decays. The integrated luminosity of 9 pb$^{-1}$ collected in the experiment corresponds to $3.5\cdot10^6$ and $7\cdot10^6$ produced $\rho$ and $\omega$ mesons, respectively. One of the goals of the experiment was the investigation of the rare process $$e^+e^-\to\rho,\omega\to\pi^0\pi^0\gamma.
\label{ppg}$$ Our preliminary study [@rhoppg] of the process (\[ppg\]) was based on 1/3 of collected statistics. Its results were the first measurement of $B(\rho\to\pi^0\pi^0\gamma)=(4.8^{+3.4}_{-1.8}\pm0.2)\cdot10^{-5}$ and the measurement of $B(\omega\to\pi^0\pi^0\gamma)=(7.8\pm2.7\pm2.0)\cdot10^{-5}$ confirming the only previous measurement of this decay by GAMS: $B(\omega\to\pi^0\pi^0\gamma)=(7.2\pm2.5)\cdot10^{-5}$ [@GAMS].
The theoretical study of the $\rho,\omega\to\pi^0\pi^0\gamma$ decays was begun by P.Singer in Ref. [@Singer1] where the transitions via $\omega\pi^0$ (Fig. \[diad\]a) and $\rho^0\pi^0$ intermediate states were suggested. The vector meson dominance (VMD) calculation with these intermediate states leads to branching ratios $\sim1 \cdot10^{-5}$ and $\sim 3\cdot10^{-5}$ for $\rho\to\pi^0\pi^0\gamma$ and $\omega\to\pi^0\pi^0\gamma$, respectively [@Bramon1].
model VMD loops, $\sigma\gamma$ total
--------------- ------------------- ------------------------- ------------------------- -- -- -- --
$\chi PT$ $1.3\cdot10^{-5}$ $1.0\cdot10^{-5}$ $2.9\cdot10^{-5}$
$U\chi PT$ $1.5\cdot10^{-5}$ $1.5\cdot10^{-5}$ $4.2\cdot10^{-5}$
$L \sigma M$ $1.3\cdot10^{-5}$ $(0.8-2.1)\cdot10^{-5}$ $(2.8-4.7)\cdot10^{-5}$
$\sigma$ pole $1.0\cdot10^{-5}$ $\sim2\cdot10^{-3}$ $\sim2\cdot10^{-3}$
: The branching ratios for different contributions to $\rho\to\pi^0\pi^0\gamma$ decay. The chiral loop or $\rho\to\sigma\gamma$ contribution is calculated in the framework of Chiral Perturbation Theory ($\chi PT$) [@Bramon3; @Bramon2], Unitarized Chiral Perturbation Theory ($U\chi PT$) [@Marco; @Oset], Linear Sigma Model ($L \sigma M$) [@Bramon2] and $\sigma$ pole model [@Gokalp1].[]{data-label="theort"}
For $\rho\to\pi^0\pi^0\gamma$ decay another mechanism through the pions loops (Fig. \[diad\]b) is also possible [@Bramon1]. The branching ratios expected for this mechanism in different models [@Bramon3; @Bramon2; @Marco; @Oset] are listed in Table \[theort\]. It was noted in Ref. [@Marco] that the $\rho\to\pi^0\pi^0\gamma$ decay via chiral loops can be interpreted as $\rho\to\sigma\gamma$, where $\sigma$ is a scalar state decaying into $\pi\pi$ pair. The dependence of $B(\rho\to\pi^0\pi^0\gamma)$ on $\sigma$ parameters was studied in Ref. [@Bramon2]. The range of $B(\rho\to\pi^0\pi^0\gamma)$ values in L$\sigma M$ model corresponds to different $\sigma$ widths.
![Two mechanism of $\rho\to\pi^0\pi^0\gamma$ decay: (a) is the VMD mechanism, (b) is the transition through pion loop.[]{data-label="diad"}](diag.eps){width="98.00000%"}
Since the amplitudes of $\rho\to\omega\pi^0\to\pi^0\pi^0\gamma$ transition (Fig. \[diad\]a) and of the pion loops (Fig. \[diad\]b) are of the same order of magnitude, their interference is substantial. The interference contribution into branching ratio is predicted to be positive. The theoretical values for the total branching ratios are also listed in Table \[theort\]. The predictions of the chiral models [@Bramon3; @Bramon2; @Marco; @Oset] are in agreement with our previous experimental result. The significantly larger value of $B(\rho\to\pi^0\pi^0\gamma)$ was obtained in Ref. [@Gokalp1] using $\sigma$ pole model. Their result contradicts to existing experimental data.
In $\omega\to\pi^0\pi^0\gamma$ decay the contribution of pion loops is *G*-parity suppressed while the contribution of kaon loop is small due to large kaon mass. Therefore, it is assumed that the $\omega\to\pi^0\pi^0\gamma$ decay proceeds through $\rho^0\pi^0$ intermediate state. The first measurement $B(\omega\to\pi^0\pi^0\gamma)=(7.2\pm2.5)\cdot10^{-5}$ [@GAMS] significantly exceeded the existent prediction of VMD model: $3\cdot10^{-5}$ [@Bramon1]. An attempt to explain this discrepancy was done in Ref. [@Singer2] where $\rho\,$–$\,\omega$ mixing was taken into account and coupling constants were extracted from experimental values of $\Gamma(\omega\to 3\pi)$ and $\Gamma(\rho^0\to \pi^0\gamma)$. As a result the estimated value of $B(\omega\to\pi^0\pi^0\gamma)$ increased up to $(4.6\pm1.1)\cdot 10^{-5}$. Similar results (4.5–4.7)$\cdot 10^{-5}$ were then obtained in Refs. [@Bramon3; @Oset]. In Ref. [@Gokalp2] the large experimental value of $B(\omega\to\pi^0\pi^0\gamma)$ was explained by additional contribution of the $\omega\to\sigma\gamma$ transition and used to extract the value of $g_{\omega\sigma\gamma}$ coupling constant.
In the present work we present the experimental results on $B(\omega\to\pi^0\pi^0\gamma)$ and $B(\rho\to\pi^0\pi^0\gamma)$ based on full SND data sample.
Event selection
===============
For analysis five-photon events with the energy deposition in the calorimeter $$E_{tot}>0.7\cdot E
\label{etot}$$ and the total momentum measured by the calorimeter $$P_{tot}<0.15\cdot E/c
\label{ptot}$$ were selected. Here $E$ is $e^+e^-$ center of mass energy.
Due to high beam background rate in 5% of events fake photons appear. This makes possible for lower photon multiplicity QED processes $e^{+}e^{-} \to 2\gamma,\,3\gamma,4\gamma$, and $\rho,\omega\to \pi^0\gamma, \eta\gamma \to 3 \gamma$ decays to imitate five-photon events producing main background contribution for the process under study. Detector response to the beam background was studied using special events recorded with a random generator trigger. The information on the fired detector channels in these events was used for simulation of the process under study and the background processes. Considerable suppression (by a factor of 8) of the background from events with fake photons was achieved by imposing the following cuts: $$E_{min} > 30 \,\mbox{MeV},\:30^{\circ}<\theta_{min}<150^{\circ},
\label{emin}$$ where $E_{min}$ and $\theta_{min}$ are the energy and polar angle of the softest photon in an event. These cuts reduce the detection efficiency for the process under study by 25%. Another background source is the $e^{+}e^{-} \to\eta\gamma \to 3\pi^0\gamma \to 7\gamma$ reaction producing five-photon events mainly due to the merging of near photons. To suppress this background, the parameter $\chi_{\gamma}$ describing transverse energy deposition profile of the detected photon [@xinm] was used. The cut $$\chi_{\gamma}<5
\label{xinm}$$ suppresses the $e^{+}e^{-}\to\eta\gamma$ background by a factor of 2 with a 5% loss of actual 5-photon events.
Further selection was based on the kinematic fitting of the events. Compatibility of the event kinematics with $e^{+}e^{-} \to 5\gamma$ and $e^{+}e^{-} \to 3\gamma$ hypotheses was checked. For the $3\gamma$ hypothesis two out of five photons were considered spurious: all 3-$\gamma$ subsets were tested and the best one with minimum $\chi^2$ value was selected. As a result of kinematic fitting the $\chi^2$ values, $\chi_{5\gamma}$ and $\chi_{3\gamma}$, were calculated for both hypotheses. The cut $$\chi_{3\gamma}>20
\label{xi3g}$$ practically eliminates $e^{+}e^{-} \to 2\gamma,\,3\gamma$ background with the loss only 2.5% of the events of the process under study. Figure \[f1\] depicts the $\chi_{5\gamma}$ distribution of the experimental and simulated events. The following cut was imposed on this parameter: $$\chi_{5\gamma}<20.
\label{xi5g}$$
![The $\chi_{\pi\pi\gamma}-\chi_{5\gamma}$ distribution. The points with error bars are experimental data. The histogram is the sum of simulated events of the process (\[ppg\]) and background processes. The shaded histogram is the contribution of the background processes. []{data-label="f2"}](xi5g.eps){width="98.00000%"}
![The $\chi_{\pi\pi\gamma}-\chi_{5\gamma}$ distribution. The points with error bars are experimental data. The histogram is the sum of simulated events of the process (\[ppg\]) and background processes. The shaded histogram is the contribution of the background processes. []{data-label="f2"}](xifg.eps){width="98.00000%"}
Finally, the events with two $\pi^0$ mesons were selected. To do this the kinematic fit in $e^{+}e^{-} \to \pi^0\pi^0\gamma$ hypothesis was performed and the following cut was imposed: $$\chi_{\pi\pi\gamma}-\chi_{5\gamma}<10.
\label{xifg}$$ Here $\chi_{\pi\pi\gamma}$ is the $\chi^2$ value of the kinematic fit for the $e^{+}e^{-} \to \pi^0\pi^0\gamma$ hypothesis. The $\chi_{\pi\pi\gamma}-\chi_{5\gamma}$ distributions for the experimental and simulated events are shown in Fig.\[f2\].
The difference between distributions of the background events and events of the process under study (Figs. \[f1\] and \[f2\]) was used to estimate the accuracy of the background calculation. The experimental distribution was fitted by the sum of simulated distributions for the process (\[ppg\]) and background. As a result of the fit the ratio $K=N_{exp}^{bkg}/N_{calc}^{bkg}=0.7\pm0.2$ was determined. For softer selection criteria without cuts (\[emin\],\[xi3g\],\[xinm\]) this ratio is equal to $1.30 \pm 0.05$ and does not depend on the beam energy at our level of statistical accuracy. From this we conclude that the systematic error of the background estimation does not exceed 50%.
The total of 310 events were selected with estimated background of $15\pm7$ events. The main sources of the residual background are $e^+e^-\to\eta\gamma$ and $e^+e^-\to 4\gamma$ processes. The distribution of selected events and calculated background over center of mass energy is given in the Table \[t1\]. No events were detected below 600 MeV. The uncertainty of the center of mass energy, integrated luminosity, and detection efficiency are listed in Table \[t1\] for each energy point. The uncertainty of the center of mass energy includes the beam energy spread and the energy shift between 1998 and 2000 scans. The integrated luminosity was measured using $e^+e^-\to\gamma\gamma$ process. The statistical error of the luminosity in each energy point does not exceed 1% and is not included in the table. Its systematic error was estimated to be 3%. The detection efficiency for the process (\[ppg\]) was determined by simulation. The differential cross section of the $e^+e^-\to\pi^0\pi^0\gamma$ process calculated in VMD model [@ompn] was used for simulation. The systematic error of the detection efficiency, including the model error due to possible contribution from $\sigma\gamma$ intermediate state was estimated to be 5%.
$E$,MeV $\sigma_E$,MeV L, nb$^{-1}$ $N_{exp}$ $N_{bkg}$ $\varepsilon$ $1+\delta$ $\sigma^{exp}$,nb
--------- ---------------- -------------- ----------- ----------- --------------- ------------ ----------------------------
600.1 0.29 88.3 0 0.1 0.273 0.912 $-0.005_{-0.002}^{+0.052}$
630.1 0.30 116.1 0 0.1 0.269 0.906 $-0.004_{-0.002}^{+0.041}$
660.2 0.25 271.6 2 0.3 0.273 0.900 $ 0.025_{-0.019}^{+0.040}$
690.2 0.29 167.2 2 0.2 0.263 0.895 $ 0.046_{-0.033}^{+0.067}$
720.3 0.26 588.5 1 0.8 0.251 0.892 $ 0.002_{-0.007}^{+0.018}$
750.2 0.32 219.0 3 0.1 0.259 0.897 $ 0.057_{-0.032}^{+0.057}$
760.2 0.31 238.9 2 0.3 0.251 0.896 $ 0.032_{-0.024}^{+0.049}$
764.2 0.32 250.4 5 0.2 0.254 0.892 $ 0.085_{-0.038}^{+0.060}$
770.2 0.31 284.4 8 0.3 0.253 0.877 $ 0.122_{-0.044}^{+0.063}$
774.2 0.34 217.1 7 0.2 0.252 0.855 $ 0.145_{-0.055}^{+0.081}$
778.1 0.34 247.9 6 0.5 0.261 0.820 $ 0.104_{-0.045}^{+0.068}$
780.2 0.35 319.5 16 1.2 0.263 0.807 $ 0.218_{-0.059}^{+0.075}$
781.1 0.33 339.6 20 1.2 0.267 0.807 $ 0.257_{-0.061}^{+0.076}$
782.1 0.31 656.3 34 1.0 0.257 0.815 $ 0.240_{-0.042}^{+0.050}$
783.2 0.30 473.4 30 2.0 0.253 0.833 $ 0.280_{-0.055}^{+0.066}$
784.2 0.32 346.2 24 0.7 0.261 0.857 $ 0.301_{-0.063}^{+0.077}$
785.3 0.24 212.3 12 0.4 0.257 0.890 $ 0.238_{-0.070}^{+0.094}$
786.1 0.33 267.7 11 0.4 0.255 0.914 $ 0.170_{-0.052}^{+0.071}$
790.1 0.34 191.4 4 0.3 0.258 1.006 $ 0.074_{-0.039}^{+0.064}$
794.2 0.34 206.7 1 0.2 0.256 1.044 $ 0.014_{-0.015}^{+0.042}$
800.2 0.32 276.8 10 0.3 0.255 1.053 $ 0.130_{-0.042}^{+0.057}$
810.2 0.34 279.5 3 0.4 0.240 1.043 $ 0.037_{-0.024}^{+0.030}$
820.1 0.36 315.2 2 0.3 0.244 1.035 $ 0.021_{-0.016}^{+0.033}$
840.2 0.35 677.5 8 0.8 0.247 1.025 $ 0.042_{-0.016}^{+0.023}$
880.0 0.41 376.0 7 0.5 0.222 1.001 $ 0.078_{-0.031}^{+0.045}$
919.9 0.44 478.6 8 0.3 0.256 0.916 $ 0.069_{-0.025}^{+0.035}$
939.9 0.43 469.0 22 0.7 0.248 0.856 $ 0.214_{-0.047}^{+0.058}$
949.7 0.32 261.7 20 0.3 0.261 0.855 $ 0.338_{-0.076}^{+0.095}$
957.7 0.32 233.9 13 0.2 0.263 0.858 $ 0.242_{-0.067}^{+0.089}$
969.7 0.34 251.5 29 0.5 0.250 0.865 $ 0.524_{-0.099}^{+0.119}$
: Center of mass energy $E$, its standard deviation $\sigma_E$, integrated luminosity $L$, number of selected events $N_{exp}$, calculated number of background events $N_{bkg}$, detection efficiency $\varepsilon$, radiative correction $1+\delta$ and the $e^+e^-\to\pi^0\pi^0\gamma$ cross section $\sigma^{exp}$. []{data-label="t1"}
Fitting of the cross section
============================
The fitting procedure maximizes the logarithmic likelihood function $$L=\sum_i \ln{P_i(N_i^{exp}, N_i^{th})},$$ where $P_i$ is a Poisson probability to detect observed number of events $N_i^{exp}$ in the $i$-th energy bin with a theoretical expectation of $N_i^{th}$. The theoretical expectations were calculated as $$N_i^{th}=\varepsilon_i L_i \sigma(E_i) (1+\delta(E_i))+N_i^{bkg},$$ where $N_i^{bkg}$ is a calculated number of background events, $\varepsilon_i$ is a detection efficiency, $L_i$ is the integrated luminosity, $\sigma(E)$ is $e^+e^-\to\pi^0\pi^0\gamma$ cross section depending on a set of approximation parameters, $\delta$ is a radiative correction. The radiative correction, which is a functional of the cross-section energy dependence $\sigma(E)$ [@radc], was determined within the fitting procedure. The values of radiative correction evaluated for each experimental energy point are listed in Table \[t1\]. The model error of the $(1+\delta)$ value does not exceed 3%. The values of the experimental cross section calculated as $$\sigma_i^{exp}=\frac{N_i^{exp}-N_i^{bkg}}{\varepsilon_i L_i (1+\delta(E_i))}$$ are shown in Fig.\[f6\] and listed in Table \[t1\]. The systematic error of the cross section is determined by the the errors of the detector efficiency, integrated luminosity, and radiative correction. It was estimated to be 7%.
![The cross section of the $e^+e^-\to\pi^0\pi^0\gamma$ process. Points with error bars are experimental data. Solid line is a result of fitting in the model 3 of Table \[t4\]. The dashed line corresponds to the fit with $B(\rho\to\sigma\gamma)=0$. []{data-label="f6"}](cst03.eps){width="90.00000%"}
To calculate the cross section $\sigma(E)$ the amplitude of the $e^+e^-\to\pi^0\pi^0\gamma$ process was parametrized as $$A_{\pi\pi\gamma}=A_{\rho\omega\pi}
(BW_\rho+\alpha_1 BW_{\rho^\prime}
+\alpha_2 BW_{\rho^{\prime\prime}})+
\beta A_{\rho\sigma\gamma}BW_\rho +
\gamma A_{\omega}BW_\omega, \label{ampl}$$ The first term in Eq.(\[ampl\]) is the amplitude of the $e^+e^-\to \rho,\rho^\prime,\rho^{\prime\prime} \to \omega\pi^0$, where $\rho^\prime$ and $\rho^{\prime\prime}$ are excitations of the $\rho(770)$. Second and third terms are $e^+e^-\to\rho(770)\to\sigma\gamma\to\pi^0\pi^0\gamma$ and $e^+e^-\to\omega\to\pi^0\pi^0\gamma$ amplitudes. Each amplitude is written in a factorized form. The functions $A_{\rho\omega\pi^0}$, $A_{\rho\sigma\gamma}$, $A_{\omega}$ depending on the momenta of final particles describe the dynamics of vector mesons decays. The functions $BW_i$ describe the Breit-Wigner resonance shapes: $$BW_i=\frac{m_i^2}{m_i^2-E^2-iE\Gamma_i(E)},\;\; i=\rho, \rho^\prime, \rho^{\prime\prime}, \omega.$$ Here $m_i$ and $\Gamma_i(E)$ are resonance mass and energy dependent width. The cross section is calculated from Eq.(\[ampl\]) by integration over the phase space of final particles: $\sigma(E)=\int |A_{\pi\pi\gamma}|^2 d \Pi$. At the energy above $\omega\pi$ threshold the Breit-Wigner functions of $\rho$ mesons are modified $BW_{\rho_i} \to BW_{\rho_i} C_{\rho_i\omega\pi}$, where $C_{\rho_i\omega\pi}$ are Blatt-Weisskopf factors, restricting fast growth of the $\Gamma_{\rho_i\omega\pi}$ partial widths [@Clegg]: $$C_{\rho\omega\pi}=\frac{1}{\sqrt{1+(Rq_\omega(E))^2}},\:
C_{\rho_i\omega\pi}=
\sqrt{\frac{1+(Rq_\omega(m_{\rho_i}))^2}
{1+(Rq_\omega(E))^2}},\; \rho_i=\rho^\prime,\rho^{\prime\prime}.
\label{Compi}$$ Here $q_\omega$ is the $\omega$ meson momentum in $\rho_i\to\omega\pi$ decay. The range parameter $R$ is supposed to be the same for $\rho, \rho^\prime, \rho^{\prime\prime}$ mesons. The main decay modes of $\rho$ mesons were taken into account for calculation of the energy dependence of the resonance widths. For instance, in the case of $\rho(770)$ we use the following expression: $$\Gamma_{\rho}(E)=\Gamma_{\rho}(m_\rho)
\biggl(\frac{m_\rho}{E}\biggr)^2
\biggl(\frac{q_\pi(E)}{q_\pi(m_\rho)}\biggr)^3 C_{\rho\pi\pi}^2+
\frac{g_{\rho\omega\pi}^2}{12\pi}\, q_\omega^3(E)\, C_{\rho\omega\pi}^2.
\label{grho}$$ Here $q_\pi$ is a pion momentum in the $\rho\to 2\pi$ decay. The Blatt-Weisskopf factor $C_{\rho\pi\pi}$ is expressed by the formula similar to Eq.(\[Compi\]).
The amplitude of the $e^+e^-\to\omega\pi$ process is described by formulas from Ref. [@ompn; @phiompi] and depends on 8 parameters. The data from energy region below 1 GeV are insufficient to determine them. Therefore we used additional measurements of the $e^+e^-\to\omega\pi^0$ cross section in the 1–1.4 GeV energy range by SND [@phiompi] and of the $\omega\pi^0$ spectral function in $\tau\to\omega\pi\nu_\tau$ decay by CLEO [@CLEOt]. The spectral function can be converted to corresponding production cross section in $e^+e^-$ collisions using CVC hypothesis [@CVC]. The all-data fit on $e^+e^-\to\pi^0\pi^0\gamma$ cross section is shown in Fig.\[f7\].
![The cross section of the $e^+e^-\to\pi^0\pi^0\gamma$ reaction. Points with error bars are experimental data. Curve is a fit result in the model 3 of Table \[t4\].[]{data-label="f7"}](cst01.eps){width="90.00000%"}
For $E>1\,\mathrm{GeV}$ experimental data are well described by three models with parameters listed in Table \[t2\].
$g_{\rho\omega\pi}$ $m_{\rho^\prime}$ $\Gamma_{\rho^\prime}$ $\alpha_1$ $m_{\rho^{\prime\prime}}$ $\Gamma_{\rho^{\prime\prime}}$ $\alpha_2$
--- --------------------- ------------------- ------------------------ -------------- --------------------------- -------------------------------- --------------
1 14.3-15.8 — — — 1630-1710 630-1000 -(0.19-0.24)
2 14.1-15.7 1400 500 -(0.04-0.06) 1580-1620 420-580 -(0.14-0.18)
3 15.4-16.6 1400 500 -(0.39-0.42) 1560-1640 380-780 0.24-0.30
: Parameters of the models describing $e^+e^-\to\omega\pi\to\pi^0\pi^0\gamma$ cross section for $E>1\,\mathrm{GeV}$.[]{data-label="t2"}
The ranges of parameter values correspond to the variation of the parameter $R$ from 0 to 2 GeV$^{-1}$. The statistical errors are not shown because they are significantly smaller than model biases. For models with two excited $\rho$ states the $\rho^\prime$ mass and width were fixed to 1400 MeV and 500 MeV. These values are close to $\rho^\prime$ parameters from $\pi^\pm \pi^0$ spectral function data [@ALEPH; @CLEO2pi].
Below $\omega\pi^0$ threshold the amplitude of the $e^+e^-\to\omega\pi^0$ process drops rapidly and the product $|BW_\rho|^2\int |A_{\rho\omega\pi}|^2 d\Pi$ in contrast with the corresponding product for $\rho\to \sigma\gamma$ transition does not demonstrate resonance behavior. This allows to separate contributions of the two $\rho$ decay mechanisms by measurement of energy dependence of the $e^+e^-\to\pi^0\pi^0\gamma$ cross section. The $\rho\to\sigma\gamma$ decay amplitude was described by the $\chi PT$ and $L\sigma M$ models from Ref. [@Bramon2]. The three sets of $\sigma$ parameters [@E791; @CLEO; @Bramon2] used in the $L\sigma M$ model are listed in Table \[t4\]. The $\chi PT$ model corresponds to $m_\sigma \to \infty$. The $\beta$ parameter represents the difference between observed value of the $\rho\to\sigma\gamma$ decay amplitude and theoretical prediction.
For $\omega\to\pi^0\pi^0\gamma$ decay the variation of the final state phase in the CMS energy interval of the $\omega$ meson is small, so we cannot separate different decay mechanisms studying the cross section energy dependence. Thus the amplitude of the $\omega\to\pi^0\pi^0\gamma$ decay was written according to VMD model [@ompn]. The $\rho$-$\omega$ mixing was taken into account following Ref. [@Bramon2]. Possible contributions of other mechanisms would result in a deviation of the complex parameter $\gamma$ from 1.
Full description of the energy dependence of the cross section below 1 GeV requires extra four parameters e.g. absolute values and phases of $\beta$ and $\gamma$: $|\beta|$, $\phi_\beta$, $|\gamma|$, $\phi_\gamma$. But we prefer two other sets of parameters: $B(\rho\to\sigma\gamma)$, $\phi_\beta$, $B(\omega\to\pi^0\pi^0\gamma)$, $\phi_\gamma$ or $B(\rho\to\pi^0\pi^0\gamma)$, $\phi_\beta$, $B(\omega\to\pi^0\pi^0\gamma)$, $\phi_\gamma$. The branching ratios are related to $\beta$ and $\gamma$ as $$B(\rho\to\pi^0\pi^0\gamma)=\frac{m_\rho^2}{\Gamma_\rho^2}
\frac{1}{\sigma_\rho} \int |A_{\rho\omega\pi}(m_\rho)+
\beta A_{\rho\sigma\gamma}(m_\rho)|^2 d\Pi,$$ $$B(\rho\to\sigma\gamma)=\frac{m_\rho^2}{\Gamma_\rho^2}
\frac{1}{\sigma_\rho} \int |
\beta A_{\rho\sigma\gamma}(m_\rho)|^2 d\Pi,$$ $$B(\omega\to\pi^0\pi^0\gamma)=\frac{m_\omega^2}{\Gamma_\omega^2}
\frac{1}{\sigma_\omega}\int |
\gamma A_{\omega}(m_\rho)|^2 d\Pi,$$ where $\sigma_V=12\pi B(V\to e^+e^-)/M_V^2$ is a total vector meson production cross section in $e^+e^-$ collisions.
Characteristic feature of the process under study is a large interference between the contributions of $\rho$ and $\omega$ decays. For instance the cross section of $e^+e^-\to\omega\to\pi^0\pi^0\gamma$ process at $E=m_\omega$ evaluated using the table value of $B(\omega\to\pi^0\pi^0\gamma)=(7.2\pm 2.5)\cdot10^{-5}$ is equal to 0.12 nb. The interference with $\rho$ meson increases this value up to approximately 0.25 nb (Fig.\[f6\]). The experimental data on the energy dependence of the cross section are insufficient for determination of unambiguous solution for interference phases $\phi_\beta$, $\phi_\gamma$. There are four solutions listed in Table \[t3\].
$B(\omega\to\pi^0\pi^0\gamma)$ $B(\rho\to\pi^0\pi^0\gamma)$ $B(\rho\to\sigma\gamma)$ $P(\chi^2)$
--- --------------------------------------------------- ------------------------------ -------------------------- --------------------- ----- -- --
1 $\phi_\beta\approx 0$, $\phi_\gamma\approx 0$ $6.3_{-1.3}^{+1.4}$ $4.1_{-0.9}^{+1.0}$ $1.9_{-0.8}^{+0.9}$ 35%
2 $\phi_\beta\approx \pi$, $\phi_\gamma\approx 0$ $12.3_{-1.6}^{+2.3}$ $3.8_{-0.8}^{+0.9}$ $4.4\pm1.0$ 30%
3 $\phi_\beta\approx 0$, $\phi_\gamma\approx \pi$ $25.5_{-2.3}^{+2.4}$ $5.1_{-0.9}^{+1.0}$ $1.9\pm1.0$ 6%
4 $\phi_\beta\approx \pi$, $\phi_\gamma\approx \pi$ $15.8\pm 2.3$ $4.7_{-0.8}^{+0.9}$ $5.6_{-1.0}^{+1.1}$ 8%
: The branching ratios of $\rho$ and $\omega$ decays ($B\times 10^{5}$) and $P(\chi^2)$ values obtained as a result of cross section fitting with different values of $\phi_\beta$, $\phi_\gamma$. The $\omega\pi$ amplitude was described by the Model 3 from Table \[t2\]. Model 1 from Table \[t4\] was used for description of $\rho\to\sigma\gamma$ amplitude. Only statistical errors of parameters are shown. []{data-label="t3"}
The third and fourth ones correspond to a large destructive contribution into $\omega$ decay from mechanisms other than $\rho^0\pi^0$. The $B(\omega\to\pi^0\pi^0\gamma)$ values obtained in this case disagree with existing experimental value $B(\omega\to\pi^0\pi^0\gamma)=(7.2\pm 2.5)\cdot10^{-5}$. The solution with $\phi_\beta\approx \pi$, $\phi_\gamma\approx 0$ can be ruled out for two reasons: $B(\omega\to\pi^0\pi^0\gamma)$ exceeds the table value by 1.7 standard deviations and consistency of the calculated spectrum of the recoil photon with the experimental one is poor. The analysis of the photon spectrum is described in the next section.
For the only survivor solution with both phases close to zero, the model dependence of the fit parameters was studied. Three models of excited $\rho$ states (Table \[t2\]) and four sets of $\sigma$ parameters (Table \[t4\]) were tested. The $\phi_\beta$ was found ranging within $(20\div 80)^\circ \pm 80^\circ$. These values are in agreement with theoretically expected zero value [@Bramon2]. Therefore the final fitting was performed with $\phi_\beta=0$. The phase $\phi_\gamma$ was considered as a floating parameter to take into account its possible shift due to the contribution of mechanisms other than $\omega\to\rho\pi^0$. The fitted $\phi_\gamma=-(2\div 20)^\circ\pm 20^\circ$ is consistent with zero.
The probabilities of the $\rho$ and $\omega$ decays into $\pi^0\pi^0\gamma$ obtained with different parameters of $\sigma$ meson are listed in Table \[t4\].
------------ ------------ ----------------- -------------------------------- --------- ----- --------- -----
Model $m_\sigma$ $\Gamma_\sigma$ $B(\omega\to\pi^0\pi^0\gamma)$
MeV MeV exp. th. exp. th.
L$\sigma$M 478 324 6.2-6.5 4.0-4.1 3.8 1.5-1.9 1.5
L$\sigma$M 555 540 6.3-6.6 4.2-4.3 2.8 1.9-2.3 0.8
L$\sigma$M 478 263 6.2-6.4 4.2-4.3 4.7 1.7-2.1 2.1
$\chi$PT — — 6.5-6.9 3.9-4.0 2.9 1.8-2.2 1.0
------------ ------------ ----------------- -------------------------------- --------- ----- --------- -----
: The probabilities of the $\rho$ and $\omega$ decays into $\pi^0\pi^0\gamma$ ($B\times 10^{5}$) obtained with different parameters of $\sigma$ meson. The spreads in the parameter values correspond to the models listed in Table \[t2\]. Theoretical values of branching ratios are taken from Ref. [@Bramon2]. The final results with statistical and systematic errors are given in the bottom line of the table. []{data-label="t4"}
The spreads in parameter values correspond to different models describing $e^+e^-\to\omega\pi^0$ cross section above 1 GeV. All models reproduce the experimental data well. Therefore parameter midrange was taken as a final result. Its spread was regarded as the model error. The branching ratios obtained this way with statistical and systematic errors are listed in the last row of Table \[t4\]. The systematic error includes the model error, uncertainties in the detection efficiency and integrated luminosity. The variation of the background level within its systematic error practically does not change the $B(\omega\to\pi^0\pi^0\gamma)$ central value and results in following additional uncertainties of the $\rho$ meson branching ratios: 7% for $B(\rho\to\pi^0\pi^0\gamma)$ and 12% for $B(\rho\to\sigma\gamma)$. Since these uncertainties affect statistical significance of the results they were added to statistical errors. The energy dependence of the cross section in the model with $m_\sigma=478$ MeV and $\Gamma_\sigma=324$ MeV is shown in Fig. \[f6\] together with the curve corresponding to $B(\rho\to\sigma\gamma)=0$. The $P(\chi^2)$ value for the latter model is equal to 0.5%.
The energy and angular spectra
==============================
From Table \[t1\] it is seen that selected events are mainly concentrated in two energy regions: 180 events near $\omega$ peak and 92 events in the range 920–970 MeV above the $e^+e^-\to\omega\pi^0$ reaction threshold. The angular and energy distributions in the latter region agree with $\omega\pi^0$ mechanism.
![Distribution of cosine angle between photon and $\pi^0$ meson in the $\pi^0\pi^0$ rest frame. Point with error bars are experimental data, Solid line is a simulation with $\omega$ decay into pure $\rho^0\pi^0$ state. Dashed line is a simulation with $\omega$ decaying into a mixture of $\rho^0\pi^0$ and $\sigma\gamma$ states. []{data-label="f4"}](egfg.eps){width="98.00000%"}
![Distribution of cosine angle between photon and $\pi^0$ meson in the $\pi^0\pi^0$ rest frame. Point with error bars are experimental data, Solid line is a simulation with $\omega$ decay into pure $\rho^0\pi^0$ state. Dashed line is a simulation with $\omega$ decaying into a mixture of $\rho^0\pi^0$ and $\sigma\gamma$ states. []{data-label="f4"}](difg.eps){width="98.00000%"}
Recoil photon spectrum for events from 760–800 MeV energy range is shown in Fig.\[f3\]. Although this energy region is dominated by $\omega$ peak the contributions of both $\omega$ and $\rho$ decays must be taken into account to obtain the theoretical spectrum. The spectrum calculated in model 3 from Table \[t4\] (Fig.\[f3\]) is in a good agreement with the experimental one. In this model it is supposed that $\omega\to\pi^0\pi^0\gamma$ decay proceeds through $\rho^0\pi^0$ intermediate state. Another way is to describe the $\omega$ decay by a sum of contributions of $\omega\to\rho^0\pi^0$ and $\omega\to\sigma\gamma$ mechanisms. To do this we fix $B(\omega\to\rho^0\pi^0\to\pi^0\pi^0\gamma)$ at $2.5\cdot 10^{-5}$ and fit the $\omega\to\sigma\gamma$ decay contribution to a value yielding observed $\omega\to\pi^0\pi^0\gamma$ branching ratio. In case of constructive interference this leads to $B(\omega\to\sigma\gamma\to\pi^0\pi^0\gamma)=1.3\cdot 10^{-5}$ and the photon spectrum close to that expected for $\rho^0\pi^0$ mechanism. As was shown in the previous section, assumption of destructive interference results in $B(\omega\to\pi^0\pi^0\gamma)$ inconsistent with the PDG table value.
The second theoretical spectrum in Fig.\[f3\] corresponds to the model 2 from Table \[t3\] with destructive interference of $\omega\pi$ and $\sigma\gamma$ amplitudes in $\rho$ decay. For this model the consistency between theoretical and experimental spectra calculated using Kolmogorov test [@hbook] is about 1%, which was one of the reasons to discard this model.
Additional information about mechanism of $\omega$ decay can be obtained from the analysis of angular distributions. One of such distribution is shown in Fig.\[f4\]. The same figure displays the theoretical distributions obtained under assumptions that $\omega$ decay proceeds through either pure $\rho^0\pi^0$ intermediate state or a mixture of $\rho^0\pi^0$ and $\sigma\gamma$. One can see that our limited statistics does not allow to distinguish these two models.
Summary
=======
The branching ratios measured in this work, $$B(\omega\to\pi^0\pi^0\gamma)=(6.6_{-1.3}^{+1.4}\pm0.6)\cdot10^{-5},
\label{oppg}$$ $$B(\rho\to\pi^0\pi^0\gamma)=(4.1^{+1.0}_{-0.9}\pm0.3)\cdot10^{-5}.
\label{rppg}$$ are in a good agreement with our preliminary results [@rhoppg] and GAMS measurement $B(\rho\to\pi^0\pi^0\gamma)=(7.2\pm2.5)\cdot10^{-5}$ [@GAMS], but have higher accuracy.
The probability of $\rho\to\pi^0\pi^0\gamma$ decay significantly exceeds VMD model prediction $(1.3-1.5)\cdot10^{-5}$. This excess can be explained by the contribution of the decay via scalar state $\rho\to\sigma\gamma$. Two mechanisms, $\rho\to\sigma\gamma$ and $\rho\to\omega\pi$, can be separated using difference in energy dependence of their amplitudes. Our result on $\rho\to \sigma\gamma$ decay $$B(\rho\to \sigma\gamma\to\pi^0\pi^0\gamma)=(1.9_{-0.8}^{+0.9}\pm0.4)\cdot10^{-5}$$ differs from zero by 2.4 standard deviations and is consistent with the predictions of chiral models [@Bramon2; @Oset]. The magnitude of $B(\rho\to \sigma\gamma)$ is sensitive to $\sigma$ parameters. As can be seen from Table \[t4\], the models with $\Gamma_\sigma\sim300$ MeV give the most consistent description of the experimental data.
The value of the branching ratio $B(\omega\to\pi^0\pi^0\gamma)=(6.7\pm1.2)\cdot10^{-5}$, obtained by averaging of our measurement with the GAMS result exceeds theoretical predictions, $(4.6\pm1.1)\cdot10^{-5}$ [@Bramon2; @Singer2] and $(4.7\pm0.9)\cdot10^{-5}$ [@Oset], by 1.3 standard deviations. It is necessary to make some remarks about these predictions. The result of Ref. [@Singer2] is based on table value of $\Gamma(\rho\to\pi^0\gamma)=102\pm26$ keV [@PDG2000]. It must be corrected taking into account newer measurement $\Gamma(\rho\to\pi^0\gamma)=76\pm22$ keV [@p0g], which is close to the value for charged $\rho$, $\Gamma(\rho^\pm\to\pi^\pm\gamma)=68\pm8$ keV [@PDG2000]. This decreases the predicted $B(\omega\to\pi^0\pi^0\gamma)$ and worsens agreement with the experiment. In the Refs. [@Bramon2] and [@Oset] the values of $g_{\rho\omega\pi}$ equal to 15 GeV$^{-1}$ and 15.9 GeV$^{-1}$ were used to calculate $B(\omega\to\pi^0\pi^0\gamma)\propto g_{\rho\omega\pi}^2
g_{\rho\pi\gamma}\propto g_{\rho\omega\pi}^4/g_{\rho}^2$. On the other hand, the use of these $g_{\rho\omega\pi}$ values for VMD calculation of $\Gamma(\omega\to 3\pi)$ and $\Gamma(\omega\to \pi^0\gamma)$ leads to too large values conflicting with experimental data. For example, the $g_{\rho\omega\pi}$ obtained from $\omega\to 3\pi$ width assuming intermediate $\rho\pi$ state is equal to $(14.3\pm0.2)$ GeV$^{-1}$ [@ompn]. Therefore, our opinion is that $4.6\cdot 10^{-5}$ is the maximum branching ratio acceptable within VMD model and the additional theoretical study is required to explain the large value of $B(\omega\to\pi^0\pi^0\gamma)$.
Acknowledgments
===============
This work is supported by “Russian Fund for Basic Researches” grants No. 01-02-16934-a and 00-15-96802 and STP “Integration” Fund, grant No. A0100.
[20]{} M.N. Achasov et al., Nucl. Instr. Meth. A 449 (2000) 125. M.N. Achasov et al., JETP Letters, 71 (2000) 355. D. Alde et al., Phys. Lett. B 340 (1994) 122. P. Singer, Phys. Rev. 128 (1962) 2789. A. Bramon, A. Grau, G. Pancheri, Phys. Lett. B 283 (1992) 416. A. Bramon, A. Grau, G. Pancheri, Phys. Lett. B 289 (1992) 97. A. Bramon et al., Phys. Lett. B 517 (2001) 345. E. Marco et al., Phys. Lett. B 470 (1999) 20. J. E. Palomar, S. Hirenzaki, E. Oset, e-print hep-ph/0111308. A. Gokalp, O. Yilmaz, Phys. Lett. B 508 (2001) 25. D. Guetta, P. Singer, Phys. Rev. D 63 (2001) 017502. A. Gokalp, O. Yilmaz, Phys. Lett. B 494 (2000) 69. A.V. Bozhenok, V.N. Ivanchenko, Z.K. Silagadze, Nucl. Instr. Meth. A 379 (1996) 507. M.N. Achasov et al., Nucl. Phys. B 569 (2000) 158. E.A. Kuraev, V.S. Fadin, Sov. J. Nucl. Phys. 41 (1985) 466. A.B. Clegg, A. Donanachie, Z. Phys. C 62 (1992) 455. M.N. Achasov et al., Phys. Lett. B 486 (2000) 29. K.W. Edwards et al., Phys. Rev. D 61 (2000) 072003. Y.S. Tsai, Phys. Rev. D 4 (1971) 2821. R. Barate et al., Z. Phys. C 76 (1997) 15. S. Anderson et al., Phys. Rev. D 61 (2000) 112002. E.M. Aitala et al., Phys. Rev. Lett. 86 (2001) 770. D.M. Asner et al., Phys. Rev. D 61 (2000) 012002. HBOOK Reference Manual, CERN program library Y250, 1998, p.90. Review of Particle Physics. Eur. Phys. J. C 15 (2000) 1. M.N. Achasov et al., e-print hep-ex/0109035.
[^1]: e-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We study tree metrics that can be realized as a mixture of two star tree metrics. We prove that the only trees admitting such a decomposition are the ones having only one internal edge and, moreover, certain relations among the weights assigned to all edges must hold. We also describe the fibers of the corresponding mixture map. In addition, we discuss the general framework of tropical secant varieties and we interpret our results within this setting. Finally, after discussing recent results on upper bounds on star tree ranks of metrics on $n$ taxa, we show that analogous bounds for star tree metric ranks cannot exist.'
address: 'Department of Mathematics, University of California, Berkeley, CA 94720, USA'
author:
- María Angélica Cueto
bibliography:
- 'bibliography.bib'
title: Tropical mixtures of star tree metrics
---
[^1]
Introduction {#s:Intro}
============
In the present paper we investigate tropical mixtures of two *star tree metrics*. Here star tree metrics on $n$ taxa are star trees with $n$ leaves labeled 1 through $n$, equipped with positive weights on all its edges. Tropical mixture of two metrics $D, \bar{D}$ on $n$ taxa are defined as $(D\oplus {\bar{D}})_{(i,j)}=\max\{ D_{(i,j)},
{\bar{D}}_{(i,j)}\}$ for all $1\leq i,j\leq n$, that is point-wise maxima of metrics on $n$ taxa. This study was inspired by the recent work of E. Matsen, E. Mossel and M. Steel ([@MatSt], [@MMS]), as well as by some open questions regarding tropical secant varieties of linear spaces ([@tropicalDevelin], [@tropicalDraisma].)
Matsen and his collaborators consider phylogenetic mixtures (convex combinations of site pattern frequencies) of two weighted trees of the same topological type, and they characterize the conditions under which these give rise to weighted trees of a different topology. In particular, they pay special attention to mixtures of star tree metrics on four taxa. Their interest in studying these mixtures lies in a phenomenon called “branch repulsion”. If we mix together two weighted trees on four taxa with the same topology, with short internal branch lengths and with pendant edges alternating being long and short, we are likely to get a tree with a different topology. Since tree topologies are characterized by the subtrees spanned by four of its leaves, the case of four taxa suffices to study trees on more taxa. Intuitively, from the branch repulsion behavior we see that we can approximate any tree topology we want by mixing together two star tree metrics. Thus the interest in understanding mixtures of star tree metrics.
In our case, rather than being interested in phylogenetic mixtures we focus our attention on *tropical mixtures* of star tree metrics. Tropical mixtures arise naturally in the context of Probability Theory as approximations of mixtures of multivariate Poisson distributions ([@ASCB Ch. III]) and logdet transforms defined for Markov models on trees ([@Phylog Thm 8.4.3], [@ReconstructingTrees].)
The convex mixture maps of two star tree metrics studied by Matsen et al. correspond, in the language of Algebraic Geometry, to secant varieties of the space of star tree metrics. If we allow negative weights for the edges of our star trees, then mixtures of these objects correspond to secant varieties of the space of star trees. Applying standard techniques in Tropical Geometry, one can tropicalize the convex mixture map, to obtain a piecewise linear map: the *tropical mixture map*. Its image will be the tropical first secant of the tropicalization of the space of star trees, which is an $n$-dimensional linear space ([@tropicalDevelin].)
However, when dealing with *star tree metrics* (i.e. phylogenetic star trees) we should consider tropical secants of polyhedral cones instead of tropical secants of ordinary linear spaces. The linear space, in this case, will be the linear span of the cone of star tree metrics. The extremal rays of this cone are the cut metrics $d_i$ assigning $d_i(i,j)=1$ for all $j\neq i$ and where all other distances are zero. As one can imagine, the sign constraints for the scalars reflect the difference between star trees and star tree metrics: the sign constraint on the edge weights. While star trees admit negative weights for their edges, this is forbidden in the biological framework of star tree metrics.
One interesting question that arises from these two settings involves *star tree* and *star tree metric* ranks of a symmetric matrix. By relaxing the positivity condition, one can show that every metric on $n$ taxa can be written as the tropical mixture of $n-2$ star trees ([@DM].) The analogous question for metrics and star tree metrics seems more delicate. Indeed, we prove that most *cut metrics* on $n$ taxa (defined by partitions of the $n$ taxa, assigning pairwise distance 0 to pairs of taxon on the same subset, and distance 1 otherwise) are not tropical mixtures of finitely many star tree metrics. This implies that upper bounds for star tree metric ranks cannot exist.
The paper is organized as follows. In Section 2 we provide the basic definitions on distances over finite sets and tree metrics, and we present our main result: a complete characterization of the intersection of the space of tree metrics and the image of the mixture map for star trees, together with a description of its fibers (Theorem \[thm:mainResult\].) All proofs are deferred to Section \[s:proofs\]. Since log-limits of secant varieties of toric varieties correspond to tropical secant varieties of linear spaces, one should expect similar results between convex mixtures of star tree metrics and tropical mixtures of star tree metrics. Our characterization shows this connection.
In Section \[s:TropicalSecantVarieties\] we switch gears and we describe the general framework of tropical secants of linear spaces as treated in [@tropicalDevelin]. In the case of star trees with arbitrary weights, the associated linear space is spanned by $\{r_i=\sum_{j<i} e_{ji}+\sum_{j>i}e_{ij}, 1\leq i \leq n\}$. In the spirit of [@tropicalDevelin], we can associate to this linear space a point configuration in ${\mathbb{R}}^{\binom{n}{2}}$. In this case, this configuration is nothing else than the set of vertices of the second hypersimplex $\Delta(2,n)$. We refer to [@GBandCP Ch. 9], [@GKZ Ch. 6 $\S$3] for historical background and basic properties of this well-studied polytope. Tropical secants of star tree metrics will have the same underlying point configuration, but the crucial difference between these two settings will lie in the admissible regular subdivisions of $\Delta(2,n)$, which we call *positive regular subdivisions*. We show they are different even in the simplest case of trees on four taxa. The existence of metrics with infinite star tree metric rank implies that regular subdivisions of $\Delta(2,n)$ need not be positive. Thus, the problem of characterizing metrics with finite tree metric ranks cannot be approached by studying star tree ranks (see [@ASCB p 124] for a conjecture on this topic.) We finish the paper with several open questions on tropical secant varieties of polyhedral cones, which provide the natural setting to investigate mixtures of tree metrics.
Basic definitions and main result {#s:BasicDefinition}
=================================
We begin by providing notation and basic definitions. We fix $[n]=\{1,
\ldots, n\}$.
A *tree on $n$ taxa* is an unrooted tree (i.e. a connected graph with no cycles) with leaves labeled by $[n]$. We let $E(T)$ be the edges of $T$.
It is well-known that trees are completely determined by their list of quartets, i.e. subtrees spanned by four leaves. For example, quartet $(ij|kl)$ represents the subtree in Figure \[fig:quartet\].
![Subtree determined by quartet $(ij|kl)$.[]{data-label="fig:quartet"}](tree12.mps)
Many authors have discussed this tree representation and the minimum number of quartets required to characterize the tree topology ([@314547], [@quartetRepresentation], [@treeOfLife $\S
5.4.2$].)
Given a tree $T$, we can assign nonnegative weights to all edges $e\in E(T)$ via a map $w: E(T) \to {\mathbb{R}}_+$. The pair $(T,w)$ is called a *weighted tree*. Throughout this section all trees are weighted unrooted trees on $n$ taxa. Given $i,j\in [n]$ and $(T,w)$ a weighted tree, we define the *distance* between $i$ and $j$ to be $ d_T(i,j)=\sum_{e\in (i\rightarrow j)}w(e)$, where $e$ varies along all edges in the unique path from $i$ to $j$. We omit the subscript $T$ when understood from the context.
A *dissimilarity map* is a symmetric matrix $D\in {\mathbb{R}}_+^{n\times n}$ such that $D_{ii}=0$ for all $i\in
[n]$. Alternatively, whenever it is convenient for us we encode $D$ by a vector in ${\mathbb{R}}_+^{\binom{n}{2}}$ corresponding the upper diagonal portion of the original symmetric matrix. A dissimilarity map defines a *metric* if it satisfies the triangular inequality $D_{ij}\leq D_{ik}
+D_{kj}$ for all triples $i,j,k$.
Dissimilarity maps will measure distances between pairs of taxa.
A dissimilarity map $D$ is a *tree metric* if there exists a weighted tree $(T,w)$ such that $D=d_T$. By abuse of notation we will sometimes denote a tree metric by its associated weighted tree.
From the definition, it is straightforward to check that the set of dissimilarity maps is a closed pointed rational polyhedral cone isomorphic to ${\mathbb{R}}_+^{\binom{n}{2}}$. Likewise, the set of metrics is a full dimensional closed pointed rational polyhedral cone contained in the cone of dissimilarity maps. However, the space of tree metrics is a non-convex cone, but it can be decomposed into convex cones. Each piece is a closed rational polyhedral cone corresponding to a fixed tree topology. Its extremal rays are the *compatible cut metrics* that determine the topology of the tree $T$. All such cones will be isomorphic to the orthant ${\mathbb{R}}_+^{|E(T)|-n}$ ([@ASCB Prop 2.37].)
Tree metrics are characterized algebraically by the “Four Point Condition” [@ASCB Thm 2.36]:
\[thm:4ptCondition\] (Four Point Condition) A dissimilarity map $D$ is a tree metric *if and only if* for any 4-tuple $i,j,k,l \in [n]$, the maximum among $$\{D_{ij}+D_{kl}, D_{ik}+D_{jl}, D_{il}+D_{jk}\}$$ is attained at least *twice*. Moreover, if the previous holds $D$ will be realized by a unique weighted tree. In particular, a tree metric $D$ is realized by a star tree *if and only if* $D_{ij}+D_{kl}=D_{ik}+D_{jl}=D_{il}+D_{jk}$.
Now let us consider mixtures of dissimilarity maps:
Let $D, {\bar{D}}$ be two dissimilarity maps. We define the *tropical mixture* of $D$ and ${\bar{D}}$ as $(D\oplus {\bar{D}})_{ij}=\max\{D_{ij},
{\bar{D}}_{ij}\}=D_{ij} \oplus {\bar{D}}_{ij}$ for all $i,j\in [n]$. This generalizes to mixtures of any number of dissimilarity maps in the natural way.
To simplify notation, throughout the paper we refer to this construction as the mixture of $D$ and ${\bar{D}}$. From the definition it is straightforward to see that mixtures of metrics are metrics themselves. Since the Four Point condition with non distinct indices provides the metric condition, we only need to check tuples of four *distinct* indices.
We define the *mixture map* as the function $\phi: \text{\{tree
metrics\}}^2 \to \text{\{metrics\}}$ sending $(D,{\bar{D}}) \mapsto D\oplus
{\bar{D}}$. Since $ \lambda (D \oplus {\bar{D}})= (\lambda D) \oplus (\lambda
{\bar{D}})$ for $\lambda \in {\mathbb{R}}_+$, it follows that the image of $\phi$ is also a pointed cone inside ${\mathbb{R}}_+^{\binom{n}{2}}$. The commutativity of the mixture map can be interpreted by a natural action of ${\mathbb{S}}_2$ on pairs of tree metrics. The mixture map will be constant on orbits of $\text{\{tree metrics\}}^2$ under this action. The description of a mixture in terms of its summands is in general not unique, even after considering this ${\mathbb{S}}_2$-action. We will see several examples of this behavior later on.
From now on, we will focus our attention on the restriction of $\phi$ to pairs of star trees $$\phi: \text{\{star trees\}} \times \text{\{star trees\}} \to
\text{\{metrics\}}.$$ As one can show, the image of this restriction will also be a pointed cone. The main goal of this paper is to characterize the set ${\text{im}}\,\phi \cap \text{\{tree metrics\}}$. Along the way, we determine the fibers of $\phi$. Since $D=D\oplus D$, we already know that $\text{\{star trees\}} \subset {\text{im}}\, \phi$. Thus, we need only study when a non-star tree metric belongs to the image of $\phi$. We now present the desired characterization. Our building blocks with be the cases of four and five taxa.
\[lm:n=4\] Assume $n=4$ and let $T$ be quartet $(ij|kl)$ with edge weights as in Figure \[fig:quartetLabels\]. Then $T$ is a mixture of two star tree metrics *if and only if* $e_i,e_j>g$ or $e_k,e_l>g$. Moreover for each pair of inequalities, the fiber over $T$ consists of four pairwise disjoint 2-dimensional families of pairs $(D,{\bar{D}})$, which correspond to 2-dimensional polygons in ${\mathbb{R}}_+^8$, together with their orbits under the ${\mathbb{S}}_2$-action. All orbits will have size 2. On the contrary, if both inequalities hold, the two collections of families are disjoint and thus the fibers consist of eight families of 2-dimensional polygons, together with their orbits. Moreover, there is a one-to-one correspondence between the two sets of four families, after conveniently relabeling the leaves of the resulting trees.
Note that the characterization involves open conditions. If we do not require the inequalities to be strict, the fibers over boundary points will consist of degenerate star trees because some edges will have weight zero. The families describing the fibers will intersect at these pairs of degenerate star trees.
It is interesting to compare the result of the previous lemma with the analogous one obtained in [@MMS Prop 2.2] for *convex mixtures* of two star trees on four taxa. Tree metrics that are convex mixture of two star tree metrics are exactly the star trees and the quartets where the internal branch length is shorter than the sum of the branch lengths for *any* pair of non-adjacent edges.
\[lm:n=5\] Assume $n=5$ and let $T$ be a non-star weighted tree on five taxa. Denote by $e_i$ the weight of the edge attached to leaf $i$. Then, $T$ is a mixture of two star tree metrics *if and only if* $T$ has *only one internal edge* (labeled $g$), and $e_i,e_j>g$, where $i,j$ are the leaves of the unique cherry attached to one endpoint of the internal edge (see Figure \[fig:5taxaLemma\].) Moreover, the fiber over such a tree $T$ is the union of four disjoint ${\mathbb{S}}_2$-orbits of 2-dimensional polygons corresponding to the pairs $(D,{\bar{D}})$.
In particular, from the lemma we know that no trivalent tree metric is a mixture of two star tree metrics. The characterization for general $n$ follows from the previous two lemmas.
\[thm:mainResult\] Let $T$ be a non-star tree metric. Then $T$ is a mixture of two star tree metrics *if and only if*
1. $n=4$, $T$ is quartet $(ij|kl)$ and $e_i,e_j>g$ or $e_k,e_l>g$, where $g$ denotes the weight of the internal edge (see Figure \[sf:mainThma\]);
2. $n\geq 5$, $T$ has only one internal internal edge with weight $g$, and if $I$ and $J$ are a partition of $n$ into labels of leaves attached to the left-hand side and right-hand side of $T$, respectively, and we assume $|I|\leq |J|$, we have two cases:
1. $|I|=2$ and $e_{i_1}, e_{i_2}>g$ (see Figure \[sf:mainThmb\]);
2. $|I|\geq 3$, and $e_i>g$ for all $i\in [n]$ (see Figure \[sf:mainThmc\]).
Moreover, for each one of these trees we can describe the fiber in terms of four or eight disjoint ${\mathbb{S}}_2$-orbits of 2-dimensional polygons corresponding to families of pairs $(D,{\bar{D}})$.
Proof of Lemmas \[lm:n=4\], \[lm:n=5\], and of Theorem \[thm:mainResult\] {#s:proofs}
==========================================================================
Before presenting the proofs of the two main lemmas, we need an algebraic characterization of the topology of three metrics that equal $D\oplus {\bar{D}}$ in terms of the entries of $D$ and ${\bar{D}}$. Following the approach in [@MatSt], our description involves relative differences among the entries of both metrics. We denote by $a,c,d,e$ the weights of the edges in $D$, whereas ${\bar{a}}, {\bar{c}},{\bar{d}},{\bar{e}}$ are used for the edges in ${\bar{D}}$ as illustrated by Figure \[sf:ejemploBLabelingStar\]. We label the middle edge of $(12|34)$ by $b$ (see Figure \[sf:ejemploALabelingQuartets\].)
Since the topology of star trees is highly symmetric we need only treat the case of fibers of the mixture map over the quartet $(12|34)$. The general case will follow by relabeling. As an aside remark, note that since the set ${\text{im}}\, \phi\, \cap \{T:
T=(12|34)\}$ is a cone, we should expect our characterization to be invariant under multiplication by a positive scalar. Indeed, this will be the case.
From the labeling of edges in the star trees associated to $D$ and ${\bar{D}}$ we obtain two $4\times 4$ nonnegative symmetric matrices corresponding to these tree metrics. To simplify notation we only show the coefficients above the diagonal: $$D:=
\left(\begin{array}{cccc}
0 & a+d & a+c & a+e \\
& 0 & d+c & d+e \\
& & 0 & c+e\\
& & & 0
\end{array}
\right)
\; ; \;
{\bar{D}}:=\left(\begin{array}{cccc}
0 & {\bar{a}}+{\bar{d}}& {\bar{a}}+ {\bar{c}}& {\bar{a}}+ {\bar{e}}\\
& 0 & {\bar{d}}+ {\bar{c}}& {\bar{d}}+ {\bar{e}}\\
& & 0 & {\bar{c}}+{\bar{e}}\\
& & & 0
\end{array}
\right).$$ Consider six new indeterminates $s,t,x,y,u,w$ and the following equations relating the entries in $D$ and $\bar{D}$, which express the relative differences between these two matrices. $$\left\{
\begin{aligned}
{\bar{a}}+ {\bar{c}}& = a + c + s,\\
{\bar{d}}+ {\bar{e}}& = d +e + t,\\
{\bar{a}}+ {\bar{e}}& = a +e +x,\\
{\bar{d}}+ {\bar{c}}& = d + c + y,\\
{\bar{a}}+ {\bar{d}}& = a +d + u, \\
{\bar{c}}+ {\bar{e}}& = c +e + w.
\end{aligned} \right.\label{eq:3}$$
Denote $s^+=\max\{s,0\}$ and similarly for the other five new variables. Observe that $\max\{D_{13}, {\bar{D}}_{13}\} = D_{13}+s^+$. Thus $T=D\oplus {\bar{D}}$ will satisfy:
$$\left\{
\begin{aligned}
T_{12} +T_{34} & =
a+d + u^+ + c+e + w^+,\\
T_{13} + T_{24}& =
a+c + s^+ + d+e + t^+,\\
T_{14} + T_{23} & =
a+e + x^+ + d+c + y^+.
\end{aligned}\right.
\label{eq:1}$$
From and the fact that $D$ and ${\bar{D}}$ are star trees, these new variables also satisfy two linear equations: $
s+t=x+y=u+w$. The next result will be crucial in the proof of Lemma \[lm:n=4\].
\[pr:charact(12|34)\] With the above notation, $T$ is the quartet $(12|34)$ *if and only if* $$u^+ + w^+ < s^+ + t^+ = x^+ + y^+.$$
Follow immediately from the Four Point Condition.
Our strategy goes as follows. We assume $T=D\oplus {\bar{D}}$ to be the quartet $(12|34)$, and we find necessary and sufficient conditions on the weights of $T$ for this to hold. Our unique restrictions are given by requiring all values of edge weights to be positive because we consider non-degenerate star tree metrics. We show that these conditions suffice to explicitly construct $D$ and ${\bar{D}}$ from $T$.
Since the mixture map is symmetric, without loss of generality we may assume $s\geq 0$ (if not, switch the role of $D$ and ${\bar{D}}$.) Thus, we need to analyze two cases according to the sign of $t$. In what follows, we make full use of Proposition \[pr:charact(12|34)\].
If $t\geq 0$, we need $s+t=x+y$ and $s+t = x^+ + y^+$. Thus $x,y\geq
0$. Likewise $s+t=u+w$ but $u+w \leq u^++ w^+ <s+t$, a contradiction. Hence $t<0$. Switching the roles of $s$ and $t$ we also conclude that they both are nonzero.
By symmetry on the leaves of $(12|34)$, to expect this topology for $T$ we must have *both* pairs $(s,t), (x,y)$ with coordinates of *opposite* sign. Hence, without loss of generality we may assume $s> 0$ and $t<0$. In addition, since $x+y=s+t<s$ and $s=x^++ y^+$ we must have either $x=s,y=t$ or $x=t,y=s$. Assume the first conditions hold. By analyzing we conclude: $$s=x > 0,\, t=y<0 \; \Rightarrow\; {\bar{c}}=c+\frac{w}{2},\,
{\bar{e}}=e+\frac{w}{2},\, {\bar{a}}=a+ s-\frac{w}{2},\, {\bar{d}}=d+ y-\frac{w}{2}.$$ Likewise, for the second conditions: $$s=y> 0,\, t=x<0 \; \Rightarrow\; {\bar{a}}=a+\frac{u}{2},\,
{\bar{d}}=d+\frac{u}{2},\, {\bar{c}}=c+ s-\frac{u}{2},\, {\bar{e}}=e+ x-\frac{u}{2}.$$
We now consider the sign patterns of $u$ and $w$. Assume $u,w$ have opposite sign. Then from $u+w=s+t$ we conclude $u^++w^+<s$ *if and only if* ($0\leq u<s$ and $w=s+t-u<0$) ($0\leq
w<s$ and $u=s+t-w<0$). In case both $u$ and $w$ have the same sign, we have no restrictions other than $u+w=s+t$.
Let us compute the weights of the edges in $T \in (12|34)$, following the labeling of Figure \[fig:quartetLabels\] with $i=1, j=2, k=3,
l=4$: $$\left\{
\begin{aligned}
g&:=\frac{1}{2}(T_{14}+T_{23}-(T_{12}+T_{34}))=\frac{1}{2}(s-(u^++w^+)),\\
e_1 & := \frac{1}{2}(T_{14}+T_{13}-2g-T_{34})= a+\frac{x^+}{2}+\frac{u^+}{2} ,\\
e_2 & := \frac{1}{2}(T_{24}+T_{23}-2g-T_{34})=
d+\frac{y^+}{2}-\frac{s}{2}+\frac{u^+}{2} ,\\
e_3 & := \frac{1}{2}(T_{13}+T_{23}-2g-T_{12})= c+\frac{y^+}{2}+
\frac{w^+}{2},\\
e_4 & := \frac{1}{2}(T_{14}+T_{24}-2g-T_{12})=
e+\frac{x^+}{2}-\frac{s}{2}+\frac{w^+}{2}. \\
\end{aligned}\right. \label{eq:edges}$$ We compute each indeterminate in according to our two choices for pairing $s,t,x,y$ together:
- **Case 1:** $s=x > 0$, $t=y<0$. We obtain: $$e_1 = a+\frac{s}{2} + \frac{u^+}{2} \;,\; e_2 = d-\frac{s}{2} +
\frac{u^+}{2}\; ,\; e_3 = c+ \frac{w^+}{2} \;,\; e_4 = e+
\frac{w^+}{2} .$$
- **Case 2:** $s=y > 0$, $t=x<0$. Likewise: $$e_1 = a+ \frac{u^+}{2}\; , \;
e_2 = d+ \frac{u^+}{2}\; , \;
e_3 = c+\frac{s}{2} + \frac{w^+}{2} \; , \;
e_4 = e-\frac{s}{2} + \frac{w^+}{2} .$$
We subdivide these two cases according to sing constrains: $s>u\geq 0$ and $w<0$ (cases 1.1 and 2.1), $s>w\geq 0$ and $u<0$ (cases 1.2 and 2.2) or $u,w$ with the same sign (positive in cases 1.3 and 2.3 and negative in cases 1.4 and 2.4). In this way we get expressions for all our indeterminates and we obtain necessary conditions they must satisfy.
\[rk:symmetriesCases\] Notice that Cases 1 and 2 are symmetric. More precisely, they are equivalent if we change the labels for leaves $1$ and $2$ by $3$ and $4$ respectively and we also change the variables $(a,d,u,w,x,y)$ by $(c,e,w,u,y,x)$, while leaving $s$ and $t$ fixed. One needs to make these replacements between suitable pairs: 1.1 $\&$ 2.2, 1.2 $\&$ 2.1, 1.3 $\&$ 2.3 and 1.4 $\&$ 2.4 respectively. This correspondence will be crucial for the proof of Lemma \[lm:n=5\]. Therefore, we only need to study cases 1.1 through 1.4.
**Cases 1.1-1.4:**
**Case 1.1:** Suppose $s>u\geq 0$ and $w<0$, $$\left\{
\begin{array}{lcl}
g & := & \frac{s-u}{2},\\
e_1 & := & a+\frac{s}{2} + \frac{u}{2} ,\\
e_2 & := & d-\frac{s}{2} + \frac{u}{2},\\
e_3 & := & c,\\
e_4 & := & e.
\end{array}
\right.$$ Since $a>0,u\geq 0$ the previous equations yield $e_1>g$. On the other hand, by imposing all variables ${\bar{a}},{\bar{d}},{\bar{e}},{\bar{c}}$ and $e_k$ to be positive we conclude some necessary restrictions among $a,d,c,e,u,w,s$ and $t$. Namely, $d>s-u-\frac{w}{2}=2g-\frac{w}{2}>2g$ and $c,e>-\frac{w}{2}>0$. Finally, $e_2=d-g>g$. Therefore, we conclude $e_1,e_2>g$, $d>s-u-\frac{w}{2}$ and $c,e>-\frac{w}{2}$.
**Case 1.2:** Suppose $s>w\geq 0$ and $u<0$, $$\left\{
\begin{array}{lcl}
g & := & \frac{s-w}{2},\\
e_1 & := & a+\frac{s}{2} ,\\
e_2 & := & d-\frac{s}{2} ,\\
e_3 & := & c+ \frac{w}{2} ,\\
e_4 & := & e+ \frac{w}{2}.
\end{array}
\right.$$ In this case, from $a>0$ and $w\geq 0$ we also have $e_1>g$. In a similar way to the previous case, from $u<0$ and $0 \leq w<s$ we obtain additional necessary conditions: $d>s-u-\frac{w}{2}$ (because $s-u-\frac{w}{2}>\frac{s}{2}$). Again, $e_2=d-\frac{s}{2}>g-u>g$.
**Case 1.3:** Suppose $u,w\geq 0$, $$\left\{
\begin{array}{lcl}
g&:=& \frac{s-(u+w)}{2},\\
e_1 & := & a+\frac{s}{2} + \frac{u}{2} ,\\
e_2 & := & d-\frac{s}{2} + \frac{u}{2} ,\\
e_3 & := & c+ \frac{w}{2} ,\\
e_4 & := & e+ \frac{w}{2}.
\end{array}
\right.$$ As before, we have $e_1>g$, and we obtain the extra conditions $d>s-u-\frac{w}{2}$, $e_2=d-\frac{s}{2}+\frac{u}{2}>s-u-\frac{w}{2}-\frac{s-u}{2}=g$.
**Case 1.4:** Suppose $u,w < 0$, $$\left\{
\begin{array}{lcl}
g&:=& \frac{s}{2} ,\\
e_1 & := & a+\frac{s}{2} ,\\
e_2 & := & d-\frac{s}{2} ,\\
e_3 & := & c ,\\
e_4 & := & e.
\end{array}
\right.$$ Here we have $e_1>g$ and we get $c,e>-\frac{w}{2}$, $d>s-u-\frac{w}{2}$, thus $e_2=d-\frac{s}{2}>s-u-\frac{w}{2}-\frac{s}{2}>\frac{s}{2}-u-\frac{w}{2}>g$.
Changing variables as stated in Remark \[rk:symmetriesCases\], we get analogous results for cases 2.1 through 2.4. In particular, we obtain $e_3,e_4>g$ and a lower bound for $e$ in terms of $s,w$ and $u$. Hence, if $(12|34)$ is a mixture of two non-degenerate star tree metrics we conclude $e_1,e_2>g$ or $e_3,e_4>g$.
For the converse, we assume that $T$ is quartet $(12|34)$ and it satisfies $e_1,e_2>g$ or $e_3,e_4>g$. Our goal is to construct all possible pairs $D$, ${\bar{D}}$ giving $T=D\oplus {\bar{D}}$. From the “if direction” we know that such pairs $(D,{\bar{D}})$ come in four or eight two dimensional families, depending on which conditions on the weights of $T$ hold. Moreover, we have a description of these families so we only need to check if we can find values for the indeterminates satisfying the sign restrictions.
We claim that cases 1.1 through 1.4 are admissible *if and only if* $e_1,e_2>g$. We already know the “only if direction” from the previous discussion. We now show the converse holds:
**Case 1.1:** Assume $e_1,e_2>g$. We want to determine the values of $a,c,d,e$ and ${\bar{a}},{\bar{c}},{\bar{d}},{\bar{e}}$ so that $T=D\oplus
{\bar{D}}$. From the previous equations we easily obtain $e=e_4$, $c=e_3$ and $d=e_2+g$. From the conditions $e_4,e_3>-\frac{w}{2}$ we see that $0>w>-2e_4,2e_3$, although we still do not have an exact value for $w$.
On the other hand, we have $s=u+2g$. Hence, $a=e_1-\frac{s+u}{2}=e_1-g-u$, ${\bar{a}}=e_1+g-\frac{w}{2}$, ${\bar{d}}=e_2-g+\frac{w}{2}$, ${\bar{c}}= e_3+\frac{w}{2}$ and ${\bar{e}}=
e_4+\frac{w}{2}$. The only conditions on $u,w$ are determined by the positivity of all weights in $D$ and ${\bar{D}}$, namely, $u,w$ are free parameters satisfying $0>w>-2\min\{e_2-g, e_3,e_4\}$ and $e_1-g>u\geq 0$. The sign patterns for $s,t$ will hold automatically. Hence our necessary and sufficient conditions are $e_1,e_2>g$.
**Case 1.2:** As in the previous case, assume $e_1,e_2>g$. By the same procedure we get ${\bar{c}}=e_3$, ${\bar{e}}=e_4$, $s=2g+w$, $c=e_3-\frac{w}{2}$, $e=e_4-\frac{w}{2}$, $a=e_1-g-\frac{w}{2}$, $d=e_2+g+\frac{w}{2}$, ${\bar{a}}=e_1+g$ and ${\bar{d}}=e_2-g+u$ were $u$, $w$ are two free parameters satisfying $0>u>g-e_2$ and $2\min\{e_3,e_4,e_1-g\}>w\geq 0$. Therefore in this case our necessary and sufficient conditions are $e_1,e_2>g$ as well.
**Case 1.3:** Suppose $e_1,e_2>g$. From scratch we have ${\bar{e}}=e_4$ and ${\bar{c}}=e_3$. We let $u,w\geq 0$ be two free parameters, so $c=e_3-\frac{w}{2}$, $e=e_4-\frac{w}{2}$, $s=2g+u+w$, $a=e_1-g-u-\frac{w}{2}$, ${\bar{a}}=e_1+g$, $d=e_2+g+\frac{w}{2}$ and ${\bar{d}}=e_2-g$.
In particular, the only restrictions over $u,w$ are $0\leq w<2\min\{e_3,e_4,e_1-g\}$ and $0\leq
u<e_1-g-\frac{w}{2}$.
**Case 1.4:** Assume $e_1,e_2>g$. Then $s=2g$, $a=e_1-g$, $d=e_2+g$, $c=e_3$ and $e=e_4$. Our free parameters are $u,w<0$. We get ${\bar{a}}=e_1+g-\frac{w}{2}$, ${\bar{e}}=e_4+\frac{w}{2}$, ${\bar{c}}=e_3 + \frac{w}{2}$ and ${\bar{d}}=e_2-g+u+\frac{w}{2}$. Therefore, $u,w$ must satisfy $0>w>-2\min\{e_3,e_4,e_2-g\}$ and $0>u>g-e_2-\frac{w}{2}$.
By the symmetries mentioned in Remark \[rk:symmetriesCases\], we have similar results for cases 2.1 through 2.4. The necessary and sufficient conditions are $e_3,e_4>g$. We also obtain $u, w$ as free parameters and their linear restrictions are translations of the restrictions from the corresponding previous four cases.
Therefore, we have three scenarios for the fibers of the mixture map:
1. $e_1,e_2>g$ does *not* hold. Then $e_3,e_4>g$ and the fiber of the mixture map is the union of the four families described by cases 2.1 through 2.4, and their orbits under the action of ${\mathbb{S}}_2$ (which interchanges $D$ and ${\bar{D}}$.) All four families and their orbits are disjoint. By construction, each family is irreducible in the Zariski topology (affinely isomorphic to a product of intervals, or a triangle in ${\mathbb{R}}^2$.)
2. $e_3,e_4>g$ does *not* hold. Then $e_1,e_2>g$ and the fiber of the mixture map is the union of the families from cases 1.1 through 1.4, and their orbits. As before, these eight families are irreducible and disjoint.
3. $e_1,e_2,e_3,e_4>g$. Then the fiber consists of the orbits of all eight families given by cases 1.1 through 2.4. All orbits are disjoint.
As a reference for the proof of Lemma \[lm:n=5\], we include the description of all eight cases below. The free parameters are $u,w$.
- **Case 1.1:** $e_1-g>u\geq 0$ and $0>w>-2\min\{e_2-g,e_4,e_3\}$; $$\left\{
\begin{array}{llll}
a := e_1-g-u;&
d := e_2+g;&
c := e_3;&
e := e_4;\\
\bar{a} := e_1+g-\frac{w}{2};&
\bar{d} := e_2-g+\frac{w}{2};&
\bar{c} := e_3 +\frac{w}{2};&
\bar{e} := e_4 + \frac{w}{2}.
\end{array}
\right.$$
- **Case 1.2:** $0> u>g-e_2$ and $2\min\{e_3,e_4,e_1-g\}>w\geq 0$; $$\left\{
\begin{array}{llll}
a := e_1-g-\frac{w}{2};&
d := e_2+g+\frac{w}{2};&
c := e_3-\frac{w}{2};&
e := e_4-\frac{w}{2};\\
\bar{a} := e_1+g;&
\bar{d} := e_2-g+u;&
\bar{c} := e_3;&
\bar{e} := e_4.
\end{array}
\right.$$
- **Case 1.3:** $e_1-g-\frac{w}{2}>u\geq 0$ and $2\min\{e_3,e_4,e_1-g\}>w\geq 0$; $$\left\{
\begin{array}{llll}
a := e_1-g-u-\frac{w}{2};&
d := e_2+g+\frac{w}{2};&
c := e_3-\frac{w}{2};&
e := e_4-\frac{w}{2};\\
\bar{a} := e_1+g;&
\bar{d} := e_2-g;&
\bar{c} := e_3;&
\bar{e} := e_4.
\end{array}
\right.$$
- **Case 1.4:** $0>u>g-e_2-\frac{w}{2}$ and $0>w>-2\min\{e_3,e_4,e_2-g\}$; $$\left\{\hspace{-0.2cm}
\begin{array}{llll}
a := e_1-g;&\hspace{-0.1cm}
d := e_2+g;&\hspace{-0.1cm}
c := e_3;&\hspace{-0.1cm}
e := e_4;\\
\bar{a} := e_1+g-\frac{w}{2};&\hspace{-0.1cm}
\bar{d} := e_2-g+ u+ \frac{w}{2} ;&\hspace{-0.1cm}
\bar{c} := e_3 + \frac{w}{2};&\hspace{-0.1cm}
\bar{e} := e_4 + \frac{w}{2}.
\end{array}\hspace{-0.35cm}
\right.$$
- **Case 2.1:** $ 2\min\{e_1,e_2,e_3-g\}>u\geq 0$ and $0>w>g-e_4$; $$\left\{
\begin{array}{llll}
a := e_1-\frac{u}{2};&
d := e_2-\frac{u}{2};&
c := e_3-g-\frac{u}{2};&
e := e_4+g+\frac{u}{2};\\
\bar{a} := e_1;&
\bar{d} := e_2;&
\bar{c} := e_3+g;&
\bar{e} := e_4-g+w.
\end{array}
\right.$$
- **Case 2.2:** $0>u>-2\min\{e_4-g,e_1,e_2\}$ and $e_3-g>w\geq 0$; $$\left\{
\begin{array}{llll}
a := e_1;&
d := e_2;&
c := e_3-g-w;&
e := e_4+g;\\
\bar{a} := e_1+\frac{u}{2};&
\bar{d} := e_2+\frac{u}{2};&
\bar{c} := e_3 +g-\frac{u}{2};&
\bar{e} := e_4 -g+\frac{u}{2}.
\end{array}
\right.$$
- **Case 2.3:** $2\min\{e_2,e_1,e_3-g\}>u\geq 0$ and $e_3-g-\frac{u}{2}>w\geq 0$; $$\left\{\hspace{-0.2cm}
\begin{array}{llll}
a := e_1-\frac{u}{2};&\hspace{-0.1cm}
d := e_2-\frac{u}{2};&\hspace{-0.1cm}
c := e_3-g-\frac{u}{2}-w;&\hspace{-0.1cm}
e := e_4+g+ \frac{u}{2};\\
\bar{a} := e_1;&\hspace{-0.1cm}
\bar{d} := e_2;&\hspace{-0.1cm}
\bar{c} := e_3+g;&\hspace{-0.1cm}
\bar{e} := e_4-g.
\end{array}\hspace{-0.35cm}
\right.$$
- **Case 2.4:** $0>u>-2\min\{e_1,e_2,e_4-g\}$ and $0>w>(g-e_4)-\frac{u}{2}$; $$\left\{
\begin{array}{llll}
a := e_1;&
d := e_2;&
c := e_3-g;&
e := e_4 +g;\\
\bar{a} := e_1+ \frac{u}{2};&
\bar{d} := e_2 + \frac{u}{2};&
\bar{c} := e_3 + g- \frac{u}{2};&
\bar{e} := e_4 -g + \frac{u}{2}+w.
\end{array}
\right.$$
From the description of all eight cases one can show that, even when considering the commutativity of the mixture map, the sets corresponding to each case are pairwise disjoint. Notice also that in all eight cases, one of the star trees has at least three of its branch weights independent of the values of $u,w$.
We will make full use of the results on four taxa by considering the quartets describing our non-star tree metric $T$. By symmetry on the taxa, we need only consider the two cases illustrated in Figure \[fig:n=5Trees\]. We treat each case separately analyzing the possible quartets on these trees. The overall strategy consists of describing the fibers of the mixture map on four taxa and find compatible fibers over each quartet to build the original tree by gluing the partial liftings.
\[Proof for Figure \[sf:ejemplob\]\] Assume $T$ as in Figure \[sf:ejemplob\]. As we said before, we have necessary and sufficient conditions for each quartet in $T$ to be a mixture of two star tree metrics. We work with quartets $(12|34),
(12|35)$ and $(12|45)$. Since the first two quartets involve leaves $1$ and $3$, we can always assume that $s>0$. However, for the third quartet we need to consider *both* signs for $s$.
![Edge weights in star tree and in tree $T$ for $n=5$.[]{data-label="fig:labelsEjemplob"}](tree4.mps "fig:") ![Edge weights in star tree and in tree $T$ for $n=5$.[]{data-label="fig:labelsEjemplob"}](tree5.mps "fig:")
Assume $T=D\oplus {\bar{D}}$. Let $g$ be the weight of the middle edge and $e_i$ the weight of the edge adjacent to leaf $i$, whereas we label by $a,c,d,e,f$ the weights on the star tree associated to $D$ and similarly for ${\bar{D}}$ (see Figure \[fig:labelsEjemplob\].) Therefore, by considering the restrictions to the three mentioned quartets we see that the underlying tree $T$ must satisfy the following conditions: $$e_1,e_2 > g \quad \text{ or } \quad e_3,e_4,e_5 > g.$$ Since we have a complete characterization of mixtures on four taxa, we construct our candidates for the mixtures given each quartet. To finish, we glue our partial star trees together (whenever possible) to build our candidates $D,{\bar{D}}$ on 5 taxa. We show this can be achieved by exhaustive case by case analysis of the fibers over each quartet.
Assume $e_1,e_2> g$. Therefore, we know that at least we have to analyze cases 1.1 up to 1.4 for the three quartets (Recall that the order of $D$ and ${\bar{D}}$ is determined by the sign of $s$). We also treat cases 2.1 through 2.4 in case we have $e_3,e_4>g$.
Suppose $(12|34)$ satisfies case 1.1. Then we have $$\left\{
\begin{array}{llll}
a = e_1-g-u,&
c = e_3,&
d = e_2+g,&
e = e_4,\\
\bar{a} = e_1+g-\frac{w}{2},&
\bar{c} = e_3 +\frac{w}{2},&
\bar{d} = e_2-g+\frac{w}{2},&
\bar{e} = e_4 + \frac{w}{2},
\end{array}
\right.$$ where the free parameters are $u,w$ s.t. $0>w>-2\min\{e_2-g,e_4,e_3\}$ and $0\leq u<e_1-g$.
For quartet $(12|35)$ we have the same value of $s={\bar{a}}+{\bar{c}}-a-c>0$, so we do not need to switch the order of $D$ and ${\bar{D}}$. We claim that the fiber over $(12|35)$ can only correspond to case 1.1. Assuming cases 1.2 or 1.3 we see that $e_1+g-\frac{w}{2}={\bar{a}}= e_1+g$ gives $w=0<0$, a contradiction. Likewise, for case 1.4 we have $e_1+g-\frac{w}{2}={\bar{a}}=e_1+g-\frac{w'}{2}$ implies $w'=w$ and so $
e_2-g+\frac{w}{2} ={\bar{d}}= e_2-g+u'+\frac{w'}{2}$ gives $u'=0<0$, a contradiction. Cases 2.1 and 2.3 cannot hold because they would imply $e_3+g=c=e_3+\frac{w}{2}$ ($w< 0$). Likewise cases 2.2 and 2.4 cannot hold since they would yield $e_2 =d=e_2+g$ ($g>0$.) Thus by joining the values from cases 1.1 for both quartets we get $u'=u$ and $w'=w$ with compatible values for all weights in $D$ and ${\bar{D}}$.
To analyze quartet $(12|45)$ we need to compute the difference $({\bar{a}}+{\bar{f}})-(a+f)$, since its sign *determines the order* of $D$ and ${\bar{D}}$ in expression . We obtain $a+f=e_1+e_5-g-u$ and ${\bar{a}}+{\bar{f}}= e_1+e_5+g$, thus ${\bar{a}}+{\bar{f}}-a-f=2g+u>0$. Therefore, we do *not* need to switch the star trees in our computation of the edge lengths for this quartet.
As before, there is only one possibility for quartet $(12|45)$: case 1.1 must hold. The proof is analogous to the one for the quartet $(12|34)$ *because* we know that the order of the mixture is preserved.
Therefore, if $e_1,e_2>g$ and we are in case 1.1 for the quartet $(12|34)$ we have that $T=D\oplus {\bar{D}}$ where $$\left\{\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-g-u,&
c = e_3,&
d = e_2+g,&
e = e_4,&
f = e_5,\\
\bar{a} = e_1+g-\frac{w}{2},&
\bar{c} = e_3 +\frac{w}{2},&
\bar{d} = e_2-g+\frac{w}{2},&
\bar{e} = e_4 + \frac{w}{2},&
\bar{f} = e_5 + \frac{w}{2},
\end{array}\hspace{-.3cm}
\right.$$ and $u,w$ are free parameters satisfying $0\leq
u<e_1-g$, $0>w>-2\min\{e_2-g,e_4,e_3,e_5\}$.
On the other hand, assume case 1.2 holds for $(12|34)$. Like we observed when studying the previous quartet, we know that case 1.2 must hold for $(12|35)$ (compare the weight of ${\bar{a}}$ for cases 1.1, 1.4, 2.2 and 2.4, ${\bar{d}}$ for the case 1.3 and ${\bar{c}}$ for cases 2.1 and 2.3.) Therefore we get $w'=w$ and $u'=u$ where $0>u>g-e_2$ and $2\min\{e_3,e_4,e_5,e_1-g\}> w \geq 0$ on both quartets with compatible values for all weights.
As before we need to compare ${\bar{a}}+{\bar{f}}$ with $a+f$. We have ${\bar{a}}+{\bar{f}}=a+f+(2g+w)$, and so we do not need to interchange $D$ and ${\bar{D}}$. Therefore, the only possibility for quartet $(12|45)$ is case 1.2 which is compatible with the weights obtained before. Hence, $T=D\oplus {\bar{D}}$, where $$\left\{\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-g-\frac{w}{2},&
d = e_2+g+\frac{w}{2},&
c = e_3-\frac{w}{2},&
e = e_4-\frac{w}{2},&
f = e_5-\frac{w}{2},\\
\bar{a} = e_1+g,&
\bar{d} = e_2-g+u,&
\bar{c} = e_3,&
\bar{e} = e_4,&
\bar{f} = e_5,
\end{array}\hspace{-.3cm}
\right.$$ and $u,w$ are free parameters satisfying $0\geq u>g-e_2$ and $2\min\{e_1-g,e_4,e_3,e_5\}>w\geq 0$.
Next, assume case 1.3 holds for $(12|34)$. As before, the same case also holds for $(12|35)$ and we get $u=u'$, $w=w'$ where $2\min\{e_1-g,e_4,e_3,e_5\}>w\geq 0$ and $e_1-g-\frac{w}{2}>u\geq 0$ and the values of all edge weights are compatible. In addition, ${\bar{a}}+{\bar{f}}-a-f =2g +(u+w)=s>0$, so $D$ and ${\bar{D}}$ do not switch. Hence, case 1.3 must also hold for $(12|45)$ and we conclude $T=D\oplus {\bar{D}}$ with $$\left\{\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-g-u-\frac{w}{2},&\hspace{-.15cm}
d = e_2+g+\frac{w}{2},&\hspace{-.15cm}
c = e_3-\frac{w}{2},&\hspace{-.15cm}
e = e_4-\frac{w}{2},&\hspace{-.15cm}
f = e_5-\frac{w}{2},\\
\bar{a} = e_1+g,&\hspace{-.15cm}
\bar{d} = e_2-g,&\hspace{-.15cm}
\bar{c} = e_3,&\hspace{-.15cm}
\bar{e} = e_4,&\hspace{-.15cm}
\bar{f} = e_5,
\end{array}\hspace{-.35cm}
\right.$$ and $u,w$ are free parameters satisfying $2\min\{e_1-g,e_4,e_3,e_5\}>w\geq 0$ and $e_1-g-\frac{w}{2}>u\geq 0$.
To finish, assume $(12|34)$ satisfies case 1.4. Then as before the same case holds for $(12|35)$ and we obtain compatible values for all weights. In this case, ${\bar{a}}+{\bar{f}}-a-f=2g>0$, so $(12|45)$ also satisfies case 1.4. Therefore $T=D\oplus {\bar{D}}$ with $$\left\{\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-g,&\hspace{-.15cm}
d = e_2+g,&\hspace{-.15cm}
c = e_3,&\hspace{-.15cm}
e = e_4,&\hspace{-.15cm}
f = e_5,\\
\bar{a} = e_1+g-\frac{w}{2},&\hspace{-.15cm}
\bar{d} = e_2-g+ u+ \frac{w}{2} ,&\hspace{-.15cm}
\bar{c} = e_3 + \frac{w}{2},&\hspace{-.15cm}
\bar{e} = e_4 + \frac{w}{2},&\hspace{-.15cm}
\bar{f} = e_5 + \frac{w}{2},
\end{array}\hspace{-.35cm}
\right.$$ and $u,w$ are free parameters satisfying $0>w>-2\min\{e_2-g,e_4,e_3,e_5\}$ and $0>u>g-e_2-\frac{w}{2}$.
Next, we analyze cases 2.1 through 2.4 for quartet $(12|34)$. By symmetry on the leaves $4$ and $5$, we know that cases 2.1 through 2.4 must also hold for quartet $(12|35)$.
Assume case 2.1 for quartet $(12|34)$. By looking at ${\bar{a}}=e_1$, we see that cases 2.2 or 2.4 cannot hold for $(12|35)$. Thus we have either cases 2.1 or 2.3. Suppose case 2.3 holds for $(12|35)$, we get $e_1-\frac{u}{2}=a=e_1-\frac{u'}{2}$, so $u'=u$ and from $c$ we get $e_3-g-\frac{u}{2}=e_3-g-\frac{u}{2}-w'$ so $w'=0$. Thus: $$\left\{
\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-\frac{u}{2},&\hspace{-.13cm}
d = e_2-\frac{u}{2},&\hspace{-.13cm}
c = e_3-g-\frac{u}{2},&\hspace{-.13cm}
e = e_4+g+\frac{u}{2},&\hspace{-.13cm}
f = e_5+g+\frac{u}{2},\\
\bar{a} = e_1,&\hspace{-.13cm}
\bar{d} = e_2,&\hspace{-.13cm}
\bar{c} = e_3+g,&\hspace{-.13cm}
\bar{e} = e_4-g+w,&\hspace{-.13cm}
\bar{f} = e_5-g.
\end{array}\hspace{-.3cm}
\right.$$ We have ${\bar{a}}+{\bar{f}}=e_1+e_5-g=a+f-2g$, so we *must switch* the roles of $D$ and ${\bar{D}}$ for quartet $(12|45)$. By looking at the weight ${\bar{e}}$ for cases 2.1 and 2.3 we see ${\bar{e}}=e_5+g+\frac{u''}{2}=e_5-g+w$ with $u''\geq 0$, $w<0$, $g>0$, a contradiction. Likewise, for cases 2.4 and 2.2: ${\bar{e}}= e_4+g = e_4 -g+w$, $w<0$, a contradiction. For cases 1.1 through 1.4 we use the value of ${\bar{f}},f$: ${\bar{f}}=e_5$ or $f=e_5$ compared to the known values $f=e_5+g+\frac{u}{2}$, ${\bar{f}}=e_5-g$ ($u\geq 0, g>0$.) Note that these contradictions only involve the values of ${\bar{e}}$ and ${\bar{f}},f$ that come from $(12|34)$ and $(12|35)$, and the fact that we *had to switched* $D$ and ${\bar{D}}$ for the quartet $(12|45)$.
Therefore, case 2.1 holds for both quartets $(12|34)$ and $(12|35)$, and so $$\left\{\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-\frac{u}{2},&\hspace{-.15cm}
d = e_2-\frac{u}{2},&\hspace{-.15cm}
c = e_3-g-\frac{u}{2},&\hspace{-.15cm}
e = e_4+g+\frac{u}{2},&\hspace{-.15cm}
f = e_5+g+\frac{u}{2},\\
\bar{a} = e_1,&\hspace{-.15cm}
\bar{d} = e_2,&\hspace{-.15cm}
\bar{c} = e_3+g,&\hspace{-.15cm}
\bar{e} = e_4-g+w,&\hspace{-.15cm}
\bar{f} = e_5-g+w',
\end{array}\hspace{-.3cm}
\right.$$ with $w,w'<0$ and $u\geq 0$. Now ${\bar{a}}+{\bar{f}}=e_1+e_5-g+w'=a+f +
(-2g+w')$ and $-2g+w'<0$, so we must switch again $D$ and ${\bar{D}}$ for the quartet $(12|45)$. The same proof we gave using the values of edges ${\bar{e}}$ and $f$ or ${\bar{f}}$ shows that none of the four possible cases can hold for $(12|45)$. Thus, we conclude that case 2.1 does not hold for $(12|34)$.
The remaining cases will follow the same path of ideas, using the value of the edge ${\bar{e}}$, ${\bar{c}}$ and the switching vs non-switching of $D$ and ${\bar{D}}$ on the quartet $(12|45)$. Suppose case 2.2 holds for quartet $(12|34)$. From the weight ${\bar{a}}$ we see that cases 2.1 and 2.3 cannot hold for $(12|35)$, so we need to analyze cases 2.2 and 2.4. Assume case 2.4 holds for quartet $(12|35)$. Then from the value of ${\bar{a}}$ and $c$ we see that $u'=u,
w=0$ and so $$\left\{\hspace{-.25cm}
\begin{array}{lllll}
a = e_1,&\hspace{-.3cm}
d = e_2,&\hspace{-.3cm}
c = e_3-g,&\hspace{-.3cm}
e = e_4+g,&\hspace{-.3cm}
f = e_5+g,\\
\bar{a} = e_1+\frac{u}{2},&\hspace{-.3cm}
\bar{d} = e_2+\frac{u}{2},&\hspace{-.3cm}
\bar{c} = e_3+g-\frac{u}{2},&\hspace{-.3cm}
\bar{e} = e_4 - g + \frac{u}{2},&\hspace{-.3cm}
\bar{f} = e_5-g+\frac{u}{2}+w',
\end{array}\hspace{-.3cm}
\right.$$ with $u,w'<0$. We get ${\bar{a}}+{\bar{f}}-a-f=-2g+u+w'<0$, so we need to switch the roles of $D$ and ${\bar{D}}$ in $(12|45)$.
Let us look at the value of $e$ from $(12|45)$ for each one of these cases. For case 2.1: $e=e_4+g=e_4-g+w''$ with $w''<0$; for case 2.2: $e=e_4+g=e_4-g+\frac{u''}{2}$ with $u''<0$; for case 2.3: $e=e_4+g=e_4-g$, $g>0$ and for case 2.4: $e=e_4+g=e_4-g+\frac{u''}{2}+w''$ with $u'',w''<0$. All four expressions lead to contradictions. Likewise, for cases 1.1 and 1.4: $e=e_4+\frac{w''}{2}=e_4+g$ ($w''<0$) and for cases 1.2 and 1.3: $e=e_4=e_4+g$, which cannot occur.
Therefore, case 2.2 must hold for $(12|35)$. In particular $f=e_5+g$ and ${\bar{f}}=e_5-g+\frac{u}{2}$ ($u=u'<0$). Thus, ${\bar{a}}+{\bar{f}}-a-f
=-2g+u<0$. Hence, we must switch $D$ and ${\bar{D}}$ for $(12|45)$. The same argument using the value of $e$ at $(12|45)$ leads to a contradiction. Hence case 2.2 cannot hold for $(12|34)$.
Assume case 2.3 holds for quartet $(12|34)$. By looking at edge ${\bar{a}}=e_1$ we see that cases 2.2 or 2.4 cannot hold for $(12|35)$ (because $u'<0$), leaving us with cases 2.1 and 2.3 as possible candidates. Suppose case 2.1 for $(12|35)$. Then by looking at weights $a$ and $c$ we have $u=u'$ and $w=0$. Thus $$\left\{\hspace{-.2cm}
\begin{array}{lllll}
a = e_1-\frac{u}{2},&\hspace{-.3cm}
d = e_2-\frac{u}{2},&\hspace{-.3cm}
c = e_3-g-\frac{u}{2},&\hspace{-.3cm}
e = e_4+g+\frac{u}{2},&\hspace{-.3cm}
f = e_5+g+\frac{u}{2},\\
\bar{a} = e_1,&\hspace{-.3cm}
\bar{d} = e_2,&\hspace{-.3cm}
\bar{c} = e_3+g,&\hspace{-.3cm}
\bar{e} = e_4 - g,&\hspace{-.3cm}
\bar{f} = e_5-g+w',
\end{array}
\right.\hspace{-.5cm}$$ with $u\geq 0$, $w'<0$. Then ${\bar{a}}+ {\bar{f}}-a-f=
-2g+w'<0$, so we must interchange $D$ and ${\bar{D}}$ for the quartet. If cases 2.2 or 2.4 hold for $(12|45)$ we get ${\bar{e}}=e_4-g=e_4+g$, a contradiction. Likewise if cases 2.1 or 2.3 hold, we get ${\bar{e}}=e_4-g=e_4+g+\frac{u''}{2}$ and $u''\geq 0$, again a contradiction. Therefore, our only choice is case 2.3. Note that cases 2.1 and 2.3 share the same value for ${\bar{e}}=e_4-g$, so the only thing we need to know to ensure that case 2.3 cannot hold for $(12|35)$ is to guarantee that we must switch $D$ and ${\bar{D}}$ and that cases 1.1 through 1.4 are invalid candidates for $(12|45)$.
Since $f=e_5+g+\frac{u}{2}$ and ${\bar{f}}=e_5-g$ (we know $u'=u$), we get ${\bar{a}}+{\bar{f}}-a-f=-2g$, so $D$ and ${\bar{D}}$ must switch roles in quartet $(12|45)$. For cases 1.1 and 1.4 we have ${\bar{e}}=e_4=e_4-g$, whereas for cases 1.2 and 1.3: $e=e_4=e_4+g+\frac{u}{2}$ with $u\geq 0$, a contradiction. Therefore, we conclude that case 2.3 cannot hold for $(12|34)$.
To finish, assume case 2.4 for $(12|34)$. By looking at weight ${\bar{a}}$ we know that 2.1 and 2.3 are forbidden cases for $(12|35)$. Assume case 2.2 holds. By considering weights ${\bar{c}},c$ we have $u=u'<0$ and $w'=0$. Thus $$\left\{\hspace{-.23cm}
\begin{array}{lllll}
a = e_1,&\hspace{-.29cm}
d = e_2,&\hspace{-.29cm}
c = e_3-g,&\hspace{-.29cm}
e = e_4 +g,&\hspace{-.29cm}
f = e_5+g,\\
\bar{a} = e_1+ \frac{u}{2},&\hspace{-.29cm}
\bar{d} = e_2 + \frac{u}{2},&\hspace{-.29cm}
\bar{c} = e_3 + g- \frac{u}{2},&\hspace{-.29cm}
\bar{e} = e_4 -g + \frac{u}{2}+w,&\hspace{-.29cm}
\bar{f} = e_5 - g+ \frac{u}{2}.
\end{array}\hspace{-.35cm}
\right.$$ Now ${\bar{a}}+{\bar{f}}-a-f= -2g+u<0$, so we must interchange $D$ and ${\bar{D}}$ for quartet $(12|45)$. We compare the value of weight ${\bar{e}}$ for all four possible cases for quartet $(12|45)$. For cases 2.1 and 2.3 we get ${\bar{e}}=e_4-g+\frac{u}{2}+w =e_4+g+\frac{u''}{2}$ with $u''\geq 0$ and $u,w<0$, whereas for cases 2.2 and 2.4 ${\bar{e}}=e_4-g+\frac{u}{2}+w=e_4+g$, so we get a contradiction. Likewise, for cases 1.1 and 1.4: $e=e_4+\frac{w''}{2}=e_4+g$ ($w''<0$) whereas for cases 1.2 and 1.3: $e=e_4=e_4+g$, also a contradiction. Therefore, case 2.4 must hold for $(12|35)$. In particular, we know $f=e_5+g$, $\bar{f}=e_5-g+\frac{u}{2}+w''$ ($u''=u$) and so $a +f
-a-f=u+w''-2g<0$, thus $D$ and ${\bar{D}}$ must switch roles in quartet $(12|45)$. However, since we have the same value for $e$ as before (it was computed from quartet $(12|34)$) the same proof we gave when assuming case 2.2 shows that case 2.4 is not allowed for $(12|35)$.
From the previous analysis we conclude: if $e_1,e_2>g$, then the fiber over $T$ consists of cases 1.1 through 1.4 for all three quartets. Conversely, assume $T$ is the mixture of $D$ and ${\bar{D}}$. If $e_1,e_2>g$ does not hold, this would imply that cases 2.1 through 2.4 are the only candidates for $(12|34)$. But from the construction this cannot happen. Therefore, we conclude $e_1,e_2>g$.
\[Proof for Figure \[sf:ejemploc\]\] We now change gears and we analyze the topology of $T$ as in Figure \[sf:ejemploc\]. We label the edges of $T$ and of the star trees as in Figure \[fig:labelsEjemploc\].
![Weights for edges in star tree and $T$ for $n=5$.[]{data-label="fig:labelsEjemploc"}](tree4.mps "fig:") ![Weights for edges in star tree and $T$ for $n=5$.[]{data-label="fig:labelsEjemploc"}](tree6.mps "fig:")
We will show that no tree with this topology is a mixture of star trees.
Assume the contrary, and let $T$ be such a point. Then by considering quartet $(12|34)$ we know that $e_1,e_2>g+h$ or $e_3,e_4>g+h$. Without loss of generality we may assume $e_1,e_2>g+h$. Now, consider quartet $(12|53)$. Since $e_1,e_2>g+h>g$, we have to analyze all eight cases for $(12|34)$ and $(12|35)$.
Suppose case 1.1 holds for $(12|34)$, so $d=e_2+g+h$. If cases 1.1 or 1.4 hold for $(12|35)$, we get $e_2+g=d=e_2+g+h$, because the middle edge weight for $(12|35)$ is $g$. This contradicts $h>0$. Likewise, for cases 2.1 and 2.3 we get $d=e_2+g+h=e_2-\frac{u'}{2}$ ($u'\geq
0$), whereas for cases 2.2 and 2.4: $d=e_2+g+h=e_2$, a contradiction. Hence, cases 1.2 or 1.3 must hold for $(12|35)$. For any of them we look at the weight ${\bar{c}}$ and we get ${\bar{c}}=e_3=e_3+\frac{w}{2}$ which contradicts $w<0$. Thus, case 1.1 is not allowed for $(12|34)$.
Next, suppose case 1.2 holds for $(12|34)$. Then if we assume cases 1.1 or 1.4 for quartet $(12|35)$ we get $d=e_2+g=e_3+g+h+\frac{w}{2}$ with $w\geq 0$ and $h>0$, a contradiction. Likewise, for cases 1.2 and 1.3 we get ${\bar{a}}=e_1+g+h=e_1+g$ which cannot happen. For cases 2.1 and 2.3: ${\bar{a}}=e_1+g+h=e_1$, whereas for cases 2.2 and 2.4: ${\bar{a}}=e_1+g+h=e_1+\frac{u'}{2}$ ($u'<0$), which cannot occur.
Now assume case 1.3 for $(12|34)$, so in particular ${\bar{a}}=e_1+g+h$. Similarly to the procedure for case 1.2, if we assume cases 1.1 or 1.4 for $(12|35)$ we get ${\bar{a}}=e_1+g+h=e_1+g-\frac{w'}{2}$ where $w'<0$, a contradiction. Likewise from cases 1.2 or 1.3 we see ${\bar{a}}=e_1+g+h=e_1+g$, which is also a contradiction. Since the value of ${\bar{a}}$ at $(12|34)$ for cases 1.2 and 1.3 is the same, the arguments discussed in the previous paragraph discards cases 2.1 through 2.4 for $(12|35)$.
To finish, suppose case 1.4 for $(12|34)$. As in the analysis of case 1.1, if cases 1.2 or 1.3 hold for $(12|35)$ we have ${\bar{c}}=e_3=e_3+\frac{w}{2}$, with $w<0$, a contradiction. For cases 1.1 and 1.4, we look at $d=e_2+g+h=e_2+g$, so $h=0$ which cannot occur, whereas for cases 2.1 through 2.4 we use the same argument given in the analysis of case 1.1 for quartet $(12|34)$.
We now analyze the cases 2.1 through 2.4 for the quartet $(12|34)$. By symmetry, $(12|35)$ can only satisfy cases 2.1 through 2.4 as well. For cases 2.1 and 2.3, we work with weight ${\bar{c}}=e_3+g+h$ and get contradictions for all possible expressions of quartet $(12|35)$. For cases 2.2 and 2.4 we consider ${\bar{a}}=e_1+\frac{u}{2}$ ($u<0$) to rule out cases 2.1 and 2.3. From expression ${\bar{a}}=e_1+\frac{u}{2}=e_1+\frac{u'}{2}$ (it coincides for both 2.2 and 2.4), we conclude $u=u'$, and from weight ${\bar{c}}$ we get $e_3+g+h-\frac{u}{2}=e_3+g-\frac{u}{2}$, contradicting $h>0$.
\[rk:keyn=5\] Note that in all four cases of the previous proof we found that the three quartets satisfy the same case, either 1.1, 1.2, 1.3 or 1.4 and the parameters involved on different quartets were *the same*. This observation will be *essential* to prove Theorem \[thm:mainResult\].
Assume $T \in {\text{im}}\,\phi \cap \text{\{non-star tree
metrics\}}$. Then the fiber of $\phi$ over $T$ is characterized as follows. If we fix the condition ${\bar{D}}_{13}\geq D_{13}$ (i.e. $s\geq
0$), there are four different and disjoint 2-dimensional polyhedral families giving the possible mixture decompositions of $T=D\oplus
{\bar{D}}$. They correspond to cases 1.1 through 1.4 in the proof of Lemma \[lm:n=4\], applied to each one of the quartets of $T$.
We need only prove item (ii) since (i) is contained in Lemma \[lm:n=4\]. Moreover, for $n=5$ we can only have $|I|=2$ and the result was proved in Lemma \[lm:n=5\]. Thus we need only analyze the case $n\geq 6$.
From Lemma \[lm:n=5\], it is clear that we have only one topology on $n$ taxa that is likely to be in ${\text{im}}\,\phi \cap \text{\{tree
metrics\}}$ in addition to the star topology. Namely, the one given in Figure \[fig:nGralT’\]: two star trees with labels in the disjoint sets $I=\{i_1,\ldots, i_{p}\}$ and $J=\{j_1,\ldots, j_q\}$ with $I\sqcup
J=[n]$, glued together by one internal node of positive weight. Call this weight $g$.
![Non-star candidate $T$ for a mixture of two star tree metrics on $n$ taxa.[]{data-label="fig:nGralT’"}](tree7.mps)
By symmetry on the taxa, we can assume $1,2 \in I$ and $3,4\in
J$. Moreover, assume $|I|=p\leq q=|J|$. We claim that we have only two cases to deal with: $|I|=2$ or $|I|>2$.
Restrict $T$ to the quartets $(12|34)$ and $(12|3j)$, with $j>4$. Then, as we saw in the proof of Lemma \[lm:n=5\] and stated in Remark \[rk:keyn=5\], both quartets satisfy the same case (among cases 1.1 through 1.4) and $e_1,e_2>g$. By symmetry, the same case will hold for all quartets $(12|kl)$.
Assume $|I|=2$. By gluing the previous partial solutions together we get four families of expressions for $T=D\oplus {\bar{D}}$ where ${\bar{D}}_{13}\geq D_{13}$. In particular, $T\in {\text{im}}\, \phi \cap
\text{\{non-star tree metrics\}}$ *if and only if* $e_1,e_2>g$.
On the contrary, assume $|I|>2$. Therefore, the roles of $I$ and $J$ are symmetric. We conclude that $e_i>g$ for all $i\in [n]$ and that all quartets $(kl|34)$ (resp. $(12|kl)$) satisfy the same case among cases 2.1 through 2.4 (resp. cases 1.1 through 1.4.) The correspondence between cases mentioned in Remark guarantees that we can construct $D$ and ${\bar{D}}$ as in the case of five taxa.
Tropical secant varieties and the space of star trees {#s:TropicalSecantVarieties}
=====================================================
In this section we relate tropical secant varieties and mixtures of finitely many star trees on $n$ taxa. For the basic definitions on tropical geometry we refer the reader to [@RGStT]. For applications to phylogenetics we recommend [@ASCB Ch 2-3].
It is a well-known fact that the space of star trees is the tropicalization of a toric variety. More precisely, of a torus translate of the image of the monomial map $$\label{eq:monomialMap}
\psi: {\mathbb{C}}[t_1^{\pm 1}, \ldots, t_n^{\pm 1} ] \to {\mathbb{C}}[x_{ij}^{\pm 1}: 1\leq
i<j\leq n] \quad \psi_{ij}(\underline{t})=t_it_j=\underline{t}^{e_i+e_j}.$$ Thus, the equations describing the torus translate by ${(\alpha_{ij})}_{i,j}\in ({\mathbb{C}}^*)^{\binom{n}{2}}$ of the image of $\psi$ generate the ideal $$\begin{split}
I=\langle x_{ij}-\alpha_{ij}t_it_j: 1\leq
i< j\leq n \rangle \bigcap {\mathbb{C}}[x_{ij}^{\pm 1}: 1\leq i< j\leq n ] = \\\langle
x_{ij}x_{kl}-\frac{\alpha_{ik}\alpha_{kl}}{\alpha_{ik}\alpha_{jl}}
x_{ik}x_{jl}\;;\;
x_{il}x_{jk}-\frac{\alpha_{ik}\alpha_{kl}}{\alpha_{ik}\alpha_{jl}}
x_{il}x_{jk} : i,j,k,l\rangle,
\end{split}$$ where we identify $x_{ij}$ with $x_{ji}$ and $\alpha_{ji}$ with $\alpha_{ij}$ if $j<i$, and similarly for the other variables.
From Theorem \[thm:4ptCondition\] we know that the linear equations characterizing the star tree topology are the tropicalization of the generators of the ideal $I$ above with respect to the trivial valuation. Moreover since this ideal is binomial, the tropicalization of any set of binomials generating the ideal constitutes a tropical basis.
Let $L$ be the linear space spanned by the star tree metrics. From expression we know that it is generated by the set of all vertices of the second hypersimplex $\Delta(2,n)$, i.e. the exponents of our monomial map $\psi$: $\{e_i+e_j: 1\leq i<j\leq n\}$. As an example, if $n=4$, this polytope is the octahedron and for $n=5$ it is a 4-dimensional polytope with 10 vertices. In general, $\Delta(2,n)$ is an $(n-1)$-dimensional polytope in ${\mathbb{R}}^n$ with $\binom{n}{2}$ vertices. In particular, we conclude that the dimension of $L$ equals $n$.
Although the space of star trees is the tropicalization of a toric variety, the space of positive weighted star trees (i.e. *“star tree metrics”*) is not the tropicalization of any algebraic variety. More precisely, it is characterized by equations in tropical arithmetic, but they are not tropicalizations of Laurent polynomials, despite the fact that the Four Point Condition is the tropicalization of the polynomials $x_{ij}x_{kl}+x_{il}x_{jk}+x_{ik}x_{jl}$. The metric condition causes for this misbehavior. However the space of star tree metrics is the intersection of the positive orthant with the tropical variety of tropical rank one symmetric matrices (i.e. the space of trees.) In the language of [@ASCB], it is the positive part of this tropical variety.
As we mentioned in Section \[s:Intro\], mixtures of star tree metrics are closely related to tropical secants of (tropicalizations of) toric varieties ([@tropicalDevelin].) However, *Tropical secants of cones* are the appropriate setting to analyze mixtures of finitely many star tree metrics. Understanding this construction could help us extend the results obtained in previous sections. More precisely, following the notation of [@tropicalDevelin], the linear space $L$ must be replaced by a rational polyhedral cone, the rows in $M_L$ are replaced by the extremal rays of this cone, and we need to switch from *min* to *max* convention. Among the results that can be extended to the cone setting, we have:
[@tropicalDevelin Thm 2.1] \[thm:2.1ForCones\] Let $L\subset {\mathbb{R}}^{m}$ be a polyhedral *cone* with $d$ extremal rays. Consider the associated matrix $M_L\in {\mathbb{R}}^{d\times m}$ whose rows equal the previous $d$ rays. Let $V_L=\{v_1, \ldots, v_m\}$ be the $n$-point configuration of ${\mathbb{R}}^d$ given by the columns of $M_L$. Then a vector $x=(x_1, \ldots, x_m)$ is in the $k$-th tropical secant of $L$ (with the $\max$ convention) *if and only if* the lower envelope of the polytope formed by the height vector $x$ has $k+1$ facets given by linear functionals with only *positive* coefficients whose union contains each point of $V_L$.
The positivity condition for the coefficients comes from the positivity condition for scalars describing points in cones. For simplicity, we call such regular subdivisions *positive regular subdivisions*. Other than that, the proof follows line by line [@tropicalDevelin Thm 2.1].
In our example, we consider the cone in ${\mathbb{R}}^{\binom{n}{2}}$ of star tree metrics. Following the notation in [@tropicalDevelin], our matrix $M_L \in {\mathbb{Z}}^{n\times \binom{n}{2}}$ will have columns $$\label{eq:matrixA}
M_L^{(ij)}=e_i +e_j \in {\mathbb{R}}^n \qquad \forall 1\leq
i<j\leq n.$$ Notice that this is the integer matrix associated to the monomial map $\psi$. The extremal rays of the cone are the *cut metric* corresponding to the partition $\{i\} \sqcup
([n]\smallsetminus \{i\})$, as we vary $i\in [n]$. Notice that the cone is $n$-dimensional and it has $n$ extremal rays.
In unpublished work [@DM], the authors study star tree ranks of tropical skew-symmetric $n\times n$-matrices, that is, symmetric matrices where the diagonal is ignored. This notion of rank corresponds to computing the minimum $r$ s.t. any symmetric matrix (modulo its diagonal) can be written as a mixture of $r$ star trees, where we allow negative weights. More precisely, they show that for $n\geq 3$, any dissimilarity map on $n$ taxa can be written as the tropical mixture of $n-2$ star trees, possibly with negative weights on their edges. Moreover, this bound is tight. By comparing this result with Lemma \[lm:n=4\], we see that any metric on four taxa is a tropical mixture of two star trees, but not necessarily two star tree *metrics*. Therefore, star tree metric ranks are often higher than star tree ranks. We will show that, moreover, star tree metric ranks can be infinite (e.g. a cut metric on four taxa.)
Another interesting question to study emerges also from [@tropicalDevelin]. By [@tropicalDevelin Thm 2.1, Cor 2.2] one knows that each regular subdivision in the $k$-th tropical secant variety of the space of star trees corresponds to a polyhedral cone of height vectors. In particular, this results in a decomposition of the tropical secant variety ${\mathscr{C}}$ as a polytopal complex. In the case of the $k$-th tropical secant variety of pointed cones we have:
[@tropicalDevelin Cor 2.2] The $k$-th tropical secant variety of a pointed cone $L$ is cone from $L$ over a polytopal complex, which we call the $k$-th tropical secant complex of $L$. The faces correspond to positive regular subdivisions of $V_L$ in which there exist $(k+1)$ facets containing all of the points, with a face $F$ containing a face $G$ if the regular subdivision associated to $F$ refines the one associated to $G$.
Consider the spaces of star trees, of star tree metrics, and their corresponding tropical secant complexes. One knows that the complex corresponding to star tree metrics ${\mathscr{C}}_+$ is a strictly contained in the complex corresponding to star trees ${\mathscr{C}}$. In particular, it would be desirable to study how the cell structures of both complexes relate. In addition, one can investigate how the cell structure of ${\mathscr{C}}_+$ compares to the subdivision of ${\mathscr{C}}_+$ into topological types.
To finish, we would like to make some comments on star tree metric ranks of dissimilarity maps. Since the configuration of points associated to $\Delta(2,n)$ is in convex position, [@tropicalDevelin Cor 2.3] shows that for $k\gg 0$, the $k$-th tropical secant (or *$\infty$-th tropical secant*) of the space of star trees equals the ambient space ${\mathbb{R}}^{\binom{n}{2}}$. Notice that the results of [@DM] give us $n-2$ as a sharp lower bound on such $k$. However, the space of tree metrics does not equal the $\infty$-th tropical secant variety of star tree metrics. A very simple example is given for $n=4$ taxa and the cut metric $(12|34)$, i.e. the metric defined by $D(1,3)=D(1,4) =D(2,3)=D(2,4)=1$ and $0$ otherwise. Assume $D=\bigoplus_{k=1}^N S_k$ where $S_k$ are star tree metrics, and denote $e_j^{(k)}$ the weight of the edge pendant to leaf $j$ in $S_k$. Then we have $S_k(1,2)=e_1^{(k)}+e_2^{(k)}\leq D(1,2)=0$ and $S_k(3,4)=e_3^{(k)}+e_4^{(k)}\leq D(3,4)=0$. By the positivity condition on the edge weights, we conclude $e_{j}^{(k)}=0$ for all $i,j$, a contradiction.
In general, we consider any *cut metric* on $n$ taxa, given by a partition with at least two subsets are not singletons. Then, by conveniently restricting $D$ to a metric on a four-element set we conclude that $D$ does not lie in $\infty$-th secant variety of the space of star tree metrics, so the latter is not the space of metrics. Moreover, the same example shows that we cannot cover all tree metrics by tropical mixtures of finitely many star tree metrics. In particular, these results together with [@tropicalDevelin Cor 2.3] imply that regular subdivisions of the second hypersimplex need not be positive, as one would expect from the constructions.
As we see from the previous discussion, mixtures of star trees and of star tree metrics have completely different behavior, although the underlying combinatorics in both cases are closely related. Basic questions on the latter remain open. Among these are the characterization of $k$-mixtures of star tree metrics, and membership tests for these sets. We believe that a serious study of tropical secants of cones will help us attack most of these open problems.
Acknowledgments {#sec:aknowledgements .unnumbered}
===============
This project developed from a course project in Lior Pachter’s Math 239 graduate course at UC Berkeley. I wish to thank Lior Pachter and Bernd Sturmfels for their guidance and suggestions, and also Dustin Cartwright for useful discussions.
[^1]: The author was supported by a UC Berkeley Chancellor’s Fellowship and the Laboratory for Mathematical and Computational Biology at UC Berkeley.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
We present an approach to type theory in which the typing judgments do not have explicit contexts. Instead of judgments of shape $\Gamma \vdash A : B$, our systems just have judgments of shape $A : B$. A key feature is that we distinguish free and bound variables even in pseudo-terms.
Specifically we give the rules of the [[‘[Pure Type System]{}’]{}]{} class of type theories in this style. We prove that the typing judgments of these systems correspond in a natural way with those of Pure Type Systems as traditionally formulated. I.e., our systems have exactly the same well-typed terms as traditional presentations of type theory.
Our system can be seen as a type theory in which all type judgments share an identical, infinite, typing context that has infinitely many variables for each possible type. For this reason we call our system [$\Gamma_\infty$]{}. This name means to suggest that our type judgment $A : B$ should be read as ${\ensuremath{\Gamma_\infty}}\vdash A : B$, with a fixed infinite type context called [$\Gamma_\infty$]{}.
author:
- Herman Geuvers
- Robbert Krebbers James McKinna Freek Wiedijk
bibliography:
- 'Gi.bib'
title: '[Pure Type Systems without Explicit Contexts]{}'
---
Introduction
============
\#1\#2\#3[0=to0[$\hss\displaystyle{\strut{#1}\over\strut\box0}\rlap{\raise1pt\hbox{$\,#3$}}\hss$]{}]{} \#1
Problem
-------
One of the important insights type theory gives us is the need to be aware of the *context* in which one works. This was already stressed by de Bruijn in his 1979 paper *Wees contextbewust in WOT* [@bru:79], Dutch for [[“[Be context aware in the mathematical vernacular]{}”]{}]{}. In type theory a term always is considered with respect to a context $\Gamma$, which gives the types of the variables occurring free in the term. This is also apparent in the shape $\Gamma \vdash M : A$ of the judgments of type theory, where the context $\Gamma$ is made explicit. Thus a [[‘[bound]{}’]{}]{} variable is bound *locally* in a term, while a [[‘[free]{}’]{}]{} variable actually is *globally* bound, namely by the context. In customary presentations of first order predicate logic [@mendelson:intro; @vanDalen:LandS for example], and in fact in the presentation of most other logics as well, free variables are not treated in such a way. In these logics free variables are really *free*. They are taken from an infinite supply of variables that are just available to be used in formulas and terms, without them having to be declared first.
This difference between type theory and predicate logic means that when we model predicate logic in type theory, actually we do not get the customary version of predicate logic, but instead get a version called *free* logic [@lambert:free-logic1963]. In traditional treatments the formula $(\forall x.\, P(x)) \to (\exists x.\, P(x))$ is usually provable. For instance a natural deduction proof of this formula would look like: $$\der{
\der{
\der{
\leaf{[\forall x.\, P(x)]}
}
{P(y)}{\forall E}
}
{\exists x.\, P(x)}{\exists I}
}
{(\forall x.\, P(x)) \to (\exists x.\, P(x))}{{\to}I}$$ This proof uses the free variable $y$. If one cannot use any other variables than those introduced by earlier rules, then this proof fails. Indeed, the type corresponding to the formula is not inhabited: there is no term $M$ such that the following judgment is derivable: $$D : {*},\, P : D \to {*} \vdash M : (\Pi x : D.\, P(x)) \to (\Sigma x : D.\, P(x))$$ because we cannot avoid the case in which the domain $D$ is empty, where the formula is false.
Now there are two things one can do to bridge this gap between type theory and traditional logic:
- Make predicate logic more like type theory, by explicitly keeping track in the judgments of the set of variables that can be used in the proof.
- Make type theory more like predicate logic, by having a version of type theory that does not need contexts in the judgments, i.e., in which free variables are just taken from some infinite supply.
Although the first option is interesting too, especially in categorical treatments of logic [@lambek-scott:hocl1988 for example], in this paper we focus on the second. We originally thought that the dependent types in type theory would prevent a version of type theory without contexts from being a viable option, but to our surprise it turns out that one *can* present type theory in a style where there are no contexts and in which therefore free variables are really free, provided we are prepared to pay the small price of labelling variables in a rather involved manner.
In those type theories actually implemented in interactive theorem provers, the context always consists of a part holding global *definitions* and parameters and a part holding the *free variables* in the term (as in [@CSC:MKM2004; @severi-poll:DPTS for example]). For simplicity of exposition, and for the sake of proving an exact correspondence between a standard presentation of type theory and the variant we propose, in this paper we consider only the second part of such contexts. We believe, however, that the other part can be treated in exactly the same way.
There is another reason why it is interesting to look at a version of type theory where there are no explicit contexts. One of the most popular architectures for proof assistants is the *LCF architecture*, named after the LCF system from the seventies [@LCF-book:1979]. In the original form of such a system there is an abstract data type called [`term`]{}, whose elements can only be created by a small number of functions exported from the type-checking kernel. Elements of this datatype always correspond to type-correct terms, and those terms can contain free variables.
A system using this approach has a kernel interface containing a function:
> `app : term term -> term`
When this function is called, the kernel of the system makes sure that the types of the arguments are compatible, i.e., that the result is again a type-correct term.
This is how the HOL family of theorem provers is implemented. These systems have a logical foundation that is based on a typed lambda calculus. However, in these systems the free variables in the terms are not recorded in a context of variables. The only context in these systems is the context of [definitions]{}, which is kept track of in a stateful variable. Definitions are never allowed to be removed from this context, as that would compromise the safety of the kernel. Hence, these systems are stateful, although they can be made functional using a variant of the approach presented here [@stateless].
There are two classes of systems that can be said to implement a *type theory*, a typed lambda calculus:
- The simpler type theories, in which no dependent types are allowed. They often are a form of simple type theory with some enhancements, such as some form of polymorphism or type classes. These include the systems from the HOL family: HOL4, HOL Light, ProofPower and Isabelle. These systems can be, and are, implemented following the LCF architecture just outlined. In these systems variables come from an infinite [[‘[sea]{}’]{}]{} of free variables, and in the logical theory there is *no* context keeping track of the variables.
- More advanced type theories, often called type theory with *dependent types*. These come from the Dutch <span style="font-variant:small-caps;">Automath</span> systems, the Swedish tradition of Martin-Löf type theory, the French tradition of the Calculus of Constructions and variations on the Edinburgh Logical Framework. Their implementations include Coq, NuPRL, Twelf, Agda, Lego, Plastic and Epigram. In the logical theory of these systems there *is* a context keeping track of the variables.
For this second style of type theory the pure LCF approach is not attractive. The [`app`]{} function will need to check whether the contexts in which the terms live are compatible, which will be very expensive, if it needs to be done for the type-checking of each function application.
For this reason actual type-theoretical proof assistants have a kernel with a different kind of interface. In such a kernel there is no abstract datatype of *terms* (there just is a—non-abstract—type of *pseudo-terms*). Instead there is an abstract type of well-typed *contexts*, which we call [`context`]{} here. (There is also a type [`pseudocontext`]{}, but that is irrelevant here.) The interface then looks like:
> `mkApp : pseudoterm pseudoterm -> pseudoterm`\
> `add_constant : string pseudoterm -> context -> context`
where [`mkApp`]{} is just the constructor of the data type of pseudoterms, whereas [`add_constant`]{} is a function that does the type-checking: a pseudoterm will only be added to the context after it has been type-checked. (These are the actual names of the functions in the kernel of the Coq system. The types of those functions in Coq are essentially what is presented here. The type [`pseudoterm`]{} is called [`constr`]{} in Coq, while the type [`context`]{} is called [`safe‘_environment`]{}.) The system also has a global variable
> `global_env : context ref`
corresponding to the *state* of the system in which the user works. It is not part of the kernel (and in fact is changed back by an undo operation), but as there are no interesting operations combining two different [`context`]{}s, only one global [`context`]{} is ever relevant, the one given by the contents of this variable.
Although the architecture with contexts that we described is purely functional (as is the Coq kernel), the fact that the actual implementation has this global variable means that it is used in a rather [[‘[stateful]{}’]{}]{} way. The desire to investigate a possible LCF-style kernel for type theory that is [[‘[less stateful]{}’]{}]{} motivated this research. In the conclusions we will address the question whether the style of type theory we present here will lead to such a type-checking architecture.
Approach
--------
The approach we will follow here is to imagine there to be an [[‘[infinite context]{}’]{}]{} called [$\Gamma_\infty$]{}. For each type-correct type $A$ this context will have infinitely many variables $x_i^A$. It should be stressed that this $A$ should be considered to just be a label, a *string*. Reduction will never happen inside these $A$s. Also, $x_i^A$ and $x_i^B$ will be *different* variables, even when $A$ and $B$ are convertible, or even if they are $\alpha$-equivalent. Note that the (free) variables in $A$ themselves will also be of shape $y_j^B$: this means that the variables themselves, as well as the terms, have a recursive tree-like structure.
For example, a variable corresponding to the successor function on natural numbers looks like: $$s^{N^* \to N^*}$$ If we use numbered names for the variables, this might become: $$x_0^{x_0^* \to x_0^*}$$ So the [[“[small price]{}”]{}]{} alluded to above is that a free variable $x_i^A$ in a well-typed term carries with it the (well-founded) history of how it comes to be well-typed; that is, the label $A$ witnesses the validity of the context extension $\Gamma, x_i : A$.
Now our systems will have judgments $A : B$, which should be interpreted as ${\ensuremath{\Gamma_\infty}}\vdash A : B$. For this reason we call the general approach to type theory introduced here [[‘[[$\Gamma_\infty$]{}]{}’]{}]{} (reusing the name of the context as the name of the system). Note that [$\Gamma_\infty$]{} is not a single system: each type theory will have a [[‘[[$\Gamma_\infty$]{}-variant]{}’]{}]{}.
The [$\Gamma_\infty$]{} approach has the essential feature that there are two different classes of variables. There are the variables that come from the [$\Gamma_\infty$]{} context (the [[‘[free]{}’]{}]{} variables), and then there is another kind of [[‘[bound]{}’]{}]{} variables. When thinking about our systems one might imagine de Bruijn indices for the bound variables, although the presentation we give here uses named variables for them as well.
Although we expect many type theories to have a [$\Gamma_\infty$]{}-variant, here we only consider the class of type theories called Pure Type Systems [@bar-lics-hbk] (PTSs). That way we keep everything concrete, and it allows us to prove a precise correspondence between PTSs in the traditional style and our version in [$\Gamma_\infty$]{} style.
One should note that in [$\Gamma_\infty$]{} any type will be inhabited, simply because there are free variables of every type: in particular, just as in traditional treatments of logic, all domains are assumed to be inhabited. This is in essence the same as in the version with contexts, except that such variables are there explicitly recorded in the context.
Related Work
------------
To our surprise, we found little published work investigating such an approach to *dependent* type theory. In Church’s original formulation of *simple* type theory [@church:stt1940], variables, both free *and* bound, are annotated with their types, writing for example $\lambda
x^{\alpha}. f^{\alpha{{\rightarrow}}\alpha}\, x^{\alpha}$. (whereas in our formulation we would write $\lambda x {{:}}{\alpha^*}. f^{\alpha^* {{\rightarrow}}\alpha^*}\,
x$.) Girard adopted [[‘[Church-style]{}’]{}]{} in the account of System F in his thesis [@girard:thesis]. In neither system do *term* variables occur in types, while *type* variables are not regulated by an explicit context. In these non-dependent type theories, contexts are not strictly needed, because one can define the different syntactic classes — types, terms — in stages. One can regard our approach as extending that of Church to dependent types, but optimised to avoid the need to consider substitution in labels on *bound* variables which otherwise might arise in the application rule.
Conor McBride (private communication) observed that Pollack’s <span style="font-variant:small-caps;">LEGO</span> implementation already supported the [$\Gamma_\infty$]{} idea, and this idea was then used in his <span style="font-variant:small-caps;">OLEG</span> extensions, and subsequently in the architecture of <span style="font-variant:small-caps;">Epigram 1</span>. However, this approach has not been treated theoretically as we do here.
The explicit distinction between free and bound variables on a syntactic level already can be found in [@mckinna-pollack:jar1999]. The motivation there was to avoid capture during substitution while keeping a close correspondence with the informal presentation of PTSs with named variables, rather than how to give a [[‘[[$\Gamma_\infty$]{}]{}’]{}]{} presentation, as here. Various approaches to representing binding are discussed in [@Vaughan06areview], which considers named free variables and de Bruijn index bound variables one of the best options for mechanisation. Indeed, in ongoing work [@krebbers:gammainf] the second author has formalised one half of the correspondence proved in Section \[sec:thms\] in such a style. We expand on the niceties of this formalisation in Section \[sec:conclusion\].
Elsewhere, Pollack considered presentations of type theory separating the typing judgment from that for context well-formedness [@randy:closure-under1993], although judgments are still [[‘[in context]{}’]{}]{}. This allows a subtle range of issues to be explored, especially regarding closure under $\alpha$-conversion, here treated only informally.
Most significantly, but starting from a rather different point, Sacerdoti Coen considered the problem of proof-checking in the setting of a distributed library, and hence the problem of how to *reconstruct* a context in which a given term may be successfully type-checked [@CSC:MKM2004]. This work (elaborated in [@CSC:PhD], and forming the basis of the Matita system) goes beyond the standard PTS setting considered in this paper. It identifies a subtle problem which arises when attempting to merge contexts (including definitions) in the presence of global constraints (such as universe levels).
*Added in proof.* Between acceptance of the final version of this paper and this final version, our attention was drawn to the recent work of Matthias Boespflug [@boespflug], which itself references an earlier (2009) account of our ideas. By contrast with our presentation, which is purely first-order, Boespflug uses higher-order abstract syntax (HOAS), with a view to implementation.
Contribution
------------
We present a different approach to type theory, much closer to the way logical systems usually are presented than the standard presentation, in which free variables are not bound in a finite context but are taken to be really free.
We validate our approach by proving two theorems, \[thm:PTStoGinf\] and \[thm:GinftoPTS\] below, establishing a straightforward correspondence between the standard presentation and the variant presented here.
Outline
-------
The structure of the paper is as follows. In Section \[sec:pts\] we recall the PTS rules and some of its theory. In Section \[sec:Gi\] we present the [$\Gamma_\infty$]{}-variant of the PTS rules, in which the judgments do not have contexts. In Section \[sec:thms\] we show that both systems correspond to each other in a natural way. In Section \[sec:conclusion\] we conclude with a prospectus for an implementation based on our variant of the PTS rules.
Pure Type Systems in the traditional style {#sec:pts}
==========================================
Pure Type Systems (PTSs) generalize many existing type systems and thus the class of PTSs contains various well-known systems, like systems F and F$\omega$, dependent type theory $\lambda$P and the Calculus of Constructions.
\[def.pts\] For $\SS$ a set (the set of *sorts*), ${{\cal A}}\subseteq \SS\times\SS$ (the set of *axioms*) and ${{\cal R}}\subseteq \SS\times \SS \times \SS$ (the set of *rules*), the *Pure Type System* $\lambda(\SS ,{{\cal A}},{{\cal R}})$ is the typed $\lambda$-calculus with inference rules as in Figure \[fig:pts\].
[$$\begin{array}{|llr|}
\hline
(\mbox{sort})
&
\prooftree
\justifies
\vdash s_1 : s_2
\endprooftree &\mbox{if }(s_1 ,s_2)\in {{\cal A}}\\
&&\\
(\mbox{var})
&
\prooftree
\Gamma \vdash A : s
\justifies
\Gamma ,x{{:}}A\vdash x : A
\endprooftree &\mbox{if }x\notin\Gamma \\
&&\\
(\mbox{weak})
&
\prooftree
\Gamma \vdash A : s\;\;\;\;
\Gamma \vdash M : C
\justifies
\Gamma ,x{{:}}A \vdash M : C
\endprooftree &\mbox{if }x\notin\Gamma \\
&&\\
(\Pi)
&
\prooftree
\Gamma \vdash A : s_1\;\;\;\;
\Gamma ,x{{:}}A \vdash B : s_2
\justifies
\Gamma \vdash \Pi x{{:}}A .B : s_3
\endprooftree & \mbox{if }(s_1 ,s_2 ,s_3) \in {{\cal R}}\\
&&\\
(\lambda)
&
\prooftree
\Gamma ,x{{:}}A \vdash M : B\;\;\;\;
\Gamma \vdash \Pi x{{:}}A .B : s
\justifies
\Gamma \vdash \lambda x{{:}}A .M : \Pi x{{:}}A.B
\endprooftree &\\
&&\\
(\mbox{app})
&
\prooftree
\Gamma \vdash M : \Pi x{{:}}A.B\;\;\;\;
\Gamma \vdash N : A
\justifies
\Gamma \vdash MN : B[x := N]
\endprooftree &\\
&&\\
(\mbox{conv})
&
\prooftree
\Gamma \vdash M : A\;\;\;\;
\Gamma \vdash B : s
\justifies
\Gamma \vdash M : B
\endprooftree & A =_{\beta} B\\[1em]
\hline
\end{array}$$ ]{}
In the rules, the expressions $M,N,A,B,C$ are taken from the set of *pseudo-terms* ${{\cal T}}$ defined by\[pts-pseudo-terms\] $${{\cal T}}::=
s \bar {{\cal V}}\bar \Pi {{\cal V}}{{:}}{{\cal T}}.{{\cal T}}\bar \lambda {{\cal V}}{{:}}{{\cal T}}.{{\cal T}}\bar {{\cal T}}{{\cal T}}.$$ with ${{\cal V}}$ a set of variables, and the $\Gamma$ taken from the set of *pseudo-contexts* $$x_1 : A_1, \ldots , x_n :A_n \quad (x_i \in {{\cal V}}, A_i\in{{\cal T}}, 1\le i\le n)$$ with the $x_i$ all distinct. (We leave the choice of variable *names* unspecified at this point, as this does not matter as long as ${{\cal V}}$ is countably infinite, but below we will take a specific choice of names.)
There is a lot of theory about PTSs and various systems have been studied in the context of PTSs. We do not give a complete overview but refer to [@bar-lics-hbk; @bargeu-ar-hbk; @geuv93] for details and explanation. Here we just give the results that we use in the rest of the paper to prove the equivalence between a PTS and its [$\Gamma_\infty$]{}-variant.
The pseudo-term $A$ is called *well-typed* if a pseudo-context $\Gamma$ and pseudo-term $B$ exist such that $\Gamma\vdash A:B$ or $\Gamma\vdash B:A$ is derivable. A pseudo-context $\Gamma$ is *well-formed* if pseudo-terms $A$ and $B$ exist such that $\Gamma\vdash A:B$ is derivable; a *context* is a well-formed pseudo-context. The set of variables declared in pseudo-context $\Gamma$ is called the *domain* of $\Gamma$, ${{\rm dom}}(\Gamma)$. For $x\in{{\rm dom}}(\Gamma)$, let ${{\rm type}}_{\Gamma}(x)$ denote the [[‘[type]{}’]{}]{} assigned to $x$ in $\Gamma$: if $x:A
\in\Gamma$, then ${{\rm type}}_{\Gamma}(x) = A$. The expression ${{\rm type}}(\Gamma)$ denotes the set of such [[‘[types]{}’]{}]{} occurring in $\Gamma$. The set of well-typed terms of $\lambda (\SS,{{\cal A}},{{\cal R}})$ is denoted by ${{\rm Term}}(\lambda (\SS,{{\cal A}},{{\cal R}}))$.
We adopt the usual notions of bound and free variable, $\alpha$-conversion ($\equiv$), substitution ($B[x := N]$, used in the rule ()), $\beta$-reduction (${\rightarrow_{\beta}}$) and $\beta$-equality (${=_{\beta}}$, used in the rule ()) on pseudo-terms.
The following are well-known properties of PTSs. The relation $\Gamma\subseteq\Delta$ denotes inclusion between $\Gamma$ and $\Delta$ regarded as sets of variable assignments. The third, Permutation, is a corollary of Strengthening.
\[prop.metatheoryPTS\]
Thinning
: If $\Gamma \vdash M:A$ and $\Delta \supseteq \Gamma$ is well-formed, then $\Delta \vdash M:A$.
Strengthening
: If $\Gamma, x:B,\Delta \vdash M:A$ and $x\notin{{\rm FV}}({{\rm type}}(\Delta),M,A)$, then $\Gamma, \Delta \vdash M:A$.
Permutation
: If $\Gamma, x:B,y:C,\Delta \vdash M:A$ and $x\notin{{\rm FV}}(C)$, then $\Gamma, y:C, x:B,\Delta \vdash M:A$.
In proving the equivalence between a PTS and its [$\Gamma_\infty$]{}-variant, we need to merge two contexts to create a new one. Therefore we introduce the following:
\[def.comp\] Let $\Gamma$ and $\Delta$ be two pseudo-contexts. We say $\Gamma$ and $\Delta$ are *compatible*, notation $\Gamma{\mathrel{{|}{|}{|}}}\Delta$, if $$\forall x\in {{\rm dom}}(\Gamma)\cap{{\rm dom}}(\Delta)( {{\rm type}}_{\Gamma}(x) \equiv {{\rm type}}_{\Delta}(x)).$$ The *merge* of $\Gamma$ and $\Delta$, notation $\Gamma{\ltimes}\Delta$, is the pseudo-context $\Gamma ,
(\Delta\setminus\Gamma)$. This is $\Gamma$ followed by the declarations $x:B\in\Delta$ for which $x\notin{{\rm dom}}(\Gamma)$.
Note the strong requirement in $\Gamma{\mathrel{{|}{|}{|}}}\Delta$ that the types of $x$ in $\Gamma$ and $\Delta$ should be $\alpha$-equal, and not just $\beta$-convertible.
If $\Gamma$ and $\Delta$ are contexts and $\Gamma {\mathrel{{|}{|}{|}}}\Delta$, then $\Gamma {\ltimes}\Delta$ is well-formed.
We write $x_1:B_1, \ldots, x_n:B_n$ for $\Delta$. $\Gamma {\ltimes}\Delta$ is the pseudo-context $\Gamma, (\Delta\setminus\Gamma)$. As $\Gamma$ is well-formed, we only have to consider the part $\Delta\setminus\Gamma = x_{i_1} : B_{i_1}, \ldots, x_{i_n} : B_{i_n}$. But $x_1:B_1, \ldots, x_{i_1 -1}: B_{i_1 -1} \subseteq \Gamma$, so by Thinning $\Gamma \vdash B_{i_1} :s$ for some sort $s$, so $\Gamma, x_{i_1} :B_{i_1}$ is well-formed.\
The same reasoning applies to $x_{i_2}:B_{i_2}, \ldots, x_{i_n}:B_{i_n}$, so we conclude that $\Gamma, (\Delta\setminus\Gamma)$ is well-formed.
In our system [$\Gamma_\infty$]{}, the free variables will have special names, as they are labelled by their types. Of course, consistently renaming the free variables in a judgment $\Gamma \vdash M:A$ does not change its meaning, as the free variables in $M$ and $A$ are actually bound in $\Gamma$. For clarity, we introduce this notion explicitly.
\[def:alpha-judg\] The judgment $\Gamma \vdash M:A$ is $\alpha$-equivalent to $\Gamma ' \vdash M':A'$ in case one can be obtained from the other by renaming bound variables, where we consider the free variables in $M$ and $A$ to be bound by their declaration in $\Gamma$.
Pure Type Systems in the [$\Gamma_\infty$]{} style {#sec:Gi}
==================================================
We now make the set of variables ${{\cal V}}$ explicit. We have two kinds: variables $\dot x$ with a dot on top, intended to be bound by $\lambda$ and $\Pi$ binders; and variables $x^A$ tagged with a pseudo-term $A$, intended to be bound in the context. This means we take ${{\cal V}}$ in Definition \[def.pts\] as follows, where ${{\cal X}}$ supplies the *names* of variables: $$\begin{array}{rcl}
{{\cal V}}&::=& \dot{{\cal X}}\bar {{\cal X}}^{{\cal T}}\\
{{\cal X}}&::=& x \bar y \bar z \bar \dots \bar x_0 \bar x_1 \bar x_2 \bar \dots
\end{array}$$ Clearly the rules for ${{\cal T}}$ and ${{\cal V}}$ are mutually recursive.
Note that although the variables are *intended* to be used in a certain way (made precise in Definition \[def.typeannotated\] below), in the PTSs as defined above both kinds of ${{\cal V}}$ can be used for all purposes. In particular $\dot x$ can be put in a context, $x^A$ can be bound, and the label $A$ of $x^A$ need not correspond to the type of $x^A$.
Note also that the definition of substitution, and hence the relation of $\beta$-equality, is agnostic about the structure of the annotations of the variables. (The definition of $=_\beta$ in Section \[sec:pts\] takes ${{\cal V}}$ just as a set). This means that although $$(\lambda \dot A:{*}. \dot A)\,B^{*} =_\beta B^{*}$$ we have that $$x^{(\lambda \dot A:{*}. \dot A) B^{*}} \ne_\beta x^{B^{*}}$$
We will now define the rules of [$\Gamma_\infty$]{}. To do this we first have to introduce the notion of *hereditarily free variables of the types of the free variables* in a term. We first motivate the need for this notion.
In a PTS, the context takes care that one can only abstract over a variable if nothing else depends on that variable. In the rules, this is formalised by requiring that the $x:A$ abstracted over in the $\Pi$ or $\lambda$ rule must be the last declaration in the context. This ensures that $x$ does not occur free in any of the other types in the context. In [$\Gamma_\infty$]{} we do not have contexts, so we have to replace this by another side condition on the rules. We would like to have a $\Pi$ rule as follows: $${
A : s_1
\quad\enskip
B : s_2
\over
\Pi \dot y {:} A.\, B[x^A := \dot y] : s_3
}
\rlap{ \small $(s_1,s_2,s_3)\in{\cal R}$}$$ but that is wrong, because then we would be able to form (in PTS terminology) the $\Pi$-type $${
\ldots \vdash A:* \quad \enskip A:*, P:A{{\rightarrow}}*, Q:\Pi x{{:}}A. P\, x {{\rightarrow}}*, a:A, h:P\, a \vdash Q \, a\, h :*
\over
A:*, P:A{{\rightarrow}}*, Q:\Pi x{{:}}A. P\, x {{\rightarrow}}*, h:P\, a \vdash \Pi y :A.Q\,y\, h : *
}$$ But this cannot be correct, because $h$ is not of type $P\, y$ in the conclusion. In [$\Gamma_\infty$]{}, this derivation would read (adding some brackets for readability): $${
A^*:* \quad \enskip
Q^{\Pi \dot x{{:}}A^*. (P^{A^*{{\rightarrow}}*}\, \dot x) {{\rightarrow}}*} \, a^{A^*}\, h^{P^{A^*{{\rightarrow}}*}\, a^{A^*}} :*
\over
\Pi \dot y :A^*. Q^{\Pi \dot x{{:}}A^*. (P^{A^*{{\rightarrow}}*}\, \dot x) {{\rightarrow}}*} \, \dot y\, h^{P^{A^*{{\rightarrow}}*}\, a^{A^*}} :*
}$$ which would be derivable according to the $\Pi$-rule above, but clearly undesirable.
\[def:HFV\] Given $M\in {{\cal T}}$, we define the *hereditarily free variables in $M$*, denoted ${{\rm hfv}}(M)$, as follows: $$\begin{aligned}
{{\rm hfv}}(s) ={{\rm hfv}}(\dot x) &=& \emptyset\\
{{\rm hfv}}(x^{A})&=& \{x^{A}\} \cup{{\rm hfv}}(A)\\
{{\rm hfv}}(F\,N) &=& {{\rm hfv}}(F)\cup{{\rm hfv}}(N)\\
{{\rm hfv}}(\lambda \dot x {{:}}A.N) = {{\rm hfv}}(\Pi \dot x{{:}}A.N) &=& {{\rm hfv}}(A)\cup{{\rm hfv}}(N)\end{aligned}$$
So, where $=_\beta$ basically ignores the structure of the type labels of free variables, we now take them seriously, collecting the variables (hereditarily) free in the type labels as well.
We put as a side condition in the $\Pi$-rule that $x^A$ should not occur free in any of the *types of the free variables in $B$*, and similarly for the $\lambda$ rule. We give an explicit definition of this notion.
\[def:HFVT\] Given $M\in {{\cal T}}$, we define the *hereditarily free variables of the types of the free variables in $M$*, denoted ${{\rm hfvT}}(M)$, as follows: $$\begin{aligned}
{{\rm hfvT}}(s) ={{\rm hfvT}}(\dot x) &=& \emptyset\\
{{\rm hfvT}}(x^{A})&=& {{\rm hfv}}(A)\\
{{\rm hfvT}}(F\,N) &=& {{\rm hfvT}}(F)\cup{{\rm hfvT}}(N)\\
{{\rm hfvT}}(\lambda \dot x {{:}}A.N) = {{\rm hfvT}}(\Pi \dot x{{:}}A.N) &=& {{\rm hfvT}}(A)\cup{{\rm hfvT}}(N)\end{aligned}$$
So, for example ${{\rm hfvT}}(h^{P^{A^*{{\rightarrow}}*}\, a^{A^*}}) = \{ P^{A^*{{\rightarrow}}*} , a^{A^*}, A^*\}$. An easy corollary of the definition is that $${{\rm hfvT}}(M) \subseteq {{\rm hfv}}(M)$$
We now give the derivation rules of the system.
The derivation rules of [$\Gamma_\infty$]{} are given by the inference rules in Figure \[fig:gammainf\].
[$$\begin{array}{|llr|}
\hline
(\mbox{\rm sort})
&
\prooftree
\justifies
s_1 : s_2
\endprooftree &\mbox{if }(s_1 ,s_2)\in {{\cal A}}\\
&&\\
(\mbox{\rm var})
&
\prooftree
A : s
\justifies
x^A : A
\endprooftree & \\
&&\\
(\Pi)
&
\prooftree
A : s_1\;\;\;\;
B : s_2
\justifies
\Pi \dot x{{:}}A .B[y^A := \dot x] : s_3
\endprooftree & \hspace{-2em}\mbox{if }(s_1 ,s_2 ,s_3) \in {{\cal R}}\mbox{ and }y^A \notin{{\rm hfvT}}(B)\\
&&\\
(\lambda)
&
\prooftree
M : B\;\;\;\;
\Pi \dot x{{:}}A .B[y^A := \dot x] : s
\justifies
\lambda \dot x{{:}}A .M[y^A := \dot x] : \Pi \dot x{{:}}A.B[y^A := \dot x]
\endprooftree & \mbox{if }y^A \notin{{\rm hfvT}}(M)\cup{{\rm hfvT}}(B)\\
&&\\
(\mbox{\rm app})
&
\prooftree
M : \Pi \dot x{{:}}A.B\;\;\;\;
N : A
\justifies
MN : B[\dot x := N]
\endprooftree &\\
&&\\
(\mbox{\rm conv})
&
\prooftree
M : A\;\;\;\;
B : s
\justifies
M : B
\endprooftree & A =_\beta B\\[1em]
\hline
\end{array}$$ ]{}
The $\Pi$ and $\lambda$ rules have the side condition that $\dot x$ should not be *captured* in $B$ or $M$ when doing the substitution. That is, $\dot x$ should not be bound by a binder under which $y^A$ occurs.
The side condition on the $\Pi$ and $\lambda$ rules is no restriction as you can go to an $\alpha$-equivalent version of the term afterwards.
The correspondence theorems {#sec:thms}
===========================
We now prove that a PTS and its [$\Gamma_\infty$]{}-variant correspond to each other. For a [$\Gamma_\infty$]{} judgment $M:A$ we generate a context $\Gamma$ such that $\Gamma\vdash M:A$ is PTS-derivable. Conversely, if $\Gamma \vdash M : A$ is PTS-derivable, we always have an $\alpha$-equivalent judgment (see Definition \[def:alpha-judg\]) $\Gamma'\vdash M':A'$ such that $M' : A'$ is derivable in [$\Gamma_\infty$]{}. This specific form $\Gamma' \vdash M' : A'$ we call a *type annotated* judgment.
\[def.typeannotated\] A *type annotated context* in a PTS is a context of the form $$x^{B_1}_1 : B_1, \ldots, x^{B_n}_n:B_n$$ where we moreover assume that all bound variables in the $B_i$ are of the form $\dot x$.
\[def.typeannotated1\] A *type annotated judgment* in a PTS is one of the form $$x^{B_1}_1 : B_1, \ldots, x^{B_n}_n:B_n \vdash M:A$$ where: $x^{B_1}_1 : B_1, \ldots, x^{B_n}_n:B_n$ is a type annotated context; all free variables in $M$ and $A$ are of the form $x^{B_i}_i$; and all bound variables are of the form $\dot x$.
We now first show the easy direction of our correspondence result:
Every judgment $\Gamma \vdash M:A$ in a PTS is $\alpha$-equivalent to a type annotated judgment $\Gamma' \vdash M':A'$.
From left to right we rename the variables in the context (and of course also in $M$ and $A$) to the [[‘[standard]{}’]{}]{} names $$x_1^{B_1}:B_n,\dots,x_n^{B_n}:B_n$$ (Here $x_i$ is not a meta-variable for a variable name, but really the *explicit* variable name [[“[$x_i$]{}”]{}]{} in ${{\cal X}}$.) We also $\alpha$-rename any bound variable of the form $x^A$ to a fresh variable of the form $\dot x$.
As an example, consider the PTS judgment $$A : {*},\, a : A \,\vdash (\lambda x{{:}}A.\,x)\, a : A$$ This does not fit the variable names from our ${{\cal V}}$, so this does not conform to the definitions in this paper. Instead using our variables it should be something like: $$\dot A : {*},\, a^{*} : \dot A \,\vdash (\lambda x^{\dot B}{{:}}\dot A.\,x^{\dot B})\, a^{*} : \dot A$$ This of course is not a *type annotated* judgment, the annotations make no sense at all, but still this is a perfectly fine PTS judgment as defined in Definition \[def.pts\].
Now according to the theorem this is $\alpha$-equivalent to a judgment that *is* type annotated. And it is, for example it is $\alpha$-equivalent to $$x_1^{*} : {*},\, x_2^{x_1^{*}} : x_1^{*} \,\vdash (\lambda \dot x{{:}}x_1^{*}.\,\dot x)\, x_2^{x_1^{*}}$$ Or, if one uses more readable names, to $$A^{*} : {*},\, a^{A^{*}} : A^{*} \,\vdash (\lambda \dot x{{:}}A^{*}.\,\dot x)\, a^{A^{*}}$$
The other part of the easy direction of our correspondence is that a type annotated PTS judgment essentially is the same as a [$\Gamma_\infty$]{} judgment:
\[thm:PTStoGinf\] If the type annotated PTS judgment $\Gamma \vdash M : A$ is derivable, then $M : A$ is a derivable judgment of the corresponding [$\Gamma_\infty$]{} type theory.
By induction on the size of the derivation of $\Gamma \vdash M : A$. We do a case split on the last rule used in the derivation:
- Immediate.
- Trivial, because if $\Gamma,x{{:}}A$ is a type annotated context, then certainly $\Gamma$ is a type annotated context.
- By induction we have $A : s$ derivable in [$\Gamma_\infty$]{}, and because the context is type annotated the variable name must be of the form $x^A$. Hence $x^A : A$ in [$\Gamma_\infty$]{}.
- We know that $\Gamma \vdash M : A$ and $\Gamma \vdash B : s$. Now $A$ need not have bound variables of the form $\dot x$, but one can rename them to obtain $A' \equiv A$ such that $\Gamma \vdash M : A'$ *will* be type annotated. Then $M : A'$ and $B : s$ in [$\Gamma_\infty$]{} by induction and therefore also $M : B$ in [$\Gamma_\infty$]{}.
- Again, we might need to change the bound variable in $\Gamma \vdash M : \Pi x{{:}}A.B$ to the dotted kind, to get a type annotated judgment. Apart from that this case is trivial, just like the previous one.
- The conclusion will be of the form $\Gamma \vdash \Pi\dot x{{:}}A.B : s_3$. Now if we take $y^A$ a completely fresh variable, then $\Gamma,y^A:A \vdash B[\dot x := y^A] : s_2$ will be a type annotated judgment, as well as being $\alpha$-equivalent to $\Gamma,\dot x:A \vdash B : s_2$. Accordingly, let $B' := B[\dot x := y^A]$, so that $B \equiv B'[y^A := \dot x]$.
Clearly now $y^A \notin{{\rm hfvT}}(B')$ because $\Gamma,y^A:A \vdash B' : s_2$, so all type annotations will be typable in $\Gamma$, which does not contain $y^A$.
By induction $A : s_1$ and $B' : s_2$ in [$\Gamma_\infty$]{} and because $y^A \notin{{\rm hfvT}}(B')$ we get that $\Pi\dot x{{:}}A.B'[y^A := \dot x] : s_3$ in [$\Gamma_\infty$]{}. But this is precisely $\Pi\dot x{{:}}A.B : s_3$.
- This case essentially follows that of the previous one.
The other direction of our correspondence—from [$\Gamma_\infty$]{} to traditional PTS style—is a bit more involved. We need to *synthesise* an appropriate context, and because we build this context by recursion over the type derivation, we need to merge these synthesised contexts using ${\ltimes}$. For this we need a number of lemmas involving type annotated contexts in PTSs.
\[lem.annotctxt\] If $\Gamma$ and $\Delta$ are type annotated contexts, then $\Gamma{\mathrel{{|}{|}{|}}}\Delta$.
If $x^A \in {{\rm dom}}(\Gamma)\cap{{\rm dom}}(\Delta)$, then $x^A : A\in\Gamma$ and $x^A :A \in \Delta$, of course, $A \equiv A$ and this is what we need to prove according to Definition \[def.comp\].
\[lem.annothfvt\] If $\Gamma$ is a type annotated context with $x^A\in{{\rm dom}}(\Gamma)$, $\Gamma \vdash M:B$ and $x^A \notin{{\rm hfvT}}(M,B)$, then $$\exists \Delta \subset \Gamma (\Delta, x^A:A\vdash M:B)$$
Write $\Gamma =\Gamma_1, x^A :A, \Gamma_2$, and suppose $y^C \in
{{\rm dom}}(\Gamma_2)$ with $x^A\in{{\rm FV}}(C)$. If $y^C \in {{\rm FV}}(M,B)$, then $
x^A\in {{\rm hfvT}}(M,B)$, contradiction. So $y^C \notin{{\rm FV}}(M,B)$. This means that all declarations $y^C:C$ in $\Gamma_2$ for which $x^A\in{{\rm FV}}(C)$ can be removed by Strengthening (Proposition \[prop.metatheoryPTS\]), starting from the rightmost declaration in $\Gamma_2$. We end up with a judgment $$\Gamma_1, x^A :A, \Gamma'_2 \vdash M:B$$ with $\Gamma'_2 \subseteq
\Gamma_2$ and $x^A\notin{{\rm type}}(\Gamma'_2)$. Using Permutation (Proposition \[prop.metatheoryPTS\]), we conclude that $\Gamma_1,
\Gamma'_2, x^A :A \vdash M:B$ and we take $\Gamma_1, \Gamma'_2$ for $\Delta$.
\[cor.annothfv\] If $\Gamma \vdash M:B$ is a type annotated judgment, there is a $\Delta \subseteq \Gamma$ such that $\Delta \vdash M:A$ and ${{\rm dom}}(\Delta) \subseteq {{\rm hfv}}(M,B)$.
Let $x^A\in {{\rm dom}}(\Gamma)$ and $x^A\notin {{\rm hfv}}(M,B)$. Then $x^A\notin
{{\rm hfvT}}(M,B)$, so (according to Lemma \[lem.annothfvt\]), there is a $\Delta\subseteq
\Gamma$ such that $\Delta, x^A:A\vdash M:B$. But also $x^A\notin
{{\rm FV}}(M,B)$, so by Strengthening (Proposition \[prop.metatheoryPTS\]), $\Delta\vdash M:B$.
So, in $\Gamma \vdash M:B$, we can always make the context $\Gamma$ so small that its domain is within the set of hereditarily free variables of $M,B$. The other way around, the hereditarily free variables of $M,B$ should be in ${{\rm dom}}(\Gamma)$:
\[lem.hfvdom\] If $\Gamma \vdash M:B$ is type annotated, then ${{\rm hfv}}(M,B) \subseteq {{\rm dom}}(\Gamma)$.
We prove $\Gamma \vdash M:B \Rightarrow {{\rm hfv}}(M) \subseteq
{{\rm dom}}(\Gamma)$, by induction on the derivation and then we are done, because if $\Gamma \vdash M:B$, then $\Gamma \vdash B:s$ for some sort $s$, or $B$ is a sort.
- Immediate.
- By induction, ${{\rm hfv}}(A) \subseteq{{\rm dom}}(\Gamma)$, so ${{\rm hfv}}(x^A) \subseteq {{\rm dom}}(\Gamma, x^A : A)$.
- By induction, ${{\rm hfv}}(M)\subseteq{{\rm dom}}(\Gamma)$, so we are done.
- By induction, ${{\rm hfv}}(F) \subseteq{{\rm dom}}(\Gamma)$ and ${{\rm hfv}}(M)\subseteq{{\rm dom}}(\Gamma)$, so ${{\rm hfv}}(F\,M) \subseteq{{\rm dom}}(\Gamma)$.
- By induction, ${{\rm hfv}}(A)\subseteq{{\rm dom}}(\Gamma)$ and ${{\rm hfv}}(B)\subseteq{{\rm dom}}(\Gamma, y^A:A)$, so ${{\rm hfv}}(\Pi \dot x{{:}}A. B[y^A := \dot x]) \subseteq {{\rm dom}}(\Gamma)$.
- By induction, ${{\rm hfv}}(M)\subseteq{{\rm dom}}(\Gamma, y^A:A)$ and ${{\rm hfv}}(\Pi \dot x{{:}}A. B[y^A := \dot x]) \subseteq {{\rm dom}}(\Gamma)$, so ${{\rm hfv}}(A) \subset {{\rm dom}}(\Gamma)$, and hence ${{\rm hfv}}(\lambda \dot x{{:}}A. M[y^A := \dot x]) \subseteq {{\rm dom}}(\Gamma)$.
The more difficult direction of our correspondence now follows:
Let $M : A$ be a derivable [$\Gamma_\infty$]{} judgment. Then all free variables in $M$ and $A$ have the form $x^A$ and all bound variables have the form $\dot x$.
By induction on the derivation of $M : A$.
\[thm:GinftoPTS\] Let $M : A$ be a derivable [$\Gamma_\infty$]{} judgment. Then there is a type annotated judgment $\Gamma \vdash M:A$ derivable in the associated PTS, such that $\Gamma$ contains exactly the variables in ${{\rm hfv}}(M) \cup {{\rm hfv}}(A)$.
By induction on the derivation of $M:A$ we show there exists a type annotated context $\Gamma(M,A)$ such that $\Gamma(M,A) \vdash M:A$. Note that $\Gamma(M,A)$ depends on the *derivation* of the judgment $M:A$, not just on the terms $M$ and $A$.
- Immediate.
- By induction, $\Gamma(A,s) \vdash A:s$. So $\Gamma(A,s), x^A : A \vdash x^A : A$.
- By induction, $\Gamma(M,A) \vdash M:A$ and $\Gamma(B,s) \vdash B:s$, and we also know that $A{=_{\beta}}B$.\
So $\Gamma(M,A){\ltimes}\Gamma(B,s)\vdash M : B$ by Thinning and the (conv) rule.
- By induction, $\Gamma(F,\Pi \dot x{{:}}A.B) \vdash F:\Pi \dot x {{:}}A. B$ and $\Gamma(M,A) \vdash M:A$.\
So $\Gamma(F,\Pi \dot x{{:}}A.B){\ltimes}\Gamma(M,A)\vdash F\,M : B[\dot x := M]$ by Thinning and the (app) rule.
- By induction, $\Gamma(A,s_1) \vdash A:s_1$ and $\Gamma(B,s_2) \vdash B:s_2$.\
If $y^A \notin\Gamma(B,s_2)$, then $\Gamma(A,s_1){\ltimes}\Gamma(B,s_2), y^A:A_i \vdash B:s_2$, so by Thinning and the ($\Pi$) rule we have $\Gamma(A,s_1){\ltimes}\Gamma(B,s_2)\vdash \Pi \dot x{{:}}A. B[y^A := \dot x] : s_3$.\
If $y^A \in{{\rm dom}}(\Gamma(B,s_2))$, then $\Delta , y^A:A \vdash B:s_2$ for some $\Delta \subset \Gamma(B,s_2)$. So by Thinning and the ($\Pi$) rule we have $\Gamma(A,s_1){\ltimes}\Delta\vdash \Pi \dot x{{:}}A. B[y^A := \dot x] : s_3$.
- By induction, we obtain $\Gamma(M,B) \vdash M:B$ and $\Gamma(\Pi \dot x{{:}}A. B[y^A := \dot x],s) \vdash \Pi \dot x{{:}}A.\penalty100 B[y^A := \dot x]:s$.\
If $y^A \notin{{\rm dom}}(\Gamma(M,B))$, then $\Gamma(A,s_1){\ltimes}\Gamma(M,B),
y^A:A_i \vdash M:B$. So by Thinning and the ($\lambda$) rule, we have $\Gamma(\Pi \dot x{{:}}A. B[y^A :=
\dot x],s){\ltimes}\Gamma(M,B)\vdash \lambda \dot x{{:}}A. M[y^A :=
\dot x]: \Pi \dot x{{:}}A. B[y^A := \dot x]$.\
If $y^A \in{{\rm dom}}(\Gamma(M,B))$, then $\Delta , y^A:A \vdash M:B$ for some $\Delta \subset \Gamma(M,B)$, by Lemma \[lem.annothfvt\]. So by Thinning and the ($\lambda$) rule, we have $\Gamma(\Pi \dot x{{:}}A. B[y^A :=
\dot x],s){\ltimes}\Delta\vdash \lambda \dot x{{:}}A. M[y^A := \dot x]:
\Pi \dot x{{:}}A. B[y^A \penalty 100 := \dot x]. $
By Lemma \[lem.hfvdom\], ${{\rm dom}}(\Gamma(M,A)) \supseteq {{\rm hfv}}(M,A)$. Corollary \[cor.annothfv\] lets us strengthen the context to $\Delta \subseteq \Gamma(M,A)$ such that $\Delta \vdash
M:A$ and ${{\rm dom}}(\Delta) ={{\rm hfv}}(M,A)$.
Let $$M : A$$ be a derivable [$\Gamma_\infty$]{} judgment. Take all variables of the form $x^A_i$ occurring in ${{\rm hfv}}(M) \cup {{\rm hfv}}(A)$, and put them in *any* order $x^{A_1}_{i_1}, \dots, x^{A_n}_{i_n}$ such that if $x^{A_k}_{i_k}$ occurs in $A_l$ then $k < l$. Then the following judgment is derivable in the PTS: $$x^{A_1}_{i_1} : A_1, \dots, x^{A_n}_{i_n} : A_n \vdash M : A$$
From the previous Theorem using Permutation.
Conclusion and Further work {#sec:conclusion}
===========================
There are three obvious continuations of this work:
1. The first is to investigate to what extent other type theories than the PTSs admit a [$\Gamma_\infty$]{} presentation.
2. The second is to see how well the approach presented here can be used as a basis of an LCF-style kernel for type theory.
3. The third is to formally develop the theory presented in this paper in a proof assistant.
With respect to the first point: we expect most type theories to have a [$\Gamma_\infty$]{}-variant, although the observations about universe inconsistency [@CSC:MKM2004] arising from merging contexts may complicate the picture for applied type systems such as that of Coq. More important is to investigate how our approach needs to be adapted to support type theories with definitions. As previously noted, in any real implementation, the definitions for defined constants form a more significant part of the contexts $\Gamma$ we are eliminating than the free variables.
We are currently investigating the second point, developing an OCaml implementation for the PTS $\lambda P$ extended with definitions (a system corresponding to the logical framework LF) along the lines of this paper. The main question is how expensive, computationally, the two following operations are:
- The substitutions $[y^A := \dot x]$ that occur in the $\lambda$ and $\Pi$ rules.
- The check of the side-condition $y^A \notin{{\rm hfvT}}(M,B)$ in the $\lambda$ and $\Pi$ rules.
The first is in some sense [[‘[local]{}’]{}]{}, because it does not look inside constant definitions in the environment. To make the second reasonably efficient will be harder. It is possible we need to consider *three* kinds of variables, distinguishing (1) $\dot x$ bound variables, (2) $y^A$ variables to be substituted with bound variables later (essentially, the *eigenvariables* of the ($\Pi$) and ($\lambda$) rules), and (3) $x^A$ variables that do remain free, so they may be considered as [[‘[axiomatic constants]{}’]{}]{} of the system.
In such a system, it is essential that the implementation language can use pointer equality to efficiently determine equality of terms (and in particular, equality of annotations in variable occurrences). This motivated our choice of OCaml, which is such a language. Although in OCaml the comparison function [[“[[`=`]{}]{}”]{}]{} does not have this feature (because floating points NaNs are not taken to be equal to themselves, the system never looks at pointer equality when evaluating [[“[[`x = y`]{}]{}”]{}]{}), the comparison function [[“[[`fun x y -> Pervasives.compare x y = 0`]{}]{}”]{}]{} does.
We are currently working on the third point as well. In ongoing work [@krebbers:gammainf] the second author has formalised a large part of the theory presented in this paper up to the first direction of the correspondence theorem in the proof assistant Coq. However, the presentation in this formal development is slightly different from that of this paper.
- Firstly, we distinguish bound from free variables at the level of PTS pseudo-terms, following existing practice, established since the third author’s work with Pollack in the 1990s [@mckinna-pollack:tlca1993; @mckinna-pollack:jar1999]. Our informal presentation above uses named variables in each case, the so-called *locally named* approach, whereas the formalisation uses the *locally nameless* representation: de Bruijn indices for bound variables and names for free variables. Because we make this difference at the level of PTS pseudo-terms already, we have a canonical representative for each term and therefore need not worry about $\alpha$-equivalence. For further details of both approaches, see [@mckinna-pollack:jar1999; @aydemir:metatheory for example].
- Secondly, variable binding in the ($\Pi$) and ($\lambda$) rules of [$\Gamma_\infty$]{} is handled by substituting free variables for bound variables, rather than bound for free as in Section \[sec:Gi\]. This choice has been used successfully in other formalisations (*ibid.*), and emphasises the conceptual priority of free variables over bound. Recent work by Pollack and Sato compares the two approaches, in a detailed account of canonical representation for languages with binding [@SatoPollack:JSC2010].
Based on the methodology described in [@aydemir:metatheory], we have combined the locally nameless presentation with co-finite quantification to obtain strong induction principles. To be sure that our co-finite presentation is adequate we have proved it to be equivalent to an exists-fresh presentation.
For example the ($\Pi$) rule, in respectively exists-fresh and co-finite presentation, is as follows. $$\begin{array}{llr}
(\Pi\mbox{ exists-fresh})
&
\prooftree
A : s_1\;\;\;\;
B[0:=x^A] : s_2
\justifies
\Pi A .B : s_3
\endprooftree & \mbox{if }(s_1 ,s_2 ,s_3) \in {{\cal R}}\mbox{ and }x^A \notin{{\rm hfv}}(B)\\
& & \\
(\Pi\mbox{ co-finite})
&
\prooftree
A : s_1\;\;\;\;
\forall x \notin L\ .\ B[0:=x^A] : s_2
\justifies
\Pi A .B : s_3
\endprooftree & \mbox{if }(s_1 ,s_2 ,s_3) \in {{\cal R}}\mbox{ and }L \subset_{\mbox{finite}} {{\cal V}}\\
\end{array}$$ Observe that we require $x^A \notin{{\rm hfv}}(B)$ instead of $x^A \notin{{\rm hfvT}}(B)$ (this was also observed by John Boyland). The reason for this is that we have to bind every occurrence of the free variable $x^A$ in $B[0:=x^A]$, hence $x^A$ should not be in ${{\rm FV}}(B)$ either. While the condition is (potentially) more expensive to check, it removes the need for Definition \[def:HFVT\], in favour of the (conceptually) simpler Definition \[def:HFV\].
The other direction of the correspondence theorem uses the second property, Strengthening, of Proposition \[prop.metatheoryPTS\] in an essential way. This presents two difficulties: a practical one, since the existing formalisations of this lemma are highly non-trivial [@mckinna-pollack:jar1999]; and a theoretical one, namely that we may *not* be able to establish a correspondence between traditional and [$\Gamma_\infty$]{} presentations of a given type theory without first establishing strengthening.
More interestingly, from the point of view of the pragmatics of formalisation, for this direction it is essential to abstract over the kinds of free variables used in the definition of PTS judgments. At first this does not seem troublesome, however, many definitions and theorems do not just depend on the kind of free variables but also on finite sets of free variables. Hence we are also required to abstract over various operations on such finite sets. The recently developed finite set library [@lescuyer:containers], based on the new type classes feature in Coq, might be very useful in implementing this abstraction.
#### Acknowledgments
Thanks to Jean-Christophe Filliâtre for details about the architecture of the Coq kernel. We are grateful to the anonymous referees, and to the audience at LFMTP, especially John Boyland, Brigitte Pientka and Andrew Pitts, who made several helpful remarks, especially concerning related work. This research is partially funded by the NWO BRICKS/FOCUS project [[“[ARPA: Advancing the Real use of Proof Assistants]{}”]{}]{} and the NWO cluster <span style="font-variant:small-caps;">Diamant</span>.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
To explain the pulsed emission of the rotation powered pulsars from radio to gamma-ray, the polar cap models, the slot gap models, and the outer gap models are proposed. The recent observations suggest that these models are likely to co-exist in the same magnetosphere. If so, their mutual relation is known to be troublesome [@b25] due to the boundary conditions and the direction of the current which are properly assumed in each acceleration models.
We performed a particle simulation for the global magnetospheric structure. Based on the simulation, we present a new picture of the global structure of the pulsar magnetosphere. It is found that a new dead zone is formed along the current neutral line which separates the oppositely directed current. We shall call this the current- neutral zone. We suggest that the polar cap accelerators and the slot gaps locate above the current-neutral zone, and the outer gap exist between the current neutral zone and the traditional dead zone. We also give an estimate of the super-rotation region.
author:
- 'Shinya <span style="font-variant:small-caps;">Yuki</span> and Shinpei <span style="font-variant:small-caps;">Shibata</span>'
title: 'A Particle Simulation for the Pulsar Magnetosphere: Relationship of Polar Cap, Slot Gap, and Outer Gap'
---
Introduction
============
The mechanism of the pulsed emission from pulsars has been an important unsolved problem for more than forty years. The recent observations of the $\textit{Fermi}$ Gamma-ray Space Telescope has promoted an understanding of the emission from gamma-ray pulsars. The pulse shapes show substantial diversity, but roughly 75$\%$ of the gamma-ray profiles have two peaks, separated by $\gtrsim 0.2$ of rotation phase [@b1]. Additionally, most of the radio-loud gamma-ray pulsars have the first gamma-ray peak lagging the main radio component. This indicates that radio and gamma-ray beams are coming from different emitting regions. Also, the high-confidence gamma-ray from a millisecond pulsar is detected first by $\textit{Fermi}$. After that, it was found that most of the millisecond pulsars have the features of the gamma-ray light curves and spectra like normal pulsars.
Polar cap (PC) (eg, [@b2]), slot gap (SG) (eg, [@b3; @b4]) and outer gap (OG) (eg, [@b5; @b6]) models are promising particle acceleration models to explain pulsed emission from pulsars. In PC models, particle acceleration and resultant gamma-ray emission takes place in the open field line region above the magnetic poles. The gamma-rays undergo the magnetic pair production, which leads to super-exponential spectral cutoff. Then, the produced pairs screen out the accelerating electric field above the so-called pair formation front (PFF). It is generally believed that the observed radio emission is basically coming from upper part of the PC region. The SG, which is the rim of the PC, extends along the last-open-field line. There is no sufficient acceleration to emit gamma-rays producing electron-positron pairs at the low altitude because the field-aligned electric field near the last open field line is weak. Therefore, particle acceleration lasts up to high altitudes of many stellar radii. For this reason, the SG can provide extended high-energy emission above the last-open-field lines. The OG is located between a neighborhood of the null charge surface and the light cylinder along the open field lines. The accelerating electric field exists because of charge depletion, and photon-photon pair production occurs in the gap. The OG can also provide extended high energy emission. In addition, SG and OG models have roughly simple exponential spectral cutoff, which is more gradual than that of PC models.
SG and/or OG models can explain the double-peaked gamma-ray light curves (eg, [@b7; @b8]). On the other hand, PC models have much more difficulty reproducing the observed wide pulse profile because the beam size from the PC is too small. Furthermore, the possibility of gamma-ray emission from the PC, which would exhibit super-exponential cutoff due to magnetic pair attenuation, was eliminated by the spectral observation such as MAGIC and $\textit{Fermi}$ (eg, [@b9; @b1]). Also, there are not only the study of light curves and averaged spectrum but also that of phase-resolved spectra (eg, [@b10; @b11]).
In PC, SG and OG models, acceleration is caused by an electric field parallel to the magnetic field line, and the accelerating electric field is calculated by the Poisson equation for the non-corotational potential: $$- \nabla^2 \Phi = 4 \pi (\rho - \rho_{\mathrm{GJ}}),$$ where $\boldsymbol{E} = -(\boldsymbol{\Omega} \times \boldsymbol{r})
\times \boldsymbol{B}/c - \nabla \Phi$; $\rho$ and $\rho_{\mathrm{GJ}} = - \boldsymbol{\Omega} \cdot \boldsymbol{B} /
(2 \pi c) [1 - (\boldsymbol{\Omega} \times \boldsymbol{r} / c)^2]$ are the charge density and Goldreich-Julian charge density [@b12], respectively; $\boldsymbol{\Omega}$ is the angular momentum; and $\boldsymbol{r}$ is the position vector. Thus, the field-aligned electric field $E_{||} = -\nabla \Phi \cdot \hat{\boldsymbol{B}}$, with $\hat{\boldsymbol{B}}$ unit vector along the magnetic field line, is formed by the difference between local charge density and Goldreich-Julian density.
The boundary conditions of these models, which are required to solve the Poisson equation, are summarized as follows. Most of the PC and SG models assume $E_{||}=0$, as well as $\Phi=0$ at the stellar surface, and $\Phi=0$ along the last open field line. In addition, the condition $E_{||}=0$ is imposed at an upper boundary, ie, PFF. The OG models, which elongate along open field lines, usually have upper and lower boundaries along the field lines; an inner boundary, which is on the star side; and an outer boundary, which is located near the light cylinder. In particular, the lower boundary is supposed to be the last open field line. A standard boundary condition is such that $\Phi=0$ on the upper and lower boundaries, $\Phi=0$ and $E_{||}=0$ on the inner boundary, and $E_{||}=0$ on the outer boundary. The boundary conditions are very important because they affect the structure of the accelerating electric field. However, the screening process on each boundary has not been well modeled or simulated. Therefore, it is not clear whether $\Phi=0$ along the last open field line or not.
There is also a problem with the directions of the current flow. The directions of the current along the magnetic field line above the last open field line are different in these acceleration models. When $\boldsymbol{\mu} \cdot \boldsymbol{\Omega} > 0$, where $\boldsymbol{\mu}$ is the magnetic moment, the direction of the current in the PC and SG is inward, and while the current in the OG is outward. When $\boldsymbol{\mu} \cdot \boldsymbol{\Omega} < 0$, the directions are reversed. Looking at various pulse shapes from radio to gamma-rays, people sometimes speculate all these types of acceleration co-exist. If all of these acceleration models exist above the same last open field line, the directions of the currents in these models are opposite. Therefore, it seems to be impossible that polar-slot gap and OG exist above the same last open field lines consistently from the theoretical prospective.
On the other hand, most of the observed radio and gamma-ray features of both normal and millisecond pulsars are well reproduced by radio emission from upper part of the PC region and gamma-ray emission from the SG or OG (eg, [@b13; @b8]). It is required on the observational side that the PC and SG and/or OG co-exist.
To understand the formation mechanism of the PC, SG and OG in view of electrodynamics, particle simulation for the global magnetosphere would be a powerful way. By using GRAPE [@b14], a massively-parallel special-purpose computer for astronomical N-body simulations at National Astronomical Observatory of Japan, we performed a simulation particularly aimed at understanding the boundary conditions for the PC, SG and OG; the relationship among the models; and the current directions.
The organization of this paper is as follows: Section 2 describes the method of our simulation; Section 3 shows the simulation results and describes the structure of the global pulsar magnetosphere; Section 4 discusses the position relationships of the PC, SG, and OG, and summarizes the results of our simulation. Our main conclusion, that both polar-slot gap and OG can exist in a magnetosphere, is also presented in Section 4.
Method of Simulation
====================
It is necessary to handle formation of the field-aligned electric field $E_{||}$, so that the simulation must not be the usual MHD simulation. A breakthrough can be found by a global particle simulation, which can handle cross-field drift motions due to radiation drag and particle inertia by solving each particle motion. (, hereafter WS) have succeeded in overcoming this heavy computation by using GRAPE. Typical use of GRAPE is for stellar dynamics. However, it should be noted that GRAPE, in which sign-bit is available, can calculate Coulomb force as well as gravitational force because both of them are represented by the inverse-square law: $$\boldsymbol{E}_i = - \sum_{j \neq i}^N q_j \frac{\boldsymbol{r}_j - \boldsymbol{r}_i}{( |\boldsymbol{r}_j - \boldsymbol{r}_i|^2 + \epsilon^2)^{3/2}},$$ where the subscripts $i$, $j$ represent $i/j$th particle; and $N$,$\boldsymbol{r}$, $\epsilon$, $q$, and $\boldsymbol{E}$ are the total particle number, position vector, softening parameter, charge, and electric field, respectively.
In our simulation, we assumed that a star is an aligned rotator and a perfect conductor. To specify the charge sign, we chose aligned ($\boldsymbol{\mu} \cdot
\boldsymbol{\Omega} > 0$), not counter aligned ($\boldsymbol{\mu} \cdot
\boldsymbol{\Omega} < 0$), ie, the poles are negatively charged, while the equator is positively charged. The numerical method follows WS. Magnetospheric plasmas are represented as super-particles. The electric field on the $i$th particle is calculated by the superposition of the vacuum solution $\boldsymbol{E}_{\mathrm v}$ and the space-charge solution $\boldsymbol{E}_{\mathrm q}$, where $$\begin{aligned}
\boldsymbol{E}_{\mathrm v} (\boldsymbol{r}_i) = - \left[ \frac{\Omega \mu R^2 (3\cos^2 \theta -1)}{c r^4_i} - \frac{Q_{\mathrm {sys}}}{r^2_i} \right] \boldsymbol{e}_{\mathrm r} \nonumber
\\ - \left( \frac{2 \Omega \mu R^2 \sin \theta \cos \theta }{c r^4_i} \right) \boldsymbol{e}_{\mathrm \theta},\end{aligned}$$ $$\begin{aligned}
\boldsymbol{E}_{\mathrm q} (\boldsymbol{r}_i) = \sum_{j \neq i}^N \biggl[
\frac{\boldsymbol{r}_i - \boldsymbol{r}_j }{|\boldsymbol{r}_i - \boldsymbol{r}_j|^3}
- \frac{R}{r_j} \frac{\boldsymbol{r}_i - (R/r_j)^2 \boldsymbol{r}_j
}{|\boldsymbol{r}_i - (R/r_j)^2 \boldsymbol{r}_j|^3} \nonumber
\\ - \left( 1 - \frac{R}{r_j} \right) \frac{\boldsymbol{r}_i}{r^3_i} \biggr], \label{eq:01}\end{aligned}$$ where $\boldsymbol{e}_{\mathrm r}$ and $\boldsymbol{e}_{\mathrm \theta}$ are unit vectors in spherical coordinates; $R$, $Q_{\mathrm {sys}}$, and $c$ are the stellar radius, system charge and speed of light, respectively. To satisfy the boundary condition for a perfect conductor, the second and third terms in (\[eq:01\]) representing the effect of mirror charge exist.
As an improvement from WS, we take into account the modification of the magnetic field from the dipole field. By using GRAPE, the magnetic field is calculated by the Biot-Savart law: $$\begin{aligned}
\boldsymbol{B} (\boldsymbol{r}_i) =
B_* \left( \frac{R}{r_i} \right)^3 (\cos \theta \boldsymbol{e}_{\mathrm r} + \frac{1}{2} \sin \theta \boldsymbol{e}_{\mathrm \theta})
\\ - \sum_{j \neq i}^N \frac{q_j \boldsymbol{v}_j \times (\boldsymbol{r}_j - \boldsymbol{r}_i)}{( |\boldsymbol{r}_j - \boldsymbol{r}_i|^2 + \epsilon^2)^{3/2}},\end{aligned}$$ where $\boldsymbol{v}_j$ is the velocity of the $j$th particle.
We made the following assumptions as WS did: (1) particles are emitted freely from the stellar surface; (2) the particles are subject to the radiation drag force; and (3) pair creation occurs where the field-aligned electric field $E_{||}$ is stronger than a critical value $E_{\mathrm {cr}}$. Since we are looking for an axisymmetric steady state, we ignored the effect of the time variation of the fields. Therefore, the problem is electro-static and magneto-static, and whereby GRAPE can be used. Our simulation is as follows: starting from vacuum around the star, the surface charge is pulled out from the star and, as it spreads throughout the magnetosphere, radiation drag takes place; and pairs are created, if $E_{||} > E_{\mathrm {cr}}$. This simulation proceeds in the following steps:
1. Start the calculation from the vacuum around the star.
2. Replace the surface charges on the star by particles.
3. Compute the electromagnetic field at the particles’ positions.
4. Compute the force exerted on the particles, and advance the particle positions and velocities by the equations of motion.
5. Create electron-positron pairs where $E_{||} > E_{\mathrm {cr}}$.
6. Remove particles which fall back to the star or leave the simulation box through the outer boundary.
7. Go back to step (2) unless the steady state is established.
Our simulation is started from the vacuum state. Then, the surface charges are pulled out by the induced electromotive force which is much larger than the work function. They are emitted as charged particles above the stellar surface. The emission is expected to last unless $E_{||} \neq 0$. Therefore, we emitted the surface charge $\sigma$ per unit area which is defined as $E_{||} = 4 \pi (\sigma - \sigma_{\mathrm {GJ}}) (\boldsymbol{n} \cdot \hat{ \boldsymbol{B}})$, where $\sigma_{\mathrm {GJ}}$ is the surface charge density when the corotational electric field exists both inside and outside of the star; $\boldsymbol{n}$ is the unit normal vector to the stellar surface; and $\hat{ \boldsymbol{B}}$ is the unit vector of the magnetic field on the stellar surface. The surface charges are emitted $\sim 230$ times in a rotation period.
We solve the equation of motion, $$\frac{{\mathrm {d}} \boldsymbol{p}_i}{{\mathrm {d}} t}
=
q_i \left( \boldsymbol{E} + \boldsymbol{\beta}_i
\times \boldsymbol{B} \right)
+ \boldsymbol{F}_{\mathrm {rad}},$$ where $\boldsymbol{\beta}_i$ is the velocity of the $i$th particle in units of the speed of light, $\boldsymbol{p}_i = \gamma_i m_i \beta_i c$, and $\gamma_i = (1 - \beta_i^2)^{-1/2}$; $\boldsymbol{F}_{\mathrm {rad}} = (2/3) (q_i^2 / R_{\mathrm c}^2) \gamma_i^4
(\boldsymbol{p}_i / |\boldsymbol{p}_i|)$ (if $\beta_i \simeq 1$) is the radiation reaction force, and $R_{\mathrm c}$ is the curvature radius. This approximate expression for the radiation drag force can be used when the force due to the external field is larger than the radiation drag force.
For pair creation, grid points $\boldsymbol{r}_{i,j} = (r_i, \theta_j)$ are provided on the meridional plane. If $E_{||}$ is larger than the critical value $E_{\mathrm {cr}}$, $\alpha |E_{||}(\boldsymbol{r}_{i,j})| / E_{\mathrm {cr}}$ pairs whose initial velocities are zero are generated in the grid spaces. Here the multiplicity $\alpha$ is set to $\alpha = 2$ and $E_{\mathrm {cr}} = 0.25 B_{\mathrm L}$, where $B_{\mathrm L} = \mu/R_{\mathrm L}^3$ is the light cylinder magnetic field, in the present simulation.
The volume of the grid spaces is in proportion to $r^{-2}$, so that pair creation rate per unit volume is in proportion to $r^{^2}$. This reproduces the fact that the density of target photons supplied from the stellar surface decreases in proportion to $r^{2}$. A time interval for the pair creation is also the parameter of the calculation. In the present calculation, it is such that pairs are created $\sim 75$ times in a rotation period.
The outer boundary is located at 10 $R_{\mathrm L}$. Magnetospheric plasmas are expressed as super-particles, and therefore the exaggerated electric force causes a bound motion like positronium. We remove such bound pairs because they behave as if they were a neutral particle, and they would not affect the structure of the magnetosphere. In the steady state, creation and loss of the particles in the simulation balance with each other. Thus, the total charge of the system is determined automatically.
As for normalization, the position vector, magnetic field, electric field, and velocity are normalized as follows: $\bar{ \boldsymbol{r} } = \boldsymbol{r} / R$, $\bar{ \boldsymbol{B} } = \boldsymbol{B} / B_*$, $\bar{ \boldsymbol{E} } = \boldsymbol{E} / B_*$, and $\boldsymbol{\beta} = \boldsymbol{v} / c$, where $B_*$ is the magnetic field strength on the poles; the variables with bar are non-dimensional in this paper. In our simulation, the strength of the magnetic field is normalized by $B_*$, so that our simulation can treat pulsars which have various magnetic field strengths basically. The other major simulation parameters are $\bar{ \Omega } = R/R_{\mathrm L} =
0.2$, $\bar{m} = 10^{-10}$, and $\bar{q} = 10^{-5}$, where $R_{\mathrm L}$ is the light cylinder radius; $\bar{m}$ and $\bar{q}$ are the mass and charge of a super-particle, respectively, in non-dimensional form.
The simulation parameters should be chosen so that what takes place in the simulation reproduces the actual phenomena in the pulsar magnetosphere. At the same time, we have a restriction of the computational power. Our parameters include the mass and charge of the super-particles $\bar{m}$ and $\bar{q}$, time steps of integration, angular velocity $\bar{\Omega}$, frequency of pair creation, critical electric field for pair creation $E_{\mathrm {cr}}$, pair multiplicity $\alpha$, frequency of injection from the stellar surface, and softening parameter of Coulomb force. It is notable that the total number of particles in the simulation box is determined self-consistently but can be controlled by $\bar{q}$ and the parameters of pair creation. We do not have any grid points for the electromagnetic field. We give typical scales of time and length for the simulation, normal pulsars, and MSPs in Table \[tab:00\] and \[tab:01\], where $\gamma_1$ and $\gamma_2$ are the Lorentz factors in units of $10$ and $10^2$, respectively; $P_{-3}$ and $P_{1}$ are the stellar rotation periods in units of $10^{-3}$ sec and $1$ sec, respectively; $B_{8}$ and $B_{12}$ are the magnetic field strength in units of $10^{8}$ G and $10^{12}$ G, respectively; $R_6$ is the stellar radius in units of $10^6$ cm; and $M_2$ is the multiplicity which is defined as $M \equiv n / n_{\mathrm {GJ}}$ in units of $10^2$. The time step was taken to be much smaller than the Larmor period to reproduce drift motions due to the electric field, the gradient of the magnetic field, the centrifugal force, and the radiation drag, and so on as seen in Table \[tab:00\]. To consider the screening effect and the formation of the field-aligned electric field due to inertia, the time step also has to be much smaller than the the period of plasma oscillation. This requirement is also satisfied as seen in Table \[tab:00\]. According to Table 2, the Larmor radii and inertial lengths are enough smaller than the typical scale of the system. The ratio of the plasma frequency $\omega_p$ to the gyro frequency $\omega_g$ changes in such a way that $\omega_p / \omega_g \ll 1$ at the magnetic poles, $\omega_p / \omega_g < 1$ at the light cylinder, and $\omega_p / \omega_g > 1$ at the outer boundary in both simulation and actual pulsars. Thus, the magnetization property of plasma is also simulated well in the simulation and actual pulsars.
Results
=======
----------------------------- -----------------------------
(80mm,80mm)[figure1a.eps]{} (80mm,80mm)[figure1b.eps]{}
(80mm,80mm)[figure1c.eps]{} (80mm,80mm)[figure1d.eps]{}
----------------------------- -----------------------------
Magnetospheric plasmas are represented by $\sim 1 \times 10^5$ super-particles. After integration of $\sim 10$ periods of time, the amount of particles produced at the stellar surface and the gaps, and the amount of removed particles are balanced, so that the total particle number becomes constant in the simulation box. The net charge of the system also converges to a constant value, and therefore we can consider the magnetosphere is in a steady state. The top left panel of figure \[fig:res01\] shows particle distribution and velocity field on the meridional plane in the steady state. Most of the positive and negative particles are located around the equatorial plane and polar regions, respectively. In addition, a near vacuum region appears around the middle-altitudes.
The bottom left panel of figure \[fig:res01\] shows the strength of the field-aligned electric field $E_{||}$. The strong $E_{||}$, which is $\gtrsim 0.25 B_{\mathrm L}$, exists around the null charge surface and above the poles. In total $\sim 3.6 \times 10^4$ electron-positron pairs (super-particle pairs) are generated per rotation.
The former region is identified as the OG (see WS). However, note that the strong $E_{||}$ above the poles still has artificial effects because it was found that it decreases with increasing the frequency of the particle emission on the stellar surface. When we changed this emitting frequency from $\sim 45$ times to $\sim 230$ times per rotation, the $|E_{||}|$ above the poles decreased from $\sim 0.036$ to $\sim 0.016$. This indicates that if the emission frequency becomes higher, the value of $E_{||}$ would approach $E_{||} \sim 0$. However, it is difficult to remove such an artificial effect in the global particle simulation due to a limited computational power. It is not clear at the moment whether the $|E_{||}|$ above the poles includes the essential electric field or not.
The pairs generated in the OG are accelerated in opposite directions by the $E_{||}$. The positive particles are accelerated outward, and flowed out as a pulsar wind. On the other hand, the negative particles are accelerated toward the star by the outward-directed $E_{||}$. If the star solely absorbs the negative particles continuously, it would be charged up negatively. To prevent such charging up, the negative particles are re-emitted from the stellar surface around the magnetic poles and spread into the magnetosphere. In this way, a current passing through the PC originates the negative particles generated in the OG.
In our simulation, particles are traced up to the distance of $10 R_{\mathrm L}$. Most of them flow out beyond the outer boundary as a pulsar wind, but some negative particles emitted from the PC region return to the star. This circulating flow can cross the magnetic field due to the $\boldsymbol{F}_{\mathrm {rad}} \times \boldsymbol{B}$ drift. In order to understand how modification of the magnetic field lines affects the structure of the magnetosphere, we have also performed the simulation with the magnetic field fixed to the dipole field. In the case of the dipole field, the magnetic flux crossing the equatorial plane beyond $R_{\mathrm L}$ and $10 R_{\mathrm L}$ are $\Psi_{\mathrm L}=0.1$ and $\Psi_{\mathrm 10L}=0.01$, respectively. In contrast, in the case of the modified magnetic field, the values are $\Psi_{\mathrm L} \sim 0.14$ and $\Psi_{\mathrm 10L} \sim 0.03$, respectively. The closed magnetic fields are pushed out and the magnetic flux passing through the outer boundary is increased. This indicates that the open flux is increased. The top right panel of figure \[fig:res01\] shows the particle distribution and velocity field in the case of the dipole field. In comparison with the dipole case, it was found that the circulating flow diminishes in the case of the modified magnetic field. As for the total number of particles going through the outer boundary, it is $\sim 3.3 \times 10^4$ per rotation, which corresponds to $\sim 0.6 \dot{N}_{\mathrm {GJ}}$, where $\dot{N}_{\mathrm {GJ}} = (\Omega^2 R^3 B_* ) / (2ec)$ is a current of primary particles flowing out from the single polar cap in Goldreich & Julian model, in the case of the modified magnetic field.
The non-corotational potential $\Phi$ represents deviation of the electric field from the corotational one. We assume $\Phi = 0$ on the stellar surface. The bottom right panel of figure \[fig:res01\] gives the map of $\Phi$, where if the region is white, then $\Phi \approx 0$ (i.e. the corotation region). In our results, the polar regions are reddish due to the potential drop just above the stellar surface, while the low-latitude region, except for the equatorial closed magnetic flux, is bluish due to the outer gap.
Note that the regions other than the white region in the bottom right panel of figure \[fig:res01\] do not always mean $E_{||} \neq 0$. Even if $\Phi \neq 0$, the field-aligned electric field can be zero when $\boldsymbol{B} \cdot \nabla \Phi = 0$. On the other hand, the white region where $\Phi = 0$ is connected with the star through the magnetic field lines on which $E_{||} = 0$, and corotates with the star.
There are two distinguishable corotation white regions. One is the traditional dead zone with closed magnetic flux (the reason why it is rather small shall be discussed later). The other is in the middle latitudes extending along the field lines. This elongated corotation region separates the OG regions with out-going current from the PC regions with in-going current. In other words, the middle-latitude corotation region is formed where the directions of the currents change. Let us call this region “the current-neutral zone”. In our numerical experiment, there is no significant flow of both positive and negative particles along the magnetic field lines in the current-neutral zone. The cloud of negative particles is naturally formed there. The current-neutral zone is formed in both the dipole and modified magnetic field cases, although the locations were slightly different because it is formed along the magnetic field lines.
(80mm,80mm)[figure2.eps]{}
The traditional dead zone is defined by the magnetic field lines closed within the light cylinder, but the one in our simulation does not reach the light cylinder, and is obviously small. Additionally, the shape of the disk does not follow the magnetic field lines. As shown in figure \[fig:res02\], the inner part of the disk corotates with the star, but on the other hand the super-rotation, which is caused by the $E_{||}$ in the gap, took place at the outer part of the disk. The Lorentz factor increases to saturate by radiation drag, and it is $\sim 10$ for the super-particle. This Lorentz factor would correspond in the actual pulsar to $\sim 10^7 P_1^{-1/4} B_{12}^{1/4} R_6^{3/4}$, at which the Lorentz force and the radiation drag are equal to each other near the light cylinder. For this reason, $\boldsymbol{F}_{\mathrm {rad}} \times \boldsymbol{B}$ drift causes cross-field motion within the light cylinder, so that the particles leak from the outer part of the disk toward the light cylinder. The primary force on the particles are the Lorentz force, and therefore shows gyro-motion, thus the approximation formula for the radiation drag force can be applied. The secondary force is radiation drag and causes the drift motion. Thus, the size of the corotational disk becomes smaller. As for the ratio of the magnitude of the electric field to the magnitude of the magnetic field, the electric field is dominant in the wedge-shaped region beyond the light cylinder (figure \[fig:res03\]). This feature, whose apex angle is $\approx 28^\circ$, is similar to the feature found in the force-free solution by Uzdensky (2003), whose apex angle is $\approx 62^\circ$.
(80mm,80mm)[figure3.eps]{}
Discussion and Conclusion
=========================
(80mm,80mm)[figure4.eps]{} \[fig:dis01\]
Let us discuss the structure of the global magnetosphere on the basis of our results. In figure \[fig:dis01\], we summarize electromagnetic properties of each magnetospheric region. There is a possibility that multiple acceleration regions exist in a different way from the traditional picture because of the current-neutral zone.
First, we discuss Region A and Region B. In PC and SG models, the side boundary condition is that $\Phi=0$ along the last open field lines. In our simulation, the side boundary of the polar flows correspond to the rim of Region B. Therefore, the surface of Region B plays the role of the side boundary of PC and SG models, instead of the traditional dead zone. The boundary condition on the neutron star surface depends on conditions such as the surface temperature, the surface magnetic field strength, and the atomic number of matter in the surface layer. In the case of most pulsars, one may assume $E_{||}=0$ and $\Phi=0$ for the boundary conditions on the stellar surface. As described in the previous section, the strong $E_{||}$ just above the magnetic poles (figure \[fig:dis01\]) includes the artificial effects, which are caused by the limited emitting frequency of the particles from the stellar surface. If the emitting frequency is increased, the value of the $E_{||}$ will likely approach zero.
In space-charge limited flow models, in which $E_{||}=0$ and $\Phi=0$ on the stellar surface, particles are accelerated and emit gamma-rays. Then magnetic pair production occurs, and the PFF is formed. We applied a simple calculation method for generating pairs in order to avoid heavy computations. However, if we could apply a more accurate and realistic calculation, the PFF would be properly treated, and the PC and SG would appear in a region bounded by the stellar surface and Region B. Additionally, we also performed a simulation in which pair creation was suppressed above the magnetic poles. Even in this simulation, we obtain an steady state where the OG is formed. This simulation corresponds to a pair-starved polar cap (PSPC) [@b21; @b22], in which the amount of pair creation is not enough to completely screen the accelerating electric field. It was found in this simulation that the structure of Region B was almost unchanged as compared with the previous simulation. This indicates that the current-neutral zone is concrete. In summary, the acceleration, like polar-slot gap or PSPC, may exist in a region bounded by Region B. It is noted that the pair creation in the OG is the key to make the magnetosphere active even in the aligned rotator, although the diocotron instability is also proposed to play a similar role (see, eg, [@b16; @b17]). If pair creation does not occur anywhere, the dome-disk magnetosphere, such as the magnetosphere in @b18, is formed in our simulation. In contrast, if the pair creation occurs in the OG, the generated electrons go back to the star and are re-emitted around the magnetic poles, and then, they flow out to the magnetosphere. In this way, pair creation in the OG can change a static magnetosphere into an active magnetosphere without diocotron instability.
Next, let us discuss the region around the OG. The corotation Region D, where $\Phi=0$, is identified as a traditional dead zone, but Region D is smaller than the traditional one and does not reach the light cylinder. Instead, Region E exists between Region D and the light cylinder. In Region E, $E_{||}=0$, but $\Phi \neq 0$ as seen in figure \[fig:dis01\] because the magnetic field lines which pass through Region E go through the OG, where $E_{||} \neq 0$. The super-rotation arises in Region E due to the gap. Note that the super-rotation appears even if $E_{||} = 0$ in the disk. This means that rotation speeds of particles approach the speed of light within the light cylinder (see figure \[fig:res02\]), and the particles in the region can leak out due to the cross-field motion. The $E_{||}$ in the OG is $\gtrsim 0.25 B_{\mathrm L}$, and the potential drop in the OG is $\sim 20\%$ of the effective electromotive force in the simulation. In the case of actual young pulsars, the pair creation rate is higher, so that the strength of the $E_{||}$ in the gap should be smaller. In addition, super-particles can cross the field lines easier due to their large masses. Therefore, it is thought that the size of Region E in the actual magnetosphere is much smaller than that of the simulation.
(80mm,80mm)[figure5.eps]{}
The azimuthal velocity in the closed field region originates from trans-field potential difference via $\boldsymbol{E}_\bot \times \boldsymbol{B}$ drift motion, which may be given by $$v_\phi = c \varpi \frac{{\mathrm d} \phi}{{\mathrm d} \Psi},
\label{eq:dis02}$$ where $\phi$ is the scalar potential. Let us define the corotating dead zone by $\Psi \le \Psi_{\mathrm {in}}$, and $\phi_{\mathrm {in}} = (\Omega /c) \Psi_{\mathrm {in}}$ on its surface (see figure \[fig:dis02\]). Outside of this corotating dead zone, but within the field line tangent to the light cylinder, denoted by $\Psi_{\mathrm L}$, we have a super-rotating region with closed field lines ($\Psi_{\mathrm {in}} < \Psi < \Psi_{\mathrm L}$). The electric potential on the field line $\Psi_{\mathrm L}$ is $\phi_{\mathrm L} = (\Omega/c) \Psi_{\mathrm L}$ on the stellar surface, but owing to the potential drop $\Delta \Phi$ along the field line in the gap, the potential on the same field line and on the equatorial plane is reduced to $\phi_{\mathrm L} -\Delta \Phi$. Thus, the potential difference between $\Psi_{\mathrm {in}}$ and $\Psi_{\mathrm L}$ is increased to provide super-rotation: $$\begin{aligned}
v_\phi & \approx & c \varpi \frac{\phi_{\mathrm {in}} - (\phi_{\mathrm L} - \Delta \Phi)}{\Psi_{\mathrm {in}} - \Psi_{\mathrm L}} \\
& = & \varpi \Omega \left( 1 + \frac{c}{\Omega} \frac{\Delta \Phi}{\Psi_{\mathrm {in}} - \Psi_{\mathrm L}} \right) > \varpi \Omega,
\label{eq:dis03}\end{aligned}$$ which indicates that somewhere within the light cylinder, the rotation velocity formally exceeds the light speed, say at $\varpi = \varpi_1$. In reality, radiation drag in the azimuthal direction drives trans-field motion in a radial direction, where just within $\varpi_1$, and beyond $\varpi_1$. In our simulation, the thickness of the super-rotation region is artificially enlarged because of the somewhat large value of $E_{\mathrm {cr}}$. The actual thickness of this region $\Delta \varpi = R_{\mathrm L} -
\varpi_1$, can be estimated by (\[eq:dis03\]). Providing $\Delta \varpi \sim R_{\mathrm L} (\Psi_{\mathrm {in}} - \Psi_{\mathrm L}) \Psi_{\mathrm L} \ll R_{\mathrm L}$, we have $\Delta \varpi / R_{\mathrm L} \sim (\Delta \Phi / \phi_{\mathrm L})^{1/2} \approx (\mu/10^{30} {\mathrm G} {\mathrm {cm}}^3)^{1/2} P$, where $\Delta \Phi \sim 10^{13}$ Volt is assumed. In the Crab pulsar, $\Delta \varpi \sim 0.03 R_{\mathrm L}$. However, in old pulsars, where $\Phi$ becomes comparable to $\phi_{\mathrm L}$, the super-rotation region must be significant.
There would be no pair creation on the magnetic field flux between $\varpi_1$ and $\varpi_{\mathrm {in}}$, and the inner part of the OG. If pair creation occurred there, the pairs would not be able to cross the closed magnetic field lines and accumulate. If so, it is expected that the inner part of the OG is filled with the pairs, and the gap would disappear. However, this is inconsistent with the initial assumption that the gap exists. Thus, the pair-starved region must exist in the inner part of the OG.
In the super-rotational region of the actual pulsar, some positrons which generated in the OG flow out at an amount approximate to the amount of the GJ current to the outside of the magnetosphere by the $\boldsymbol{F} \times \boldsymbol{B}$ drift. On the other hand, some electrons return to the star, and the amount of this current is thought to be that of the GJ current, at most. However, the number of pairs generated in the OG is $\sim 100$ times larger than that of the primary particles. This indicates that most of the pairs do not become a current. They might leak as a plasmoid when the particle energy density becomes larger than that of the magnetic field (see figure \[fig:dis02\]). We later suggest magnetic reconnection at the top of Region E.
Region E mostly consists of positrons. The positron flow is divided into two kinds of flows in the simulation with the magnetic field fixed to be dipole. One is a flow along the equatorial plane to escape from the magnetosphere, and the other one is a flow circulating the magnetosphere to return back to the star from the poles. However, in the case of the modified magnetic field, the circulating component disappears, and most of the positrons flow out as a pulsar wind. In the present simulation, the mean free path for pair creation is set at zero for simplification, and the generated pairs do not have initial outward velocities. This approximation works well at acceleration regions where pair creation occur actively, but does not work well around the acceleration regions whose structure would not be reproduced correctly in our simulation. Generated pairs are immediately charge-separated by the accelerating electric field in the gap, and only positrons can escape out of the magnetosphere. In the actual pulsar, the mean free path has a finite value, so that gamma-rays can cross the field and produce pairs outside the OG. Thus, not only positrons, but also electron-positron pairs can escape outside the OG. As pointed out by @b19, gamma-rays from the gap always propagate in the convex sides of the magnetic field lines, and subsequently, copious electron-positron pairs must quench the field-aligned electric field above the OG (Region C and Region F). In addition to the trans-field effect, some pairs generated in the outer part of the OG can escape to Region E and Region F due to the initial outward velocities. In this way, Region C, Region E, and Region F should be filled with quasi-neutral plasmas, so that the condition $E_{||}=0$ would be satisfied there. If the pair creation rate is made higher than the present value so that the current reaches the GJ value, some magnetic field lines would be opened. In that case, the magnetic neutral sheet around the equatorial plane (Region G) would be formed. The inner edge of the neutral sheet, which is the open-close boundary of the magnetic field structure, is called Y-point. @b20 studied the Y-point via an axisymmetric particle-in-cell simulation. They demonstrated that the magnetic reconnection occurred quasi-periodically and thereby plasma was heated and accelerated. They also pointed out that the electric-field dominant region ($|\boldsymbol{E}| > |\boldsymbol{B}|$) was formed around the equatorial plane. Our simulation treats steady state, so that time-dependent phenomena (eg, the magnetic reconnection) do not appear. However, Region G also holds the condition $|\boldsymbol{E}| > |\boldsymbol{B}|$ in our simulation (figure \[fig:res03\]).
As for the directions of the currents, the direction in PC, SG, and PSPC models is inward and the direction in the OG model is outward. The mutual location of the accelerating regions with opposite currents and how the return current is maintained are troublesome problems. One possible picture may be indicated in our simulation. Note that the inward and outward currents in our simulation exist along the different field lines separated by the current-neutral zone (Region B).
The size of the OG is determined by balance of the amount of supply and loss of pairs. We find in our simulation that the size of the OG decreases with increasing supply of pairs. In the actual pulsars, the supply is influenced by the pair creation rate, i.e., the gamma-ray emissivity and the mean free path for photon-photon collision to make pairs. On the other hand, the loss is due to escaping of pairs from the closed field region. However, the loss process is somewhat uncertain, depending on the structure of the closed-open boundary and on the super-rotation. The super-rotation is affected by the mass-charge ratio of the super-particle. As we discussed previously, the actual size of the super-rotation region and the gap size in the simulation can be artificially enlarged. Since the pair creation rate is treated as a model parameter and the particle loss is merely included quantitatively, it is at the moment difficult to discuss actual gap size as function of the pulsar parameters such as the period, field strength and X-ray flux from the surface.
Some observational results support the aforementioned picture. According to @b23, *Fermi*-LAT have detected pulsations from PSR B1509-58 up to 1 GeV with a light curve presenting two gamma-ray peaks P1 and P2, which was observed at phases $0.96 \pm 0.01$ and $0.33 \pm 0.02$, respectively. P2 was consistent with the outer magnetosphere geometry although a sharp cutoff was not well explained. Considering the extension to 1 GeV, and the magnetic pair and photon-splitting attenuation limits, P1, which precedes the radio peak, must also originate in the outer magnetosphere. However, its phase location could be explained by neither two-pole caustic (TPC) models (eg, [@b24]), which might be realized in SG acceleration models, nor OG models. PSPC models can produce the gamma-ray peak preceding the radio peak [@b8], so that PSPC emission might be able to explain the phase location of P1. Thus to explain both peaks, both pair-starved and non-pair-starved (gap) models are required to co-exist [@b23]. As mentioned, we also performed the simulation without pair creation above the PC, and found that the structure around the OG was not affected by eliminating pairs in the PCs. This indicates that both pair-starved and non-pair-starved gaps exist simultaneously.
@b1 discovered that the distribution of the separation of the two gamma-ray peaks appears to be bimodal with no strong dependence on pulsar $\dot{E}$ (or age). The $\Delta$ distributions of the separation for each PC, TPC, and OG models are calculated by @b13. However, the distribution does not show the bimodal feature, but if two acceleration regions exist in a magnetosphere, this bimodal feature could be explained. The bimodal distribution might indicate that we look at different gamma-ray emitting regions from various viewing angles. From this viewpoint, we can naturally understand the fact that there is no relationship between $\Delta$ and $\dot{E}$.
It is considered that the radio emission from most pulsars come from above the magnetic poles. The radio emission is thought to be related to pair creation although the mechanism is still not fully understood. Therefore, an acceleration region to produce pairs is required above the magnetic poles. The possibility that gamma-rays originate from the PC was ruled out by recent observations (eg, [@b9; @b1]). Nevertheless, acceleration regions above the magnetic poles, and polar cap cascade are required to explain the radio emission. Peak separations between radio peaks and gamma-ray peaks indicate that acceleration regions are located at different regions. These facts also support the idea that at least two kinds of acceleration regions exist in a pulsar magnetosphere. Explaining both radio and gamma-ray light curves would require both low- and high- altitude acceleration regions.
In conclusion, the above observational facts strongly suggest that both OG and polar-slot gap/PSPC exist. From our simulation, these acceleration regions are located at above and below the current-neutral zone. Our present simulation is not enough to resolve the structure of the PC and SG because the sizes of these regions are insignificant compared to the global magnetosphere. As a future work, high-accuracy simulations, which have a larger number of particles, or follow more realistic pair creation processes, are required to discuss the PC and SG in more detail. Also, it would be challenging to study the magnetospheric structure in the case of oblique rotators.
Acknowledgments {#acknowledgments .unnumbered}
===============
Numerical computations were carried out on GRAPE system at Center for Computational Astrophysics, CfCA, of National Astronomical Observatory of Japan.
[99]{} Abdo, A. A. 2010a, , 714, 927 Abdo, A. A. 2010b, , 187, 460 Aliu, E. 2008, Sci, 322, 1221 Cheng, K. S., Ho, C., & Ruderman, M. 1986, , 300, 500 Daugherty, J. K., & Harding, A. K. 1996, , 458, 278 Dyks, J., & Rudak, B. 2003, , 598, 1201 Goldreich, P., & Julian, W. H. 1969, , 157, 869 Harding, A. K. 2009, in Neutron Stars and Pulsars, ed. Becker, W. (Berlin and Heidelberg: Springer-Verlag) 521 Harding, A. K., Stern, J. V., Dyks, J., & Frackowiak, M. 2008, , 680, 1378 Hirotani, K. 2008, , 688, L25 Krause-Polstorff, J., & Michel, F. C. 1985, , 144, 72 Makino, J., Fukushige, T., Koga, M., & Namura, K. 2003, , 55, 1163 Muslimov, A. G., & Harding, A. K. 2003, , 588, 430 Muslimov, A. G., & Harding, A. K. 2004, , 606, 1143 Muslimov, A. G., & Harding, A. K. 2004, , 617, 471 Muslimov, A. G., & Harding, A. K. 2009, , 692, 140 Pétri, J. 2009, , 503, 1 Romani, R. W., & Watters, K. P. 2010, , 714, 810 Spitkovsky, A., & Arons, J. 2002, in Neutron Stars in Supernova Remnants, ed. Slane, P. O., & Gaensler, B. M., (San Francisco: ASP) 81 Takata, J., & Chang, H. K. 2007, , 670, 677 Takata, J., Shibata, S., & Hirotani, K. 2004, , 354, 1120 Umizaki, M., & Shibata, S. 2010, , 62, 131 Venter, C., Harding, A. K., & Guillemot, L. 2009, , 707, 800 Wada, T., & Shibata, S. 2007, , 376, 1460 Watters, K. P., Romani, R. W., Weltevrede, P., & Johnston, S. 2009, , 695, 1289
| {
"pile_set_name": "ArXiv"
} |
---
author:
- 'Shahram Esmaeilsabzali[^1]'
- Rupak Majumdar
- Thomas Wies
- 'Damien Zufferey[^2]'
bibliography:
- 'main.bib'
subtitle: Extended Version
title: Dynamic Package Interfaces
---
[^1]: Shahram Esmaeilsabzali was at MPI-SWS when this work was done.
[^2]: Damien Zufferey was at IST Austria when this work was done.
| {
"pile_set_name": "ArXiv"
} |
---
abstract:
- '[This erratum contains the full corrected version of the paper [*Complete set of Feynman rules for the Minimal Supersymmetric Standard Model*]{} [@PRD41]. The complete set of Feynman rules for the R-parity conserving MSSM is listed, including the most general form of flavour mixing. Propagators and vertices are computed in t’Hooft-Feynman gauge, convenient for perturbative calculations beyond the tree level.]{}'
- '[The complete set of Feynman rules for the $R$-parity conserving MSSM is listed, including the most general form of flavour mixing. Propagators and vertices are computed in t’Hooft-Feynman gauge, convenient for perturbative calculations beyond the tree level.]{}'
author:
- |
Janusz Rosiek[^1]\
Institute of Theoretical Physics, Warsaw University\
Ho[ż]{}a 69, 00-681 Warsaw, POLAND
date: 'November 6, 1995'
title: |
\
\
Complete set of Feynman rules for the MSSM – [*erratum*]{}
---
Instead of putting on the web next version of the [*“erratum”*]{}, with few more errors in the original paper corrected, I decided to resubmit the “integrated” version, i.e. full paper text with all necessary corrections included - it should be easier to use in this way. I also used this opportunity to correct the most irritating features of the notation used in the original [*Phys. Rev.*]{} [**D**]{} paper, making it, wherever possible, closer to commonly used naming conventions. However, I kept unchanged the “final” matrix notation for the interactions vertices in the mass eigenstates basis, as it proved to be very useful in compactifying many complicated loop calculations.
Most of the expressions for mass matrices, mixing angles and vertices listed in [@PRD41] have been checked during the calculations of the 1-loop radiative corrections in the gauge and Higgs sectors of the MSSM [@CPR; @DELTAR] and in calculations of various CP violation/FCNC processes [@POROSA; @MIPORO; @BCRS] The 1-loop corrections were calculated in on-shell renormalization scheme, which provide a very strict test of correctness of all formulae entering the expressions for the renormalized quantities: most of the errors in Feynman rules lead immediately to non-cancellation of the divergencies. Only the most exotic vertices like 4-sfermion couplings, several rarely used 2 Higgs boson-2 sfermion couplings were not used and did not pass this test yet. Other vertices can be with good probability considered as checked.
Formulae for diagonalization of mass matrices and most of the vertices listed in [@PRD41] are accessible also as the ready FORTRAN codes. They are part of the bigger library for calculation of the 1-loop radiative corrections in on-shell renormalization scheme to the MSSM neutral Higgs production and decay rates. This library can be found at:\
[*http://www.fuw.edu.pl/rosiek/physics/neutral\_higgs.html*]{}
[**In order to avoid too many replacements in hep-ph archive, and to speed up the process of introducing further corrections if any were found, I will always put the most recent version of this collection of Feynman rules on my private web page. It will be available at:**]{}\
[*http://www.fuw.edu.pl/rosiek/physics/prd41.html*]{}\
[**I will not update the hep-ph version of the [*erratum*]{} any more!**]{}
[**Acknowledgments**]{}
I wish to thank to P. Chankowski, A. Belayev, J. R. Espinosa, J. Hagelin, M.Laine, M. Misiak, P. Slavich, M. Rauch and others who called my attention to some of the errors listed in this erratum. Especially, I would like to thank T. Ewerth for very careful reading of its integrated version.
[**Complete set of Feynman rules for the MSSM**]{}
[published in [*Phys. Rev.*]{} [**D41**]{} (1990) 3464]{}
[Janusz Rosiek]{}
Institute of Theoretical Physics, Warsaw University\
Ho[ż]{}a 69, 00-681 Warsaw, POLAND
Introduction {#sec:intro}
============
Since many years the minimal supersymmetric extension of the standard model (MSSM) has been the subject of intensive studies. Practical calculations in the MSSM are usually tedious because of its complexity. For easier reference, in this paper we complete all Feynman rules for the MSSM in the t’Hooft-Feynman gauge, convenient for calculations of loop corrections. We put special emphasis on including the most general form of flavour mixing allowed in the $R$-parity conserving MSSM.
Some of the mass matrices and vertices given in the paper, mainly related to the Higgs and -inos sectors, has been listed in other papers (see e.g. [@ref1; @ref2]). For completeness, we write down those formulae once again, in order to collect in one place and fixed convention the full set of rules needed to calculate any process in the frame of the MSSM. The spinor conventions used in the paper follow those given in [@ref1].
The paper has the following structure. Section \[sec:general\] contains short review of the rules of constructing SUSY Yang-Mills theories. In section \[sec:fields\] we define fields and parameters present in the MSSM Lagrangian. The physical content of the theory - the mass eigenstates fields are given in section \[sec:eigen\]. In section \[sec:gauge\] we define the gauge used throughout the paper. Section \[sec:lagr\] contains the MSSM Lagrangian expressed in terms of the physical fields. A short summary and comments on the choice of the parameters of the model are given in section \[sec:comment\]. In \[sec:app\_a\] we write down the MSSM Lagrangian in terms of the initial fields, before the $SU(2)\times U(1)$ symmetry breaking. \[sec:app\_b\] contains the full set of Feynman rules corresponding to the Lagrangian of section \[sec:lagr\].
General structure of the SUSY models {#sec:general}
====================================
Supersymmetric Yang-Mills theories contain two basic types of fields - gauge multiplets $(\lambda^{a},V_{{\mu}}^{a})$ in the adjoint representation of a gauge group $G$ and matter multiplets $(A_i,\psi_i)$ in some chosen representations of $G$. By $\lambda$ and $\psi$ we denote here fermions in two-component notation, $A_i$ are complex scalar fields and $V_{\mu}^a$ are spin-1 real vector fields (the spinor notation and conventions used in the paper are the same as those explained in Appendix A of ref. [@ref1]). To construct the Lagrangian of such a theory we follow the rules given in ref. [@ref1]. In the strictly supersymmetric case one has the following terms (summation convention is used unless stated otherwise):
1. Kinetic terms.
2. Self interaction of gauge multiplets: three- and four-gauge boson vertices plus additional interaction of gauginos and gauge fields: igf\_[abc]{} \^[a]{} \^ |\^b V\_\^c
3. Interactions of the gauge and matter multiplets ($T^{a}$ is the Hermitian group generator in the representation corresponding to the given multiplet): &-gT\_[ij]{}\^[a]{} V\_\^[a]{} (|\_[i]{} |\^ \_[j]{} +iA\_[i]{}\^ \_ A\_[j]{}),&\
&ig T\_[ij]{}\^[a]{} (\^[a]{} \_[j]{} A\_[i]{}\^ -|\^[a]{} |\_[i]{} A\_[j]{}),&\
&g\^[2]{} (T\^[a]{} T\^[b]{})\_[ij]{} V\_\^[a]{} V\^[b]{} A\_[i]{}\^ A\_[j]{}.&
4. Self interactions of the matter multiplets. For technical reasons it is convenient to define the so called “superpotential” $W$ as at most cubic gauge-invariant polynomial which depends on scalar fields $A_i$, but not on $A_i^{\star}$ [@ref1] (alternatively, the superpotential can be defined as a function of the superfields). Introducing two auxiliary functions: &F\_i = W/A\_i&\
&D\^a = g A\_i\^ T\_[ij]{}\^a A\_j& one can write the scalar supersymmetric potential as: V = D\^a D\^a + F\_i\^ F\_i Yukawa interactions are given by: - ( [\^[2]{} WA\_i A\_j]{} \_i\_j + )
In the case of semisimple groups $G=G_1 \times \ldots \times G_n$ in the expressions above one should substitute terms of the form $gVT$ by sums $\Sigma g_i V_i T_i$ and similarly for gauginos. There are also $n$ vertices $\lambda-\bar\lambda-V$ and $n$ terms $\frac{1}{2}D^2$ in the scalar potential. For $U(1)$ factors there are no gaugino-gaugino-gauge interaction, and, by convention, the product $g
T_{ij}^a V_{\mu}^a$ is replaced by $\frac{1}{2}g y_i \delta_{ij}
V_{\mu} $ (no sum over $i$), where $y_i$ is the $U(1)$ quantum number of the matter multiplet $(A_i,\psi_i)$ (similarly for gauginos). In general the $U(1)$ $D$-field may be shifted by the so called Fayet-Iliopoulos term $\xi$ [@ref4]: D = g y\_i A\_i\^ A\_i + $\xi\neq 0$ may introduce dangerous quadratic divergences into the theory. In most realistic models (including MSSM) this term is absent.
This completes the construction of a strictly supersymmetric theories. To build models which preserve the most important feature of such theories - absence of quadratic divergences - and which simultaneously are experimentally acceptable, it is necessary to add to the above Lagrangian explicit soft SUSY breaking terms. The most general form of appropriate expressions can be written down as [@ref5]: m\_1 A\^2 + m\_2 A\^2 + y(A\^3 + ) + m\_3 (\^[a]{} \^[a]{} + ) $A^2$ and $A^3$ denote symbolically all possible gauge invariant combinations of scalar fields. These terms split the masses of scalars and fermions present in the SUSY multiplets and introduce new, non-supersymmetric trilinear scalar couplings.
MSSM field and coupling structure {#sec:fields}
=================================
To obtain the realistic supersymmetric version of the Standard Model one should extend the field content of the theory by adding appropriate scalar or fermionic partners to the ordinary matter and gauge fields. As stated in the previous section, the superpotential can only be constructed as a function of fields and not of their complex conjugates. Therefore it is not possible to give masses to all fermions using only one Higgs doublet - at least two with opposite $U(1)$ quantum numbers are necessary. The full field content of the MSSM is listed below:
1. Multiplets of the gauge group $SU(3) \times SU(2) \times U(1)$:
- $B_{\mu},\lambda_B$ - weak hypercharge gauge fields, coupling constant $g_1$
- $A_{\mu}^i,\lambda_A^i$ - weak isospin gauge fields, coupling constant $g_2$
- $G_{\mu}^a,\lambda_G^a$ - QCD gauge fields, coupling constant $g_3$
2. Matter multiplets - we assume that three matter generations exist, so the index $I$ (and similarly all capital $I,J,K\ldots$ indices in the rest of the paper) runs from $1$ to $3$ (such notation can be immediately generalized to the case of $N$ generations).
----------------------------------------- ----------------------------- ----------------
Scalars Fermions $U(1)$ charge
\[3mm\] $ L^I = \left( $\Psi_L^I = \left( $-1$
\begin{array}{c} \begin{array}{c}
\tilde{\nu}^I \\ \nu^I \\
\tilde{e}_L^{-I} e^{-I}
\end{array} \end{array}
\right)$ \right)_L$
\[3mm\] $R^I = \tilde{e}_R^{+I}$ $\Psi_R^I = (e_L^{-I})^c$ $2$
\[3mm\] $ Q^I = \left( $ \Psi_Q^I = \left( $\frac{1}{3}$
\begin{array}{c} \begin{array}{c}
\tilde{u}^{I} \\ u^{I} \\
\tilde{d}_{L}^{I} d^{I}
\end{array} \end{array}
\right) $ \right)_L $
\[3mm\] $D^I = \tilde{d}_R^{I\star}$ $\Psi_D^I = (d_{L}^{I})^c $ $\frac{2}{3}$
\[3mm\] $ U^I = \tilde{u}_{R}^{I\star}$ $\Psi_U^I = (u_{L}^{I})^c $ $-\frac{4}{3}$
\[3mm\] $ H^1 = \left( $\Psi_H^1 = \left( $-1$
\begin{array}{c} \begin{array}{c}
H^{1}_{1} \\ {\Psi}^1_{H1} \\
H_{2}^{1} {\Psi}^1_{H2}
\end{array} \end{array}
\right)$ \right)$
\[3mm\] $ H^2 = \left( $ \Psi_H^2 = \left( $1$
\begin{array}{c} \begin{array}{c}
H^{2}_{1} \\ {\Psi}^2_{H1} \\
H_{2}^{2} {\Psi}^2_{H2}
\end{array} \end{array}
\right)$ \right) $
----------------------------------------- ----------------------------- ----------------
The $SU(3)$ indices are not written explicitly. We assume that $Q$ quarks and squarks are QCD triplets, $D$ and $U$ fields - QCD antitriplets.
In order to define the theory we have to write down the superpotential and introduce the soft SUSY breaking terms (without them even using two Higgs doublets it is impossible to break spontaneously the gauge symmetry). The most general form of the superpotential which does not violate gauge invariance and the SM conservation laws is: W = \_[ij]{} H\^1\_i H\^2\_j + \_[ij]{} Y\_l\^[IJ]{} H\^1\_i L\_j\^I R\^J + \_[ij]{} Y\_d\^[IJ]{} H\^1\_i Q\_j\^I D\^J + \_[ij]{} Y\_u\^[IJ]{} H\^2\_i Q\_j\^I U\^J Some bilinear and trilinear terms (e.g. $\varepsilon_L^I
{\epsilon}_{ij} H_i^2 L_j^I$) are gauge invariant but break lepton and/or barion number conservation, so we do not include them. In general, presence of such terms can be forbidden by requiring the preservation of additional global symmetry of the model, so called $R$-parity (for review see e.g. [@rparity]): R= (-1)\^[L+3B+2S]{} Soft breaking terms can be divided into several classes:
1. Mass terms for the scalar fields. &- m\_[H\_1]{}\^2 H\_i\^[1]{} H\_i\^1 - m\_[H\_2]{}\^2 H\_i\^[2]{} H\_i\^2 - (m\_L\^2)\^[IJ]{} L\_i\^[I]{} L\_i\^J - (m\_R\^2)\^[IJ]{} R\^[I]{} R\^J&\
&- (m\_Q\^2)\^[IJ]{} Q\_i\^[I]{} Q\_i\^J - (m\_D\^2)\^[IJ]{} D\^[I]{} D\^J - (m\_U\^2)\^[IJ]{} U\^[I]{} U\^J &
2. Mass terms for gauginos. M\_1 \_B \_B + M\_2 \_A\^i \_A\^i + M\_3 \_G\^a \_G\^a +
3. Trilinear couplings of the scalar fields corresponding to the Yukawa terms in the superpotential. m\_[12]{}\^2 \_[ij]{} H\_i\^1 H\_j\^2 + \_[ij]{} A\_l\^[IJ]{} H\_i\^1 L\_j\^I R\^J + \_[ij]{} A\_d\^[IJ]{} H\_i\^1 Q\_j\^I D\^J + \_[ij]{} A\_u\^[IJ]{} H\_i\^2 Q\_j\^I U\^J +
4. Trilinear couplings of the scalar fields, different in the from from the Yukawa terms in the superpotential (sometimes called “non-analytic terms” as they involve charge conjugated Higgs fields). Usually such couplings are not considered as they are not generated in the most popular SUSY-breaking models. A\_l\^[’IJ]{} H\_i\^[2]{} L\_i\^I R\^J + A\_d\^[’IJ]{} H\_i\^[2]{} Q\_i\^I D\^J + A\_u\^[’IJ]{} H\_i\^[1]{} Q\_i\^I U\^J +
In general constants $\mu$, $m_{12}^2$, Yukawa matrices, squark and gaugino masses and the trilinear soft couplings may be complex. One can perform three operations which eliminate unphysical degrees of freedom. First, it is possible to change globally the phase of one of the Higgs multiplets in such a way, that the constant $m_{12}^2$ becomes a real number. Then the equations for the vacuum expectations values of the Higgs fields involve only real parameters (at least on the tree level). Second, one can redefine simultaneously the phases of all fermions in the model removing the complex phase from one of gaugino mass parameter (see e.g. [@POROSA]). Paralelly one should redefine the phases of other couplings to absorb the change of the phase of the Higgs multiplet. The last operation is the same as in the SM - by the rotation of the fields
------------------------------------------------------ -- -------------------
$(Q_i^I, \Psi_{Qi}^I) \ra V_{Qi}^{IJ} (Q_i^J, (no sum over $i$)
\Psi_{Qi}^J)$
\[3mm\] $(U^I,\Psi_U^I) \ra V_U^{IJ} (U^J,\Psi_U^J)$
\[3mm\] $(D^I,\Psi_D^I) \ra V_D^{IJ} (D^J,\Psi_D^J)$
\[3mm\]
------------------------------------------------------ -- -------------------
and similarly for leptons, one can diagonalize the matrices $Y_l^{IJ}$, $Y_u^{IJ}$ and $Y_d^{IJ}$ obtaining Yukawa couplings of the form $ {\epsilon}_{ij} Y_l^{I} H_{i}^{1} L_{j}^{I} R^{I} $ etc. Simultaneous rotations of quark and squark fields lead to the so-called super-KM basis (see e.g. [@MIPORO]). After the proper redefinition of the parameters the matrices $V_Q$, $V_U$, $V_D$, $V_L$ and $V_R$ disappear from the Lagrangian leaving as their trace the Kobayashi-Maskawa matrix $K$, appearing in many expressions containing both quarks and squarks: K = V\_[Q\_1]{}\^ V\_[Q\_2]{} Of course, sfermion mass matrices in the super-KM basis may be still non-diagonal, i.e. Yukawa and soft matrices need not to be diagonalized by the same rotations.
In the rest of the paper, in particular in section \[sec:eigen\] and in \[sec:app\_a\] we use the [*rotated*]{} soft breaking parameters. For example, if the initial Yukawa and trilinear scalar couplings were $Y_u^{(0)}$ and $A_u^{(0)}$, respectively, the rotation of the quark and squark fields to the super-KM basis lead to diagonal Yukawa coupling $Y_u = V^{\dagger}_{Q1}Y_u^{(0)}V_U$ and new trilinear scalar coupling $A_u = V^{\dagger}_{Q1}A_u^{(0)} V_U$, which are then used in other expressions. The procedure of the redefinition of the squark mass matrices remains some freedom. We have chosen the redefined left squark mass matrix parameter in such a way that the Kobayashi-Maskawa matrix multiplies $m_Q^2$ in the up-squark mass matrix.
The full Lagrangian written in terms of the initial fields (before $SU(2)\times U(1)$ symmetry breaking) is given in \[sec:app\_a\].
Physical spectrum of the MSSM {#sec:eigen}
=============================
In the previous section we defined the field content and all parameters of the MSSM. To obtain the physical spectrum of particles present in the theory one should carry out the standard procedure of gauge symmetry breaking via vacuum expectation values of the neutral Higgs fields and find the eigenstates of the mass matrices for all fields. The VEV’s of the Higgs fields satisfy the following equations ($\theta$ denotes the Weinberg angle, $s_W=\sin\theta$, $c_W=\cos\theta$, $e = g_2 s_W = g_1 c_W$): <H\^1> = [1]{} (
[c]{} v\_1\
0
) <H\^2> = [1]{} (
[c]{} 0\
v\_2
) v\_1 &=& - m\_[12]{}\^2 v\_2\
v\_2 &=& - m\_[12]{}\^2 v\_1 Parameters of the above set of equations are constrained by the condition that $v_1$ and $v_2$ should reproduce the proper values of the gauge boson masses.
The physical fields of the MSSM can be identified as follows:
1. Gauge bosons. Eight gluons $g_{\mu}^a$ and the photon $F_{\mu}$ are massless, bosons $W_{\mu}^{\pm}$ and $Z_{\mu}$ have masses M\_Z &=& [e 2 s\_W c\_W]{} (v\_1\^2 +v\_2\^2)\^\
M\_W &=& [e 2 s\_W]{} (v\_1\^2 +v\_2\^2)\^
2. Charged Higgs scalars. Four charged Higgs scalars exist, two of them with the mass M\_[H\_1\^]{}\^2 = M\_W\^[2]{} + m\_[H\_1]{}\^2 + m\_[H\_2]{}\^2 + 2||\^2 and the other two massless. In the physical (unitary) gauge $H_2^{\pm} (\equiv G^{\pm})$ are eaten by $W$ bosons and disappear from the Lagrangian. Fields $H_1^+$ and $H_2^+$ are related to the initial Higgs fields by the rotation matrix $Z_H$ : (
[c]{} H\_2\^[1]{}\
H\_1\^2
) = Z\_H (
[c]{} H\_1\^+\
H\_2\^+
)\
Z\_[H]{} = (v\_1\^2+v\_2\^2)\^[- ]{} (
[rr]{} v\_2 &-v\_1\
v\_1 &v\_2
)
3. Neutral Higgs scalars. If the Lagrangian contains only real parameters the neutral Higgs particles have well defined CP eigenvalues - two of them are scalars, the other two are pseudoscalars. This is no longer true if some parameters are complex. Nevertheless, in both cases it is convenient to divide neutral Higgses into two classes.
- “Scalar” particles $H_i^0$ , $i=1,2 $, defined as:
$\sqrt{2} \Re H_i^i = Z_R^{ij} H_j^0 + v_i$ (no sum over $i$)
The matrix $Z_R$ and the masses of $H_i^0$ can be obtained by diagonalizing the $M_R^2$ matrix: Z\_R\^T (
[cc]{} - m\_[12]{}\^2 [v\_2v\_1]{} + [e\^2 v\_1\^2 4 s\_W\^2 c\_W\^2]{} & m\_[12]{}\^2 - [e\^2 v\_1 v\_2 4 s\_W\^2 c\_W\^2]{}\
m\_[12]{}\^2 - [e\^2 v\_1 v\_2 4 s\_W\^2 c\_W\^2]{} & - m\_[12]{}\^2 [v\_1v\_2]{} + [e\^2 v\_2\^2 4 s\_W\^2 c\_W\^2]{}
) Z\_R = (
[cc]{} M\^2\_[H\^0\_1]{} &0\
0 & M\^2\_[H\^0\_2]{}
)
- “Pseudoscalar” particles $A_i^0$ , $i=1,2$:
$ \sqrt{2} \Im H_i^i = Z_H^{ij} A_j^0$ (no sum over $i$)
$A_1^0 (\equiv A^0)$ has the mass $M_A^2 = m_{H_1}^2 + m_{H_2}^2 +
2|\mu|^2$, $A_2^0 (\equiv G^0)$ is the massless Goldstone boson which disappears in the unitary gauge. The $Z_H$ matrix is the same as in the case of the charged Higgs bosons.
Matrix notation used in the paper is convenient in the case of non-unitary gauge, when the Goldstone bosons are explicitly present in the Lagrangian and enter the calculations together with the physical Higgs particles. In order to compare our expressions with the more commonly used notation one should substitute: Z\_H = (
[cc]{} &-\
&
) Z\_R = (
[cc]{} &-\
&
) where: &=& [v\_2v\_1]{}\
2&=& 2\
It is also worth remembering that due to the SUSY structure of the model the Higgs boson masses fulfill two interesting tree-level relations: M\_[H\_1\^+]{}\^2 &=& M\_A\^2 + M\_W\^2\
M\_[H\_1\^0]{}\^2 + M\_[H\_2\^0]{}\^2 &=& M\_A\^2 + M\_Z\^2
4. Matter fermions (quarks and leptons) have masses (note that $Y_l^I, Y_d^I$ are defined as negative): m\_\^I = 0 & & m\_e\^I = - [v\_1 Y\_l\^I ]{}\
m\_d\^I = - [v\_1 Y\_d\^I]{} & & m\_u\^I = [v\_2 Y\_u\^I]{}
5. Charginos. Four 2-component spinors $(\lambda_A^1,
\lambda_A^2, \Psi_{H2}^1, \Psi_{H1}^2)$ combine to give two 4-component Dirac fermions $\chi_1, \chi_2$ corresponding to two physical charginos. The chargino mixing matrices $Z_+$ and $Z_-$ are defined by the condition: (Z\_-)\^T (
[cc]{} M\_2 & [ev\_2 s\_W]{}\
[ev\_1 s\_W]{} &
) Z\_+ = (
[cc]{} M\_[\_1]{} & 0\
0 & M\_[\_2]{}
) The unitary matrices $Z_-,Z_+$ are not uniquely specified - by changing their relative phases and the ordering of the eigenvalues it is possible to choose $M_{\chi_i}$ to be positive and $M_{\chi_2} >
M_{\chi_1}$. The fields $\chi_i$ are related to the initial spinors as below: \_[H1]{}\^2 &=& Z\_+\^[2i]{} \_i\^+\
\_[H2]{}\^1 &=& Z\_-\^[2i]{} \_i\^- \_[i]{} = (
[c]{} \^[+]{}\_[i]{}\
|\_[i]{}\^[-]{}
)\
\_A\^ && [\^1\_A i\^2\_A ]{} = iZ\_\^[1i]{} \_i\^
6. Four 2-component spinors $(\lambda_B, \lambda_A^3,
\Psi_{H1}^1, \Psi_{H2}^2)$ combine into four Majorana fermions $\chi_i^0$, $i =1 \ldots 4 $, called neutralinos. The formulas for mixing and mass matrices are the following: Z\_N\^T (
[cccc]{} M\_1 & 0 & [-ev\_[1]{} 2c\_W]{} & [ev\_[2]{} 2c\_W]{}\
0 & M\_2 & [ev\_[1]{} 2s\_W]{} & [-ev\_[2]{} 2s\_W]{}\
[-ev\_[1]{} 2c\_W]{} & [ev\_[1]{} 2s\_W]{} & 0 & -\
[ev\_[2]{} 2c\_W]{} & [-ev\_[2]{} 2s\_W]{} &- & 0
) Z\_N &=& (
[ccc]{} M\_[\^0\_1]{} &&0\
&&\
0&& M\_[\^0\_4]{}
) \_B &=& iZ\_N\^[1i]{} \_i\^0\
\_A\^3 &=& iZ\_N\^[2i]{} \_i\^0\
\_[H1]{}\^1 &=& Z\_N\^[3i]{} \_i\^0 \_[i]{}\^[0]{} = (
[c]{} \_[i]{}\^[0]{}\
|\_[i]{}\^[0]{}
)\
\_[H2]{}\^[2]{} &=& Z\_N\^[4i]{} \_[i]{}\^[0]{}
7. $SU(3)$ gauginos do not mix. In four component notation one has eight gluinos $\Lambda_G^a$ with masses $|M_3|$. \_G\^a = (
[r]{} -i\^a\_G\
i|\_G\^a
)
8. Three complex scalar fields $L_1^I$ form three sneutrino mass eigenstates $\tilde\nu^I$ with masses given by diagonalization of a matrix ${\cal M}_{\nu}^2$ : L\_1\^I &=& Z\_\^[IJ]{} \^J\
Z\_\^ [M]{}\_\^2 Z\_ &=& (
[ccc]{} M\_[\_1]{}\^2 &&0\
&&\
0&&M\_[\_3]{}\^2
)\
[M]{}\_\^2 &=& [e\^2(v\^2\_1-v\^2\_2) 8 s\_W\^2 c\_W\^2]{} 1 + m\_L\^2 Sneutrinos are neutral but [*complex*]{} scalars.
9. Fields $L_2^I$ and $R^I$ mix to give six charged selectrons $L_i$, $i=1 \ldots 6$: L\_2\^I = Z\_L\^[Ii]{} L\^-\_i R\^I = Z\_L\^[(I+3)i]{} L\^+\_i Z\_L\^ (
[cc]{} ([M]{}\_L\^2)\_[LL]{} & ([M]{}\_L\^2)\_[LR]{}\
([M]{}\_L\^2)\_[LR]{}\^ & ([M]{}\_L\^2)\_[RR]{}
)Z\_L = (
[ccc]{} M\_[L\_1]{}\^2 &&0\
&&\
0&&M\_[L\_6]{}\^2
) ([M]{}\_L\^2)\_[LL]{} &=& [e\^2(v\_1\^2-v\_2\^2)(1-2c\_W\^2) 8 s\_W\^2 c\_W\^2]{} [1]{} + [v\_1\^2 Y\_l\^22]{} + (m\_L\^2)\^T\
([M]{}\_L\^2)\_[RR]{} &=& - [e\^2(v\_1\^2-v\_2\^2) 4c\_W\^2]{} [1]{} + [v\_1\^2 Y\_l\^22]{} + m\_R\^2\
([M]{}\_L\^2)\_[LR]{} &=& [1]{} (v\_2 (Y\_l \^ - A\_l\^[’]{}) + v\_1 A\_l )
10. Fields $Q_{1}^{I}$ and $U^{I}$ turn into six up squarks $U_{i}$. Q\_1\^I = Z\_U\^[Ii]{} U\^+\_i U\^I = Z\_U\^[(I+3)i]{} U\^-\_i Z\_U\^T (
[cc]{} ([M]{}\_U\^2)\_[LL]{} & ([M]{}\_U\^2)\_[LR]{}\
([M]{}\_U\^2)\_[LR]{}\^ & ([M]{}\_U\^2)\_[RR]{}
) Z\_U\^ = (
[ccc]{} M\_[U\_1]{}\^2 &&0\
&&\
0&&M\_[U\_6]{}\^2
) ([M]{}\_U\^2)\_[LL]{} &=& - [e\^2(v\_1\^2-v\_2\^2)(1-4c\_W\^2) 24 s\_W\^2 c\_W\^2]{} [1]{} + [v\_2\^2 Y\_u\^2 2]{} + (Km\_Q\^2K\^)\^T\
([M]{}\_U\^2)\_[RR]{} &=& [e\^2(v\_1\^2-v\_2\^2) 6c\_W\^2]{} [1]{} + [v\_2\^2 Y\_u\^2 2]{} + m\_U\^2\
([M]{}\_U\^2)\_[LR]{} &=& - [1]{} (v\_1 (A\_u\^[’]{} + Y\_u \^) + v\_2 A\_u) One should note that $Z_U$ is defined with the complex conjugate comparing to definitions of $Z_L$ (and $Z_D$ below). With such a definition, all positively charged sfermion fields in the MSSM Lagrangian in section \[sec:eigen\] are multiplied by $Z_X^{ij}$, negatively charged by $Z_X^{ij\star}$. This makes easier to control correctness of various calculations including complex parameters.
11. Finally one has six down-squarks $D_i$ composed from fields $Q_2^I$ and $D^I$: Q\_2\^I = Z\_D\^[Ii]{} D\^-\_i D\^I = Z\_D\^[(I+3)i]{} D\^+\_i Z\_D\^ (
[cc]{} ([M]{}\_D\^2)\_[LL]{} & ([M]{}\_D\^2)\_[LR]{}\
([M]{}\_D\^2)\_[LR]{}\^ & ([M]{}\_D\^2)\_[RR]{}
)Z\_D = (
[ccc]{} M\_[D\_1]{}\^2 &&0\
&&\
0&&M\_[D\_6]{}\^2
) ([M]{}\_D\^2)\_[LL]{} &=& - [e\^2(v\_1\^2-v\_2\^2)(1+2c\_W\^2) 24 s\_W\^2 c\_W\^2]{} [1]{} + [v\_1\^2 Y\_d\^2 2]{} + (m\_Q\^2)\^T\
([M]{}\_D\^2)\_[RR]{} &=& - [e\^2(v\_1\^2-v\_2\^2) 12c\_W\^2]{} [1]{} + [v\_1\^2 Y\_d\^2 2]{} + m\_D\^2\
([M]{}\_D\^2)\_[LR]{} &=& [1]{} (v\_2 (Y\_d \^ - A\_d\^[’]{}) + v\_1 A\_d )
We have now completely defined all the physical fields existing in the MSSM:\
------------------------ ------------------------------- ----------------- --------------------
Photon $F_{\mu}$
Gauge bosons $Z^0_{\mu}, W_{\mu}^{\pm}$
Gluons $g_{\mu}^a$ $a=1 \ldots 8$
Gluinos $\Lambda_G^a$ $a=1 \ldots 8$ (Majorana spinors)
Charginos $\chi_i$ $i=1,2$ (Dirac spinors)
Neutralinos $\chi_i^0$ $i =1\ldots 4$ (Majorana spinors)
Neutrinos $\nu^I$ $I=1 \ldots 3$ (Dirac spinors)
Electrons $e^I$ $I=1 \ldots 3$ (Dirac spinors)
Quarks $u^I, d^I$ $I =1 \ldots 3$ (Dirac spinors)
Sneutrinos ${\tilde\nu}^I$ $I=1 \ldots 3$
Selectrons $ L_i^{\pm}$ $i=1 \ldots 6$
Squarks $ U^{\pm}_i, D^{\pm}_i$ $i =1 \ldots 6$
Higgs particles:
charged $H_1^{\pm}$ ($\equiv H^\pm$)
neutral “scalar” $H_1^0, H_2^0$ ($\equiv H,h$)
neutral “pseudoscalar” $ A^0_1$ ($\equiv A^0$)
------------------------ ------------------------------- ----------------- --------------------
Not for all the possible values of input parameters one can obtain reasonable sets of particle masses. For instance, for some choices of the Higgs sector data the $SU(2)\times U(1)$ symmetry would not be broken or, on the opposite, incorrect values of squark sector parameters can lead to negative values of their masses and in consequence to the color symmetry breaking [@QCD_BREAK]
Choice of the gauge {#sec:gauge}
===================
As long as one considers various processes in the spontaneously broken gauge theory in the tree approximation, the most natural and preferred choice is the unitary gauge in which the unphysical Goldstone bosons are absent from the Lagrangian and Feynman rules. When one wants to calculate higher order corrections, one must include the ghost loops suitable for the given gauge. In such case it is much more efficient to use the t’Hooft-Feynman gauge, in which the Goldstone fields appear explicitly in the calculations, but ghost vertices are relatively simple. For our model the appropriate choice for the gauge fixing terms is: L\_[GF]{} &=& - [12]{} (\^ G\_\^a )\^2 - [12]{} (\^ A\_\^3 + M\_Z c\_W G\^0 )\^2 - [12]{} (\^ B\_ - M\_Z s\_W G\^0 )\^2\
&-& [12]{} (\^ A\_\^[1]{} + [i]{} M\_W (G\^+ - G\^-))\^2 - [12]{} (\^ A\_\^2 - [1]{} M\_W (G\^+ + G\^-))\^2\
&=& - [12]{} (\^ G\_\^a )\^2 - [12]{} (\^ Z\_ )\^2 - [12]{} (\^ F\_ )\^2 - (\^ W\^+\_) (\^ W\^-\_ )\
&-& M\_Z G\^0 \^ Z\_ - iM\_W (G\^+ \^ W\^-\_ - G\^- \^ W\^+\_) - M\_Z\^2 (G\^0)\^2 - M\_W\^2 G\^+ G\^-
In some calculations it may be convenient to use even more complicated version of the above expression, with different gauge fixing parameters for various gauge groups (see e.g. [@CPR]).
The interaction Lagrangian {#sec:lagr}
==========================
Although we consider only the minimal extension of the standard model, the full set of Feynman rules for such a theory in the gauge described in section \[sec:gauge\] is very complicated. In this section we write down the interaction part of the MSSM Lagrangian. The propagators and vertices suitable for the chosen gauge are collected in \[sec:app\_b\]. Of course, one can obtain from them rules for tree calculations in the unitary gauge by setting $H_{2}^{{\pm}}$ and $A_{2}^{0}$ to zero and neglecting the ghost terms.
It is convenient to divide all terms in the Lagrangian into classes corresponding to the different types of particles taking part in the interactions (the quark, squark and gluino vertices which contain the QCD coupling constant $g_3$ are collected together as the separate class).
We start from two technical remarks explaining the notation used through the rest of the paper. First, the expression “$ + \mathrm{
H.c.}$” always refers only to the line in which it was used. Second, after the diagonalization of the superpotential the Yukawa matrices $Y_l^{IJ}, Y_u^{IJ}, Y_d^{IJ}$ change into $Y_l^{I} {\delta}^{IJ},
Y_u^{I} {\delta}^{IJ}, Y_d^{I} {\delta}^{IJ}$ and simultaneously sums of the type $A^{IJ} Y_l^{JK} B^{KL}$ convert into $A^{IJ} Y_l^{J}
B^{JL}$ etc., containing the capital indices $I,J,K\ldots$ more than twice; nevertheless, one should always use the summation convention in those cases. This should not lead to any misunderstandings.
.
Interactions of gauge bosons and superscalars.
i\) quark-squark-gauge interactions (the color indices are not written explicitly): &-& e|[u]{}\^I \^ u\^I F\_ - ie(U\_[i]{}\^[-]{} \^ U\_[i]{}\^[+]{}) F\_ + e|[d]{}\^I \^ d\^I F\_ + ie(D\_[i]{}\^[+]{} \^ D\_[i]{}\^[-]{}) F\_\
&-& [e 2s\_Wc\_W]{} |[u]{}\^I \^ (P\_L - s\_W\^2) u\^I Z\_ - [ie 2s\_Wc\_W]{} (Z\_U\^[Ii]{} Z\_U\^[Ij]{} - s\_W\^2\^[ij]{}) (U\_[i]{}\^[-]{} \^ U\_[j]{}\^[+]{}) Z\_\
&+& [e 2s\_Wc\_W]{} |[d]{}\^I \^ (P\_L - s\_W\^2) d\^I Z\_ + [ie 2s\_Wc\_W]{} (Z\_D\^[Ii]{} Z\_D\^[Ij]{} - s\_W\^2\^[ij]{})(D\_[i]{}\^[+]{} \^ D\_[j]{}\^[-]{}) Z\_\
&-& [e s\_W]{} K\^[JI]{} |[d]{}\^I \^ P\_L u\^J W\_\^- - [ie s\_W]{} Z\_D\^[Ii]{} Z\_U\^[Jj]{} K\^[JI]{} (D\_[i]{}\^[+]{} \^ U\_[j]{}\^[+]{})W\_\^[-]{} +\
&+& e\^2 F\_ F\^ U\_[i]{}\^[-]{} U\_[i]{}\^[+]{} + [2e\^2 3s\_Wc\_W]{}(Z\_U\^[Ii]{} Z\_U\^[Ij]{} - \^[ij]{} s\_W\^2) Z\_ F\^ U\_[i]{}\^[-]{} U\_[j]{}\^[+]{}\
&+& [e\^23c\_W\^2]{} Z\_ Z\^ U\_[i]{}\^[-]{} U\_[j]{}\^[+]{}\
&+& e\^2 F\_ F\^ D\_[i]{}\^[+]{} D\_[i]{}\^[-]{} + [e\^2 3s\_Wc\_W]{}(Z\_D\^[Ii]{} Z\_D\^[Ij]{} - \^[ij]{}s\_W\^2) Z\_ F\^ D\_[i]{}\^[+]{} D\_[j]{}\^[-]{}\
&+& [e\^23c\_W\^2]{} Z\_ Z\^ D\_[i]{}\^[+]{} D\_[j]{}\^[-]{}\
&+& [e\^2 2s\_W\^2]{} Z\_U\^[Ii]{} Z\_U\^[Ij]{} W\_\^[+]{} W\^[-]{} U\_[i]{}\^[-]{} U\_[j]{}\^[+]{} + [e\^2 2s\_W\^2]{} Z\_D\^[Ii]{} Z\_D\^[Ij]{} W\_\^[+]{} W\^[-]{} D\_[i]{}\^[+]{} D\_[j]{}\^[-]{}\
&-& [e\^2 6s\_Wc\_W]{} Z\_D\^[Ii]{} Z\_U\^[Jj]{} K\^[JI]{} D\_[i]{}\^[+]{} U\_[j]{}\^[+]{} (Z\^ s\_W - F\^c\_W) W\_\^[-]{} +
ii\) lepton-slepton-gauge interactions &-& [e 2s\_Wc\_W]{} |\^I \^ P\_L \^I Z\_ - [ie 2s\_Wc\_W]{} (\^[I]{} \^ \^I) Z\_ + e|[e]{}\^I \^ e\^I F\_ + ie(L\_[i]{}\^[+]{} \^ L\_[i]{}\^[-]{})F\_\
&+& [e 2s\_Wc\_W]{} |[e]{}\^I \^ (P\_L - 2s\_W\^2) e\^I Z\_ + [ie 2s\_Wc\_W]{} (Z\_L\^[Ii]{} Z\_L\^[Ij]{} - 2s\_W\^2 \^[ij]{}) (L\_[i]{}\^[+]{} \^ L\_[j]{}\^[-]{})Z\_\
&-& [e s\_W]{} |\^I \^ P\_L e\^I W\_\^[+]{} - [ie s\_W]{} Z\_\^[IJ]{} Z\_L\^[Ii]{} (L\_[i]{}\^[+]{} \^ \^J) W\_\^[-]{} +\
&+& [e\^2 4s\_W\^2c\_W\^2]{} Z\_ Z\^ \^[I]{} \^I + e\^2 F\_ F\^ L\_[i]{}\^[-]{} L\_[i]{}\^[+]{} + [e\^2s\_Wc\_W]{} (Z\_L\^[Ii]{} Z\_L\^[Ij]{} - 2 \^[ij]{} s\_W\^2 ) F\_ Z\^ L\_[i]{}\^[-]{} L\_[j]{}\^[+]{}\
&+& [e\^2c\_W\^2]{} (\^[ij]{} s\_W\^2 + [1-4s\_W\^2 4s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{} )Z\_ Z\^ L\_[i]{}\^[-]{} L\_[j]{}\^[+]{}\
&+& [e\^2 2s\_W\^2]{} W\_\^[+]{} W\^[-]{} \^[I]{} \^I + [e\^2 2s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{} W\_\^[+]{} W\^[-]{} L\_[i]{}\^[-]{} L\_[j]{}\^[+]{}\
&+& [e\^2 s\_Wc\_W]{} Z\_\^[IJ]{} Z\_L\^[Ii]{} \^J L\_[i]{}\^[+]{} W\_\^[-]{} (Z\^ s\_W - F\^ c\_W) +
.
Interactions of the Higgs particles and gauge bosons. For more concise notation, we define the auxiliary matrices A\_M\^[ij]{} = Z\_R\^[1i]{} Z\_H\^[1j]{} - Z\_R\^[2i]{} Z\_H\^[2j]{} C\_R\^i = v\_1 Z\_R\^[1i]{} + v\_2 Z\_R\^[2i]{} The Higgs-gauge interaction has the form: &+& [e 2s\_Wc\_W]{} A\_M\^[ij]{} (H\^0\_i \^ A\^0\_j) Z\_ + [ie2 s\_W c\_W]{} (H\^+\_i \^ H\^-\_i) ( (c\_W\^2 - s\_W\^2) Z\_ + 2 s\_W c\_W F\_)\
&-& [ie 2s\_W]{} A\_M\^[ij]{} (H\^0\_i \^ H\^-\_j) W\_\^[+]{} - [e 2s\_W]{} (A\^0\_i \^ H\^-\_i) W\_\^[+]{} +\
&+& [e\^2 2s\_W\^2]{} C\_R\^i (W\_\^[+]{} W\^[-]{} + [1 2c\_W\^2]{} Z\_ Z\^) H\^0\_i - ( [e M\_Wc\_W]{} (Z\^ s\_W - F\^c\_W) W\_\^[+]{} H\_[2]{}\^[-]{} + )\
&+& [e\^24 s\_W\^2 c\_W\^2]{} ( (c\_W\^2 - s\_W\^2)\^2 Z\_ Z\^ + 4 s\_W c\_W (c\_W\^2 - s\_W\^2) Z\_ F\^ + 4 s\_W\^2 c\_W\^2 F\_ F\^) H\^+\_i H\^-\_i\
&+& [e\^2 4s\_W\^2]{} (W\_\^[+]{} W\^[-]{} + [1 2c\_W\^2]{} Z\_ Z\^)(H\^0\_i H\^0\_i + A\^0\_i A\^0\_i) + [e\^2 2s\_W\^2]{} W\_\^[+]{} W\^[-]{} H\^+\_i H\^-\_i\
&+& [e\^2 2s\_Wc\_W]{} A\_M\^[ij]{} (Z\^ s\_W - F\^ c\_W) W\_\^[+]{} H\^-\_j H\^0\_i - [ie\^2 2s\_Wc\_W]{} (Z\^ s\_W - F\^c\_W) W\_\^+ H\^-\_i A\^0\_i +
.
Higgs-lepton and Higgs-quark interactions. &+& [1]{} Y\_l\^I Z\_R\^[1i]{} |[e]{}\^I e\^I H\^0\_i - [i]{} Y\_l\^I Z\_H\^[1i]{} |[e]{}\^I \_5 e\^I A\^0\_i - Y\_l\^I Z\_H\^[1i]{} (|[e]{}\^I P\_L \^I H\^-\_i + )\
&+& [1]{} Y\_d\^I Z\_R\^[1i]{} |[d]{}\^I d\^I H\^0\_i - [1]{} Y\_u\^I Z\_R\^[2i]{} |[u]{}\^I u\^I H\^0\_i - [i]{} Y\_d\^I Z\_H\^[1i]{} |[d]{}\^I \_5 d\^I A\^0\_i + [i]{} Y\_u\^I Z\_H\^[2i]{} |[u]{}\^I \_5 u\^I A\^0\_i\
&+& |[d]{}\^I (-Y\_d\^I Z\_H\^[1i]{} P\_L + Y\_u\^J Z\_H\^[2i]{} P\_R) K\^[JI]{} u\^J H\^-\_i +
.
Interactions of the charginos and neutralinos with the gauge bosons: &-& e|\_i \^ \_i F\_ - [e 2s\_Wc\_W]{} |\_i \^ (Z\_+\^[1i]{} Z\_+\^[1j]{} P\_L + Z\_-\^[1i]{} Z\_-\^[1j]{} P\_R + (c\_W\^2 - s\_W\^2) \^[ij]{} ) \_j Z\_\
&+& [e s\_W]{} |\_j \^ \^0\_i W\^+\_ +\
&+& [e 4 s\_W c\_W]{} |\_i\^0 \^ ( (Z\_N\^[4i]{} Z\_N\^[4j]{} - Z\_N\^[3i]{} Z\_N\^[3j]{}) P\_L - (Z\_N\^[4i]{} Z\_N\^[4j]{} - Z\_N\^[3i]{} Z\_N\^[3j]{}) P\_R)\^0\_j Z\_
.
Interactions of the charginos and neutralinos with the superscalars.
i\) interactions with squarks (superscript “C” denotes the charge conjugated spinor): &+& U\^-\_i |\^0\_j u\^I +\
&+& D\^+\_i |\^0\_j d\^I +\
&+& U\^+\_i |[d]{}\^I K\^[JI]{} \_j\^C +\
&-& D\^+\_i |\_j K\^[JI]{} u\^J +
ii\) interactions with sleptons &+& [e s\_Wc\_W]{} Z\_\^[IJ]{} (Z\_N\^[1i]{} s\_W - Z\_N\^[2i]{} c\_W)\^[J]{} |\^0\_i P\_L \^I +\
&+& |\^0\_j e\^I L\^+\_i +\
&-& |\_i\^C ( [e s\_W]{} Z\_+\^[1i]{} P\_L + Y\_l\^I Z\_-\^[2i]{} P\_R) Z\_\^[IJ]{} e\^I \^[J]{} +\
&-& ( [e s\_W]{} Z\_L\^[Ii]{} Z\_-\^[1j]{} + Y\_l\^I Z\_L\^[(I+3)i]{} Z\_-\^[2j]{})|\_j P\_L \^I L\_i\^+ +
.
Interactions of the charginos and neutralinos with the Higgs particles. &+& [e 2s\_Wc\_W]{} |\^0\_i \^0\_j H\^0\_k\
&-& [ie 2s\_Wc\_W]{} |\^0\_i \^0\_j A\^0\_k\
&-& [e s\_W]{} |\_i \_j H\^0\_k\
&+& [ie s\_W]{} |\_[i]{} \_j A\^0\_k\
&+& [e s\_Wc\_W]{} |\_[j]{} \^0\_i H\^+\_k +
.
Self-interactions of the gauge bosons. &-& ie (W\^+\_ W\^-\_ (\^ F\^ - \^ F\^) + F\_ (W\_\^[-]{} \^ W\^[+]{} - W\^+\_ \^ W\^[-]{} + W\^+\_ \^ W\^[-]{}) )\
&-& [ie c\_Ws\_W]{} (W\_\^[+]{} W\_\^[-]{} (\^ Z\^ - Z\^ F\^) + Z\_ (W\_\^[-]{} \^ W\^[+]{} - W\_\^[+]{} \^ W\^[-]{} + W\_\^[+]{} \^ W\^[-]{}) )\
&+& [e\^2 2s\_W\^2]{} (g\^ g\^ - g\^ g\^)W\_\^[+]{} W\_\^[+]{} W\_\^[-]{} W\_\^[-]{} + e\^2 (g\^ g\^ - g\^ g\^)F\_ F\_ W\_\^[+]{} W\_\^[-]{}\
&+& [e\^2c\_W\^2s\_W\^2]{} (g\^ g\^ - g\^ g\^) Z\_ Z\_ W\_\^[+]{} W\_\^[-]{} + [e\^2c\_Ws\_W]{} (g\^ g\^ + g\^ g\^ - 2g\^ g\^) Z\_ F\_ W\_\^[+]{} W\_\^[-]{}
.
Ghost terms. &-& [ies\_W]{} (Z\_ c\_W + F\_ s\_W) \[(\^ |\^-) \^- - (\^ |\^+) \^+\]\
&+& [ies\_W]{} W\_\^[+]{} \[(\^ |\_Z c\_W + \^ |\_F s\_W) \^- - (\^ |\^+) (\_Z c\_W + \_F s\_W)\]\
&+& [ies\_W]{} W\_\^[-]{} \[-( \^ |\_Z c\_W + \^ |\_F s\_W) \^+ + (\^ |\^-) (\_Z c\_W + \_F s\_W)\]\
& -& [ e\^2 4s\_W\^2]{} v\_i Z\_R\^[ij]{} ( [1 c\_W\^2]{} |\_Z \_Z + |\^[+]{} \^[+]{} + |\^[-]{} \^[-]{} ) H\^0\_j + [ieM\_W 2s\_W]{} (|\^[-]{} \^[-]{} - |\^[+]{} \^[+]{}) A\^0\_2\
& +& [eM\_W 2s\_Wc\_W]{}(|\_Z \^[-]{} - (c\_W\^2 - s\_W\^2) |\^+ \_Z - |\^[+]{} 2 s\_W c\_W \_F) H\^+\_2\
& +& [eM\_W 2s\_Wc\_W]{}(|\_Z \^[+]{} - (c\_W\^2 - s\_W\^2) |\^- \_Z - |\^[-]{} 2 s\_W c\_W \_F) H\^-\_2
.
Scalar potential of the Higgs particles. This is the first part of the huge and complicated quartic potential of the 27 scalar fields appearing in the theory. To streamline notation we define four further auxiliary matrices: A\_H\^[ij]{} &=& Z\_H\^[1i]{} Z\_H\^[1j]{} - Z\_H\^[2i]{} Z\_H\^[2j]{} A\_R\^[ij]{} = Z\_R\^[1i]{} Z\_R\^[1j]{} - Z\_R\^[2i]{} Z\_R\^[2j]{}\
A\_P\^[ij]{} &=& Z\_R\^[1i]{} Z\_H\^[2j]{} + Z\_R\^[2i]{} Z\_H\^[1j]{} B\_R\^[i]{} = v\_1 Z\_R\^[1i]{} - v\_2 Z\_R\^[2i]{} The Higgs potential is expressed as: &-& [e\^2 8s\_W\^2c\_W\^2]{} A\_R\^[ij]{} B\_R\^k H\^0\_i H\^0\_j H\^0\_k - [e\^2 8s\_W\^2c\_W\^2]{} A\_H\^[ij]{} B\_R\^k A\^0\_i A\^0\_j H\^0\_k\
&-& ( [e\^2 4s\_W\^2c\_W\^2]{} A\_H\^[ij]{} B\_R\^k + [eM\_W 2s\_W]{} (A\_P\^[kj]{} \^[1i]{} + A\_P\^[ki]{} \^[1j]{})) H\^+\_i H\^-\_j H\^0\_k\
&+& [ieM\_W 2s\_W]{} \_[ij]{} \^[1k]{} H\^+\_i H\^-\_j A\^0\_k + [ie\^2 4s\_W\^2]{} A\_P\^[ij]{} \_[kl]{} H\^0\_i A\^0\_j H\^+\_k H\^-\_l\
&-& [e\^2 32s\_W\^2c\_W\^2]{} A\_R\^[ij]{} A\_R\^[kl]{} H\^0\_i H\^0\_j H\^0\_k H\^0\_l - [e\^2 32s\_W\^2c\_W\^2]{} A\_H\^[ij]{} A\_H\^[kl]{} A\^0\_i A\^0\_j A\^0\_k A\^0\_l\
&-& [e\^2 16s\_W\^2c\_W\^2]{} A\_R\^[ij]{} A\_H\^[kl]{} H\^0\_i H\^0\_j A\^0\_k A\^0\_l - [e\^2 8s\_W\^2c\_W\^2]{} A\_H\^[ij]{} A\_H\^[kl]{} H\^+\_i H\^+\_k H\^-\_j H\^-\_l\
&-& [e\^2 4s\_W\^2]{} ( [1 2c\_W\^2]{} A\_R\^[ij]{} A\_H\^[kl]{} + A\_P\^[ik]{} A\_P\^[jl]{}) H\^0\_i H\^0\_j H\^+\_k H\^-\_l\
&-& [e\^2 4s\_W\^2]{} ( [1 2c\_W\^2]{} A\_H\^[ij]{} A\_H\^[kl]{} + \_[ik]{} \_[jl]{}) A\^0\_i A\^0\_j H\^+\_k H\^-\_l
It is easy to see that the couplings $H^+_i H^-_j A^0_k$ and $H^+_i
H^-_j A^0_k H^0_l$ disappear in the unitary gauge.
.
Interactions of the sleptons and Higgs bosons.
i\) three scalar (two sleptons and one Higgs) couplings: &-& [e\^2 4s\_W\^2c\_W\^2]{} B\_R\^i \^[I]{} \^I H\^0\_i\
&+& Z\_\^[IJ]{} ( [-e\^2 4s\_W\^2]{} C\_R\^j Z\_L\^[Ii]{} + (A\_l\^[IK]{} Z\_L\^[(K+3)i]{} + [v\_1]{} (Y\_l\^I)\^2 Z\_L\^[Ii]{}) Z\_H\^[1j]{}.\
&& . + (A\_l\^[’IK]{} Z\_L\^[(K+3)i]{} - \^ Y\_l\^I Z\_L\^[(I+3)i]{}) Z\_H\^[2j]{} )\^J L\^+\_i H\^-\_j +\
&+& [i]{} ((A\_l\^[IJ]{} Z\_L\^[Ij]{} Z\_L\^[(J+3)i]{} - A\_l\^[IJ]{} Z\_L\^[Ii]{} Z\_L\^[(J+3)j]{})Z\_H\^[1k]{} + (A\_l\^[’IJ]{} Z\_L\^[Ij]{} Z\_L\^[(J+3)i]{}.\
&& . - A\_l\^[’IJ]{} Z\_L\^[Ii]{} Z\_L\^[(J+3)j]{})Z\_H\^[2k]{} + Y\_l\^I (\^ Z\_L\^[Ii]{} Z\_L\^[(I+3)j]{} - Z\_L\^[Ij]{} Z\_L\^[(I+3)i]{}) Z\_H\^[2k]{} )L\_[i]{}\^[-]{} L\_[j]{}\^[+]{} A\^0\_k\
&+& ( [e\^2 2c\_W\^2]{} B\_R\^k (\^[ij]{} + [1-4s\_W\^2 2s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{}) - (Y\_l\^I)\^2 v\_1 Z\_R\^[1k]{} (Z\_L\^[Ii]{} Z\_L\^[Ij]{} + Z\_L\^[(I+3)i]{} Z\_L\^[(I+3)j]{}).\
&& - [1]{}Z\_R\^[1k]{} (A\_l\^[IJ]{} Z\_L\^[Ij]{} Z\_L\^[(J+3)i]{} + A\_l\^[IJ]{} Z\_L\^[Ii]{} Z\_L\^[(J+3)j]{})\
&& + [1]{}Z\_R\^[2k]{} (A\_l\^[’IJ]{} Z\_L\^[Ij]{} Z\_L\^[(J+3)i]{} + A\_l\^[’IJ]{} Z\_L\^[Ii]{} Z\_L\^[(J+3)j]{})\
&&. - [1]{} Y\_l\^I Z\_R\^[2k]{} (\^ Z\_L\^[Ii]{} Z\_L\^[(I+3)j]{} + Z\_L\^[Ij]{} Z\_L\^[(I+3)i]{})) L\^-\_i L\^+\_j H\^0\_k
ii\) two slepton–two Higgs couplings: &-& [e\^2 8s\_W\^2c\_W\^2]{} A\_R\^[ij]{} \^[I]{} \^I H\^0\_i H\^0\_j - [e\^2 8s\_W\^2c\_W\^2]{} A\_H\^[ij]{} \^[I]{} \^I A\^0\_i A\^0\_j\
&+& Z\_\^[KJ]{} Z\_\^[KI]{} (e\^2 [c\_W\^2 - s\_W\^24 s\_W\^2 c\_W\^2]{} A\_H\^[ij]{} - (Y\_l\^K)\^2 Z\_H\^[1i]{} Z\_H\^[1j]{})\^[J]{} \^I H\^-\_i H\^+\_j\
&+& [i]{}Z\_L\^[Ii]{} Z\_\^[IJ]{} ( [e\^2 2s\_W\^2]{} A\_H\^[jk]{} - (Y\_l\^I)\^2 Z\_H\^[1j]{} Z\_H\^[1k]{} )\^J L\^+\_i H\^-\_j A\^0\_k +\
&+& [1]{}Z\_L\^[Ii]{} Z\_\^[IJ]{} ( [-e\^2 2s\_W\^2]{} (Z\_H\^[1j]{} Z\_R\^[1k]{} + Z\_H\^[2j]{} Z\_R\^[2k]{}) + (Y\_l\^I)\^2 Z\_H\^[1j]{} Z\_R\^[1k]{} )\^J L\^+\_i H\^-\_j H\^0\_k +\
&+& ( [e\^2 4c\_W\^2]{} A\_H\^[kl]{} (\^[ij]{} + [1-4s\_W\^2 2s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{}).\
&&. - (Y\_l\^I)\^2 Z\_H\^[1k]{} Z\_H\^[1l]{} (Z\_L\^[Ii]{} Z\_L\^[Ij]{} + Z\_L\^[(I+3)i]{} Z\_L\^[(I+3)j]{})) L\^-\_i L\^+\_j A\^0\_k A\^0\_l\
&+& ( [e\^2 4c\_W\^2]{} A\_R\^[kl]{} (\^[ij]{} + [1-4s\_W\^2 2s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{}).\
&& . - (Y\_l\^I)\^2 Z\_R\^[1k]{} Z\_R\^[1l]{} (Z\_L\^[Ii]{} Z\_L\^[Ij]{} + Z\_L\^[(I+3)i]{} Z\_L\^[(I+3)j]{})) L\^-\_i L\^+\_j H\^0\_k H\^0\_l\
&+& ( [e\^2 2c\_W\^2]{} A\_H\^[kl]{} (\^[ij]{} - [1+2s\_W\^2 2s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{}) - (Y\_l\^I)\^2 Z\_H\^[1k]{} Z\_H\^[1l]{} Z\_L\^[(I+3)i]{} Z\_L\^[(I+3)j]{} ) L\^-\_i L\^+\_j H\^-\_k H\^+\_l
.
Interactions of the squarks and Higgs bosons:
i\) three scalar couplings: &+& [i]{} ((A\_u\^[IJ]{} Z\_U\^[Ij]{} Z\_U\^[(J+3)i]{} - A\_u\^[IJ]{} Z\_U\^[Ii]{} Z\_U\^[(J+3)j]{})Z\_H\^[2k]{} + (A\_u\^[’IJ]{} Z\_U\^[Ii]{} Z\_U\^[(J+3)j]{}.\
&& - . A\_u\^[’IJ]{} Z\_U\^[Ij]{} Z\_U\^[(J+3)i]{})Z\_H\^[1k]{} + Y\_u\^I (Z\_U\^[Ii]{} Z\_U\^[(I+3)j]{} - \^ Z\_U\^[Ij]{} Z\_U\^[(I+3)i]{}) Z\_H\^[1k]{} ) U\^-\_i U\^+\_j A\^0\_k\
&+& ( [-e\^2 3c\_W\^2]{} (\^[ij]{} + [3-8s\_W\^2 4s\_W\^2]{} Z\_U\^[Ii]{} Z\_U\^[Ij]{}) B\_R\^k . - (Y\_u\^I)\^2 v\_2 Z\_R\^[2k]{} (Z\_U\^[Ii]{} Z\_U\^[Ij]{} + Z\_U\^[(I+3)i]{} Z\_U\^[(I+3)j]{})\
&& + [1]{} Z\_R\^[2k]{} (A\_u\^[IJ]{} Z\_U\^[Ii]{} Z\_L\^[(J+3)j]{} + A\_u\^[IJ]{} Z\_U\^[Ij]{} Z\_U\^[(J+3)i]{})\
&& + [1]{}Z\_R\^[1k]{} (A\_u\^[’IJ]{} Z\_U\^[Ii]{} Z\_U\^[(J+3)j]{} + A\_u\^[’IJ]{} Z\_U\^[Ij]{} Z\_U\^[(J+3)i]{})\
&& . + [1]{} Y\_u\^I Z\_R\^[1k]{} (\^ Z\_U\^[Ij]{} Z\_U\^[(I+3)i]{} + Z\_U\^[Ii]{} Z\_U\^[(I+3)j]{})) U\^-\_i U\^+\_j H\^0\_k\
&+& [i]{} ((A\_d\^[IJ]{} Z\_D\^[Ij]{} Z\_D\^[(J+3)i]{} - A\_d\^[IJ]{} Z\_D\^[Ii]{} Z\_D\^[(J+3)j]{})Z\_H\^[1k]{} + (A\_d\^[’IJ]{} Z\_D\^[Ij]{} Z\_D\^[(J+3)i]{}.\
&& . - A\_d\^[’IJ]{} Z\_D\^[Ii]{} Z\_D\^[(J+3)j]{})Z\_H\^[2k]{} + Y\_d\^I (\^ Z\_D\^[Ii]{} Z\_D\^[(I+3)j]{} - Z\_D\^[Ij]{} Z\_D\^[(I+3)i]{}) Z\_H\^[2k]{} ) D\^-\_i D\^+\_j A\^0\_k\
&+& ( [e\^2 6c\_W\^2]{} (\^[ij]{} + [3-4s\_W\^2 2s\_W\^2]{} Z\_D\^[Ii]{} Z\_D\^[Ij]{}) B\_R\^k - (Y\_d\^I)\^2 v\_1 Z\_R\^[1k]{} (Z\_D\^[Ii]{} Z\_D\^[Ij]{} + Z\_D\^[(I+3)i]{} Z\_D\^[(I+3)j]{}) .\
&& - [1]{}Z\_R\^[1k]{} (A\_d\^[IJ]{} Z\_D\^[Ij]{} Z\_D\^[(J+3)i]{} + A\_d\^[IJ]{} Z\_D\^[Ii]{} Z\_D\^[(J+3)j]{})\
&& + [1]{}Z\_R\^[2k]{} (A\_d\^[’IJ]{} Z\_D\^[Ij]{} Z\_D\^[(J+3)i]{} + A\_d\^[’IJ]{} Z\_D\^[Ii]{} Z\_D\^[(J+3)j]{})\
&& . - [1]{} Y\_d\^I Z\_R\^[2k]{} (\^ Z\_D\^[Ii]{} Z\_D\^[(I+3)j]{} + Z\_D\^[Ij]{} Z\_D\^[(I+3)i]{}) ) D\^-\_i D\^+\_j H\^0\_k\
&+& U\^+\_i D\^+\_j H\^-\_k +
ii\) four scalar couplings: &+& ([-e\^2 6c\_W\^2]{} A\_H\^[kl]{} (\^[ij]{} + [3-8s\_W\^2 4s\_W\^2]{} Z\_U\^[Ii]{} Z\_U\^[Ij]{}) .\
&& . - (Y\_u\^I)\^2 Z\_H\^[2k]{} Z\_H\^[2l]{} (Z\_U\^[Ii]{} Z\_U\^[Ij]{} + Z\_U\^[(I+3)i]{} Z\_U\^[(I+3)j]{})) U\^-\_i U\^+\_j A\^0\_k A\^0\_l\
&+& ([-e\^2 6c\_W\^2]{} A\_R\^[kl]{} (\^[ij]{} + [3-8s\_W\^2 4s\_W\^2]{} Z\_U\^[Ii]{} Z\_U\^[Ij]{}) .\
&& . - (Y\_u\^I)\^2 Z\_R\^[2k]{} Z\_R\^[2l]{} (Z\_U\^[Ii]{} Z\_U\^[Ij]{} + Z\_U\^[(I+3)i]{} Z\_U\^[(I+3)j]{})) U\^-\_i U\^+\_j H\^0\_k H\^0\_l\
&+& ([-e\^2 3c\_W\^2]{} A\_H\^[kl]{} (\^[ij]{} - [3+2s\_W\^2 4s\_W\^2]{} Z\_U\^[Ii]{} Z\_U\^[Ij]{}) .\
&& . - (Y\_u\^I)\^2 Z\_H\^[2k]{} Z\_H\^[2l]{} Z\_U\^[(I+3)i]{} Z\_U\^[(I+3)j]{} - (Y\_d\^I)\^2 Z\_H\^[1k]{} Z\_H\^[1l]{} K\^[JI]{} K\^[KI]{} Z\_U\^[Ki]{} Z\_U\^[Jj]{} ) U\^-\_i U\^+\_j H\^-\_k H\^+\_l\
&+& ([e\^2 12c\_W\^2]{} A\_H\^[kl]{} (\^[ij]{} + [3-4s\_W\^2 2s\_W\^2]{} Z\_D\^[Ii]{} Z\_D\^[Ij]{}) .\
&& . - (Y\_d\^I)\^2 Z\_H\^[1k]{} Z\_H\^[1l]{} (Z\_D\^[Ii]{} Z\_D\^[Ij]{} + Z\_D\^[(I+3)i]{} Z\_D\^[(I+3)j]{})) D\^-\_i D\^+\_j A\^0\_k A\^0\_l\
&+& ([e\^2 12c\_W\^2]{} A\_R\^[kl]{} (\^[ij]{} + [3-4s\_W\^2 2s\_W\^2]{} Z\_D\^[Ii]{} Z\_D\^[Ij]{}) .\
&& . - (Y\_d\^I)\^2 Z\_R\^[1k]{} Z\_R\^[1l]{} (Z\_D\^[Ii]{} Z\_D\^[Ij]{} + Z\_D\^[(I+3)i]{} Z\_D\^[(I+3)j]{})) D\^-\_i D\^+\_j H\^0\_k H\^0\_l\
&+& ([e\^2 6c\_W\^2]{} A\_H\^[kl]{} (\^[ij]{} - [3-2s\_W\^2 2s\_W\^2]{} Z\_D\^[Ii]{} Z\_D\^[Ij]{}) .\
&& . - (Y\_d\^I)\^2 Z\_H\^[1k]{} Z\_H\^[1l]{} Z\_D\^[(I+3)i]{} Z\_D\^[(I+3)j]{} - (Y\_u\^K)\^2 Z\_H\^[2k]{} Z\_H\^[2l]{} K\^[KI]{} K\^[KJ]{} Z\_D\^[Ji]{} Z\_D\^[Ij]{} ) D\^-\_i D\^+\_j H\^-\_k H\^+\_l\
&+& [1]{} K\^[JI]{} ([-e\^2 2s\_W\^2]{}(Z\_H\^[1k]{} Z\_R\^[1l]{} + Z\_H\^[2k]{} Z\_R\^[2l]{})Z\_U\^[Ji]{} Z\_D\^[Ij]{} - A\_[P]{}\^[lk]{} Y\_u\^J Y\_d\^I Z\_U\^[(J+3)i]{} Z\_D\^[(I+3)j]{}.\
&& . + ( (Y\_u\^J)\^2 Z\_H\^[2k]{} Z\_R\^[2l]{} + (Y\_d\^I)\^2 Z\_H\^[1k]{} Z\_R\^[1l]{}) Z\_U\^[Ji]{} Z\_D\^[Ij]{} ) U\^+\_i D\^+\_j H\^-\_k H\^0\_l +\
&+& [i]{} K\^[JI]{} ([e\^2 2s\_W\^2]{} A\_H\^[kl]{} Z\_U\^[Ji]{} Z\_D\^[Ij]{} - \_[kl]{} Y\_u\^J Y\_d\^I Z\_U\^[(J+3)i]{} Z\_D\^[(I+3)j]{} .\
&& . + ( (Y\_u\^J)\^2 Z\_H\^[2k]{} Z\_H\^[2l]{} - (Y\_d\^I)\^2 Z\_H\^[1k]{} Z\_H\^[1l]{}) Z\_U\^[Ji]{} Z\_D\^[Ij]{} ) U\^+\_i D\^+\_j H\^-\_k A\^0\_l +
.
Four-scalar interactions of four sleptons or two sleptons and two squarks. &-& [e\^2 8s\_W\^2c\_W\^2]{} \^[I]{} \^[J]{} \^I \^J - ( [- e\^2 2c\_W\^2]{} (\^[ij]{} + [1-4s\_W\^2 2s\_W\^2]{} Z\_L\^[Ki]{} Z\_L\^[Kj]{})\^[IJ]{} .\
&&. + ( [e\^2 2s\_W\^2]{} Z\_L\^[Ki]{} Z\_L\^[Lj]{} + Y\_l\^K Y\_l\^L Z\_L\^[(K+3)i]{} Z\_L\^[(L+3)j]{}) Z\_\^[LI]{} Z\_\^[KJ]{} )\^[J]{} \^I L\^-\_i L\^+\_j\
&-& [e\^2 3c\_W\^2]{} (\^[ij]{} + [3-8s\_W\^2 4s\_W\^2]{} Z\_U\^[Ji]{} Z\_U\^[Jj]{})\^[I]{} \^I U\^-\_i U\^+\_j\
&+& [e\^2 6c\_W\^2]{} (\^[ij]{} + [3-4s\_W\^2 2s\_W\^2]{} Z\_D\^[Ji]{} Z\_D\^[Jj]{})\^[I]{} \^I D\^-\_i D\^+\_j\
&-& Z\_\^[JI]{} Z\_U\^[Li]{} K\^[LK]{} ( [e\^2 2s\_W\^2]{} Z\_D\^[Kj]{} Z\_L\^[Jk]{} + Y\_l\^J Y\_d\^K Z\_D\^[(K+3)j]{} Z\_L\^[(J+3)k]{} )\^I U\^-\_i D\^-\_j L\^+\_k +\
&-& ([e\^2(1+8s\_W\^2) 8s\_W\^2c\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Jj]{} Z\_L\^[Jl]{} Z\_L\^[Ik]{} + [e\^2 2c\_W\^2]{} \^[jl]{} (\^[ik]{} - 3Z\_L\^[Ii]{} Z\_L\^[Ik]{}).\
&&. + Y\_l\^I Y\_l\^J Z\_L\^[Jj]{} Z\_L\^[Ik]{} Z\_L\^[(I+3)i]{} Z\_L\^[(J+3)l]{} )L\_[i]{}\^[-]{} L\_[j]{}\^[-]{} L\_[k]{}\^[+]{} L\_[l]{}\^[+]{}\
&+& [e\^2 6c\_W\^2]{} ( [3+12s\_W\^2 2s\_W\^2]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{} Z\_U\^[Jk]{} Z\_U\^[Jl]{} - 6\^[kl]{} Z\_L\^[Ii]{} Z\_L\^[Ij]{} - 5 \^[ij]{} Z\_U\^[Ik]{} Z\_U\^[Il]{} + 4 \^[ij]{} \^[kl]{} ) L\^-\_i L\^+\_j U\^-\_k U\^+\_l\
&+& L\^-\_i L\^+\_j D\^-\_k D\^+\_l
.
Vertices containing the strong coupling constant $g_3$. There are two basic types of such interactions - couplings of the quarks and squarks with the gluons and gluinos and four squarks interactions. By $Y^a$ we denote the matrices of the $SU(3)$ generators in $\mathbf{3}$ representation (generally $a, b, c
\ldots$ are indices corresponding to $\mathbf{8}$ (adjoint) representation, $\alpha, \beta, \gamma \dots$ - to $\mathbf{3}$ (basic) representation of the QCD gauge group).
&-& g\_3 |[d]{}\^I Y\^a \^ d\^I g\_\^a - ig\_3 (D\^+\_i Y\^a \^ D\^-\_i) g\_\^a + g\_3\^2 D\^+\_i Y\^a Y\^b D\^-\_i g\_\^a g\^[b]{}\
&-& eg\_3 D\^+\_i Y\^a D\^-\_i g\_\^a F\^ + [eg\_3 s\_Wc\_W]{} (-Z\_D\^[Ii]{} Z\_D\^[Ij]{} + \^[ij]{} s\_W\^2) D\^+\_i Y\^a D\^-\_j g\_\^a Z\^\
&-& g\_3 |[u]{}\^I Y\^a \^ u\^I g\_\^a - ig\_3 (U\^-\_i Y\^a \^ U\^+\_i) g\_\^a + g\_3\^2 U\^-\_i Y\^a Y\^b U\^+\_i g\_\^a g\^[b]{}\
&+& e g\_3 U\^-\_i Y\^a U\^+\_i g\_\^a F\^ + [eg\_3 s\_Wc\_W]{}(Z\_U\^[Ii]{} Z\_U\^[Ij]{} - \^[ij]{}s\_W\^2) U\^-\_i Y\^a U\^+\_j g\_\^a Z\^\
&+& [eg\_3 s\_W]{} Z\_U\^[Jj]{} Z\_D\^[Ii]{} K\^[JI]{} D\^+\_i Y\^a U\^+\_j g\_\^a W\^[-]{} +\
&+& g\_3 U\^-\_i Y\^a |\_[G]{}\^[a]{} (-Z\_U\^[Ii]{}P\_L + Z\_U\^[(I+3)i]{} P\_R) u\^I +\
&+& g\_3 D\^+\_i Y\^a |\_[G]{}\^[a]{} (-Z\_D\^[Ii]{}P\_L + Z\_D\^[(I+3)i]{} P\_R) d\^I +\
&+& ig\_3 f\_[abc]{} |\_G\^a \^ \_G\^b g\^c\_ + g\_3 f\_[abc]{} (\^ |\_G\^a) \_G\^b g\^c\_\
&+& g\_3 f\_[abc]{} (\_ g\^a\_ - \_ g\_\^a) g\^[b]{} g\^[c]{} - g\_3\^2 f\_[abc]{} f\_[ade]{} g\^b\_ g\^[d]{} g\^c\_ g\^[e]{}
In the next terms it is necessary to write down explicitly the color indices $\alpha, \beta, \gamma \ldots$ We also define following abbreviations:
------------------------------------------------------------------ -- -----------------------------------------------------
$R_U^{ij}=Z_U^{Ii\star}Z_U^{Ij}$ $R_D^{ij}=Z_D^{Ii}Z_D^{Ij\star}$
$X_U^{ij}=\delta^{ij} - 2R_U^{ij}$ $X_D^{ij}=\delta^{ij} - 2R_D^{ij}$
$Y_U^{ij}=5R_U^{ij}-4\delta^{ij}$ $Y_D^{ij}=2\delta^{ij} - R_D^{ij}$
$V_U^{ijkl}=Y_u^I Y_u^J Z_U^{(I+3)i\star} Z_U^{Ij} Z_U^{Jk\star} $V_D^{ijkl}= Y_d^I Y_d^J Z_D^{(I+3)i} Z_D^{Ij\star}
Z_U^{(J+3)l}$ Z_D^{Jk} Z_D^{(J+3)l\star}$
------------------------------------------------------------------ -- -----------------------------------------------------
&-&[14]{}{ \_ \_ .\
&+& . \_ \_} U\^-\_[i]{} U\^-\_[k]{} U\^+\_[j]{} U\^+\_[l]{}\
&-&[14]{}{ \_ \_ .\
&+&. \_ \_} D\^-\_[i]{} D\^-\_[k]{} D\^+\_[j]{} D\^+\_[l]{}\
&-&U\^-\_[i]{} U\^+\_[j]{} D\^-\_[k]{} D\^+\_[l]{}
Summary {#sec:comment}
=======
The model described in the previous sections contains a large number of free parameters which considerably limit its predictive power. There are some commonly used ways to reduce the number of free constants in this theory. The most often employed method is to obtain values of the parameters at the electroweak scale by RGE running from the coupling constants generated at high scale by some SUSY breaking scenario. Usually such theories are much more unified and contain typically only few free numbers. Of course, there are many constraints originating from experimental data. Obviously, the masses of the superpartners are bounded from below by negative result of direct SUSY searches in collider experiments. One can find also many indirect limits: for example, if the masses of the superscalars are not very high, then the non-diagonal soft Yukawa couplings $A_d^{IJ},
A_d^{'IJ}$ etc. can lead by loop corrections to too strong flavor changing neutral currents in the quark sector.
It is worth remembering that although in the superpotential one can have only three matrices of Yukawa couplings (four if right neutrino exists), six such matrices for the soft SUSY breaking terms are allowed, three additional ones describing interactions of the superscalars with the complex conjugated Higgs doublets. Those three new couplings can affect various processes, in particular they are present in the formulas for the scalar mass matrices. In most SUSY breaking scenarios such terms are not generated, but in principle they are not forbidden by MSSM symmetries.
[**Acknowledgments.**]{}\
The author would like to thank professor S. Pokorski for many helpful discussions.
MSSM Lagrangian before gauge symmetry breaking {#sec:app_a}
==============================================
app\_a
Feynman rules {#sec:app_b}
=============
app\_b
[99]{}
J. Rosiek, [*Phys. Rev.*]{} [**D41**]{} (1990) 3464.
P. Chankowski, S. Pokorski, J. Rosiek [*Nucl. Phys.*]{} [**B423**]{} (1994) 437; [*Nucl. Phys.*]{} [**B423**]{} (1994) 497; V. Driesen, W. Hollik, J. Rosiek, [*Zeit. Phys.*]{} [**C71**]{} (1996) 259; S. Heinemeyer, W. Hollik, J. Rosiek, G. Weiglein, [*Eur.Phys.J.*]{} [**C19**]{} (2001) 535-546.
P. Chankowski et al. [*Nucl. Phys.*]{} [**B417**]{} (1994) 101.
S. Pokorski, J. Rosiek, C. Savoy [*Nucl. Phys.*]{} [**B570**]{} (2000) 81-116.
M. Misiak, S. Pokorski, J. Rosiek [*“Supersymmetry and the FCNC effects”*]{}, [*hep-ph*]{}/9703442, published in [*Adv.Ser.Direct.High Energy Phys.*]{} [**15**]{} (1998) 795-828.
A. Buras, P. Chankowski, J. Rosiek, [Ł]{}. S[ł]{}awianowska, [*Nucl.Phys.*]{} [**B619**]{} (2001) 434-466.
H. E. Haber and G. L. Kane, [*Phys. Rep.*]{} [**117**]{} (1985) 75.
J. F. Gunion and H. E. Haber, [*Nucl. Phys.*]{} [**B272**]{} (1986) 1.
P. Fayet and J. Iliopoulos, [*Phys. Lett.*]{} [**51B**]{} (1974) 461.
L. Girardello and M. T. Grisaru, [*Nucl. Phys.*]{} [**B194**]{} (1982) 65.
H. Dreiner, [*“An Introduction to explicit R-parity violation”*]{}, [*hep-ph*]{}/9707435, published in \*Kane, G.L. (ed.): Perspectives on supersymmetry\* 462-479.
J. A. Casas, [*“Charge and color breaking”*]{}, [*hep-ph*]{}/9707475, published in \*Kane, G.L. (ed.): Perspectives on supersymmetry\* 378-401.
A. Denner , H. Eck, O. Hahn, J. Kublbeck, [*Phys.Lett.*]{} [**B291**]{} (1992) 278-280; A. Denner, H. Eck, O. Hahn, J. Kublbeck, [*Nucl.Phys.*]{} [**B387**]{} (1992) 467-484.
[^1]: e-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'The spectrometer SPI, one of the two main instruments of the INTEGRAL spacecraft, has strong capabilities in the field of Gamma-Ray Burst (GRB) detections. In its 16$^o$ field of view (FoV) SPI is able to trigger and to localize GRBs with an accuracy for strong bursts better than 1$^o$. The expected GRB detection rate is about one per month.'
author:
- 'Andreas von Kienlin, Nikolas Arend, Giselher Lichti, Andrew Strong'
- Paul Connell
title: 'Gamma-Ray Burst Detection with INTEGRAL/SPI'
---
\#1[[*\#1*]{}]{} \#1[[*\#1*]{}]{} =
\#1 1.25in .125in .25in
Introduction
============
The detection and investigation of cosmic gamma-ray bursts (GRBs) is one of the important scientific topics of ESA’s INTEGRAL mission, which will be launched on Oct. 17, 2002. From all instruments of the INTEGRAL payload, the two main instruments, the spectrometer SPI and the imager IBIS, and the two monitors, JEM-X and OMC, contributions to the science of GRBs are expected. Especially the sensitivity of INTEGRAL’s imager and spectrometer above 20 keV will distinguish them from other missions. Although INTEGRAL is not equipped with an on-board burst-alert system it has strong capabilities to detect bursts thanks to the ground-based INTEGRAL burst-alert system IBAS (Mereghetti 2001). IBAS is an automatic software system for the near-real time distribution of GRB alerts detected by INTEGRAL to the scientific community. The telemetry stream of the satellite, which is linked via the mission operation center (MOC) at Darmstadt, Germany, to the INTEGRAL science data center (ISDC) at Versoix, Switzerland, will be monitored continuously for the occurrence of GRBs. The expected transmission time of the telemetry from the satellite to ISDC will be approximately 30 s. Additionally the processing time at ISDC and the time needed to distribute the data to the clients has to be added. In total an observer could expect the first alert message within less than 1 min after the onset of the burst. Thus, in the case of a long-duration burst, follow-up observations are possible even during the active phase of the burst.
Burst detection in SPI’s field of view
======================================
The aim of the spectrometer SPI is to perform high-resolution spectroscopy in the energy range between 20 keV and 8 MeV. The imaging capability will be good, but it is exceeded by that of the imager IBIS which complements SPI with its high imaging resolution, but with less spectroscopic resolving power. So the expected yield and discoveries in the field of GRB research will be of a different kind for each of the four INTEGRAL instruments. With SPI new spectroscopic discoveries are possible, supposing that previously-unkown spectral features exist. IBIS will provide more precise GRB locations to the community, which is important for the observation of GRB afterglows. The minor sensitivity of SPI for GRB detection is compensated by its larger FoV. So the expected rate/year of detected GRBs is about the same for IBIS and SPI.
The camera of SPI, which consists of 19 cooled high-purity germanium detectors, residing in a cryostat, is shielded on the side walls and rear side by a large anticoincidence shield (ACS). The field of view (FoV) of the camera is defined by the upper opening of the ACS. A detailed description of the ACS and its capabilities for the burst detction can be found in (von Kienlin 2001) and (von Kienlin 2003). The imaging capability of the instrument is attained by a passive-coded mask mounted 1.7 m above the camera. By observing a series of nearby pointing directions around the source(“dithering”) the imaging capability of SPI is improved by reducing ambiguity effects. For the normal mode of operation two dithering patterns are planned, a quadratic $5 \times 5$ and a hexagonal with 7 pointings. The pointing on one grid point will last for about 20-30 min with a short 5 min slew between succeeding pointings. Due to the short duration of the bursts, dithering will not improve the imaging of SPI for bursts. But during its short duration a burst will be in most cases the brightest $\gamma$-ray source for SPI in the whole FoV, so imaging with SPI is still possible. Despite the modest angular resolution of SPI, which is in the order of 2$^o$ , it is possible to locate the direction of bursts down to a few arcminutes.
The triggering algorithm which will be used for the burst search in SPI’s FoV will be similiar to that used inside IBAS for the IBIS-ISGRI instrument. The first algorithm is based on imaging only. It looks for the appearance of new sources by comparing the actual image of the sky with those obtained previously. This algorithm is more sensitive to slowly rising bursts, since the imaging routine needs some time. The second is expected to be faster, because it monitors the overall event rate in all detectors together as function of time. In the case of a trigger the second algorithm uses the imaging algorithm for verification.
Every interaction of a $\gamma$-ray in one of the Ge-detectors is downlinked event-by-event, wrapped into packets within the telemetry stream. The event-by-event packets contain information on the detector ID which was hit, the energy channel of the recorded pulse-height signal and the time of interaction, with an accuracy of about 100 $\mu$s. Two kinds of interactions of $\gamma$-rays in the Ge-camera have to be distinguished, in the first case a $\gamma$-ray only interacts within one of the Ge-detectors (single-detector events), in the second case the $\gamma$-ray is interacting with more than one detector (multiple-detector events). The condition for the occurrence of a multiple-detector event is defined by a coincidence window inside the digital front-end electronics (DFEE) of SPI. Additional information on the detector interaction is available through the science house keeping data (HK-data). The rate for each detector is recorded on a 1 s basis inside the DFEE and sent to ground, but without energy information. The advantage of these data is that they are always available, even in the case of telemetry limitations.
Simulations and calculations have been performed (Skinner 1997) in order to investigate SPI’s capabilities for the burst detection in its FoV. For bursts with a fluence of $1.1 \times 10^{-6}$ erg/cm$^2$ and a duration of 1 s (only accounting for single detector events in the energy range from 20 to 300 keV) the positioning errors inside the FoV ($< 9^o$) will be between 1.5’ and 20’ and for bursts with a larger offset ($< 15^o$) the position error will still be less than $1^o$. Problems arising from ambiguities and systematic errors in the determination of the position are more serious towards the edge of the FoV. In summary GRB source locations to 10’ are possible over a wide FoV. For weak bursts the detection could be problematic, the imperfect imaging together with random fluctuations can lead to a spurious peak appearing to be stronger than the true one. The best-estimate position can then be far from the correct location. The ambiguity problem can occur for bursts with a strength of less than $10^{-7}$ erg/cm$^{2}$. As described above, one method for detecting bursts in the SPI data is the monitoring of the overall event rate in all detectors together as function of time. The minimum-size burst detectable for a backgound rate of 0.25 cts cm$^{-2}$ s$^{-1}$ in the 50 to 300 keV regime for a 1 s burst is about 0.2 photons (Skinner 1997), corresponding to $4~ \times 10^{-8}$ erg/cm$^{2}$. This is about the same as the weakest bursts detected by BATSE. The rate at which GRBs are detected by an instrument depends on the FoV as well as on the sensitivity. SPI can detect bursts at off-axis angles up to at least 15$^o$, so the useful FoV is quite large, 0.2 sr or 680 square degrees and consequently the expected detection rate of about 1 burst per month is comparable for IBIS and SPI.
GRB detection in SPI’s FoV with IBAS
====================================
Figure 1 gives a simplified overview on the structure of the whole IBAS system. After the near-real-time data receipt of the telemetry from MOC, several GRB detection programs are searching in the telemetry stream for the occurrence of a burst. Three of them are responsible for the spectrometer SPI, there are the two programs monitoring the camera events ([ibas\_smonitor\_img]{} and [ibas\_smonitor\_rate]{}) and the monitor program for the ACS overall veto rate ([ibas\_acsmonitor]{}). The other half of the branches monitor the science data of IBIS ISGRI ([ibas\_imonitor\_img]{} and [ibas\_imonitor\_rate]{}) and JEMX ([ibas\_jmonitor]{}). The program [ibasalertd]{} receives the GRB triggers from the individual branches and performs the final alert verification and distribution of the alert messages to subscribed users. All IBAS processes are multithreaded applications and run as daemon processes. So IBAS is able to perform several subtasks at the same time, with the advantage for the burst search to monitor the telemetry in different energy bands and with different time binning. Each of the GRB detection processes uses the frontend library to read out the telemetry packets provided by the data receipt. Then the required instrument data (Ge single/multiple events, Ge detector count rates, ACS veto rates) are decoded from the telemetry packets and put into a time-sorted queue, which is large enough to hold several minutes of data.
The ACS detection program ([ibas\_acsmonitor]{}), which has already been delivered to ISDC, is explained in detail in (von Kienlin 2003) . The SPI-FoV software, which is comprising the two modules [ibas\_smonitor\_img]{} and [ibas\_smonitor\_rate]{}, is still under development and the delivery of the first version is expected soon. From the telemetry packets the information on the $\gamma$-ray interaction time, the detector ID, and energy channel are extracted. Up to now single and multiple events are analysed. This is sufficient to do burst imaging in the instrument system, the sky coordinates can be computed subsequently from the pointing information. The energy channel is used to perform a basic energy selection, and the data will be accumulated for time intervals optimized for the detection of long and short bursts. The parameters should be as similar as possible to those for IBIS. They include: accumulation time, significance level, and energy channel range.
von Kienlin, et al. 2003 in SPIE Conf. Proc., 4851, X-ray and Gamma-ray Telescopes and Instruments for Astronomy, in press
Mereghetti, et al. 2001 in Proc., Gamma-Ray Bursts in the Afterglow Era, ed. E. Costa, F. Frontera, & J. Hjorth (Berlin Heidelberg: Springer), 363
von Kienlin, et al. 2001 in Proc., Gamma-Ray Bursts in the Afterglow Era, ed. E. Costa, F. Frontera, & J. Hjorth (Berlin Heidelberg: Springer), 427
Skinner, G., et al. 1997 in ESA Conf. Proc., 382, 2$^{nd}$ INTEGRAL Workshop, ed. C. Winkler, T. Courvoisier, & Ph. Durouchoux (ESA), 487
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Sunspot observations and counting are carried out at the Specola Solare Ticinese in Locarno since 1957 when it was built as an external observing station of the Zurich observatory. When in 1980 the data center responsibility was transferred from ETH Zurich to the Royal Observatory of Belgium in Brussels, the observations in Locarno continued and Specola Solare Ticinese got the role of pilot station. The data collected at Specola cover now the last 6 solar cycles.
The aim of this presentation is to discuss and give an overview about the Specola data collection, the applied counting method and the future archiving projects. The latter includes the publication of all data and drawings in digital form in collaboration with the ETH Zurich University Archives, where a parallel digitization project is ongoing for the document of the former Swiss Federal Observatory in Zurich collected since the time of Rudolph Wolf.
---
Historical overview
===================
The long tradition in Switzerland of systematic observations and counting of sunspots was started by Rudolf Wolf in 1847 [@Friedli2016 (Friedli, 2016)], first in Bern and then in Zurich, where in 1855 he was nominated professor for astronomy both at the Zurich University and at the Swiss Federal Institute of Technology (ETH-Zurich). To study the solar cycle, that was discovered by Samuel Schwabe few years before, Wolf introduced the notorious empirical index well know as [*Wolf number*]{} or as [*Zurich relative sunspot number*]{}, which is defined as: $$\label{eqwolf}
R= k (10 g + s)$$ where $g$ is the number of sunspot groups, $s$ the number of single sunspots and $k$ a normalization factor which depends on the observer. Being Wolf the reference, his observations were originally normalized with the factor $k=1$. The sunspot observations and the determination of the daily value of the Zurich relative sunspot number continued at the Swiss Federal Observatory in Zurich with Wolf’s successors Alfred Wolfer, William Brunner and Max Waldmeier, who had a normalizing factor $k=0.6$ [@2014SSRv..186...35C (Clette et al., 2014)].
In 1936 the retired engineer Karl Rapp, grounder of BMW, moved to Locarno in Southern Switzerland and started regular observations of sunspots, collaborating with Brunner and Waldmeier. The favorable weather conditions in Locarno allowed often to fill the gaps of data when in Zurich clouds didn’t allow to observe. Based on this positive experience, Waldmeier managed in the International Geophysical Year (1957) to ground an external observing station in Locarno, that became the Specola Solare Ticinese [@Cortesi2016 (Cortesi et al., 2016)]. Sergio Cortesi and Araldo Pittini were hired as Waldmeier’s assistants, in order to carry out observations at the Specola.
After Waldmeier’s retirement in 1979, ETH-Zurich decided to abandon the research activity related to the sunspot counting. The Royal Observatory of Belgium in Brussels agreed to take over this service. In January 1981, the new world data center in Brussels, previously called SIDC and now SILSO, took over the responsibility to determine the daily Wolf number, from then on called [*international sunspot number*]{} (SSN) [@Stenflo2016 (Stenflo, 2016)]. At that point, thanks to experience gained with Waldmeier, Specola Solare Ticinese became the pilot station with the role to guarantee the long term stability of the SSN. Observations continued under the direction of Cortesi, that acted as main observer. After some years of training, in 2011, Marco Cagnotti became the new main observer at Specola.
The sunspot drawings of the Specola Solare Ticinese
===================================================
Observations in Locarno are obtained with a Coudé-Zeiss refractor with an aperture of 15 cm. This is diaphragmed to 8 cm, in order to obtain a better visual contrast. The refractor projects a solar mirrored image with a diameter of 25 cm on a metallic plate where a paper sheet is fixed, where the observer can draw the sunspots with a pencil (Fig. 1). Originally, sunspot counting is done following the rules taught by Waldmeier to the Specola observers [@Waldmeier48 (Waldmeier, 1948)], where sunspots are weighted according to their size. For instance, small point-like sunspots are counted as 1, larger sunspots as 2, a sunspot with penumbra of normal size is counted as 3, while a very large sunspot with penumbra is counted as 5.
![Example of a sunspot drawing obtained at Specola Solare Ticinese in Locarno. The solar image is mirrored. In the top-right table, for each sunspot group, it is reported the serial number, the weighted counting of the sunspots, the Zurich classification and the unweighted counting.[]{data-label="fig1"}](loc-d20141222.eps){width="\linewidth"}
In 2012, it has been proposed to apply in parallel to the original counting method, an unweighted counting where each sunspot is counted as 1 independently of its size. This allowed comparison studies of the two methods [@2017SoPh..292...34S (Svalgaard et al., 2017)] that were used in recent works aiming at a better homogenization of the SSN in the past. In particular the results of this study led to a better understanding of the “Waldmeier jump” occurred around 1947 when Waldmeier became the new director of the Swiss Federal Observatory in Zurich instead of Brunner. These studies were taken into account in the revision of the SSN introduced by SILSO on July 2015 by SILSO [@2016SoPh..291.2629C (Clette & Lef[è]{}vre, 2016)]. One of the decisions that was taken for this revision was to use the unweighted counting of Locarno as the new reference[^1]. Since August 2014 both weighted and unweighted countings of the Specola, are reported officially on the sunspot drawings and communicated to SILSO. Meanwhile, for further studies, we started to determine the unweighted counting from the past drawings of the Specola. This work is still in progress.
Sunspot data and drawings archiving
===================================
The digitized Sunspot drawings of the Specola Solare Ticinese are published daily on the WEB at the URL <http://specola.ch/e/drawings>. At the same address one can find the archive that includes all digitized drawings made since 1981, when the Specola became independent from ETH-Zurich. The drawings made at Specola from 1957 to 1980 are stored in the ETH Zurich University Archives at the ETH-library, together with all the material collected by the former Swiss Federal Observatory in Zurich.
For the safe long term preservation of the sunspot drawings and data collected in Locarno, Specola Solare Ticinese and ETH Zurich University Archives are starting a project together. This foresees that all the drawings, including the ones made previously than 1981, will be professionally digitized and made openly accessible through the E-manuscripta digital platform for manuscript material from Swiss libraries and archives accessible at <http://www.e-manuscripta.ch>. All drawings will have their own digital object identifier (DOI). The digitization project of the Specola drawings will complement a parallel project done by the ETH Zurich University Archives, in which the full sunspot drawings collection of the former Swiss Federal Observatory in Zurich is being digitized and made available on the same digital platform E-manuscripta.
In addition, according to the project, it is foreseen that all the Specola drawings made until now will be safely stored at the ETH Zurich University Archives. Furthermore in the next years it is planned to prepare a digital sunspot group database that will be published with open access. The database will contain information for each sunspot group about observing date, observer, image quality, group ID serial number, Zurich classification, latitude, original weighted counting. It is also foreseen to perform a post-unweighted counting and to add this information to the database. Giving open access, we hope that the database will be a useful data source for all researchers for further studies of the data series obtained at Specola.
Acknowledgments
===============
The archiving and digitization project of the Specola data and drawings will be supported in the next years by the Federal Office of Meteorology and Climatology MeteoSwiss, in the framework of GCOS Switzerland. The regular research activity at Specola Solare Ticinese is mainly financed by Canton Ticino through the Swisslos found.
The authors acknowledge also the personnel of ETH Zurich University Archives for the collaboration in the archiving and digitization project, in particular Ms. Evelyn Boesch and Mr. Christian Huber.
Clette, F., Svalgaard, L., Vaquero, J. M., & Cliver, E. W. 2014, , 186, 35
Clette, F., & Lef[è]{}vre, L. 2016, , 291, 2629
2016, , 291, 3075
Friedli, T. K. 2016, , 291, 2505
Stenflo, J. O. 2016, , 291, 2487
Svalgaard, L., Cagnotti, M., & Cortesi, S. 2017, , 292, 34
Waldmeier, M. 1948, 100 Jahre Sonnenfleckenstatistik, Astron. Mitt. Eidgenöss. Sternwarte Zür. 152, 1.
[^1]: Another important change in the SSN revision adopted by SILSO in July 2015 was to take Wolfer as reference, instead of Wolf for the normalization factor $k=1$ present in equation \[eqwolf\]. [@2016SoPh..291.2629C (Clette & Lef[è]{}vre, 2016)]
| {
"pile_set_name": "ArXiv"
} |
---
abstract: |
Let $V$ be a potential on ${\mathbb R}^3$ that is smooth everywhere except at a discrete set ${\mathcal S}$ of points, where it has singularities of the form $Z/\rho^2$, with $\rho(x) = |x - p|$ for $x$ close to $p$ and $Z$ continuous on ${\mathbb R}^3$ with $Z(p) > -1/4$ for $p \in {\mathcal S}$. Also assume that $\rho$ and $Z$ are smooth outside ${\mathcal S}$ and $Z$ is smooth in polar coordinates around each singular point. We either assume that $V$ is periodic or that the set ${\mathcal S}$ is finite and $V$ extends to a smooth function on the radial compactification of ${\mathbb R}^3$ that is bounded outside a compact set containing ${\mathcal S}$. In the periodic case, we let $\Lambda$ be the periodicity lattice and define ${\mathbb T}:= {\mathbb R}^3/
\Lambda$. We obtain regularity results in weighted Sobolev space for the eigenfunctions of the Schrödinger-type operator $H = -\Delta +
V$ acting on $L^2({\mathbb T})$, as well as for the induced ${ \mathbf{k} }$–Hamiltonians ${H_{{ \mathbf{k} }}}$ obtained by restricting the action of $H$ to Bloch waves. Under some additional assumptions, we extend these regularity and solvability results to the non-periodic case. We sketch some applications to approximation of eigenfunctions and eigenvalues that will be studied in more detail in a second paper.
address:
- 'Eugenie Hunsicker, Department of Mathematical Sciences, Loughborough University, Loughborough, Leicestershire, LE11 3TU, UK '
- 'Hengguang Li, Department of Mathematics, Wayne State University, Detroit, MI 48202, USA'
- 'V. Nistor, Pennsylvania State University, Math. Dept., University Park, PA 16802, USA, and Inst. Math. Romanian Acad. PO BOX 1-764, 014700 Bucharest Romania'
- 'Ville Uski, Department of Mathematical Sciences, Loughborough University, Loughborough, Leicestershire, LE11 3TU, UK '
author:
- Eugenie Hunsicker
- Hengguang Li
- Victor Nistor
- Ville Uski
date:
-
-
title: 'Analysis of Schrödinger operators with inverse square potentials I: regularity results in 3D'
---
Introduction and statement of main results
==========================================
We study in this paper regularity and decay properties of the eigenfunctions of Schrödinger type operators with inverse-square singularities. We either assume that the potential is periodic or that it has a nice behavior at infinity and only finitely many singularities. In order to explain our assumptions and results in more detail, we organize our Introduction in subsections, concentrating on the case of periodic potentials, the non-periodic case being similar, but simpler. We first introduce the operators ${H_{{ \mathbf{k} }}}$ obtained from the Hamiltonian $-\Delta + V$ acting on Bloch waves. In the second subsection of the Introduction, we explain our assumptions on the potential $V$. Finally, we state our main results and we summarize the contents of the paper.
This paper is written to put on a solid foundations the numerical methods developed in [@HLNU2] and [@HLNU4]. We have thus written this paper with an eye to the numerical analyst. More theoretical results on Hamiltonians with inverse square potentials in arbitrary dimensions will be included in the third part of this paper.
We have to mention Kato’s ground breaking papers [@Kato51], where the self-adjointness of Schrödinger type Hamiltonians was proved and [@Kato57], where boundedness properties of the eigenfunctions and eigenvalues of these Hamiltonian operators was proved. Moreover, see [@Bespalov; @Westphal; @Felli2; @Felli1; @MorozSchmidt; @Zuazua; @Sprung] for other papers studying Hamiltonians with inverse square potentials, both from the point of view of physical and numerical applications. See also [@Dauge; @ferrero; @Flad2; @Flad3; @Flad1; @Fournais1; @Hamaekers; @Siedentop; @Schwab; @VasyReg; @Yserentant3] for some related results.
The Hamiltonian ${H_{{ \mathbf{k} }}}$
--------------------------------------
Let $V$ be a periodic potential on ${\mathbb R}^3$ with Bravais lattice (of translational symmetries) $\Lambda \cong \mathbb{Z}^3$. Assume that $V$ is smooth except at a set of points ${\mathcal S}$, which is thus necessarily also periodic with with respect to the lattice $\Lambda$. We assume that there are only finitely many elements of ${\mathcal S}$ in any fundamental domain ${\mathbb P}$ of $\Lambda$. Let $p \in {\mathcal S}$ be a singular point and $\rho(x) = |x-p|$ for $x$ close to $p$ and $\rho$ smooth outside ${\mathcal S}$. We assume that around $p$ the potential $V$ has a singularity of the form $Z/\rho^2$, where $Z$ is continuous across $p$ and smooth in polar coordinates around $p$. We shall study numerically Hamiltonian operators of the form $$\label{eq.def.H}
H := -\Delta + V.$$ Systems with such potentials have been studied as theoretical models both from the viewpoint of classical mechanics and from the quantum mechanical viewpoint. In addition, they arise in a variety of physical contexts, such as in relativistic quantum mechanics from the square of the Dirac operator coupled with an interaction potential, or from the interaction of a polar molecule with an electron [@MorozSchmidt].
A standard method for studying these operators is through their action on Bloch waves. For any ${ \mathbf{k} } \in {\mathbb R}^3$, the Bloch waves of $H$ with wave vector ${ \mathbf{k} }$ are elements of $L^2_{loc}({\mathbb R}^3)$ that satisfy the semi-periodicity condition that, for all $X \in \Lambda$, $$\label{eq.Block}
\psi_{{ \mathbf{k} }}(x+X) = e^{i{{ \mathbf{k} }}\cdot X} \psi_{{ \mathbf{k} }}(x).$$ (It is enough to consider ${ \mathbf{k} }$ in the first Brillouin zone ${\mathbb P}_*$ of the reciprocal lattice to $\Lambda$. Also, the equality is that of two $L^2_{loc}$ functions, and hence it holds only almost everywhere in $x$.) A Bloch wave with wavevector ${ \mathbf{k} }$ can be written as $$\label{eq.Block2}
\psi_{{ \mathbf{k} }}(x) = e^{i{{ \mathbf{k} }}\cdot x} u_{{ \mathbf{k} }} (x)$$ for a function $u_{{ \mathbf{k} }}$ that is truly periodic with respect to $\Lambda_t$ and thus can be considered as living on the three-torus ${\mathbb T}:= {\mathbb R}^3 /\Lambda \simeq (S^1)^3$ (obtained by identifying points in ${\mathbb R}^3$ that are equivalent under the action of the lattice $\Lambda$ by translations). Note that the periodicity condition that a Bloch wave satisfies prevents it from being in $L^2({\mathbb R}^3)$, thus a nontrivial Bloch wave that satisfies the equation $$\label{eq.Bloch.e}
H \psi_{{ \mathbf{k} }} = \lambda \psi_{{ \mathbf{k} }}$$ is not, properly speaking, an eigenfunction of the Hamiltonian operator $H$. Rather, it is a generalized eigenfunction, corresponding to a value in the continuous spectrum of $H$. If $\psi_{{ \mathbf{k} }}$ is a Bloch wave that is a generalized eigenfunction of $H$ with generalized eigenvalue $\lambda$, then the function $u_{{ \mathbf{k} }} := e^{ -i{{ \mathbf{k} }}\cdot x} \psi_{{ \mathbf{k} }}(x)$ will then be an actual $\lambda$-eigenfunction of the ${ \mathbf{k} }$–Hamiltonian ${H_{{ \mathbf{k} }}}$ on $L^2({\mathbb T})$ defined by $$\label{eq:hamk}
{H_{{ \mathbf{k} }}}:= - \sum_{j=1}^3 (\partial_j + ik_j)^2 + V.$$ Indeed, this follows from the equation $$\label{eq.H-Hk}
H (e^{i{{ \mathbf{k} }}\cdot x} u_{{ \mathbf{k} }} (x)) = e^{i{{ \mathbf{k} }}\cdot x} {H_{{ \mathbf{k} }}}u_{{ \mathbf{k} }} (x).$$ Thus, it is useful to understand the regularity of eigenfunctions $u_{{ \mathbf{k} }}$ for the operators ${H_{{ \mathbf{k} }}}$, as well as to arrive at theoretical estimates for the accuracy of various schemes to estimate them and their associated eigenvalues.
Assumptions on the potential $V$
--------------------------------
In this paper, we extend and test the results of [@HunsickerNistorSofo] to deal with the more singular potentials that have inverse-square singularities. More precisely, we extend the results of the aforementioned paper from potentials where $\rho V$ is smooth in polar coordinates to potentials where $\rho^2V$ is smooth in polar coordinates and continuous on ${\mathbb T}$. (Recall that $\rho$ is a function that locally gives the distance to the singular point.) In particular, we obtain regularity results in weighted Sobolev spaces that will then permit us to derive estimates for the accuracy of two approximation schemes that we design and which are studied in detail in the forthcoming second and fourth parts of this paper. The first scheme is a finite element method with a mesh graded towards the singular points as in [@HunsickerNistorSofo; @BNZ3D2]. The second scheme is an augmented plane-wave method, similar to a “muffin-tin” method [@Martin].
In order to state our results, we first need to set some notation and introduce our assumptions on the potential $V$. Let ${\mathcal S}\subset {\mathbb T}:= {\mathbb R}^3/\Lambda$ be the finite set of points where $V$ has singularities. By abuse of notation, we shall denote by $|x-y|$ the induced distance between two points $x, y \in {\mathbb T}$. Let then $\rho :
{\mathbb T}\to [0, 1]$ be a nonnegative continuous function smooth outside ${\mathcal S}$ such that $$\label{eq.rho}
\rho(x) = |x - p|\ \text{ for } x \text{ close to } p \in {\mathcal S},$$ as before, and further assume also that $\rho(x) = 1$ for $x$ far from ${\mathcal S}$.
Our first assumption on $V$ is that $\rho^2V$ be smooth in polar coordinates up to $\rho=0$ near each singularity. Let us explain this in more detail. We first replace each singular point $p \in {\mathcal S}$ with a 2-sphere in a smooth way, thus obtaining a space denoted $\overline{{{\mathbb T}\smallsetminus {\mathcal S}}}$. This is the usual procedure of blowing up the singularities. We think of stretching out the holes where the singularities of $V$ are and compactifying the result using boundary spheres. It would be possible to carry out analysis similar to the calculations in this paper with only the assumption that $\rho^2 V$ be smooth on $\overline{{{\mathbb T}\smallsetminus {\mathcal S}}}$. To simplify some calculations and to obtain closed form results, however, we will further require the resulting function $Z$ to be constant on the blow up of each point in ${\mathcal S}$, which can be reformulated as saying that $\rho^2 V$ is also [*continuous*]{} on ${\mathbb T}$. Our first assumption on the potential $V$ is therefore $$\label{eq.def.Z}
\text{\bf Assumption 1}: \qquad\ Z := \rho^2 V \in
{\mathcal C}^\infty(\overline{{{\mathbb T}\smallsetminus {\mathcal S}}}) \cap {\mathcal C}({\mathbb T}).$$ Assumption 1, more precisely the continuity of $Z$ at ${\mathcal S}$, allows us to formulate our second assumption. Namely, $$\label{eq.Z1/4}
\text{\bf Assumption 2}: \qquad\
\eta_0 := \min_{p \in {\mathcal S}} Z(p) > -1/4 .$$ Therefore, the constant $$\label{eq.eta}
\eta := \sqrt{1/4 + \eta_0},$$ which will play an important role in this paper, is a [*positive real number*]{}. This constant will appear in many results below. We will use Assumptions 1 and 2 throughout the paper, except in Section \[sec1\], where we prove more general forms of our results, not requiring Assumption 2.
Regularity and approximation results
------------------------------------
The domains of all the Hamiltonian operators considered in this paper will be contained in weighted Sobolev spaces on ${\mathbb T}\smallsetminus
{\mathcal S}$. We define these spaces by: $$\label{eq.def.ws}
{\mathcal K}^m_a({{\mathbb T}\smallsetminus {\mathcal S}}) := \{v : {\mathbb T}\smallsetminus
{\mathcal S}\to {\mathbb C}, \ \rho^{|\beta|-a} \partial^\beta v \in L^2({\mathbb T}),
\ \forall\ |\beta| \leq m\}.$$ These spaces have been considered in many other papers, most notably in Kondratiev’s groundbreaking paper [@kondratiev67]. They can be identified with the b-Sobolev spaces of [@meaps] (associated to a manifold with boundary), but with a different indexing and notation. These spaces were generalized in [@AIN] to more general manifolds with corners with additional structure (Lie manifolds).
To formulate the stronger regularity for eigenvalues, we shall need the following notation. For each point $p \in {\mathcal S}$, let $$\label{eq.nu_0}
\nu_0(p) = \left\{
\begin{array}{ll}
2 & Z(p) \geq \frac{3}{4}\\
1 + \sqrt{1/4 + Z(p)} & Z(p) \in
(-\frac{1}{4},\frac{3}{4})\\
1 & Z(p) \leq -\frac{1}{4}\,,
\end{array}
\right.$$ and $$\label{eq.nu0}
\nu_0 = \min_{p \in {\mathcal S}} \nu_0(p).$$ For each point $p \in {\mathcal S}$ for which $Z(p) \in (-1/4,3/4]$, define a smooth cutoff function $\chi_p$ that is equal to $1$ in a small neighborhood of $p$ and is zero outside another small neighborhood of $p$, so that all the functions $\chi_p$ have disjoint supports. Define the space $W_{s}$ to be the complex linear span: $$\label{eq.def.Ws}
W_{s} = \sum_{Z(p) \in (-1/4,3/4]} {\mathbb C}\chi_p \rho^{\sqrt {1/4
+ Z(p)} - 1/2}.$$ Using also the notation introduced in the previous subsection, we then have the following result, whose proof follows from the proof of Theorem \[theorem1.gen\] below.
\[theorem1\] Consider a potential $V$ satisfying Assumptions 1 and 2. Then the Hamiltonian operator $H_{\bf k}$ acting as an unbounded operator on $L^2({\mathbb T})$ has a distinguished self-adjoint extension with domain $${\mathcal D}(H_{\bf k}) = {\mathcal K}_{2}^{2}({{\mathbb T}\smallsetminus {\mathcal S}}) + W_{s} \subset
{\mathcal K}^2_{\nu}({{\mathbb T}\smallsetminus {\mathcal S}}), \quad \nu< \nu_0 = \min_{p \in {\mathcal S}} \nu_0(p)
\in (0,2].$$ In particular, if $\eta_0 :=\eta_0 := \min_{p \in {\mathcal S}} Z(p) \geq
3/4$, then ${H_{{ \mathbf{k} }}}$ is in fact essentially self-adjoint and, if $\eta_0 >
3/4$, then ${\mathcal D}(H_{\bf k}) = {\mathcal K}_2^2({{\mathbb T}\smallsetminus {\mathcal S}})$.
The importance of the above theorem is the following corollary, which says that under Assumptions 1 and 2, the Hamiltonian operators ${H_{{ \mathbf{k} }}}=
-\sum_{j=1}^3 (\partial_j +ik_j)^2 +V$ acting on $L^2({\mathbb T})$ can be completely understood through their eigenfunctions and eigenvalues.
\[corollary2\] Under the assumptions of Theorem \[theorem1\], there exists a complete orthonormal basis of $L^2({\mathbb T})$ consisting of eigenfunctions of ${H_{{ \mathbf{k} }}}$.
We can now state a regularity theorem for the eigenfunctions of ${H_{{ \mathbf{k} }}}$ near a point $p \in {\mathcal S}$, or equivalently, for Bloch waves associated to the wavevector ${ \mathbf{k} }$. Recall the functions $\chi_p$ supported near points of ${\mathcal S}$ and used to define the spaces $W_{s}$.
\[theorem2\] Assume that $V$ satisfies Assumptions 1 and 2. Let ${H_{{ \mathbf{k} }}}u = \lambda
u$, where $u \in {\mathcal D}({H_{{ \mathbf{k} }}})$, $u \neq 0$. Then, for any $m \in {\mathbb Z}_+$, $$u \in {\mathcal K}^{m+1}_{a + 1}({{\mathbb T}\smallsetminus {\mathcal S}}), \quad \forall a < \eta := \min_{p
\in {\mathcal S}} \sqrt{1/4 + Z(p)}.$$ Moreover, we can find constants $a_p \in {\mathbb R}$ such that $$u - \sum_{p \in {\mathcal S}} \chi_p \rho^{\sqrt{1/4 + Z(p)} - 1/2} \in
{\mathcal K}^{m+1}_{a' +1}({{\mathbb T}\smallsetminus {\mathcal S}}), \quad \forall a' < \min_{p \in {\mathcal S}}
\sqrt{9/4 + Z(p)}\,.$$
The next result, which is the last we will mention in this introduction, will permit us to construct approximation schemes for the solutions of equations of the form $(\lambda + {H_{{ \mathbf{k} }}})u=f$.
\[theorem1.5\] Let us use the notation of Theorem \[theorem1\] and both Assumptions 1 and 2. Then there exists $C_0 > 0$ such that $\lambda + {H_{{ \mathbf{k} }}}:
{\mathcal K}_{a+1}^{m+1}({{\mathbb T}\smallsetminus {\mathcal S}}) \to {\mathcal K}_{a-1}^{m-1}({{\mathbb T}\smallsetminus {\mathcal S}})$ is an isomorphism for all $m \in {\mathbb Z}_{\ge 0}$, all $|a| < \eta$, and all $\lambda >
C_0$. In particular, $H_k$ is symmetric and bounded below, thus has a Friedrichs extension, which is equal to the closed extension considered in Theorem \[theorem1\] above.
From now on, we shall write ${H_{{ \mathbf{k} }}}$ for the Friedrichs extension of the original operator defined in Equation and ${\mathcal D}({H_{{ \mathbf{k} }}})$ for its domain.
We observe additionally that with the exception of Corollary \[corollary2\], all of the results above extend to Hamiltonian operators on ${\mathbb R}^3$ associated to a non-periodic potential with a finite number of inverse square singularities satisfying Assumptions 1 and 2 and radial limits at infinity. This is because the techniques employed to obtain the results are local, and a Hamiltonian operator over ${\mathbb R}^3$ with a smooth potential that has radial limits at infinity is always essentially self-adjoint. Of course, in this situation, there are only isolated eigenvalues below the continuous spectrum, and the bulk of the spectrum is continuous.
The paper is organized as follows. In Section \[sec1\] we prove Theorem \[theorem1\] identifying a closed self-adjoint extension of the operator $H_k$, and Theorem \[theorem2\] giving regularity results for eigenfunctions of this closure. Some of the results of this section do not rely on Assumption 2. Beginning with Section \[sec2\], however, we shall require that Assumption 2 be satisfied. In that section, we prove that for $\eta>-1/4$, $H_k$ is bounded from below, and we can thus identify the closure from Section \[sec1\] as the Friedrichs extension of ${H_{{ \mathbf{k} }}}$. In Section \[sec.last\] we discuss how our results extend to the the nonperiodic case and how to use them in numerical methods.
Acknowledgements {#acknowledgements .unnumbered}
----------------
We would like to thank Alexander Strohmaier, Joerg Seiler, Thormas Krainer, Jorge Sofo, and Anna Mazzucato for useful discussions. We also thank the Leverhulme Trust whose funding supported the fourth author during this project. This project was started while Hunsicker and Nistor were visiting the Max Planck Institute for Mathematics in Bonn, Germany, and we are grateful for its support.
Regularity and singular values\[sec1\]
======================================
The regularity analysis of the operators ${H_{{ \mathbf{k} }}}$ is done locally in the neighborhood of each $p \in {\mathcal S}$. Let us recall that $Z := \rho^2 V
\in {\mathcal C}^\infty(\overline{{{\mathbb T}\smallsetminus {\mathcal S}}}) \cap {\mathcal C}({\mathbb T})$, which is our Assumption 1, which we will require to hold true throughout this paper. In this section, we shall mention explicitly when Assumption 2 is used, since some of the results hold in greater generality.
For simplicity of the notation, we shall assume that ${\mathcal S}$ consists of a single point $p$. The results for potentials with several singularities with different values of $Z(p)$ can then be pieced together from local versions of the result in the one singularity case. The proofs of Theorems \[theorem1\] and Theorem \[theorem2\] (which we prove in this section in more general forms not requiring Assumption 2), rely on the pseudodifferential operator techniques of the b-calculus and b-operators [@aln2; @grieser; @meaps; @Lesch; @Schulze98]. A review of these basic tools is contained in [@HunsickerNistorSofo], so we will not go into detail about them again here. Throughout this paper, we will refer to b-operators and the b-calculus, although the properties can be equivalently described in terms of cone operators and the cone calculus, and in fact, in some of the references in this section, they are referred to in this way. For a discussion of the equivalence of the b- and cone calculi, see [@Joerg].
The boundary spectral set
-------------------------
In order to use the b-calculus, we study the associated b-differential operators $$P_{{\bf k},\lambda}:= -\rho^2({H_{{ \mathbf{k} }}}- \lambda).$$ We can write such an operator in polar coordinates around $p \in {\mathcal S}$ as $$\label{Hkinpolar}
P_{{\bf k},\lambda} = (\rho \partial_\rho)^2 + \rho \partial_\rho +
\Delta_{S^{n-1}} - \rho^2 V - \rho B_{{\bf k},\lambda},$$ where $$B_{{\bf k},\lambda}:= \rho\left(\sum_{j=1}^n (-2i\partial_j+k_j^2) -
\lambda \right)$$ is a first order b-operator.
The operator $P_{{ \mathbf{k} },\lambda}$ is an elliptic b-operator on $\overline{{{\mathbb T}\smallsetminus {\mathcal S}}}$. We calculate the indicial family of $P_{{ \mathbf{k} },\lambda}$ at a point $p \in {\mathcal S}$, denoted $
(\widehat{P_{{ \mathbf{k} },\lambda}})_p(\tau)$, by replacing $\rho
\partial_\rho$ with $\tau$ in Equation , and by replacing the coefficients with their values at $\rho=0$. Doing this, we find that the indicial families for $P_{{ \mathbf{k} },\lambda}$ at $p \in
{\mathcal S}$ are, in fact, independent of $\lambda$ and ${ \mathbf{k} }$, since the dependence on ${ \mathbf{k} }$ and $\lambda$ affects only $B_{{ \mathbf{k} }, \lambda}$, and $\rho$ vanishes at the singular points. We summarize this discussion in the following lemma.
\[lemma.ind.fam\] Let $P:=P_{0,0}$. Then the indicial family of $P_{{ \mathbf{k} }, \lambda}$ at $p \in {\mathcal S}$ is given by $$\label{Pindicial}
(\widehat{P_{{ \mathbf{k} },\lambda}})_p(\tau)= \hat{P}_p(\tau)= \tau^2 +
\tau + \Delta_{S^2} - Z(p).$$
Lemma \[lemma.ind.fam\] allows us to calculate the boundary spectral set $\mbox{Spec}_b(P_{p})$ for $P_{{ \mathbf{k} }, \lambda}$ at a given $p$. The [*boundary spectral set*]{} for $P_{{ \mathbf{k} }}$ is then defined by ([@meaps]) $$\mbox{Spec}_b(P_{p}):= \{(\tau,n) \mid \hat{P}_{p}(\tau)^{-1} \mbox{
has a pole of order } n+1 \mbox{ at } \tau\}.$$ By Lemma \[lemma.ind.fam\], the set $\mbox{Spec}_b(P_{p})$ will also be independent of ${ \mathbf{k} }$ and $\lambda$. To calculate the spectral set $\mbox{Spec}_b(P_p)$ explicitly, recall that the eigenvalues of $\Delta_{S^2}$ are $-l(l+1)$, for $l \in {\mathbb Z}_{\geq 0}$, and define $$\begin{aligned}
\beta_{l,p}&:= &\frac{\sqrt{(1+2l)^2 + 4Z(p)} - 1}{2}\,, \quad l\in
{\mathbb Z}_{\geq 0 } \\
\alpha_{l,p}&:=& \frac{-\sqrt{(1+2l)^2 + 4Z(p)} - 1}{2}\,, \quad l
\in {\mathbb Z}_{\geq 0} .\end{aligned}$$ By an abuse of notation, we take $\sqrt{(1+2l)^2 + 4Z(p)}$ to denote the positive imaginary root when the quantity under the root is negative. Our discussion gives the following.
\[lemma.Spec\_b\] If $Z(p) \notin \{ -(1/2+ l)^2\}_{l=0}^\infty$, we have $$\label{specPk}
\mbox{Spec}_b(P_{p}) = \bigcup_{l \in {\mathbb Z}_{\geq 0}} \left\{
\left(\beta_{l,p} ,0 \right) , \left(\alpha_{l,p} ,0 \right)
\right\},$$ and, if $Z(p) = -(1/2 + l_p)^2$, for some $l_p\geq 0$, $l_p \in {\mathbb Z}$, then $$\label{specPk2}
\mbox{Spec}_b(P_{p}) = \{(-1/2,1)\} \cup \bigcup_{l \in
{\mathbb Z}_{\geq 0}} \left\{ \left(\beta_{l,p} ,0 \right) ,
\left(\alpha_{l,p} ,0 \right) \right\}.$$
When $Z(p) < -1/4$, a finite number of these values of $\alpha$ and $\beta$ will be complex with real part equal to $-1/2$. This is one of the reasons why we have introduced Assumption 2, which states that $Z(p) > -1/4$ for all $p \in {\mathcal S}$. Of course, if Assumption 2 is satisfied, Spec${}_b(P_p)$ is given only by Equation . The case when $Z(p)$ is close to $-1/4$ is important because it gives to some interesting numerical phenomena and also because in various applications, it has an interesting interpretation, see for instance [@MorozSchmidt].
The machinery of the b-calculus now gives us information about closed self-adjoint extensions of ${H_{{ \mathbf{k} }}}$ (see [@Lesch; @GM; @GKM; @SchroheSI] for details). Note that in this setting, we are considering extensions of our operator acting on the core consisting of smooth functions supported away from the points of ${\mathcal S}$. For $Z(p) < 3/4$, there will be several possible self-adjoint extensions. Compare this to the case of extending the Laplacian operator on ${\mathbb T}$ from acting on the core consisting of all smooth functions. In this case, there is a unique self-adjoint extension. This is because the core is larger than in our case. The extension obtained using the larger core is one of the possible extensions obtained using the smaller core, but it is not the only one. This is why one does not see the issue of choosing a self-adjoint extension arising when the potential is of the form $Z\rho^\alpha$ for $\alpha>-2$, for instance, in the Coulomb case considered in [@HunsickerNistorSofo].
With the above lemmas in place, we can now prove Theorem \[theorem1\]. In fact, we shall prove a stronger result that does not require Assumption 2.
\[theorem1.gen\] Consider a potential $V$ satisfying Assumption 1 and assume that ${\mathcal S}$ consists of just one point $p$. Then the Hamiltonian operator $H_{\bf k}$ acting as an unbounded operator on $L^2({\mathbb T})$ has distinguished self-adjoint extension with domain ${\mathcal D}(H_{\bf k})
\subset {\mathcal K}^2_{\nu}({\mathbb T}\smallsetminus {\mathcal S})$ for all $\nu< \nu_0
\in (0,2]$. In particular, if $Z(p)\geq 3/4$, then ${H_{{ \mathbf{k} }}}$ is in fact essentially self-adjoint and, if $Z(p) > 3/4$, then ${\mathcal D}(H_{\bf k})
= {\mathcal K}_2^2({{\mathbb T}\smallsetminus {\mathcal S}})$. If Assumption 2 is satisfied, we also have $${\mathcal D}(H_{\bf k}) = {\mathcal K}_2^2({{\mathbb T}\smallsetminus {\mathcal S}}) + {\mathbb C}\chi \rho^{\eta -
1/2}, \quad \eta = \sqrt{1/4 + Z(p)}.$$ where $\chi$ is a cutoff function that is zero outside some neighborhood of $p$ and equals 1 close to $p$.
For each ${{ \mathbf{k} }}$ and $\lambda$, the operator ${H_{{ \mathbf{k} }}}- \lambda$ is a symmetric, unbounded b-operator on $L^2({\mathbb T})$ (see [@meaps; @Lesch; @GM]). Define the operator $A= \rho^{1/2} {H_{{ \mathbf{k} }}}\rho^{-1/2}$. Then $A$ is a symmetric unbounded b-operator on $\rho^{-1}L^2_b({{\mathbb T}\smallsetminus {\mathcal S}})=\mathcal{K}^0_{1/2}({{\mathbb T}\smallsetminus {\mathcal S}})$. The self-adjoint extensions of $A$ correspond exactly to those of ${H_{{ \mathbf{k} }}}- \lambda$ with domains shifted by weight $\rho^{1/2}$, so we will study the self-adjoint extensions of $A$ as the calculations are somewhat easier in this case.
By Lemma \[lemma.ind.fam\], the indicial roots of $A$ are the roots of ${H_{{ \mathbf{k} }}}$ shifted by 1/2. So we let $$\label{eq:alpha}
\tilde \beta_l = \sqrt{(l + 1/2)^2 +Z(p)} \quad \text{and} \quad
\tilde \alpha_l = - \tilde \beta_l.$$ Note that $0 \not= \tilde \beta_l\in {\mathbb R}$ if $Z(p)>-(l+\frac{1}{2})^2$, there is a double root at $\tilde
\beta_l=0$ if $Z(p)=-(l+\frac{1}{2})^2$, and $0 \neq \tilde \beta_l
\in i{\mathbb R}$ if $Z(p)<-(l+\frac{1}{2})^2$. The critical strip for self-adjointness of unbounded operators on $\rho^{-1}L^2_b({{\mathbb T}\smallsetminus {\mathcal S}})$ is $(-1,1)$, that is, an operator is essentially self-adjoint if and only if it has no indicial roots with real part in this interval (see [@Lesch; @GM]). Recalling that $l \in {\mathbb Z}_{\geq0}$, we see that for $Z(p) \geq \frac{3}{4}$, there are no roots in the critical strip so the operator is essentially self-adjoint. If further $Z(p) >
\frac{3}{4}$ we get the somewhat stronger result that ${\mathcal D}(H_{\bf k})
= {\mathcal K}^2_{2}({{\mathbb T}\smallsetminus {\mathcal S}})$. For $Z(p) \in
\left(-\frac{1}{4},\frac{3}{4}\right)$ we get two real roots in the critical strip corresponding to $l=0$, for $Z(p)= -\frac{1}{4}$, we get a double root at 0 in the critical strip corresponding to $l=0$, and for $Z(p)< -\frac{1}{4}$, we get a finite number of complex conjugate imaginary root pairs and possibly two real roots or a double root at 0 in the critical strip corresponding to some finite set of $l$.
By the theory in [@GM], the space ${\mathcal E}:= {\rm Dom}(A_{\rm
max})/{\rm Dom}(A_{\rm min})$ is finite dimensional and spanned by functions local around $p$ of the form: $$\label{eq:defset}
\bigcup_{|\Re(\tilde \beta_l)| \in (0,1)} \bigcup_{m=-l}^l \{
w\rho^{\tilde \beta_l}\psi^m_l, w\rho^{-\tilde \beta_l}\psi^m_l \}
\cup\bigcup_{\tilde \beta_l=0} \bigcup_{m=-l}^l \{w\psi^m_l, w\ln
\rho \psi^m_l\},$$ where $w$ is a local cutoff function that equals $1$ near $p$ and $0$ for $\rho$ large, and where the $\psi^m_l$ are an orthonormal basis for spherical harmonics with eigenvalue $l(l+1)$. Further, the operator $A$ with domain ${\mathcal D}:= {\rm Dom}(A_{\rm min}) +
\operatorname{span}(u_1, \ldots, u_n)$ is self adjoint if, and only if, linear combinations of these basis functions form a maximal set on which the pairing, $[u,v]_A$ is trivial, where $$\label{Apairing}
[u,v]_A := \frac{1}{2\pi} \oint_\gamma \hat{A} \hat{u}(\sigma)
\cdot_{S^2} \overline{\hat{v}( \overline{\sigma})} \,d\sigma.$$ Here $\gamma$ is a simple closed loop around the indicial roots of $A$ in the critical strip, $\hat{}$ represents the Mellin transform and $\cdot_{S^2}$ denotes the standard $L^2$ paring on $S^2$. Since the $\psi^m_l$ are orthonormal, this pairing reduces to a sum of loop integrals of the form: $$\label{pair}
[u_l,v_l]_A = -\frac{1}{2\pi} \oint_\gamma (\sigma^2 + \tilde
\beta_l^2) \hat{u}_l(\sigma)\overline{\hat{v}_l({\overline{\sigma}
})} \,d\sigma,$$ where $u_l = u_+ w \rho^{\tilde \beta_l} + u_- w \rho^{-\tilde
\beta_l}$ and $v_l = v_+ w \rho^{\tilde \beta_l} + v_-w
\rho^{-\tilde \beta_l}$ if $\tilde \beta_l \neq 0$ and $u_l = u_+ w +
u_- w \log(\rho)$ and $v_l = v_+ w + v_-w \log(\rho)$ if $\tilde
\beta_l = 0$.
We can consider three cases: $\tilde \beta_l >0$, $\tilde \beta_l =0$ and $\tilde \beta_l \in i{\mathbb R}$. First, as in [@GM], define $$\Phi(\sigma) = \widehat{-\rho\partial_\rho w}(\sigma) := - \int_0^\infty
\rho^{-i\sigma}w'(\rho) d\rho.$$ Then we get: $\overline{\hat{\Phi}(\bar{\sigma})} =
\hat{\Phi}(-\sigma)$ and $\Phi(0)=1$. Also, using the properties of the Mellin transform, we find that for any $\sigma \in {\mathbb C}$, $$\widehat{w\rho^{\pm \tilde \beta_l}}(\sigma) = \frac{\Phi(\sigma \pm
i\tilde \beta_l)}{\sigma \pm i \tilde \beta_l}.$$
Now consider the case $\tilde \beta_l>0$. Carrying out the loop integral by evaluating residues, we arrive at the equation $$[u_l,v_l]_A = k(u_+\bar{v}_- - u_-\bar{v}_+)$$ for a constant $k \neq 0$. If we set $[u_l,u_l]=0$, this reduces to $\arg(u_+)=\arg(u_-)$. Thus to get a self-adjoint boundary condition at $p$ for $A$ we can fix any ratio of $|u_+|$ to $|u_-|$. We will choose to enlarge the minimal domain by the set with $|u_-|=0$, so spanned by $\{w \rho^{\tilde \beta_l} \psi^m_l\}_{m=-l}^l$. Note that if $Z(p)>-\frac{1}{4}$, we have $l=0$, so to create a self-adjoint extension of $A = \rho^{1/2}{H_{{ \mathbf{k} }}}\rho^{-1/2}$, we only need to expand the minimal domain by the span of $w \rho^{\eta} = w \rho^{\sqrt{1/4
+Z(p)}}$.
Next consider the case when $\tilde \beta_l =0$. In this case we get $$\hat{w}(\sigma) = \frac{\Phi(\sigma)}{\sigma} \quad \mbox{and} \quad
\widehat{w\log \rho} = \frac{\Phi(\sigma)}{\sigma^2} -
\frac{\Phi'(\sigma)}{\sigma}.$$ Carrying out the loop integral in this case, we arrive at the equation $$[u_l,v_l]_A = k(u_+\bar{v}_- + u_-\bar{v}_+).$$ Again setting $[u_l,u_l]_A=0$, we arrive this time at the condition $u_+\bar{u}_- \in i{\mathbb R}$. So we may this time again choose to fix $u_-=0$ and enlarge the minimal domain by the set spanned by $\{w
\psi^m_l\}_{m=-l}^l$.
Finally, consider the case when $\tilde \beta_l = ia$. This time we get the equation $$[u_l,v_l]_A = k(u_+\bar{v}_+ - u_-\bar{v}_-).$$ Setting $[u_l,u_l]=0$ we arrive at the condition $|u_+| = |u_-|$. We can choose $u_+=1$ and $u_- = -1$ to get a self-adjoint condition by enlarging the minimal domain by the set spanned by $\{w \cos (a \log
\rho) \psi^m_l\}_{m=-l}^l$.
In order to get back to the corresponding choice of self-adjoint extension of ${H_{{ \mathbf{k} }}}- \lambda$, we multiply each basis element by $\rho^{-1/2}$. Since each of these basis functions is in $\rho^\mu
L^2({\mathbb T})$ for all $\mu<1$, overall we find that $\mathcal{D}(H)
\subset \rho^\mu L^2({\mathbb T})$ for all $\mu<1$. This completes the proof of Theorem \[theorem1\].
Some corollaries
----------------
We now prove some consequences of Theorem \[theorem1.gen\]. First, its proof implies the following stronger result:
If $\eta_0 > 0$, then ${\mathcal D}(H_{\bf k}) \subset H^2({\mathbb T}) \cap
\rho^\epsilon C^0({\mathbb T})$, for some $\epsilon>0$.
We now deduce Corollary \[corollary2\] from Theorem \[theorem1\] and its proof. This corollary is specific to the periodic case.
(of Corollary \[corollary2\]). The extension of ${H_{{ \mathbf{k} }}}$ is self-adjoint, hence has real spectrum, so we get that for $\lambda \notin {\mathbb R}$, the operator $$\lambda - {H_{{ \mathbf{k} }}}: {\mathcal D}({H_{{ \mathbf{k} }}}) \rightarrow L^2({\mathbb T})$$ is a bounded invertible operator with bounded inverse $Q_{{ \mathbf{k} },\lambda}$, which is the resolvent of ${H_{{ \mathbf{k} }}}$ at $\lambda$. By Theorem \[theorem1\], by the definition of the weighted Sobolev spaces ${\mathcal K}_a^m({{\mathbb T}\smallsetminus {\mathcal S}})$, and by the b-Rellich lemma [@AIN; @meaps], we have for some $\epsilon > 0$ that $${\mathcal D}({H_{{ \mathbf{k} }}}) \subset {\mathcal K}^2_{\epsilon}({{\mathbb T}\smallsetminus {\mathcal S}}) \Subset L^2({\mathbb T}).$$ (Recall that $\Subset$ means “compactly embedded”.) Thus for $\lambda \notin {\mathbb R}$, the resolvent of ${H_{{ \mathbf{k} }}}$, $$\mathcal{R}_\lambda({H_{{ \mathbf{k} }}}): L^2({\mathbb T}) \rightarrow L^2({\mathbb T})$$ is a compact operator. By standard results of functional analysis, if a self-adjoint operator has compact resolvent, then $L^2$ has a complete orthonormal basis consisting of eigenfunctions for this operator.
Singular functions expansion
----------------------------
To prove Theorem \[theorem2\], we again use results of the b-calculus, this time primarily from [@meaps]. Again, we prove a more general statement that does not require the Assumption 2. Let $\nu_0$ be as in Equation .
\[theorem2.gen\] Assume ${\mathcal S}=\{p\}$ and $Z:= \rho^2V$ satisfies Assumption 1. Assume ${H_{{ \mathbf{k} }}}u = \lambda u$ where $u \in {\mathcal D}({H_{{ \mathbf{k} }}})$. Then for any $m\in {\mathbb Z}_+$ and any $\nu < \nu_0$, $$u \in {\mathcal K}^{m}_{\nu}({{\mathbb T}\smallsetminus {\mathcal S}}).$$ Further, near each $p \in {\mathcal S}$, where $Z(p) \neq (l+1/2)^2$ for any $l \geq 0$, $u$ has a complete (though not unique) expansion of the form: $$\label{exp}
u = u_0 + \sum \rho^{\gamma} g_\gamma, \quad \gamma \in
\mathcal{I}_{Z(p)},\ \Re(\gamma) > - 1/2,$$ where the formula for $\mathcal{I}_{Z(p)}$ is given in Equation below, $u_0$ is smooth up to $\rho=0$ in polar coordinates and vanishes to all orders there, hence is in fact smooth on ${\mathbb T}$ and vanishes to all orders at $p$, and the coefficient functions $g_\gamma$ are smooth functions on $S^2$. Under the additional Assumption 2, when $Z(p)>-1/4$, the first coefficient $g_{\eta - 1/2}$ is a constant function.
Any eigenfunction $u$ of ${H_{{ \mathbf{k} }}}$ must be in its domain, thus in ${\mathcal K}^2_\nu({{\mathbb T}\smallsetminus {\mathcal S}})$ for all $\nu< \nu_0$. Our first goal is to improve the degree of smoothness from $2$ to $m$ for $m \in \mathbb{N}$. To do this, we use the fact that any $\lambda \in {\mathbb C}$, the operator $H_{{ \mathbf{k} }} -\lambda$ is Fredholm as a map between weighted Sobolev spaces: $$\label{Fredholm}
{H_{{ \mathbf{k} }}}- \lambda: {\mathcal K}^m_a({\mathbb T}\smallsetminus{\mathcal S}) \rightarrow
{\mathcal K}^{m-2}_{a-2}({\mathbb T}\smallsetminus{\mathcal S})$$ for all $a \in {\mathbb R}$ such that $a \notin \cup_{l \in {\mathbb Z}_{\geq 0 }}
\{\beta_{l,p} + \frac{3}{2}, \alpha_{l,p}+\frac{3}{2}\}=
\mbox{Spec}_b(H_{{ \mathbf{k} }}) + 3/2$. By general b-calculus theory, the set $\mbox{Spec}_b(H_{{ \mathbf{k} }})$ is a discrete subset of ${\mathbb C}$ and furthermore, for any $\gamma_0$ and $\eta$, it has only a finite number of elements in the strip $\gamma_0 \leq \Re(z) \leq \eta$. Thus for any $\nu_0$, there exist arbitrarily close $\nu<\nu_0$ such that the condition on $a$ is satisfied for $a=\nu + 2s$, where $s \in
\mathbb{N}$. Together with standard bootstrapping arguments, this allow us to improve the regularity of eigenfunctions of ${H_{{ \mathbf{k} }}}$ in terms of weighted Sobolev spaces to ${\mathcal K}^m_{\nu}({{\mathbb T}\smallsetminus {\mathcal S}})$ for all $m$ and $\nu < \nu_0$.
Next, to obtain the expansion in \[theorem2\] we use a general result in the b-calculus literature, see [*e.g.*]{} [@meaps], that implies that any $u \in \cup_{m,a}{\mathcal K}^m_a({\mathbb T}\smallsetminus{\mathcal S})$ which is an eigenfunction for ${H_{{ \mathbf{k} }}}$ in some weighted $L^2({\mathbb T})$ in fact has much stronger regularity: it polyhomogeneous in $\rho$ near each $p \in {\mathcal S}$ with index set $\mathcal{I}_{Z(p)}$. If $Z(p)
\notin \{-(l+\frac{1}{2})^2\}_{l=0}^\infty$ (for instance if $Z(p)>-1/4$), then the index set is simply a set of complex numbers that is finite in any strip $\gamma_0 \leq \Re(z) \leq \eta$: $$\label{index}
\mathcal{I}_{Z(p)} = \bigcup_{n=0}^\infty \{\beta_{l,p} + n,
\alpha_{l,p}+n \}_{l \in {\mathbb Z}_{\geq 0 }}.$$ This means that around each $p \in {\mathcal S}$, there exist smooth coefficient functions $g_{\gamma} \in {\mathcal C}^{\infty}(S^2)$ such that for all $N$, $$\label{eq.asympt.exp}
u_N := u - \sum \rho^{\gamma} g_\gamma \in \rho^N
{\mathcal C}^N(\overline{{{\mathbb T}\smallsetminus {\mathcal S}}}), \quad \gamma \in \mathcal{I}_{Z(p)}
\Re(\gamma) \leq N$$ that is, $u_N$ is $N$ times continuously differentiable up to $\rho
=0$ in polar coordinates at $p$ and in these coordinates, vanishes with all of its derivatives up to order $\rho^N$ there. We take the limit in the topology of the smallest ${\mathcal K}^m_a$ to which $u$ belongs.
If $Z(p) \neq -(l+\frac{1}{2})^2$ for any $l \geq 0$ and if $u \in
L^2({\mathbb T})$, when we let $N \rightarrow \infty$, we find that $$u = u_0 + \sum \rho^{\gamma} g_\gamma, \quad
\gamma \in \mathcal{I}, \Re(\gamma) > -1/2,$$ where $u_0$ is smooth up to $\rho=0$ in polar coordinates and vanishes to all orders there, hence is in fact smooth on ${\mathbb T}$ and vanishes to all orders at $p$.
Since the set of $\gamma$ that appear in this expansion is discrete in ${\mathbb R}$, we get that the smallest exponent that appears will in fact be somewhat better than $-3/2$. This first exponent will be $\nu_0$ if $Z(p) \leq 3/4$. If $Z(p)>3/4$, then eigenfunctions will in fact be in a space with higher weight than ${\mathcal K}^2_2({{\mathbb T}\smallsetminus {\mathcal S}})$: the weight will be $1+\sqrt{Z(p) + 1/4}$.
Finally, we can note that the terms of the expansion of an eigenfunction for ${H_{{ \mathbf{k} }}}$ that are not in ${\mathcal K}^2_2({{\mathbb T}\smallsetminus {\mathcal S}})$ will be of the forms determined in the proof of Theorem \[theorem1\] (see, eg [@HHM] for a proof). So, for instance, if $Z(p)\geq -1/4$, the leading term of any eigenfunction $u$ will be constant in $S^2$, and $u$ minus its leading term will vanish at $p$. Further, if $Z(p)\geq
-1/4$, then the exponents $\gamma$ will all be real numbers. Thus we obtain: $$\label{exp2}
u = u_0 + \rho^{\eta - \frac{1}{2}} g_{\eta - 1/2} + \sum
\rho^{\gamma} g_\gamma, \quad \gamma \in \mathcal{I}_{Z(p)} \gamma >
\eta - 1/2,$$ where $g_{\eta - 1/2}$ a constant. This completes the proof of Theorem \[theorem2\].
Invertibility\[sec2\]
=====================
In this section we prove the boundedness and invertibility result in Theorem \[theorem1.5\]. From now on, we require both Assumptions 1 and 2 to be satisfied by our potential $V$.
Preliminary results
-------------------
We begin with a few standard results lemma.
\[lemma.Q2\] Let $m, a \in {\mathbb R}$. Then
(i) For any $f \in {\mathcal C}^{\infty}(\overline{{{\mathbb T}\smallsetminus {\mathcal S}}})$, the multiplication map $${\mathcal K}^m_a({{\mathbb T}\smallsetminus {\mathcal S}}) \ni u \to f u \in
{\mathcal K}^m_a({{\mathbb T}\smallsetminus {\mathcal S}}) = \rho^{a-3/2} H^m_b({{\mathbb T}\smallsetminus {\mathcal S}})$$ is continuous for all $m \in {\mathbb Z}_+$ and all $a \in {\mathbb R}$.
(ii) The operator ${H_{{ \mathbf{k} }}}-\lambda$ maps ${\mathcal K}_{a+1}^{m+1} ({{\mathbb T}\smallsetminus {\mathcal S}})$ to ${\mathcal K}_{a-1}^{m-1} ({{\mathbb T}\smallsetminus {\mathcal S}})$ continuously.
(iii) The operator $\rho^{-1}B_{{ \mathbf{k} }, \lambda}$ maps ${\mathcal K}_{a+1}^{m+1} ({{\mathbb T}\smallsetminus {\mathcal S}})$ to ${\mathcal K}_{a}^{m}
({{\mathbb T}\smallsetminus {\mathcal S}})$ continuously.
(iv) $\rho^{-1}B_{{ \mathbf{k} }, \lambda} : {\mathcal K}_{a+1}^{m+1} ({{\mathbb T}\smallsetminus {\mathcal S}}) \to
{\mathcal K}_{a-1}^{m-1} ({{\mathbb T}\smallsetminus {\mathcal S}})$ is compact.
The simple proofs of these results are the same as that of the analogous results in [@HunsickerNistorSofo], and follow directly from properties of b-operators [@AIN; @meaps; @Lesch].
We also need the following standard lemma (again, see [@HunsickerNistorSofo] for its proof).
\[lemma.Green\] Let $a \in {\mathbb R}$ be arbitrary and assume that $u \in
{\mathcal K}_{1+a}^2({{\mathbb T}\smallsetminus {\mathcal S}})$ and that $v \in {\mathcal K}_{1-a}^2({{\mathbb T}\smallsetminus {\mathcal S}})$. Then $(\Delta
u , v) + (\nabla u, \nabla v) = 0$.
We shall also need the following consequence of the general properties of the b-calculus [@meaps; @Schulze98].
\[prop.Fredholm\] Let us fix $\lambda \in {\mathbb C}$ and $a \notin \{\tilde \beta_{l,p} ,
\tilde \alpha_{l,p}\} = \cup_{l \in {\mathbb Z}_{\geq 0 }} \{\beta_{l,p} +
\frac{1}{2}, \alpha_{l,p}+\frac{1}{2}\}$. Let $N$ be the number of elements in the set $\{\tilde \beta_{l,p} , \tilde \alpha_{l,p}\}$ that are between $0$ and $a$, counted with multiplicity. Then the operator $H_{{ \mathbf{k} }} -\lambda$ is Fredholm as a map between weighted Sobolev spaces: $${H_{{ \mathbf{k} }}}- \lambda: {\mathcal K}^{m+1}_{a+1}({\mathbb T}\smallsetminus{\mathcal S}) \rightarrow
{\mathcal K}^{m-1}_{a-1}({\mathbb T}\smallsetminus{\mathcal S})$$ and has index $-N$ if $a > 0$, respectively $N$ if $a < 0$.
We consider again the operator $P_{0, 0} = \rho ({H_{{ \mathbf{k} }}}- \lambda) \rho$, which is a b-differential operator. It is unitarily equivalent to $\rho^{1/2} P_{0, 0} \rho^{-1/2}$ acting on b-Sobolev spaces (see the proof of Theorem \[theorem1.gen\]), which has $\{\tilde \beta_{l,p}
, \tilde \alpha_{l,p}\}$ as a b-spectrum. The result then follows from the characterization of Fredholm b-differential operators [@meaps; @KMR; @LMN1].
It remains to determine the index of ${H_{{ \mathbf{k} }}}- \lambda$. Let $m_a$ be the index of the operator for a fixed value of $a$. Then it is a standard result that $m_a - m_b$ is given by the number of singular functions with exponent between $a$ and $b$ [@KMR; @meaps; @Schulze98; @MoroianuNistor]. This is enough to complete the proof.
See [@HLNU3] for an extension of this result and for more details.
Now recall the Hardy inequality, which states that $$\label{eq.Hardy}
\int_{{\mathbb R}^N} r^{-2} |u(x)|^2 dx \le (2/(N-2))^2 \int_{{\mathbb R}^N} |\nabla
u(x)|^2 dx$$ for any $u \in H^1({\mathbb R}^N)$, $N \ge 3$, where $r$ is the distance to the origin [@GGM]. We can use this to prove the following important lemma. To simplify notation, after the lemma statement, we shall let $(u, v) := (u, v)_{L^2({\mathbb T})}$.
\[lemma.bounded\] There are constants $C, \gamma > 0$ such that for any $u \in
{\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}})$, $$({H_{{ \mathbf{k} }}}u, u)_{L^2({\mathbb T})} + C (u, u)_{L^2({\mathbb T})} \ge \gamma (u,
u)_{{\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}})} := \gamma \int_{{\mathbb T}} \big (\rho^{-2} |u(x)|^2 +
|\nabla u(x)|^2 \big ) dx.$$
For an operator $T: {\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}}) \rightarrow {\mathcal K}_{-1}^{-1}({{\mathbb T}\smallsetminus {\mathcal S}})$, we shall write $T \ge 0$ if $(Tu, u) \ge 0$ for all $u \in
{\mathcal K}_{1}^1({{\mathbb T}\smallsetminus {\mathcal S}})$. Now let $\phi \ge 0$ be a smooth function $\mathbb{T}$ that is equal to $1$ in a small neighborhood of ${\mathcal S}$ and has support on the set where $\rho(x)$ is given by the distance to ${\mathcal S}$ and let $V_0 (x) = Z(p)\phi(x) \rho^{-2}(x)$ for $x$ in the support of $\phi$ and close to $p \in {\mathcal S}$. Outside the support of $\phi$, we let $V_0 = 0$. Then Hardy’s inequality applied to $\phi^{1/2} u$, which we can think of as living now on $\mathbb{R}^3$ rather than $\mathbb{T}$, gives $$\label{eq.Hardy.local}
\big( \phi^{1/2} (-\Delta + z V_0)\phi^{1/2} u, u ) \ge 0 \quad
\mbox{and } \big( \phi^{1/2} (-\Delta)\phi^{1/2} u, u ) \ge 0.$$ We can think of this as saying that the most singular part of the operator ${H_{{ \mathbf{k} }}}$, that is, $T=\phi^{1/2} (-\Delta + z V_0)\phi^{1/2}$, satisfies $T\geq 0$. We will prove Lemma \[lemma.bounded\] by decomposing the operator ${H_{{ \mathbf{k} }}}+C$ as a sum of four operators $${H_{{ \mathbf{k} }}}= T_1+ T_2 + T_{3,C} + T_{4,C},$$ which we will show are all bounded from below for sufficiently large $C$.
Recall we can write $${H_{{ \mathbf{k} }}}= -\Delta + V_0 + V_1 + \rho^{-1}B_{\rm k. 0},$$ where $V_1 := V-V_0$ satisfies $\rho V_1 \in C(\mathbb{T})$ and $\rho^{-1}B_{\rm k. 0}$ is a first order differential operator over $\mathbb{T}$ with smooth coefficients.
Assumption 2 and Equation imply that for $\epsilon<1$, the operator $$\label{op.positive}
T_1 := (1-\epsilon)T \ge 0.$$ Fix any suitable value for $\epsilon>0$. Then we can write ${H_{{ \mathbf{k} }}}+C$ in terms of $T_1$ by decomposing in terms of the multiplication operators $\phi^{1/2}$ and $(1-\phi^{1/2}$: $${H_{{ \mathbf{k} }}}+ C = \epsilon T + T_1 -\psi^{1/2}\Delta \psi^{1/2} -
(1-\phi^{1/2})\Delta(1-\phi^{1/2}) + V_1 +R_1,$$ where $R_1$ is a first order differential operator with smooth coefficients and $\psi = 2\phi^{1/2}(1-\phi^{1/2})$.
Let $T_2 := -\psi^{1/2}\Delta \psi^{1/2} - (1-\phi^{1/2})
\Delta(1-\phi^{1/2})$. Then $T_2 \geq 0$ by the Hardy equality applied to $\psi^{1/2} u$ and to $(1-\phi^{1/2})u$. Define $T_3 :=
-\epsilon \Delta + R_1 + C/2$ and $T_4 = \epsilon \phi\rho^{-2} + V_1
+ C/2$. We claim that for $C$ large enough, $T_3 \ge 0$ and $T_4 \ge
0$, which will prove the result.
The proof that $T_4 \geq 0$ for $C>>0$ follows from a straightforward calculation minimizing the function $\epsilon \phi\rho^{-2} + V_1$. The proof that $T_3 \geq 0$ for $C>>0$ is basically the same as the proof that a Schrödinger operator with periodic Coulumb type potential is bounded below. This is proved, for example, in [@HunsickerNistorSofo].
Note that the above lemma implies that ${H_{{ \mathbf{k} }}}$ is bounded from below as an operator ${\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}})\rightarrow {\mathcal K}_{-1}^{-1}({{\mathbb T}\smallsetminus {\mathcal S}})$, which is the special case of Theorem \[theorem1.5\] when $m=a=0$. In addition, if we define the form $\alpha(u,v):=(({H_{{ \mathbf{k} }}}+C)u,v),$ where the right-hand side is the natural pairing between elements of ${\mathcal K}^{-1}_{-1}$ and ${\mathcal K}^1_1$, then this lemma implies that $\alpha(u,v)$ satisfies the assumptions of the Lax-Milgram lemma for the vector space $V={\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}})$. This and Céa’s lemma imply that for any element $u \in {\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}})$ and any finite dimensional subspace $V \subset {\mathcal K}_1^1({{\mathbb T}\smallsetminus {\mathcal S}})$ we can construct a unique (Galerkin) approximation $u_V \in V$ for $u$ that, up to a multiple independent of $u$, is the best approximation for $u$ in $V$.
If we could also use the ${\mathcal K}_1^1$ norm in our approximation results, we would now have the necessary tools to prove it. However, we need to use the slightly smaller space ${\mathcal K}_{a+1}^1$ instead. Thus we the stronger result, Theorem \[theorem1.5\] to ensure the Lax-Milgram theorem and Céa’s lemma apply to the analogous form on these spaces.
We shall also need the following regularity result.
\[prop.regularity\] Let $a, \lambda \in {\mathbb R}$, $m \in {\mathbb Z}_{+}$. There exists a constant $C
> 0$ such if $u \in {\mathcal K}_{a+1}^{1}({{\mathbb T}\smallsetminus {\mathcal S}})$ and $(\lambda + {H_{{ \mathbf{k} }}})u \in
{\mathcal K}_{a-1}^{m-1}({{\mathbb T}\smallsetminus {\mathcal S}})$ then $u \in {\mathcal K}_{a+1}^{m+1}({{\mathbb T}\smallsetminus {\mathcal S}})$ and $$\|u\|_{{\mathcal K}_{a+1}^{m+1}} \le C \big ( \|(\lambda + {H_{{ \mathbf{k} }}})
u\|_{{\mathcal K}_{a+1}^{m+1}} + \|u\|_{{\mathcal K}_{a+1}^{1}} \big ).$$
We consider again the operator $P_{0, 0} = \rho ({H_{{ \mathbf{k} }}}- \lambda) \rho$, which is a b-differential operator. Our result then follows from the regularity for b-pseudodifferential operators [@meaps; @AIN].
We now complete the proof of Theorem \[theorem1.5\] as follows.
(of Theorem \[theorem1.5\]). As in [@HMN], by regularity for b-differential operators, if we prove our result for $m = 0$, then the regularity result of Proposition \[prop.regularity\] implies it for all $m \ge 0$. We shall thus assume $m=0$ and focus on extending Lemma \[lemma.bounded\], where $a=0$, to the case when $|a| < \eta$.
Fix $C$ as in Lemma \[lemma.bounded\]. Let $D_a := C + {H_{{ \mathbf{k} }}}:
{\mathcal K}^{1}_{a+1}({{\mathbb T}\smallsetminus {\mathcal S}}) \to {\mathcal K}^{-1}_{a-1}({{\mathbb T}\smallsetminus {\mathcal S}}),$ that is to say, $C +
{H_{{ \mathbf{k} }}}$ with fixed domain and range. As usual, we may identify the dual of ${\mathcal K}_{b}^{1}({{\mathbb T}\smallsetminus {\mathcal S}})$ with the space ${\mathcal K}_{-b}^{-1}({{\mathbb T}\smallsetminus {\mathcal S}})$ using the $L^2$-inner product. Then using the symmetry of ${H_{{ \mathbf{k} }}}$, we find that $D_a^* =D_{-a}$.
By Lemma \[lemma.bounded\], the operator $D_0$ is invertible. By basic results in b-calculus, $D_a$ is Fredholm for $|a| < \eta$ since the weighted spaces in its domain and range do not correspond to an indicial root as calculated in the previous section (see Proposition \[prop.Fredholm\]). Hence, for such $a$, the family $\rho^a D_a
\rho^{-a}$ is a continuous family of Fredholm operators between the same pair of spaces. Since index is constant over such families, we have that ${\operatorname{ind}}(D_a) = 0$ for all $0\leq a < \eta$. We want to know these operators are all isomorphisms. By the index calculation, it now suffices to show they are all injective.
The inclusion ${\mathcal K}^{1}_{a+1}({{\mathbb T}\smallsetminus {\mathcal S}}) \subset {\mathcal K}^{1}_{1}({{\mathbb T}\smallsetminus {\mathcal S}})$ allows us to compute $(D_a u , u) = (\nabla u, \nabla u) + (u, u)$ for $u \in {\mathcal K}^{1}_{a+1}({{\mathbb T}\smallsetminus {\mathcal S}})$, by Lemma \[lemma.Green\]. Assume $D_a
u = 0$, then $(D_a u, u) = 0$, and hence $u = 0$. This implies that the operator $D_a$ is injective for $0 \le a < \eta$. Since it is Fredholm of index zero, it is also an isomorphism. This proves our result for $0 \le a < \eta$. To prove the result for $-\eta < a \le
0$, we take adjoints and use $D_a = (D_{-a})^*$.
By the characterization in [@GM] of the Friedrichs extension of a b-operator which is bounded below, we can see that the extension we constructed in Theorem \[theorem1\] is in fact the Friedrichs extension of ${H_{{ \mathbf{k} }}}$. The proof of Theorem \[theorem1.5\] is now complete.
The fact that the domain of the Friedrichs extension is $(C -
{H_{{ \mathbf{k} }}})^{-1}(L^2({{\mathbb T}\smallsetminus {\mathcal S}}))$ and the theorem we have just proved give us a second way to identify the domain of the Friedrichs extension of ${H_{{ \mathbf{k} }}}$. Following the method of [@HunsickerNistorSofo], we see that when ${H_{{ \mathbf{k} }}}$ is Fredholm on ${\mathcal K}_{2}^{2}({{\mathbb T}\smallsetminus {\mathcal S}})$, the domain of the Friedrichs extension of ${H_{{ \mathbf{k} }}}$ consists of the span of ${\mathcal K}_{2}^{2}({{\mathbb T}\smallsetminus {\mathcal S}})$ and of the singular functions that are in ${\mathcal K}_{1}^{1}({{\mathbb T}\smallsetminus {\mathcal S}})$ but are not in ${\mathcal K}_{2}^{2}({{\mathbb T}\smallsetminus {\mathcal S}})$. This can be used to obtain an alternative proof of Theorem \[theorem1\] if $V$ satisfies both Assumptions 1 and 2, as follows.
Let $C_0$ be as in Theorem \[theorem1.5\] and $W_s$ be as in Equation . Assume the set $\{\tilde
\beta_{l,p} , \tilde \alpha_{l,p}\}$ does not contain 1. Then for $\lambda > C_0$, the map $$\lambda + {H_{{ \mathbf{k} }}}: {\mathcal K}_{2}^{m+1}({{\mathbb T}\smallsetminus {\mathcal S}}) + W_s \to {\mathcal K}_{0}^{m-1}({{\mathbb T}\smallsetminus {\mathcal S}})$$ is an isomorphism.
Let $T := \lambda + {H_{{ \mathbf{k} }}}$ with the indicated domain and codomain. Proposition \[prop.Fredholm\] shows that $T$ is Fredholm with index zero. Since $${\mathcal K}_{2}^{m+1}({{\mathbb T}\smallsetminus {\mathcal S}}) + W_s \subset {\mathcal K}_{1}^{1}({{\mathbb T}\smallsetminus {\mathcal S}}),$$ Theorem \[theorem1.5\] shows that $T$ is injective. Hence it is also surjective, hence an isomorphism.
Extensions and numerical tests\[sec.last\]
==========================================
We now discuss the extension to the non-compact case and indicate some applications to numerical methods.
The non-compact case\[ssec.new\]
--------------------------------
Most of our results in the previous sections extend to the non-compact case. Let $\overline{{\mathbb R}^3}_{rad}$ be the radial compactification of ${\mathbb R}^3$. We assume that the set of singular points ${\mathcal S}\subset
{\mathbb R}^3$ is finite and we replace each of the points in ${\mathcal S}$ with a two-sphere (that is, we blow up the singular points). Let ${\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}}$ denote the resulting compact manifold with boundary. By $\rho$ we denote a continuous function $\rho : {\mathbb R}^3 \to [0, 1]$ that is smooth outside ${\mathcal S}$, close to each $p \in {\mathcal S}$ it has the form $\rho(x) =
|x - p|$, and it is constant equal to $1$ outside a compact set. (Thus the difference with the function $\rho$ considered in the periodic case is that now $\rho$ is constant equal to $1$ in a neighborhood of infinity.) Then in the non-compact case, our Assumption 1 on $Z :=
\rho^2 V$ is replaced with $$\label{eq.def.Zprime}
\text{\bf Assumption }1^\prime: \qquad\ Z := \rho^2 V \in
{\mathcal C}^\infty({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}}) \cap {\mathcal C}({\mathbb R}^3).$$ Assumption 2 remains unchanged.
We consider now $H = -\Delta + V$ instead of the restrictions ${H_{{ \mathbf{k} }}}$. Assumptions $1^\prime$ and 2 allow us to extend to $H$ all the results for ${H_{{ \mathbf{k} }}}$ of the previous sections, except Corollary \[corollary2\] and Proposition \[prop.Fredholm\]. The weighted Sobolev spaces ${\mathcal K}_{a}^{m}({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}})$ are defined in the same way (but using the new function $\rho$).
Let $b_e$ the infimum of $V$ on the sphere at infinity. Then Corollary \[corollary2\] must be replaced with the following characterization of the essential spectrum $\sigma_e(H)$ of $H$: $$\sigma_e(H) = [b_e, \infty).$$ To prove this result, one needs also the Fredholm conditions for operators in the scattering or SG calculus [@LMN1; @MelroseScattering; @SchroheSI; @Parenti]. Then in Proposition \[prop.Fredholm\] one has to take $\lambda < b_e$. Of course, in Theorem \[theorem1.5\] one will have $C_0 > - b_e$.
However, in the non-compact case, for applications to numerical methods, our results on eigenvalues and eigenfunctions must be complemented by decay properties at infinity. The following is proved as in [@AmmannCarvalhoNistor], Theorem 4.4. See also [@AgmonDecay; @PuriceMagnetic; @PuriceDecay]. Let $r: {\mathbb R}^3 \to {\mathbb R}$ be a smooth function such $r(x) = |x|$ for $x$ outside a compact set.
\[thm.eigen.decay\] Let $V$ be a potential satisfying Assumptions $1^\prime$ and 2. Also, let $0 < \epsilon <
V(x) - \lambda $ for $x$ outside a compact set and be $u$ an eigenvector of ${H_{{ \mathbf{k} }}}$ corresponding to $\lambda$. Then $e^{\epsilon r}
u \in {\mathcal K}_{\nu}^{m}({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}})$.
Under Assumptions $1^\prime$ and 2, a perturbation argument further yields following result on the decay properties of the eigenfunctions and the solutions of the equation $(C + {H_{{ \mathbf{k} }}})u = f$.
Let us assume that $C + {H_{{ \mathbf{k} }}}: {\mathcal K}_{1}^{1}({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}}) \to
{\mathcal K}_{-1}^{-1}({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}})$ is invertible (which is the case if $C > C_0$, with $C_0$ as in Theorem \[theorem1.5\]), then for $|a|$ and $|\epsilon|$ small $$C + {H_{{ \mathbf{k} }}}: e^{\epsilon r} {\mathcal K}_{a+1}^{m+1}({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}}) \to e^{\epsilon r}
{\mathcal K}_{a-1}^{m-1}({\overline{({\mathbb R}^3 \smallsetminus {\mathcal S})}_{rad}})$$ is again invertible.
The proof uses the same arguments used in the proof of Theorem \[theorem1.5\], the continuity of the family $e^{\epsilon r} {H_{{ \mathbf{k} }}}e^{- \epsilon r}$ in $\epsilon$, and the regularity result \[prop.regularity\].
Applications and numerical tests\[ssec3\]
-----------------------------------------
Let $u$ be an eigenvector of ${H_{{ \mathbf{k} }}}$ or the solutions of equations of the form $(\lambda + {H_{{ \mathbf{k} }}})u = f$, with $f$ smooth enough. Our results give smoothness properties for $u$. They also give decay properties of $u$ in the non-periodic case. These properties, in turn, can be used to obtain approximation properties of $u$. Standard numerical methods results (Céa’s lemma or the results reviewed in [@BabuOsborn]) then lead to error estimates in the Finite Element Method for the numerical solutions of the equation $(C + {H_{{ \mathbf{k} }}})u = f$ or for the eigenfunctions of ${H_{{ \mathbf{k} }}}$. We have tested these approximation results in the periodic case using, first, a graded mesh and, second, augmented plane waves. In both cases, the tests are in good agreement with our theoretical results. These numerical and the needed approximation results will be discussed in full detail in the second and fourth parts of our paper [@HLNU2; @HLNU4].
\#1[0=0=0 0 by1pt\#1]{}
[10]{}
S. Agmon. , volume 29 of [*Mathematical Notes*]{}. Princeton University Press, Princeton, NJ, 1982.
B. Ammann, C. Carvalho, and V. Nistor. egularity for eigenfunctions of [S]{}chrödinger operators. Lett. Math. Phys. 2012.
B. Ammann, A. Ionescu, and V. Nistor. Sobolev spaces on [L]{}ie manifolds and regularity for polyhedral domains. , 11:161–206, 2006.
B. Ammann, R. Lauter, and V. Nistor. Pseudodifferential operators on manifolds with a [L]{}ie structure at infinity. , 165(3):717–747, 2007.
D. Arroyo, A. Bespalov, and N. Heuer. On the finite element method for elliptic problems with degenerate and singular coefficients. , 76(258):509–537 (electronic), 2007.
I. Babu[š]{}ka and J. Osborn. Eigenvalue problems. In [*Handbook of numerical analysis, [V]{}ol. [II]{}*]{}, Handb. Numer. Anal., II, pages 641–787. North-Holland, Amsterdam, 1991.
C. Bacuta, V. Nistor, and L. Zikatanov. Improving the rate of convergence of high-order finite elements on polyhedra. [II]{}. [M]{}esh refinements and interpolation. , 28(7-8):775–824, 2007.
S. Bidwell, M. Hassell, and C.R. Westphal. A weighted least squares finite element method for elliptic problems with degenerate and singular coefficients. Math. Comp. (in press).
M. Dauge. , volume 1341 of [*Lecture Notes in Mathematics*]{}. Springer-Verlag, Berlin, 1988. Smoothness and asymptotics of solutions.
V. Felli, A. Ferrero, and S. Terracini. Asymptotic behavior of solutions to [S]{}chrödinger equations near an isolated singularity of the electromagnetic potential. preprint.
V. Felli, A. Ferrero, and S. Terracini. Asymptotic behavior of solutions to [S]{}chrödinger equations near an isolated singularity of the electromagnetic potential. , 13(1):119–174, 2011.
V. Felli, E. Marchini, and S. Terracini. On the behavior of solutions to [S]{}chrödinger equations with dipole type potentials near the singularity. , 21(1):91–119, 2008.
H.-J. Flad, W. Hackbusch, and R. Schneider. Best [$N$]{}-term approximation in electronic structure calculations. [II]{}. [J]{}astrow factors. , 41(2):261–279, 2007.
H.-J. Flad and Harutyunyan. Ellipticity of quantum mechanical hamiltonians in the edge algebra. preprint ArXiv:1103.0207v1.
H.-J. Flad, R. Schneider, and B.-W. Schulze. Asymptotic regularity of solutions to [H]{}artree-[F]{}ock equations with [C]{}oulomb potential. , 31(18):2172–2201, 2008.
S. Fournais, M. Hoffmann-Ostenhof, T. Hoffmann-Ostenhof, and T. [Ø]{}stergaard S[ø]{}rensen. Analytic structure of solutions to multiconfiguration equations. , 42(31):315208, 11, 2009.
F. Gazzola, H.-C. Grunau, and E. Mitidieri. Hardy inequalities with optimal constants and remainder terms. , 356(6):2149–2168 (electronic), 2004.
J. Gil, T. Krainer, and G. Mendoza. Geometry and spectra of closed extensions of elliptic cone operators. , 59(4):742–794, 2007.
J. Gil and G. Mendoza. Adjoints of elliptic cone operators. , 125(2):357–408, 2003.
M. Griebel and J. Hamaekers. Tensor product multiscale many-particle spaces with finite-order weights for the electronic [S]{}chrödinger equation. , 224:527–543, 2010.
D. Grieser. Basics of the [$b$]{}-calculus. In [*Approaches to singular analysis ([B]{}erlin, 1999)*]{}, volume 125 of [*Oper. Theory Adv. Appl.*]{}, pages 30–84. Birkhäuser, Basel, 2001.
T. Hausel, E. Hunsicker, and R. Mazzeo. Hodge cohomology of gravitational instantons. , 122(3):485–548, 2004.
B. Helffer and H. Siedentop. Regularization of atomic [S]{}chrödinger operators with magnetic field. , 218(3):427–437, 1995.
E. Hunsicker, H. Li, V. Nistor, and V. Uski. Analysis of [S]{}chrödinger operators with inverse square potentials [II]{}: [FEM]{} and approximation of eigenfunctions in the periodic case. preprint, submitted for publication.
E. Hunsicker, H. Li, V. Nistor, and V. Uski. Analysis of [S]{}chrödinger operators with inverse square potentials [III]{}: the $n$ dimensional case. ork in progress, tentative title.
E. Hunsicker, H. Li, V. Nistor, and V. Uski. Analysis of [S]{}chrödinger operators with inverse square potentials [IV]{}: plane waves and orbital functions. ork in progress, tentative title.
E. Hunsicker, V. Nistor, and J. Sofo. Analysis of periodic [S]{}chrödinger operators: regularity and approximation of eigenfunctions. , 49(8):083501, 21, 2008.
V. Iftimie, M. M[ă]{}ntoiu, and R. Purice. Magnetic pseudodifferential operators. , 43(3):585–623, 2007.
V. Iftimie and V. Purice. Eigenfunctions decay for magnetic pseudodifferential operators. preprint.
T. Kato. Fundamental properties of [H]{}amiltonian operators of [S]{}chrödinger type. , 70:195–211, 1951.
T. Kato. On the eigenfunctions of many-particle systems in quantum mechanics. , 10:151–177, 1957.
V. A. Kondratev. Boundary value problems for elliptic equations in domains with conical or angular points. , 16:227–313, 1967.
V. Kozlov, V. Mazya, and J. Rossmann. , volume 85 of [*Mathematical Surveys and Monographs*]{}. American Mathematical Society, Providence, RI, 2001.
R. Lauter, B. Monthubert, and V. Nistor. Pseudodifferential analysis on continuous family groupoids. , 5:625–655 (electronic), 2000.
R. Lauter and J. Seiler. Pseudodifferential analysis on manifolds with boundary—a comparison of [$b$]{}-calculus and cone algebra. In [*Approaches to singular analysis ([B]{}erlin, 1999)*]{}, volume 125 of [*Oper. Theory Adv. Appl.*]{}, pages 131–166. Birkhäuser, Basel, 2001.
M. Lesch. , volume 136 of [*Teubner-Texte zur Mathematik \[Teubner Texts in Mathematics\]*]{}. B. G. Teubner Verlagsgesellschaft mbH, Stuttgart, 1997.
H. Li, A. Mazzucato, and V. Nistor. Analysis of the finite element method for transmission/mixed boundary value problems on general polygonal domains. , 37:41–69, 2010.
R. Martin. . Cambridge University Press, 2004.
R.B. Melrose. , 1993.
R.B. Melrose. . Stanford Lectures. Cambridge University Press, Cambridge, 1995.
S. Moroianu and V. Nistor. Index and homology of pseudodifferential operators on manifolds with boundary. In [*Perspectives in operator algebras and mathematical physics*]{}, volume 8 of [*Theta Ser. Adv. Math.*]{}, pages 123–148. Theta, Bucharest, 2008.
S. Moroz and R. Schmidt. Nonrelativistic inverse square potential, scale anomaly, and complex extension. Preprint hep-th/0909.3477v3, 2010.
C. Parenti. Operatori pseudodifferentiali in [${\mathbb R}^n$]{} e applicazioni. , 93:391–406, 1972.
E. Schrohe. , 10(3):237–254, 1992.
B.W. Schulze. , 1998.
C. Schwab and R. Stevenson. Space-time adaptive wavelet methods for parabolic evolution problems. , 78(267):1293–1318, 2009.
A. Vasy. Propagation of singularities in many-body scattering. , 34(3):313–402, 2001.
J. Vazquez and E. Zuazua. The [H]{}ardy inequality and the asymptotic behaviour of the heat equation with an inverse-square potential. , 173(1):103–153, 2000.
H. Wu and D.W.L Sprung. Inverse-square potential and the quantum votex. , 49:4305–4311, 1994.
H. Yserentant. , volume 2000 of [*Lecture Notes in Mathematics*]{}. Springer-Verlag, Berlin, 2010.
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'We propose that the “isolated” planetary mass objects observed by Zapatero Osorio et al in the $\sigma$ Orionis cluster might actually be in orbit around invisible stellar mass companions such as mirror stars. Mirror matter is expected to exist if parity is an unbroken symmetry of nature. Future observations can test this idea by looking for a periodic Doppler shift in the radiation emitted by the planets. The fact that the observations show an inverse dependence between the abundance of the these objects and their mass may argue in favour of the mirror matter hypothesis.'
address: |
School of Physics\
Research Centre for High Energy Physics\
The University of Melbourne\
Victoria 3010 Australia\
foot, sasha, [email protected]
author:
- 'R. Foot, A. Yu. Ignatiev and R. R. Volkas'
title: |
\
Do “isolated” planetary mass objects orbit mirror stars?
---
A variety of observations strongly suggests the presence of a significant amount of dark matter (DM) in the universe. Galactic rotation curves and cluster dynamics cannot be explained using standard Newtonian gravity unless non-luminous but gravitating matter exists. Arguments from Big Bang Nucleosynthesis and theories of large scale structure formation disfavour the simple possibility that all of the DM consists of ordinary baryons. Candidates for the required exotic component in the DM abound: WIMPS, axions and mirror matter are examples. The observation of microlensing events from the Small and Large Magellanic Clouds is consistent with the existence of Massive Compact Halo Objects (MACHOs) in the halo of the Milky Way [@macho]. The inferred average mass is about $0.5 M_{\odot}$, where $M_{\odot}$ is the mass of our sun. The most reasonable conventional identification sees MACHOs as white dwarfs, although there are several strong arguments against this [@freese]. For example, the heavy elements that would have been produced by their progenitors are not in evidence [@freese]. This argues against the conventional white dwarf scenario, and in favour of exotic compact objects. In summary, there is strong evidence for exotic DM which is capable of forming compact stellar mass objects.
Mirror matter [@mirror] is an interesting candidate for some of the required exotic DM [@blinnikov]. It can be independently motivated by the desire to see the full Poincaré Group, including improper transformations (parity and time reversal), as an exact symmetry group of nature. The basic postulate is that every ordinary particle (lepton, quark, photon, etc.) is related by an improper Lorentz transformation with an opposite parity partner (mirror lepton, mirror quark, mirror photon, etc.) of the same mass. Both material particles (leptons and quarks) and force carriers (photons, gluons, $W$ and $Z$ bosons) are doubled. Mirror matter interacts with itself via mirror weak, electromagnetic and strong interactions which have the same form and strength as their ordinary counterparts (except that mirror weak interactions couple to the opposite chirality). Because ordinary matter is known to clump into compact objects such as stars and planets, mirror matter will also form compact mirror stars and mirror planets. Since mirror matter does not feel ordinary electromagnetism, it will be dark. Gravitation, by contrast, is common to both sectors. Mirror matter therefore has the correct qualitative features: it is dark, it clumps, and it gravitates.[^1] Later, we will explain why the observed inverse dependence between the abundance of the these objects and their mass may already argue in favour of the mirror matter hypothesis.
It has been speculated that MACHOs might be mirror stars [@f1], and one of us (RF) has proposed that the observed extrasolar planets [@extrasolar] might be composed of mirror matter [@f2]. The former is well motivated by the aforementioned difficulties in identifying MACHOs as conventional stellar mass objects such as white dwarfs. The latter is motivated by the fact that the detected extrasolar planets are rather massive and orbit very closely to the star, which are surprising characteristics. It is unlikely that ordinary planets of sufficient size could have condensed so close to the stars. If they are composed of ordinary matter, then they probably formed much further from the stars and then migrated in. Another possibility, though, is that they are composed of exotic material such as mirror matter. Because of the weak coupling between ordinary and mirror matter, there is no barrier to mirror planet condensation very close to an ordinary star. This idea should be testable by observing the opacity and albedo properties of the planets[@foot01].
Zapatero Osorio et al. [@zo] have recently presented strong evidence for the existence of “isolated planetary mass objects” in the $\sigma$ Orionis star cluster. These objects are more massive than Jupiter $M_{J}$, but not as massive as brown dwarfs ($\sim 5 - 15 M_J$ although there is some model dependence in the mass determination[@zo]). They appear to be gas giant planets which do not seem to be associated with any visible star. So far, eighteen such objects have been identified. Given that the $\sigma$ Orionis cluster is estimated to be between 1 million and 5 million years old, the formation of these “isolated planets” must have occured within this time scale. Zapatero Osorio et al. argue that these findings pose a challenge to conventional theories of planet formation because standard theories of substellar body formation (as well as new theories inspired by previous claims of isolated planet discovery), are unable to explain the existence of numerous isolated planetary mass objects down to masses $\sim$ few $M_J$. See Ref.[@zo] and references therein for further discussion. It is possible therefore that non-standard particle physics may be required to understand their origin.
We speculate that rather than being isolated, these ordinary matter planets actually orbit invisible mirror stars. These systems could be, in a sense, just the mirror images of those ordinary star systems which have been speculated to feature large Jovian mirror planets in close orbit. Indeed, if there really are mirror planets in orbit around ordinary stars, then it is very natural to also expect mirror solar systems to sometimes contain large ordinary planets.
It should be possible to test this idea by searching for a periodic Doppler shift in spectral lines emanating from these planets. We have that $$\frac{\Delta \lambda}{\lambda} = 2 \frac{v_r}{c},$$ where $\lambda$ is wavelength, $\Delta \lambda$ is the difference between the peak and trough of the periodic Doppler modulation of $\lambda$, $v_r$ is the maximum value of the component of the planet’s orbital velocity in the direction of the Earth, and $c$ is the speed of light. Suppose that a given planet is in a circular orbit of radius $r$ around a mirror star of mass $M$. Let $I$ be the inclination of the plane of the orbit relative to the normal direction defined by the Earth - mirror star line. Then $$v_r = \sqrt{\frac{GM}{r}} \sin I,$$ where $G$ is Newton’s constant. Combining these equations we obtain $$\frac{\Delta \lambda}{\lambda} \simeq
10^{-3} \sqrt{\frac{M}{M_{\odot}}}
\sqrt{\frac{0.04\ A.U.}{r}} \sin I$$ as the level of spectral resolution required. Note that this is a few orders of magnitude larger than the Doppler shifts observed in extrasolar planet detection. However it is certainly true that the isolated planets are much fainter sources of light than the stars whose Doppler shifts have been measured so such a measurement may not be completely straightforward. However, it is worth noting that for the case of close orbiting ordinary planets where $r \sim 0.04\ A. U.$ (analogous to the close-in extra solar planets), the Doppler shift is quite large ($\sim 10^{-3}$) with a period of only a few days which should make this interesting region of parameter space relatively easy to test. Indeed, Zapatero Osorio et al. [@zo] have taken optical and near infrared low resolution spectra of three young isolated planet candidates (S Ori 52, S Ori 56, and S Ori 47). They have obtained absorption lines (at wavelengths $\sim 900$ nm), however their resolution was 1.9 nm[@zo] which is just below that needed to test our hypothesis. The higher resolution required has been achieved in the case of brown dwarfs[@brown] so we anticipate that it should be possible to test our hypothesis in the near future.
One would also expect some ordinary matter to have accumulated in the centre of the mirror star. It is possible, but not inevitable, that this ordinary matter also observably radiates. If so, one would expect this radiation to experience a much smaller Doppler modulation compared to that from the planet. Because the planet and mirror star would not be spatially resolved, one observational signature would be that some of the spectral lines are modulated (those from the planet), while a different set are not (those from the ordinary matter pollutants in the mirror star).
If the mirror star is invisible but opaque, then one would expect to see periodic planetary eclipses for some of these systems (those with $\sin I \simeq 1$). The eclipses should of course occur once per Doppler cycle, around one of the points of zero Doppler shift within a cycle. Obviously, such eclipses (along with the information provided by Doppler shift measurements) will be useful in distinguishing a mirror star from alternatives such as faint white dwarfs or neutron stars. However, it should be mentioned that standard objects such as white dwarfs and neutron stars are extremely unlikely candidates, because the age of the $\sigma$ Orionis cluser is estimated to be only 1 million to 5 million years old, while white dwarfs and neutron stars are typically billions of years old.
Before concluding, we would like to point out an intriguing systematic in both the extrasolar planet and the Zapatero Osorio et al. data that may argue in favour of the mirror matter hypothesis. One envisages a universe that contained some admixture of ordinary and mirror matter from the earliest moments after the Big Bang. Eventually, both the smooth ordinary fluid and the mirror fluid condensed into large scale structures, stars and planets. Because gravitational condensation must be aided by non-gravitational dissipative effects to carry off kinetic energy, one does not expect the ordinary and mirror matter to have condensed in congruent locations, despite their common gravitational interaction. One expects instead a nonzero “segregation scale” $\ell$ to quantify the degree of spatial separation of condensed ordinary and mirror matter clouds or clumps. While we have too little information to theoretically calculate $\ell$, the qualitative expectation is a universe of cells of scale $\ell$, with a given cell being dominated either by ordinary matter or mirror matter. Provided that $\ell$ is much greater than a typical solar system scale, which is in fact observationally required,[^2] then the majority of hybrid ordinary-mirror systems should have disparate components: large ordinary objects with small mirror objects, or the other way around. The ordinary star plus mirror planet systems, and our proposed mirror star plus ordinary planet systems, have exactly this characteristic. Indeed one might expect the number of hybrid systems to increase as a function of the disparity between the components. Intriguingly, the observed extrasolar planets increase in number as their mass decreases. Even more interestingly, the Zapatero Osorio et al. objects also increase in number with decreasing mass: from Fig.2 of Ref.[@zo] we see that there are about as many objects between $8 M_{J}$ and $10 M_{J}$ as there are between $10 M_{J}$ and $20 M_{J}$ (taking 5 million years as the relevant lifetime). We predict, therefore, that an extended search would find greater numbers of these objects at even smaller masses. Of course if the “isolated planets” do orbit mirror stars then this suggests that the star forming region near $\sigma$ Orionis could also be a region of mirror star formation. This is certainly possible and was already envisaged many years ago by Khlopov et al.[@kh] where they argued that large molecular clouds (made of ordinary matter) could merge with large mirror molecular clouds in which case the formation of mixed systems (i.e. containing both ordinary and mirror matter) is enhanced.
In conclusion, we have proposed that the “isolated” planetary mass objects observed by Zapatero Osorio et al. might actually be planets orbiting invisible mirror stars. This idea can be tested by searching for a Doppler modulation at the level of $10^{-3}-10^{-4}$ in amplitude.
[99]{}
C. Alcock et al., MACHO Coll., Ap. J. [**486**]{}, 697 (1997); [**542**]{}, 000 (2000); T. Lasserre et al., EROS Coll., astro-ph/9909505.
See, for example, B. D. Fields, K. Freese, D. S. Graff, Ap. J. [**534**]{}, 265 (2000).
T. D. Lee, C. N. Yang, Phys. Rev. [**104**]{}, 254 (1956); R. Foot, H. Lew, R. R. Volkas, Phys. Lett. B [**272**]{}, 67 (1991).
S. Blinnikov, M. Yu. Khlopov, Sov. Astron. [**27**]{}, 371 (1983); E. W. Kolb, D. Seckel, M. S. Turner, Nature [**514**]{}, 415 (1985); M. Yu. Khlopov et al., Soviet Astronomy. 35, 21 (1991); H. M. Hodges, Phys. Rev. D [**47**]{}, 456 (1993); Z. G. Berezhiani, D. Comelli, F. L. Villante, hep-ph/0008265.
R. Foot, H. Lew, R. R. Volkas, Mod. Phys. Lett. A [**7**]{}, 2567 (1992); R. Foot, Mod. Phys. Lett. A [**9**]{}, 169 (1994); R. Foot, R. R. Volkas, Phys. Rev. D [**52**]{}, 6595 (1995); see also, Z. G. Berezhiani, R. N. Mohapatra, Phys. Rev. D [**52**]{}, 6607 (1995).
R. Foot and R. R. Volkas, Astropart. Phys. [**7**]{}, 283 (1997); Phys. Rev. D[**61**]{}, 043507 (2000).
S. L. Glashow, Phys. Lett. B [**167**]{}, 35 (1986); see also B. Holdom, Phys. Lett. B [**166**]{}, 196 (1986).
R. Foot, S. N. Gninenko, Phys. Lett. B [**480**]{},171 (2000).
Z. Silagadze, Phys.At. Nucl. [**60**]{}, 272 (1997); S. Blinnikov, Talk at “Baryonic Matter in the Universe and its Spectroscopic Studies”, Atami, Japan, 22-24 Nov. 1996, astro-ph/9801015; R. Foot, Phys. Lett. B [**452**]{}, 83 (1999).
For a review and references see the extrasolar planet encyclopaedia, http://cfa-www.harvard.edu/planets/encycl.html.
R. Foot, Phys. Lett. B [**471**]{}, 191 (1999).
R. Foot, astro-ph/0101055.
M. R. Zapatero Osorio et al., Science [**290**]{}, 103 (2000). See also, M. Tamura et al., Science [**282**]{}, 1095 (1998); P. W. Lucas. P. F. Roche, Mon. Not. R. Astron. Soc. [**314**]{}, 858 (2000).
A. Thakrah, H. Jones and M. Hawkins, MNRAS, [**284**]{}, 507 (1997).
A. Yu. Ignatiev, R. R. Volkas, Phys. Rev. D [**62**]{}, 023508 (2000).
M. Yu. Khlopov et al., in Ref. [@blinnikov]
[^1]: There are several other interesting implications of the mirror matter model. In particular, oscillations between ordinary and mirror neutrinos have been proposed as a solution to the solar and atmospheric neutrino problems [@mirrornu]. Ordinary - mirror neutrino oscillations also lead to interesting implications for early Universe cosmology[@fv]. Also photon-mirror photon kinetic mixing leads to potentially observable effects for orthopositronium[@gl] and can resolve the orthopositronium lifetime anomaly[@fg].
[^2]: For instance, one can deduce an upper bound of about $10^{-3}$ for the mirror matter content of the Earth [@ig].
| {
"pile_set_name": "ArXiv"
} |
---
abstract: 'Let $(M,\g)$ be a pseudo-Riemannian manifold. We propose a new approach for defining the conformal Schwarzian derivatives. These derivatives are 1-cocycles on the group of diffeomorphisms of $M$ related to the modules of linear differential operators. As operators, these derivatives do not depend on the rescaling of the metric $\g.$ In particular, if the manifold $(M,\g)$ is conformally flat, these derivatives vanish on the conformal group ${\mathrm{O}}(p+1,q+1),$ where $\mathrm{dim} (M)=p+q.$ This work is a continuation of [@b2; @bo2] where the Schwarzian derivative was defined on a manifold endowed with a projective connection.'
author:
- |
Sofiane BOUARROUDJ[^1]\
[ Department of Mathematics, Keio University, Faculty of Science & Technology]{}\
[3-14-1, Hiyoshi, Kohoku-ku, Yokohama, 223-8522, Japan. ]{}\
[e-mail: [email protected]]{}
title: Conformal Schwarzian derivatives and conformally invariant quantization
---
ł
.1truein \[section\] \[thm\][Lemma]{} \[thm\][Corollary]{} \[thm\][Proposition]{} \[thm\][Example]{} \[thm\][Remark]{} \[thm\][Definition]{}
Introduction
============
Let $S^1$ be the circle identified with the projective line ${\mathbb{RP}}^1.$ For any diffeomorphism $f$ of $S^1,$ the expression $$\label{clas}
S(f):= \frac{f'''(x)}{f'(x)}-\frac{3}{2} \left (\frac{f''(x)}{f'(x)}\right )^2,$$ where $x$ is an affine parameter on $S^1,$ is called Schwarzian derivative (see [@cara]).\
The Schwarzian derivative has the following properties:
\(i) It defines a 1-cocycle on the group of diffeomorphisms ${\mathrm{Diff}}(S^1)$ with values in differential quadratics (cf. [@ki; @s]).
\(ii) Its kernel is the group of projective transformations ${\mathrm{PSL}}_2({\mathbb{R}})$.\
The aim of this paper is to propose a new approach for constructing the multi-dimensional conformal Schwarzian derivative. This approach was recently used in [@b2; @bo2] to introduce the multi-dimensional “projective” Schwarzian derivative. The starting point of our approach is the relation between the Schwarzian derivative (\[clas\]) and the space of Sturm-Liouville operators (see, e.g., [@wi]). The space of Sturm-Liouville operators is not isomorphic as a ${\mathrm{Diff}}(S^1)$-module to the space of differential quadratics. More precisely, the space of Sturm-Liouville operators is a non-trivial deformation of the space of differential quadratics in the sense of Neijenhuis and Richardson’s theory of deformation (see [@nr]), generated by the 1-cocycle (\[clas\]) (see [@ga] for more details). From this point of view, the multi-dimensional Schwarzian derivative is closely related to the modules of linear differential operators. To set out our approach, let us introduce some notation.\
Let $M$ be a smooth manifold. We consider the space of linear differential operators with arguments that are $\l$-densities on $M$ and values that are $\mu$-densities on $M.$ We have, therefore, a two parameter family of ${\mathrm{Diff}}(M)$-modules denoted by ${{\mathcal{D}}}_{\l,\mu}(M).$ The corresponding space of symbols is the space ${{\mathcal{S}}}_{\d}(M)$ of fiberwise polynomials on $T^*M$ with values in $\d$-densities, where $\d=\mu-\l.$ In general, the space ${{\mathcal{D}}}_{\l,\mu}(M)$ is not isomorphic as a ${\mathrm{Diff}}(M)$-module to the space ${{\mathcal{S}}}_{\d}(M)$ (cf. [@do; @lo2]). However, we are interested in the following two cases:\
(i) If $M:={\mathbb{R}}^n$ is endowed with a flat projective structure (i.e. local action of the group ${\mathrm{SL}}_{n+1}({\mathbb{R}})$ by linear fractional transformations) there exists an isomorphism between ${{\mathcal{D}}}_{\l,\mu}({\mathbb{R}}^n)$ and ${{\mathcal{S}}}_{\d}({\mathbb{R}}^n),$ for $\d$ generic, intertwining the action of ${\mathrm{SL}}_{n+1}({\mathbb{R}})$ (cf. [@lo2]). The multi-dimensional “projective” Schwarzian derivative was defined in [@b2; @bo2] as an obstruction to extend this isomorphism to the full group ${\mathrm{Diff}}({\mathbb{R}}^n).$\
(ii) If $M:={\mathbb{R}}^n$ is endowed with a flat conformal structure (i.e. local action of the conformal group ${\mathrm{O}}(p+1,q+1),$ where $p+q=n$), there exists an isomorphism between ${{\mathcal{D}}}_{\l,\mu}({\mathbb{R}}^n)$ and ${{\mathcal{S}}}_{\d}({\mathbb{R}}^n),$ for $\d$ generic, intertwining the action of ${\mathrm{O}}(p+1,q+1)$ (cf. [@dlo; @do]). In this paper we introduce the multi-dimensional “conformal” Schwarzian derivative in this context. Recall that in the one-dimensional case these two notions coincide in the sense that the conformal Lie algebra ${\mathrm o}(2,1)$ is isomorphic to the projective Lie algebra ${\mathrm{sl}}_2({\mathbb{R}}).$
Differential operators and symbols
==================================
Let $(M,\g)$ be a pseudo-Riemannian manifold of dimension $n.$ We denote by $\Gamma$ the Levi-Civita connection associated with the metric $\g.$
Space of linear differential operators as a module
--------------------------------------------------
We denote the space of tensor densities on $M$ by ${{\mathcal{F}}}_{\l}(M),$ or ${{\mathcal{F}}}_{\l}$ for simplify. This space is nothing but the space of sections of the line bundle $(\wedge^n T^* M)^{\otimes \l}.$ One can define in a natural way a ${\mathrm{Diff}}(M)$-module structure on it: for $f\in {\mathrm{Diff}}(M)$ and $\phi \in{\cal F}_{\l},$ in a local coordinates $(x^i)$, the action is given by $$\label{den}
f^*\phi=\phi\circ f^{-1}\cdot ( {J_{f^{-1}}})^{\lambda},$$ where $J_f=\left |Df/Dx \right |$ is the Jacobian of $f$.\
By differentiating this action, one can obtain the action of the Lie algebra of vector fields ${\mathrm{Vect}}(M).$
[ ${{\mathcal{F}}}_0= C^{\infty}(M), \, {{\mathcal{F}}}_1= \Omega^{n}(M)$ (space of differential $n$-forms). ]{}
Let us recall the definition of a covariant derivative on densities. If $\phi \in {{\mathcal{F}}}_\l,$ then $\nabla \phi \in \Omega^1 (M)\otimes {{\mathcal{F}}}_\l$ given in a local coordinates by $$\nabla_i \phi= \partial_i \phi-\l \Gamma_i \phi,$$ with $\Gamma_i= \Gamma_{ti}^t.$ (Here and bellow summation is understood over repeated indices).
Consider now ${\cal D}_{\l,\mu}(M),$ the space of linear differential operators acting on tensor densities $$A:{{\mathcal{F}}}_\l\to{{\mathcal{F}}}_\m.\nonumber
\label{Conv}$$ The action of ${\mathrm{Diff}}(M)$ on ${{\mathcal{D}}}_{\l,\mu}(M)$ depends on the two parameters $\l$ and $\m$. This action is given by the equation $$f_{\l,\m}(A)=f^*\circ A\circ {f^*}^{-1},
\label{Opaction}$$ where $f^*$ is the action (\[den\]) of ${\mathrm{Diff}}(M)$ on ${{\mathcal{F}}}_\l$.
By differentiating this action, one can obtain the action of the Lie algebra ${\mathrm{Vect}}(M).$
The formul[æ]{} (\[den\]) and (\[Opaction\]) do not depend on the choice of the system of coordinates.
Denote by ${\cal D}^2_{\l,\mu}(M)$ the space of second-order linear differential operators with the ${\mathrm{Diff}}(M)$-module structure given by (\[Opaction\]). The space ${\cal D}_{\l,\mu}^2(M)$ is in fact a ${\mathrm{Diff}}(M)$-submodule of ${{\mathcal{D}}}_{\l,\m}(M).$
[The space of Sturm-Liouville operators $\frac{d^2}{dx^2}+u(x): {{\mathcal{F}}}_{-1/2}\rightarrow {{\mathcal{F}}}_{3/2}$ on $S^1,$ where $u(x)\in
{{\mathcal{F}}}_{2}$ is the potential, is a submodule of ${{\mathcal{D}}}_{-\frac{1}{2},\frac{3}{2}}^2(S^1)$ (see [@wi]). ]{}
The module of symbols
---------------------
The space of symbols, ${\mathrm{Pol}}(T^*M),$ is the space of functions on the cotangent bundle $T^*M$ that are polynomials on the fibers. This space is naturally isomorphic to the space ${\cal S}(M)$ of symmetric contravariant tensor fields on $M.$ In local coordinates $(x^i,\xi_i)$, one can write $P\in {\cal S}(M)$ in the form $$P=\sum_{l\geq 0}P^{i_1, \ldots,i_l}\xi_{i_1}\cdots \xi_{i_l},$$ with $P^{i_1, \ldots,i_l}(x)\in C^{\infty}(M).$\
We define a one parameter family of ${\mathrm{Diff}}(M)-$module on the space of symbols by $${\cal S}_{\d}(M):={\cal S}(M)\otimes {{\mathcal{F}}}_{\d}.$$ For $f\in {\mathrm{Diff}}(M)$ and $P\in {\cal S}_{\d}(M),$ in a local coordinate $(x^i)$, the action is defined by $$\begin{aligned}
\label{actsym}
f_{\d}(P)&=& f^*P\cdot (J_{f^{-1}})^{\d},\end{aligned}$$ where $J_f=|Df/Dx|$ is the Jacobian of $f,$ and $f^*$ is the natural action of ${\mathrm{Diff}}(M)$ on ${\cal S}(M).$
We then have a graduation of ${\mathrm{Diff}}(M)$-modules given by $${\cal S}_\d(M)=\bigoplus_{k=0}^\infty {\cal S}_\d^k(M),$$ where $ S_\d^k(M)$ is the space of contravariant tensor fields of degree $k$ endowed with the ${\mathrm{Diff}}(M)$-module structure (\[actsym\]).
We want to study the space of contravariant tensor fields of degree less than two, denoted by ${\cal S}_{\d,2}(M)$ (i.e. ${\cal S}_{\d,2}(M):={\cal S}_{\d}^2(M)\oplus
{\cal S}_{\d}^1(M)\oplus {\cal S}_{\d}^0(M)$).
Conformal Schwarzian derivatives
================================
Let $(M,\g)$ be a pseudo-Riemannian manifold. Denote by $\Gamma$ the Levi-Civita connection associated with the metric $\g.$
Main definition {#enfant}
---------------
It is well known that the difference between two connections is a well-defined tensor field of type $(2,1).$ It follows therefore that the difference $$\label{ell}
\ell(f):=f^*\Gamma-\Gamma,$$ where $f\in {\mathrm{Diff}}(M),$ is a well-defined $(2,1)$-tensor field on $M$.
It is easy to see that the map $$f\mapsto \ell(f^{-1})$$ defines a non-trivial 1-cocycle on ${\mathrm{Diff}}(M)$ with values in the space of tensor fields on $M$ of type $(2,1).$
Our first main definition is the linear differential operator ${\cal A}(f)$ acting from ${\cal S}_{\d}^2(M)$ to ${\cal S}_\d^1(M)$ defined by $${\cal A} (f)_{ij}^k:=
{f^*}^{-1} \left( \g^{sk}\,\g_{ij}\nabla_s \right)
-\g^{sk}\,\g_{ij}\nabla_s + c\,\left(\ell(f)^k_{ij} -\frac{1}{n}
\, \mathrm{Sym}_{i,j}\,\delta_{i}^k \,\ell(f)_{j}\right),
\label{MultiSchwar1}$$ where $$\label{qua}
c=2-\delta n,$$ and $\ell(f)_{ij}^k$ are the components of the tensor (\[ell\])
\[main\] [(i)]{} For all $\d \not= 2/n,$ the map $f\mapsto {\cal A}(f^{-1})$ defines a non-trivial 1-cocycle on ${\mathrm{Diff}}(M)$ with values in ${{\mathcal{D}}}({\cal S}_{\d}^2(M), {\cal S}_\d^1(M)).$\
[(ii)]{} The operator (\[MultiSchwar1\]) does not depend on the rescaling of the metric $\g.$ In particular, if $M:={\mathbb{R}}^n$ and $\g$ is the flat metric of signature $p-q$, this operator vanishes on the conformal group ${\mathrm{O}}(p+1,q+1).$
[**Proof.**]{} To prove (i) we have to verify the 1-cocycle condition $${\cal A}(f\circ h)={h^*}^{-1} {\cal A}(f)+ {\cal A}(h),
\quad \mbox{for all } f,h\in {\mathrm{Diff}}(M),$$ where $h^*$ is the natural action on ${{\mathcal{D}}}({\cal S}^2_{\d}(M),{\cal S}_\d^1(M)).$ This condition holds because the first part of the operator (\[MultiSchwar1\]) is a coboundary and the second part is a 1-cocycle.
Let us proof that this 1-cocycle is not trivial for $\d\not=2/n$. Suppose that there is a first-order differential operator $A^k_{ij}=u^{sk}_{ij}\nabla_s+v^k_{ij}$ such that $$\label{cn}
{\cal A}(f)={f^*}^{-1} A-A.$$ From (\[cn\]), it is easy to see that ${f^*}^{-1} v^k_{ij}-v^k_{ij}=
(2-\delta n)\left(\ell(f)^k_{ij} -\frac{1}{n}
\, \mathrm{Sym}_{i,j}\,\delta_{i}^k \,\ell(f)_{j}\right)$. The right-hand side of this equation depends on the second jet of the diffeomorphism $f,$ while the left-hand side depends on the first jet of $f,$ which is absurd.
For $\d=2/n,$ one can easily see that the 1-cocycle (\[MultiSchwar1\]) is a coboundary.
Let us prove (ii). Consider a metric $\tilde \g= F\cdot\g,$ where $F$ is a non-zero positive function. Denote by $\tilde {\cal A}(f)$ the operator (\[MultiSchwar1\]) written with the metric $\tilde \g.$ We have to prove that $\tilde {\cal A}(f)={\cal A}(f).$ The Levi-Civita connections associated with the metrics $\g$ and $\tilde \g$ are related by $$\label{lien}
\tilde \Gamma^k_{ij}=\Gamma^k_{ij}+\frac{1}{2F}\left (F_i \d^k_j +F_j \d^k_i
-F_t \,\g^{tk} \g_{ij}\right),$$ where $F_i=\partial_i F.$\
We need some formul[æ]{}: denote by $\ell(f)$ the tensor (\[ell\]) written with the connection $\tilde \g,$ then we have $$\begin{aligned}
\label{bebe}
\tilde \nabla_k P^{ij}&=&\nabla_k P^{ij}+\frac{1}{2F}\left( \mathrm{Sym}_{i,j}
P^{mi} \left(F_m \delta^j_k-F_t \,\g^{tj}\g_{km}\right) +(2-n\delta )\,
P^{ij}F_k\right),\\
\tilde \ell(f)^k_{ij}&=&\ell(f)^k_{ij}+\frac{1}{2 F\circ f}
\left(\mathrm{Sym}_{i,j} \stackrel{*}{F_i}\delta^k_j -\stackrel{*}{F}_t\,
\stackrel{*}{\g}^{tk}\stackrel{*}{\g}_{ij}\right)
-\frac{1}{2F}\left( \mathrm{Sym}_{i,j} \,F_i\, \delta^k_j -F_t\,
\g^{tk}\g_{ij}\right),
\nonumber\end{aligned}$$ where $\stackrel{*}{F}_i={f^{*}}^{-1}F_i$ and $\stackrel{*}{\g}_{ij}=
{f^{*}}^{-1}\g_{ij}$ for all $P^{ij}\xi_i \xi_j\in {\cal S}_{\d}^2 (M).$\
By substituting the formul[æ]{} (\[bebe\]) into (\[MultiSchwar1\]) we get $${\cal A} (f)_{ij}^k=\tilde {\cal A} (f)_{ij}^k+
\frac{1}{2 F} (\delta n +c -2 ) \,{\g}^{sk} \,\g_{ij}\, F_s\,
+
\frac{1} {2 F\circ f}(2-c -\delta n)
\stackrel{*}{\g}^{tk} \stackrel{*}{\g}_{ij}\, \stackrel{*}{F}_t \, \,
\cdot$$ We see that ${\cal A} (f)=\tilde {\cal A} (f)$ if and only if $c=2-\delta n .$
Let us prove that the operator (\[MultiSchwar1\]) vanishes on the conformal group ${\mathrm{O}}(p+1,q+1)$ in the case when $M:={\mathbb{R}}^n$ is endowed with the flat metric $\g_0:=\mathrm{diag}(1,\ldots,1,-1,\ldots,-1)$ whose trace is $p-q.$ Any $f\in {\mathrm{O}}(p+1,q+1)$ satisfies ${{f^*}}^{-1} {\g_0}=F\cdot{\g_0},$ where $F$ is a non-zero positive function. This relation implies $$\begin{aligned}
2 \ell(f)^k_{ij} +\mathrm{Sym}_{i,j}\,\ell(f)^s_{it}\,\g_0^{tk}\,
{\g_0}_{sj}&=&\frac{1}{F}\mathrm{Sym}_{i,j}\, \partial_i F\, \delta^k_j,
\nonumber \\
\mathrm{Sym}_{i,j}\, \ell(f)^s_{li} \, {\g_0}_{sj}&=&\frac{\partial_l F}{F}
{\g_0}_{ij},\nonumber \\
\ell(f)_t&=& \frac{n}{2}\frac{\partial_t F}{F}\cdot
\nonumber\end{aligned}$$ Sibstitute these formul[æ]{} into (\[MultiSchwar1\]). Then we obtain by straightforward computation that ${\cal A}(f)\equiv 0.$\
------------------------------------------------------------------------
Suppose now that ${\rm dim }(M)>2.$
Our second main definition is the linear differential operator ${\cal B}(f)$ acting from ${\cal S}^2_{\d}(M)$ to ${\cal S}^0_\d(M)$ defined by $$\begin{aligned}
{\cal B}(f)_{ij}&=&
{f^*}^{-1} \left(\g^{st}\, \g_{ij}\nabla_s\nabla_t \right ) -\g^{st}\,\g_{ij}
\nabla_s \nabla_t
+c_1\,\left( \ell(f)_{ij}^s -\frac{1}{n}\,
\mathrm{Sym}_{i,j}\,\delta_{i}^s \,\ell(f)_{j} \right)\nabla_s \nonumber \\
&&+\,c_2 \, \ell(f)_i\,\ell(f)_j +c_3 \n_{s}
\left ( \ell(f)_{ij}^s-\frac{1}{n}
\,\mathrm{Sym}_{i,j}\, \d_{i}^s \ell_{j}(f)\right) +c_4\, \ell(f)_{ij}^s\ell(f)_s
\nonumber \\
&& +\,c_5\, \ell(f)_{si}^u\ell(f)_{uj}^s + c_6\left(
{f^{-1}}^*(R\,\g_{ij})-R\,\g_{ij}\right ),
\label{MultiSchwar2}\end{aligned}$$ where $\ell(f)$ is the tensor (\[ell\]), $R$ is the scalar curvature of the metric $\g,$ and the constants $c_1,\ldots,c_6,$ are given by $$\begin{array}{ll}
c_1=2+n(1-2\delta ), & \displaystyle c_2=\frac{(2+n(1-2\delta ))(\d- 1)}{n},
\nonumber \\[3mm]
c_3=\displaystyle \frac{(2+n(1-2\delta ))(\d n-2)}{n-2} , &c_4=
\displaystyle \frac{(2+n(1-2\delta ))(2\d-2)}{n-2}, \\[3mm]
c_5= \displaystyle \frac{(2+n(1-2\delta ))(1-\d)n}{n-2},& c_6=
\displaystyle \frac{n(\d -1)(n\d -2)}{(n-1)(n-2)}\cdot
\nonumber
\label{ga}
\end{array}$$
\[mainp\] [(i)]{} For all $\d \not= \frac{n+2}{2n},$ the map $f\mapsto {\cal B}(f^{-1})$ defines a non-trivial 1-cocycle on ${\mathrm{Diff}}(M)$ with values in ${{\mathcal{D}}}({\cal S}^2_{\d}(M),{\cal S}^0_\d(M))$.
[(ii)]{} The operator (\[MultiSchwar2\]) does not depend on the rescaling of the metric $\g.$ In the flat case, this operator vanishes on the conformal group ${\mathrm{O}}(p+1,q+1).$
[**Proof.**]{} To prove that the map $f\mapsto {\cal B}_{ij}(f^{-1})$ is a 1-cocycle, one has to verify the 1-cocycle condition $${\cal B}(f\circ h)=
{h^*}^{-1}{\cal B}(f)+{\cal B}(h), \quad \mbox{for all }f,h\in {\mathrm{Diff}}(M),$$ where $h^*$ is the natural action on ${{\mathcal{D}}}({\cal S}^2_{\d}(M),{\cal S}_\d^0(M)).$ To do this, we use the formul[æ]{} $$\begin{aligned}
\label{comp}
\n_i\, f^*_{\d} P^{kl}&=&f^*_{\d}\n_i P^{kl}-\mathrm{Sym}_{k,l} \left(
\ell(f^{-1})_{it}^k\, f^*_{\d} \, P^{tl}\right) +\d \,\ell(f^{-1})_i\,
f^*_{\d} P^{kl},\\[2mm]
\n_u h^* \ell(f)_{ij}^k&=&h^*\n_u \ell(f)_{ij}^k - h^*\ell(f)_{ij}^t\,
\ell(h^{-1})_{ut}^k + \mathrm{Sym}_{i,j}\left( h^*\ell(f)_{it}^k \,
\ell(h^{-1})_{ju}^t \right), \nonumber \end{aligned}$$ for all $f,h \in {\mathrm{Diff}}(M)$ and for all $P^{kl}\xi_k\xi_l\in {{\mathcal{S}}}_{\d}^2(M).$\
Let us prove that this 1-cocycle is not trivial. Suppose that there exists an operator $B_{ij}:= u_{ij}^{st}\nabla_s \nabla_t+ v_{ij}^s\nabla_s +t_{ij}$ such that $${\cal B}(f)={f^*}^{-1} B -B.$$ It is easy to see that ${f^*}^{-1} v_{ij}^s- v_{ij}^s=
(2+n(1-2\delta))\left( \ell(f)_{ij}^s -\frac{1}{n}\,
Sym_{i,j}\,\delta_{i}^s \,\ell(f)_{j} \right).$ The right-hand side of this relation depends on the second jet of $f,$ while the the left-hand side depends on the first jet of $f,$ which is absurd.
For $\d= \frac{n+2}{2n},$ the 1-cocycle (\[MultiSchwar2\]) is trivial: ${\cal B}(f)_{ij}={f^*}^{-1} (\g_{ij}B) -B\,\g_{ij},$ where $B:= \g^{st}\nabla_s\nabla_t -\frac{1}{4}\frac{n-2}{n-1}R$ is the so-called Yamabe-Laplace operator (see, e.g., [@besse]).\
Let us prove (ii). Consider a metric $\tilde \g:=F\cdot \g,$ where $F$ is a non-zero positive function. Denote by $\tilde {\cal B}(f)$ the operator (\[MultiSchwar2\]) written with the metric $\tilde \g.$ We have to prove that $\tilde {\cal B}(f)={\cal B}(f).$\
The proof is similar to the proof of part (ii) of Theorem (\[main\]), by means of the equation (\[lien\]), (\[bebe\]) and $$\begin{aligned}
\tilde \nabla_l \tilde \nabla_k P^{ij}&=&\nabla_l \tilde \nabla_k P^{ij}+
\frac{1}{2F} \left( (1-\d n) F_l\,\tilde\nabla_k P^{ij}- F_k\,
\tilde \nabla_l P^{ij}+\g^{st}\, \g_{lk}\, F_s \tilde
\nabla_t P^{ij}\right ) \nonumber \\
&& +\frac{1}{2F} \mathrm{Sym}_{i,j} \tilde \nabla_k P^{mi} \left (F_m \, \delta^j_l -
{\g}^{sj}{\g}_{ml}\, F_s\right),\nonumber\\
\tilde \nabla_l \, \tilde \ell(f)^k_{ij}&=&\nabla_l \,\tilde \ell(f)^k_{ij} -
\frac{1}{2F}\, F_l\,\,\tilde \ell(f)^k_{ij}+\frac{1}{2F} \left( F_t \,
\delta^k_l -\g^{sk}\, \g_{tl}\, F_s \right ) \tilde \ell(f)^t_{ij}
\nonumber \\
&&-\frac{1}{2F} \mathrm{Sym}_{i,j} \left (F_i \delta^s_l-F_m \,{\g}^{ms}
{\g}_{il}\right)\tilde \ell(f)^k_{js}\nonumber\\
\tilde R&=& \frac{1}{F} \left ( R-(n-1)\frac{1}{F}
\left( \g^{ij}\nabla_i F_j+ (n-6)\frac{1}{4 \, F}\g^{ij}F_i
F_j \right) \right) \end{aligned}$$ for all $P^{ij}\xi_i\xi_j\in {\cal S}_\d^2(M),$ where $\tilde \nabla,$ $\tilde \ell(f)$ and $\tilde R$ are, the covariant derivative, the tensor (\[ell\]), and the scalar curvature associated with the metric $\tilde \g,$ respectively.
Cohomology of ${\mathrm{Vect}}(M)$ and Schwarzian derivatives
-------------------------------------------------------------
We will give here the infinitesimal 1-cocycle associated with the 1-cocycles ${\cal A}$ and ${\cal B}$. First, let us recall the notion of a Lie derivative of a connection. For each $X \in {\mathrm{Vect}}(M),$ the Lie derivative $$\label{mou}
L_X \nabla :=(Y,Z)\mapsto [X,\n_YZ]-\n_{[X,Y]} Z- \n_{Y} [X,Z]$$ of $\n$ is a well-known symmetric $(2,1)$-tensor field. The map $$X\mapsto L_X {\nabla}$$ defines a 1-cocycle on ${\mathrm{Vect}}(M)$ with values in the space of symmetric $(2,1)$-tensor fields on $M.$\
The linear differential operator ${\mathfrak a}$ defined by $${\mathfrak a}^k_{ij}(X):=L_X \left
(\g^{sk}\, \g_{ij}\, \nabla_s\right)
+c \left ((L_X \nabla)^k_{ij}
-\frac{1}{n}\mathrm{Sym}_{i,j}\delta_j^k \, (L_X \nabla)_i\right),$$ where the constant $c$ is as in (\[qua\]) and $L_X \n$ is the tensor (\[mou\]), acts from ${{\mathcal{S}}}_{\d}^2(M)$ to ${{\mathcal{S}}}_{\d}^1(M).$ The linear differential operator ${\mathfrak b}$ defined by $$\begin{aligned}
\label{chr}
{\mathfrak b}_{ij}(X)&:=&L_X \left (\g^{st}\, \g_{ij}\, \nabla_s\nabla_t \right)
+c_1 \left ((L_X \nabla)^k_{ij}
-\frac{1}{n}\mathrm{Sym}_{i,j}\delta_j^k \, (L_X \nabla_i)\right)
\nabla_k\nonumber\\
&& +c_2 \nabla_k \left ( (L_X \nabla)^k_{ij}
-\frac{1}{n}\mathrm{Sym}_{i,j}\delta_j^k \, (L_X \nabla)_i \right )
+c_6 L_X \left ( R \, \g_{ij}\right ),\end{aligned}$$ where the constants $c_1,c_2$ and $c_6$ are as in (\[ga\]) and $L_X(\n)$ is the tensor (\[mou\]), acts from ${{\mathcal{S}}}_{\d}^2(M)$ to ${{\mathcal{S}}}_{\d}^0(M).$
The following two propositions follow by straightforward computation.
[(i)]{} The map $X\mapsto {\mathfrak a}^k_{ij}(X)$ defines a 1-cocycle on ${\mathrm{Vect}}(M)$ with values in ${{\mathcal{D}}}({{\mathcal{S}}}_{\d}^2(M), {{\mathcal{S}}}_{\d}^1(M)).$
[(ii)]{} The operator $\mathfrak a$ does not depend on the rescaling of the metric. In the flat case, it vanishes on the Lie algebra $\mathrm{o}(p+1,q+1),$ where $p+q=n.$
[(i)]{} The map $X\mapsto {\mathfrak b}_{ij}(X)$ defines a 1-cocycle on ${\mathrm{Vect}}(M)$ with values in ${{\mathcal{D}}}({{\mathcal{S}}}_{\d}^2(M), {{\mathcal{S}}}_{\d}^0(M)).$
[(ii)]{} The operator $\mathfrak b$ does not depend on the rescaling of the metric. In the flat case, it vanishes on the Lie algebra $\mathrm{o}(p+1,q+1),$ where $p+q=n.$
In section (\[va\]), we will show that the space ${{\mathcal{D}}}_{\l,\mu}^2(M)$ can be viewed as a non-trivial deformation of the module ${\cal S}_{2,\d}(M)$ in the sense of Neijenhuis and Richardson’s theory of deformation (see also [@do; @lo1]). According to the theory of deformation, the problem of “infinitesimal” deformation is related to the cohomology group $$\label{hic}
\mathrm H^1 ({\mathrm{Vect}}(M), {\mathrm{End}}({\cal S}_{2,\d}(M))\cdot$$ To compute the cohomology group (\[hic\]) we restrict the coefficients to the space of linear differential operators on ${\cal S}_{2,\d}(M),$ denoted by ${{\mathcal{D}}}({\cal S}_{2,\d}(M)).$ This space is decomposed, as a ${\mathrm{Vect}}(M)$-module, into the direct sum $${{\mathcal{D}}}({\cal S}_{2,\d}(M))=\bigoplus_{k,m=0}^{2}{{\mathcal{D}}}({\cal S}_{\d}^k(M),
{\cal S}_{\d}^m(M)),$$ where ${{\mathcal{D}}}({\cal S}_{\d}^k(M), {\cal S}_{\d}^m(M))\subset
{\mathrm{Hom}}({\cal S}_{\d}^k(M), {\cal S}_{\d}^m(M)).$
The relation between the Schwarzian derivative (\[clas\]) and the cohomology group above is as follows: recall that in the one dimensional case the space ${{\mathcal{S}}}^k_\d (S^1)$ is nothing but ${{\mathcal{F}}}_{\d-k}.$ In this case, the problem of deformation with respect to the Lie algebra ${\mathrm{sl}}_2({\mathbb{R}})$ is related to the cohomology group $$\label{ser}
{\mathrm H}^1({\mathrm{Vect}}(S^1), {\mathrm{sl}}_2 ({\mathbb{R}}); {{\mathcal{D}}}({{\mathcal{F}}}_{\d-k}, {{\mathcal{F}}}_{\d-l})),$$ where $k,l=0,1,2.$ The cohomology group (\[ser\]) was calculated in [@bo1], it is one dimension for $k=2,\, l=0,$ and zero otherwise. The (unique) non-trivial class, for $k=2$ and$l=0,$ can be integrated to the group of diffeomorphisms ${\mathrm{Diff}}(S^1);$ it is a zero-order operator given as a multiplication by the Schwarzian derivative (\[clas\]) (see [@bo1] for more details).
In the multi-dimensional case and for $\d=0,$ the first group of differential cohomology of ${\mathrm{Vect}}(M)$, with coefficients in the space ${{\mathcal{D}}}({\cal S}^k(M),{\cal S}^m(M))$ of linear differential operators from ${\cal S}^k(M)$ to ${\cal S}^m(M)$ was calculated in [@lo2]. For $n\geq2$ the result is as follows $$\label{cal}
\mathrm H^1({\mathrm{Vect}}(M), {{\mathcal{D}}}({\cal S}^k(M),{\cal S}^m(M)))=
\left\{
\begin{array}{ll}
{\mathbb{R}}\oplus \mathrm{H}_{\mathrm{DR}}^1(M), & \mbox{if}\quad k-m=0,\\
{\mathbb{R}},& \mbox{if} \quad k-m=1,m\not=0,\\
{\mathbb{R}},& \mbox{if} \quad k-m=2,\\
0,&\hbox{otherwise.}
\end{array}
\right.$$ We believe, by analogy for the one-dimensional case, that the “infinitesimal” multi-dimensional Schwarzian derivative is a cohomology class in the cohomology group above for $k-m=2.$ This class is nothing but the operator ${\mathfrak b}$ defined in (\[chr\]).
Comparison with the projective case {#connex}
------------------------------------
Let $M$ be a manifold of dimension $n.$ Fix a symmetric affine connection $\Gamma$ on $M$ (here $\Gamma$ is any connection not necessarily a Levi-Civita one). Let us recall the notion of projective connection (see [@kn]).
A projective connection is an equivalent class of symmetric affine connections giving the same unparameterized geodesics.
Following [@kn], the symbol of the projective connection is given by the expression $$\label{connection}
\Pi_{ij}^k=\G_{ij}^k-\frac{1}{n+1}\left (\delta_i^k \G_{j}+\delta_j^k \G_{i}
\right ),$$ where $\Gamma_{ij}^k$ are the Christoffel symbols of the connection $\Gamma$ and $\Gamma_i=\Gamma_{ij}^j.$\
Two affine connection $\Gamma$ and $\tilde \Gamma$ are projectively equivalent if the corresponding symbols (\[connection\]) coincide.\
A projective connection on $M$ is called *flat* if in a neighborhood of each point there exists a local coordinate system $(x^1,\ldots,x^n)$ such that the symbols $\Pi_{ij}^k$ are identically zero (see [@kn] for a geometric definition). Every flat projective connection defines a projective structure on $M$.\
Let $\Pi$ and $\tilde \Pi$ be two projective connections on $M.$ The difference $\Pi-\tilde \Pi$ is a well-defined $(2,1)$-tensor field. Therefore, it is clear that a projective connection on $M$ leads to the following 1-cocycle on ${\mathrm{Diff}}(M)$: $${\cal C}(f^{-1})=
\left(
(f^{-1})^*\Pi_{ij}^k-\Pi_{ij}^k
\right)
dx^i\otimes dx^j\otimes\frac{\partial}{\partial x^k}
\label{ellp}$$ This formula is independent on the choice of the coordinate system.
By definition, the tensor field (\[ellp\]) depends only on the projective class of the connection $M.$ In particular if $\Pi \equiv 0,$ this tensor field vanishes on the projective group ${\mathrm{PSL}}_{n+1}({\mathbb{R}}).$
One can define a 1-cocycle on ${\mathrm{Diff}}(M)$ with values in ${{\mathcal{D}}}({{\mathcal{S}}}^{2}_\d(M), {{\mathcal{S}}}^{1}_\d(M))$ by contracting any symmetric contravariant tensor field with the tensor (\[ellp\]). Therefore, the operator (\[MultiSchwar1\]) can be viewed as the conformal analogue of the tensor field (\[ellp\]). In the same spirit, the operator (\[MultiSchwar2\]) can be viewed as the conformal analogue of the “projective” multi-dimensional Schwarzian derivative introduced in [@b2; @bo2].
Relation to the modules of differential operators {#SecComput}
=================================================
Conformally equivariant quantization {#tokyo}
------------------------------------
The quantization procedure explained in this paper was first introduced in [@do; @lo1]. By an equivariant quantization we mean an identification between the space of linear differential operators and the corresponding space of symbols, equivariant with respect to the action of a (finite dimension) sub-group $G\subset {\mathrm{Diff}}({\mathbb{R}}^n).$ Recall that in the one-dimensional case the equivariant quantization process was carried out for $G={\mathrm{SL}}_2({\mathbb{R}})$ in [@cmz] (see also [@ga]).
The following theorems are proven in [@do].
For all $\d \not=1,$ there exists an isomorphism $${\cal Q}_{\l,\mu}: {\cal S}_{\d}^1(M)\oplus {\cal S}_{\d}^0(M)
\rightarrow {{\mathcal{D}}}_{\l,\mu}^1(M),$$ given as follows: for all $P=P^i\xi_i+P_0\in~{\cal S}_{\d}^1(M)
\oplus {\cal S}_{\d}^0(M),$ one can associate a linear differential operator given by $$\begin{aligned}
\label{yah1}
{\cal Q}_{\l,\mu} (P)&=&
P^{i}\n_i +\a \n_i P^{i}+P_0,\end{aligned}$$ where $$\alpha=\displaystyle \frac{\l}{1-\delta}\cdot$$ This map does not depend on the rescaling of the metric, intertwines the action of ${\mathrm{Diff}}(M).$
For $n>2$ and for all $\d \not=\frac{2}{n},
\frac{n+2}{2n}, \frac{n+1}{n},\frac{n+2}{n},$ there exists an isomorphism $${\cal Q}_{\l,\mu}: {\cal S}_{\d}^2(M) \rightarrow {{\mathcal{D}}}_{\l,\mu}^2(M),$$ given as follows: for all $P=P^{ij}\xi_i\xi_j \in~{\cal S}_{\d}^2(M),$ one can associate a linear differential operator given by $$\begin{aligned}
{\cal Q}_{\l,\mu} (P)&=&
P^{ij}\n_i\n_j \nonumber\\[2mm]
&&+(\a_1 \n_i P^{ij}+\a_2\,\g^{ij}\, \g_{kl}\n_i P^{kl})
\n_j
\label{Tensor1}\\[2mm]
&&+\a_3\n_i\n_j P^{ij}+\a_4\, \g^{st}\, \g_{ij}\n_s\n_t P^{ij}
+\a_5 R_{ij}P^{ij}+\a_6 R \,\g_{ij}\,P^{ij},
\nonumber\end{aligned}$$ where $R_{ij}$ (resp. $R$) are the Ricci tensor components (resp. the scalar curvature) of the metric $\g,$ the constants $\a_1,\ldots,\a_6$ are given by $$\begin{array}{ll}
\a_1=\displaystyle
\frac{2(n\l+1)}
{2+n(1-\d)}, &\a_2= \displaystyle
\frac{n(\l+\mu-1)}
{(2+n(1-\delta ))
(2-n\d)},\\[3mm]
\a_3= \displaystyle
\frac{n\l(n\lambda+ 1)}
{(1+n(1-\delta))(2+n(1-\delta ))}, &
\a_5= \displaystyle
\frac{n^2\l(\mu-1)}
{(n-2)(1+n(1-\delta))} ,\\[3mm]
\a_4= \displaystyle
\frac{n\l(n^2 \mu (2-\l-\mu)+2(n\l +1)^2-n(n+1))}
{(1+n(1-\delta))(2+n(1-\d))(2+n(1-2\d))(2-n\delta )}, &
\a_6= \displaystyle
\frac{(n\d-2)}
{(n-1)(2+n(1-2\delta))}\, \a_5 \cdot\\[3mm]
&\\
\end{array}$$ and has the following properties:\
(i) It does not depend on the rescaling of the metric $\g$.\
(ii) If $M={\mathbb{R}}^n$ is endowed with a flat conformal structure, this map is unique, equivariant with respect to the action of the group ${\mathrm{O}}(p+1,q+1) \subset {\mathrm{Diff}}({\mathbb{R}}^n)$.
Before to give the formula of the conformal equivariant map in the case of surfaces, let us recall an interesting approach for the multi-dimensional Schwarzian derivative for conformal mapping [@os] (see also [@ca]). First, recall that all surfaces are conformally flat. This means that every metric can be express (locally) as $$\g=F^{-1}\psi^* \g_0,$$ where $\psi$ is a conformal diffeomorphism of $M$, $F$ is a non-zero positive function and $\g_0$ is a metric of constant curvature. The Schwarzian derivative of $\psi$ is defined in [@os] as the following tensor field $$\label{osg}
S(\psi )=\frac{1}{2F}\nabla dF-\frac{3}{4F^2}\,dF\otimes dF+\frac{1}{8F^2}
\g^{-1}(dF,dF)\, \g \cdot$$
Now we are in position to give the quantization map for the case of surfaces.
For $\d \not=1,2,\frac{3}{2},\frac{5}{2},$ and for each $P=P^{ij}\xi_i\xi_j +P^i\xi_i +P_0\in {\cal S}_{\d,2}(M)$ one associates a linear differential operator given by $$\begin{aligned}
Q(P)&=&
P^{ij}\n_i\n_j \nonumber\\[2mm]
&&+(\a_1 \n_i P^{ij}+\a_2\,\g^{ij}\, \g_{kl}\n_i P^{kl})
\n_j \label{surf}\\[2mm]
&&+\a_3\n_i\n_j P^{ij}+\a_4\, \g^{st}\, \g_{ij}\n_s\n_t P^{ij}
\nonumber\\[2mm]
&& +\frac{4\l(\mu-1)}{2\d -3}\left( S(\psi)_{ij}P^{ij}+
\frac{1}{8(\d-1)} R \,\g_{ij}\,P^{ij}\right )+P_0,
\nonumber\end{aligned}$$ where $S(\psi)$ is the tensor (\[osg\]), $R$ is the scalar curvature and the coefficients $\a_1,\ldots,\a_4$ are given as above.
[ The projectively equivariant quantization map was given in [@lo1] (see also [@b3] for the non-flat case). The multi-dimensional projective Schwarzian derivative is defined as an obstruction to extend this isomorphism to the full group ${\mathrm{Diff}}(M).$ We will show in the next section that the conformal Schwarzian derivatives defined in this paper appear as obstructions to extend the isomorphisms (\[Tensor1\]), (\[surf\]) to the full group ${\mathrm{Diff}}(M).$ ]{}
Deformation of the space of symbols ${\cal S}_{2,\d}$ {#va}
-----------------------------------------------------
The goal of this section is to explicate the relation between the 1-cocycles (\[MultiSchwar1\]), (\[MultiSchwar2\]) and the space of second-order linear differential operators ${\cal D}_{\l,\mu}^2(M).$ Since the space ${\cal D}_{\l,\mu}^2(M)$ is a non-trivial deformation of the space of the corresponding space of symbols ${\cal S}_{\d,2}(M),$ where $\d=\mu-\l$, it is interesting to give explicitly this deformation in term of the 1-cocycles (\[MultiSchwar1\]), (\[MultiSchwar2\]). Namely, we are looking for the operator $\bar f_\d={\cal Q}^{-1}_{\l,\mu} \circ f_{\l,\mu} \circ
{\cal Q}_{\l,\mu}$ such that the diagram below is commutative $$\nonumber
\begin{CD}
{\cal S}_{\d,2}(M)
@> \bar f_\d >>
{\cal S}_{\d,2}(M)
\strut\\
@V{\cal Q}_{\l,\mu}VV
@VV{\cal Q}_{\l,\mu}V \strut\\
{{\mathcal{D}}}_{\l,\mu}^2 (M) @> f_{\l,\mu} >>
{{\mathcal{D}}}_{\l,\mu}^2 (M) \strut
\end{CD}
\label{TheDiagram}$$
For all $\d \not=\frac{2}{n},\frac{n+2}{2n},1,\frac{n+1}{n},
\frac{n+2}{n},$ the deformation of the space of symbols ${\cal S}_{\d,2}(M)$ by the space ${{\mathcal{D}}}_{\l,\mu}^2(M)$ as a ${\mathrm{Diff}}(M)$-module is given as follows: for all $P=P^{ij}\xi_i \xi_j +P^i\xi_i + P_0\in {\cal S}_{\d,2}(M),$ one has $$\bar f_{\d}\cdot (P )=
T^{ij}\xi_i\xi_j+T^{i}\xi_i+T^{0},$$ where $$\begin{aligned}
T^{ij}&=&
(f_{\d}\,P)^{ij},
\nonumber\\\label{ExplAction}
T^i&=&
(f_{\d}\,P)^i\;+\;
\displaystyle
\frac{n(\mu+\l -1)}{(2+n(1-\d ))(2-n\delta)}\,{\cal A}^i_{kl}(f^{-1})
(f_{\d}\,P)^{kl},\\[2mm]
\displaystyle
T^0&=&
(f_\d\, P)_0
\;-\;
\displaystyle
\frac{n \l(\mu -1)}{(2+n(1-2\d))(1-\d )(1+n(1-\d))}\,{\cal B}_{kl}(f^{-1})
(f_{\d}\, P)^{kl},\nonumber\end{aligned}$$ and $f_\d$ is the action (\[actsym\]). \[MainAct\]
[**Proof.**]{} The proof is a simple computation using (\[comp\]) and the formul[æ]{} $$\begin{aligned}
\n_i\n_j \,{f^*_{\d}}^{-1} P^{kl}&=&{f^*_{\d}}^{-1} \n_i \n_j\, P^{kl} -
\mathrm{Sym}_{l,k}\left (
{f^*_{\d}}^{-1} \n_i\, P^{tl} \;\ell(f)_{tj}^k \right) +
{f^*_{\d}}^{-1} \n_u\, P^{kl} \; \ell(f)_{ij}^u \nonumber \\
&& +\d \left ({f^*_{\d}}^{-1} \n_i\, P^{kl} \;\ell(f)_{j}\right)
-\mathrm{Sym}_{k,l} \left( \n_j \ell(f)^k_{it}\; {f^*_{\d}}^{-1}
P^{tl}\right) \\
&&+\d \n_j \ell(f)_i \; {f^*_{\d}}^{-1} P^{kl}
-\mathrm{Sym}_{k,l}\left( \ell(f)_{it}^k \n_j \, {f^*_{\d}}^{-1}
P^{tl}\right) +\d \ell(f)_i\,\n_j {f^*_{\d}}^{-1}P^{kl}
\nonumber\\[2mm]
\n_{i} \, f^* \phi&=& f^* \n_i \phi +\l \,\ell(f^{-1})_{i}\, f^* \phi
\nonumber\\[2mm]
\n_j \n_{i} \, f^* \phi&=& f^* \n_j \n_i \phi +\ell(f^{-1})_{ji}^t \, f^*
\n_t \phi +\l \mathrm{Sym}_{i,j} \, \ell(f^{-1})_j\, f^* \n_i \phi
\nonumber\\[2mm]
&& +\left (\l \n_j\, \ell(f^{-1})_{i}
+ \l^2 \ell(f^{-1})_{j}\, \ell(f^{-1})_{i}\right) \, f^* \phi
\nonumber\\[2mm]
f^* R_{jk}-R_{jk}&=& \n_i\, \ell(f^{-1})^i_{jk}-\n_j \, \ell(f^{-1})_k
- \ell(f^{-1})^m_{sj}\, \ell(f^{-1})^s_{km}+
\ell(f^{-1})_m \, \ell(f^{-1})^m_{jk},\nonumber\end{aligned}$$ for all $\phi\in {{\mathcal{F}}}_\l$ and for all $f\in {\mathrm{Diff}}(M),$ where $R_{ij}$ are the Ricci tensor components.\
------------------------------------------------------------------------
The 1-cocycle ${\cal A},$ ${\cal B}$ and the conformally invariant quantization {#new}
-------------------------------------------------------------------------------
We will show in this section that the quantization procedure is not invariant if one consider two metrics conformally equivalent. The obstruction of the invariance is given by the 1-coycles ${\cal A}$ and ${\cal B}.$
Given two conformally equivalent metrics $\g,$ $\tilde \g.$ Denote by $\n$, $\cal A$ and $\cal B,$ the covariant derivative, the 1-cocycles (\[MultiSchwar1\]) and (\[MultiSchwar2\]) written with the metric $\g,$ respectively. We have
The quantization map ${\cal Q}_{\l,\mu}^{\g}:$ ${\cal D}^1_{\l,\mu}(M)\rightarrow {\cal S}^1_\d(M)\oplus {\cal S}^0_\d (M)$ defined in (\[yah1\]) depend only on the conformal class of the metric $\g.$
[**Proof.**]{} Let $\tilde \g$ be another metric conformally equivalent to $\g. $ That means that there exists a diffeomorphism $\psi : (M,\tilde \g )\rightarrow (M,\g)$ and a non-zero positive function $F$ such that (locally) $$\tilde \g=F^{-1}\cdot \psi^* \g.$$ The Levi-Civita of the two connections are related by $${\Gamma^k_{ij}}^{ \g}= {\Gamma^k_{ij}}^{\tilde \g}+
\frac{1}{2 F} \left (\partial_i F\,
\d_j^k
+\partial_j F\, \d_i^k-\partial_s F\, \tilde \g^{sk}\tilde \g_{ij}\right ) -
\ell(\psi^{-1})^k_{ij},$$ where $\partial_i F=F_i$ and $\ell(\psi^{-1})^k_{ij}$ are the components of the tensor (\[ell\]). This equation implies that $$\begin{array}{lcl}
\n^{\g}_i\phi&=&\n^{\tilde \g}_i\phi-\displaystyle \frac{\l n}{2}
\displaystyle \frac{F_i}{F}+
\l \,\ell(\psi^{-1})_i \, \phi,\\[2mm]
\n^{\g}_i P^i &=&\n^{\tilde \g}_i P^i+\displaystyle
\frac{n(1-\d)}{2}\frac{F_i P^i}{F}-(1-\d) \ell(\psi^{-1})_{i}P^i,
\end{array}$$ for all $\phi\in {{\mathcal{F}}}_\l$ and for all $P^i\xi_i\in {{\mathcal{S}}}_{\d}^1(M).$\
Substitute these formul[æ]{} into (\[yah1\]) we see that ${\cal Q}^{\tilde \g}_{\l,\mu}={\cal Q}^{\g}_{\l,\mu}.$
The quantization map ${\cal Q}_{\l,\mu}^{\g}:$ ${\cal D}^2_{\l,\mu}(M)\rightarrow {\cal S}^2_\d(M)$ defined in (\[Tensor1\]) has the property $${\cal Q}_{\l,\mu}^{\tilde \g} (P)={\cal Q}_{\l,\mu}^{\g} (P)+d_1 \,
{\cal A}^{s}(\psi^{-1})
(P )\nabla_s + d_2 \,\nabla_s \left ( {\cal A}^{s}(\psi^{-1})
(P) \right ) +d_3 \, {\cal B}(\psi^{-1})(P),$$ for all $P\in {\cal S}_{\d}^2(M),$ where the constants $d_1,d_2$ and $d_3$ are given by $$\begin{aligned}
d_1&=&\frac{n(\l+\mu-1)}{(2+n(1-\d))(2-n\d))}, \quad
\displaystyle d_2=\frac{n\l(\l+\mu-1)}{(2+n(1-\d))(2-n\d)(1-\d)}, \nonumber
\\[1mm]
d_3&=&\displaystyle \frac{n \l(\mu -1)}{(2+n(1-2\d))(\d -1)(1+n(1-\d))} \cdot
\nonumber \end{aligned}$$
[**Proof.**]{} The proof involves the calculation of $\nabla_i ^{\tilde \g}\nabla_j ^{\tilde \g} \phi,$ $\nabla_i ^{\tilde \g}\nabla_j ^{\tilde \g} P^{kl}$ and $R^{\tilde \g}$ wich is straightforward but quite complicated.
[ The system $d_1=d_2=d_3=0$ admits a unique solution: $(\l,\mu)=(0,1).$ The value of $\d=1$ is called “resonant”. In this case, the quantization map is not unique; there exists a one-parameter family of such isomorphism (see [@do] for more dtails.) ]{}
For all $f\in {\mathrm{Diff}}(M)$ and for all conformal map $\psi: (M,\tilde \g)
\rightarrow (M,\g)$, one has
[(i)]{} ${\cal A}^{\tilde \g}(f)=\psi^* {\cal A}^{\g}
(\psi \circ f\circ \psi^{-1}),$
[(ii)]{} ${\cal B}^{\tilde \g}(f)=\psi^* {\cal B}^{\g}
(\psi \circ f\circ \psi^{-1}).$
[**Proof.**]{} Straightforward computation.
For all conformal map $\psi (M,\tilde \g)
\rightarrow (M,\g)$ one has
[(i)]{} $\tilde {\cal A}^{\tilde \g}(\psi)=-{\cal A}^{\g}(\psi^{-1} ),$
[(ii)]{} $\tilde {\cal B}^{\tilde \g}(\psi)=-{\cal B}^{\g}(\psi^{-1} )\cdot$
[ The Corollary above shows that for a conformal map $\psi,$ the 1-cocycle $\cal B$ is still a second-order differential operator and then does not coincide with the Schwarzian derivative (\[osg\]) defined by Osgood and Stow. ]{}
Appendix
========
We will give a formula for the Schwarzian derivative for the case of surfaces. As explained in section (\[tokyo\]), all surfaces are conformally flat. That means that every metric can be express (locally) as $$\g=F^{-1}\psi^* \g_0,$$ where $\psi$ is a conformal diffeomorphism of $M$, and $F$ is a non-zero positive function, $\g_0$ is a metric of constant curvature.
The explicit formula of the Schwarzian derivative in the case of surfaces is: the following $$\begin{aligned}
{\cal B}'_2(f)_{ij}&=&
{f^*}^{-1} \left( {\g}^{st}\, {\g}_{ij}\nabla_s\nabla_t \right )
-\g^{st}\,\g_{ij}\nabla_s \nabla_t +4(1-\delta)\left( \ell(f)_{ij}^s
-\frac{1}{2}\,Sym_{i,j}\,
\delta_{i}^s \,\ell(f)_{j} \right)\nabla_s \nonumber \\[2mm]
&&+4 (1-\d)^2 \ell(f)_s \left ( \ell(f)_{ij}^s -\frac{1}{4}
Sym_{i,j}\, \delta_i^s \, \ell(f)_j\right) +
2(\d-2)(1-\d) \,Sym_{i,j}\,\n_{j}\ell_{i}(f) \nonumber\\[2mm]
&&+8\,(\d-1)^2 \left(
{f^{-1}}^*(S(\psi)_{ij})-S(\psi)_{ij}+\frac{1}{2}
\n_{s}\ell(f)_{ij}^s \right )+(\d-1)
\left ({f^{-1}}^*(R\,\g_{ij})-R\,\g_{ij}\right ),
\nonumber
\label{MultiSchwar3}\end{aligned}$$ where $S(\psi)$ is the derivative (\[osg\]), $\ell(f)$ is the tensor (\[ell\]), $R$ is the scalar curvature of the metric $\g,$ is a differential operator from from ${{\mathcal{S}}}_{\d}^2(M)$ to ${{\mathcal{S}}}_{\d}^0(M).$
Theorem (\[mainp\]) remains true for this operator.
[*Acknowledgments*]{}. It is a pleasure to acknowledge numerous fruitful discussions with Prof. V. Ovsienko. I am grateful to, the referee for his pertinent remarks, Prof. Y. Maeda and Keio University for their hospitality.
[99]{} A. L. Besse, Einstein manifolds, Springer-Verlag, Berlin Heidelberg, 1987.
S. Bouarroudj, Multi-dimensional Schwarzian derivative revisited, math. DG/0101056
S. Bouarroudj, Projectively equivariant quantization map, Lett. Math. Phy. [**51:**]{} (4) (2000), 265-274.
S. Bouarroudj & V. Ovsienko, Schwarzian derivative related to the space of linear differential operators on a locally projective manifold, Banach Cent. Pub., vol 51, 2000.
S. Bouarroudj & V. Ovsienko, Three cocycles on ${\mathrm{Diff}}(S^1)$ generalizing the Schwarzian derivative, Internat. Math. Res. Notices 1998, No.1, 25–39.
C. Caratheodory, Theory of functions, Chelsa Pub. Co. N.Y. 1960.
K. Carne, The Schwarzian derivative for conformal maps, J. Reine Angrew. Math. [**408**]{} (1990), 10- 33.
P. Cohen, Yu. Manin and D. Zagier, Automorphic pseudo-differential operators, in Algebraic Aspects of Integrable Systems, Prog. Nonlinear Differential Equations Appl. [**26**]{}, Birkhäuser, Boston, 1997, 17-47.
C. Duval, P. B. A. Lecomte & V. Ovsienko, Conformally equivariant quantization: existence and uniqueness, Ann. Inst. Fourier, [**49**]{}:6 (1999) 1999-2029.
C. Duval & V. Ovsienko, Conformally equivariant quantization, Selecta Math., N. S., Vol 7, No 3, p 291-409.
D. B. Fuks, Cohomology of infinite-dimensional Lie algebras, Contemp. Soviet. Math., Consultants Bureau, New-York, 1986.
H. Gargoubi, Sur la géométrie de l’espace des opérateurs différentiels linéaires sur ${\mathbb{R}},$ Bull. Soc. Roy. Sci. Liege. Vol. 69, 1, 2000, pp. 21-47.
A.A. Kirillov, Infinite dimensional Lie groups: their orbit invariants and representations. The geometry of moments. Lecture note in Maths. [**970**]{}, Springer
S. Kobayashi, T. Nagano, On projective connections, J. Math. Mech. [**13:2**]{} (1964) 215–235.
P. B. A. Lecomte & V. Ovsienko, Projectively invariant symbol calculus, Lett. Math. Phy. [**49**]{} (3) (1999), 173-196.
P. B. A. Lecomte & V. Ovsienko, Cohomology of vector fields Lie algebras, Compositio Mathematica. [**119:**]{} 2000.
P. B. A. Lecomte, P. Mathonet E. Tousset, Comparaison of some modules of the Lie algebra of vector fields, Indag. Math., [**7**]{} (4) (1996), 461-471
A. Neijnhuis & R. W., Jr. Richardson, Deformation of homomorphisms of Lie groups and Lie algebras, Bull. Amer. Math. Soc. [**73**]{} 1967, 175-179.
B. Osgood & D. Stowe, The Schwarzian derivative and conformal mapping of Riemannian manifolds, Duke Math. J. [**67:1**]{} (1992) 57–99.
G. B. Segal, Unitary representations of some infinite dimensional groups, Comm. Math. Phys. [**80**]{} (1981) 301–342.
E. J. Wilczynski, Projective differential geometry of curves and ruled surfaces, Leipzig - Teubner - 1906.
[^1]: Research supported by the Japan Society for the Promotion of Science.
| {
"pile_set_name": "ArXiv"
} |
**Two combined methods for the global solution of semilinear differential-algebraic equations with the use of spectral projectors and Taylor expansions**
[M. S. Filipkovska (Filipkovskaya)]{}
*Department of Mathematical Physics, B. Verkin Institute for Low Temperature Physics and Engineering of the National Academy of Sciences of Ukraine, 47 Nauky Ave., Kharkiv, 61103, Ukraine,*
[email protected]
Two combined numerical methods for solving semilinear differential-algebraic equations (DAEs) are obtained and their convergence is proved. The comparative analysis of these methods is carried out and conclusions about the effectiveness of their application in various situations are made. In comparison with other known methods, the obtained methods require weaker restrictions for the nonlinear part of the DAE. Also, the obtained methods enable to compute approximate solutions of the DAEs on any given time interval and, therefore, enable to carry out the numerical analysis of global dynamics of mathematical models described by the DAEs. The examples demonstrating the capabilities of the developed methods are provided. To construct the methods we use the spectral projectors, Taylor expansions and finite differences. Since the used spectral projectors can be easily computed, to apply the methods it is not necessary to carry out additional analytical transformations.
**Key words:** differential-algebraic equation, implicit differential equation, combined method, regular pencil, spectral projector, global dynamics
`MSC2010:` 65L80, 65L20, 65L70, 34A09, 34A12, 47N40
Introduction {#Intro}
============
Consider an implicit semilinear differential equation $$\label{DAE}
\frac{d}{dt}[Ax]+Bx=f(t,x)$$ with the initial condition $$\label{ini}
x(t_0)=x_0,$$ where $t,\, t_0 \ge 0$, $x,\, x_0 \in {{\mathbb R}^n}$, $f\in C([0,\infty) \times {{\mathbb R}^n}, {{\mathbb R}^n})$, and $A,\, B\colon {{\mathbb R}^n}\to {{\mathbb R}^n}$ are linear operators (or the corresponding $n\times n$ matrices) which may be degenerate (noninvertible). An equation of the type with a degenerate operator at the derivative is called a *differential-algebraic equation* (*DAE*), since a system of differential and algebraic equations corresponds to it. The initial value $x_0$ for the problem , with a degenerate operator $A$ must be chosen so that the initial point $(t_0,x_0)$ belongs to the manifold $$\label{L_0}
L_0 = \{(t,x) \in [0,\infty) \times {{\mathbb R}^n}\mid Q_2[Bx-f(t,x)] = 0\},$$ where $Q_2$ is the spectral projector from (see also ). If the operator $A$ is nondegenerate (invertible), then the equation can be written as an ordinary differential equation (ODE) and $Q_2= 0$. The DAE is called *semilinear* and is often written in the form $$\frac{d}{dt}[Ax]=g(t,x).$$ There are several reasons why we consider a semilinear DAE in the form . Firstly, DAEs of this form are used to describe mathematical models in radio electronics, economics, control theory, mechanics of multilink mechanisms, chemical kinetics and other fields [@Campbell-II; @Rabier-Rh; @Fox-Jen-Zom; @Morishima; @Riaza-Zufiria; @Kunkel_Mehrmann; @Engwerda-Salm-W; @Riaza; @RF2; @Brenan-C-P; @Lamour-Marz-Tisch; @Vlasenko1]. Secondly, in the equation , not only the operator $A$ but also the operator $B$ may be degenerate, and the influence of the linear part $\displaystyle\frac{d}{dt}[Ax]+Bx$ of the equation is determined by the properties of the pencil $\lambda A+B$. It is assumed that $\lambda A+B$ is a regular pencil of index not higher than 1 (see section \[IndexMeth\]). Then there exist the spectral projectors $P_1$, $P_2$, $Q_1$, $Q_2$ [@Rut; @Rut_Vlas] (see section \[IndexMeth\]) which are used in the development of the numerical methods and in the proof of theorems.
Note the following. We do not require that the equation be a regular DAE of index 1 for all $(t,x)\in [0,\infty )\times {{\mathbb R}^n}$ (or $(t,x)\in L_0$), i.e., that the pencil $\displaystyle\lambda A + B-\frac{\partial f}{\partial x}(t,x)$ be a regular pencil of index 1 for all $(t,x)$. This requirement is one of the conditions for the global solvability of a nonlinear DAE in the theorem [@Lamour-Marz-Tisch Theorem 6.7]. In [@Ascher-Petz; @Brenan-C-P; @Kunkel_Mehrmann; @HairerLR; @Hairer-W; @Knorrenschild], restrictions similar to the above requirement of index 1 for a regular DAE are used locally to prove the local solvability of DAEs. Various notions of an index for a regular DAE and the relationship between them are discussed in [@Fil.MPhAG Remark 2.1, section 2].
In what follows, for the sake of generality, *the equation with an arbitrary (not necessarily degenerate) linear operator $A\colon {{\mathbb R}^n}\to {{\mathbb R}^n}$* will be called a *semilinear DAE*, since the obtained results are also true in the case of the nondegenerate operator $A$, however, the main object of our study is the equation with the degenerate operator.
A function $x(t)$ is called a *solution* of the initial value problem , on some interval $[t_0,t_1)$, $t_1\le \infty $, if $x \in C([t_0,t_1),\, {{\mathbb R}^n})$, $Ax \in C^1([t_0,t_1),{{\mathbb R}^n})$, $x$ satisfies the equation on $[t_0,t_1)$ and the initial condition .
It should be noted that in the case of the degenerate operator $A$ a solution of a semilinear DAE of the form $$\displaystyle A\frac{d}{dt}x+Bx=f(t,x)$$ must be smoother (the solution $x(t)$ is continuously differentiable) than a solution of a semilinear DAE of the form (the solution $x(t)$ is continuous, and its component corresponding to $Ax(t)$ is continuously differentiable).
A solution $x(t)$ of the initial value problem , is called *global* if it exists on the interval $[t_0,\infty )$.
A differential-algebraic equation is also called an algebraic-differential, degenerate differential [@Vlasenko1], descriptor [@Engwerda-Salm-W] equation (or system) and a singular system of differential equations [@Campbell-II; @Boyarintzev3; @Boyar-Dan-L-Ch].
There are various ways of finding approximate solutions of DAEs. In most works, the main idea is the reduction of a DAE to an ODE or the replacement of a DAE by a stiff ODE for the further application of the known methods for solving ODEs, as well as the use of these methods directly for solving DAEs. In [@Gear_Petz84], various algorithms for reducing regular linear DAEs to ODEs have been presented and references to earlier works have been given. In addition, the BDF (backward differentiation formulas) method for solving a regular nonlinear DAE having uniform index 1 has been proposed (see also [@Brenan-C-P; @Ascher-Petz]). As a rule, the $\varepsilon$-embedding method is applied to the autonomous semi-explicit DAE $$\begin{aligned}
&\dot{y}= f(y,z), \\
&0= g(y,z)\end{aligned}$$ of index 1 (this DAE has index 1 for all $y$, $z$ such that $\left[\frac{\partial g}{\partial z}(y,z)\right]^{-1}$ exists and is bounded) [@Hairer-W; @HairerLR]. The method is as follows: The corresponding stiff system of ODEs $\dot{y}= f(y,z)$, $\varepsilon \dot{z}= g(y,z)$, $\varepsilon \to 0$, is considered, then Runge-Kutta, Rosenbrock or other suitable methods are applied to the stiff system and $\varepsilon = 0$ is put in the resulting formulas. The similar $\varepsilon$-embedding method (the Runge-Kutta methods are applied to the stiff system $\dot{y}= f(t,y,z,\varepsilon )$, $\varepsilon \dot{z}= g(t,y,z)$, $\varepsilon \to 0$, and then $\varepsilon = 0$ is put in the resulting formulas) [@Knorrenschild; @Ascher-Petz; @Brenan-C-P; @Kunkel_Mehrmann], the BDF method and general linear multi-step methods [@Ascher-Petz; @Brenan-C-P; @Kunkel_Mehrmann] are applied to the nonautonomous semi-explicit DAE $$\begin{aligned}
&\dot{y}= f(t,y,z), \\
&0=g(t,y,z)\end{aligned}$$ of index 1 ($\left[\frac{\partial g}{\partial z}(t,y,z)\right]^{-1}$ exists and is bounded). The semi-explicit DAE corresponding to the stiff ODE system is called reduced. The application of the Runge-Kutta method to the reduced semi-explicit DAEs of index 2, 3 is described in [@Hairer-W; @HairerLR] (for the autonomous DAEs) and in [@Ascher-Petz; @Brenan-C-P] (for the nonautonomous DAEs). For a regular nonlinear DAE of index 1 (this condition for the semilinear DAE was discussed above) [@Brenan-C-P; @Ascher-Petz; @Lamour-Marz-Tisch] and for a regular quasilinear DAE of the form $C(y)\dot{y}=f(y)$ with constraints providing the local solvability [@Hairer-W], the application of the BDF methods, the Runge-Kutta methods, the backward Euler method (the implicit Euler method) and the general linear multi-step method has been considered. In [@Kunkel_Mehrmann], the BDF and linear multi-step methods are used for solving regular linear DAEs with constant matrix coefficients and it is discussed why similar results cannot hold for linear DAE with variable coefficients. The collocation Runge–Kutta method and the BDF method for solving a regular strangeness-free DAE (with the strangeness index 0) of the form $$\begin{aligned}
&F_1(t,x,\dot{x})=0, \\
&F_2(t,x)=0\end{aligned}$$ are also proposed in [@Kunkel_Mehrmann]. Half-explicit methods for the strangeness-free DAEs are studied in [@Linh-Mehrmann]. In [@Boyar-Dan-L-Ch], an analog of the Euler method is applied to the nonlinear system $f(\dot{x},x,t)=0$ when special conditions are fulfilled and a DAE of the form $A(t)\dot{x}+\Phi(x,t)=0$ is considered as a particular case. The combination of the simple iteration method and the explicit Euler method is used for solving a degenerate semilinear integro-differential equation in [@Piven]. In [@Kylikov], two combined methods such as the implicit Euler method in combination with the simple iteration method and the implicit Adams method in combination with the Newton method [@Kylikov] are proposed for solving the autonomous semi-explicit DAE $\dot{y}= f(y,z)$, $z= g(y,z)$. In [@HankeMarTiWeinW], a least-squares collocation method is constructed for linear higher-index DAEs and its convergence is shown for a limited class of such DAEs. Methods of finding numerical solutions of linear DAEs by using the Obreshkov formula [@Boyar-Dan-L-Ch], the implicit Euler method [@Gear-Petz; @Boyarintzev3; @Lamour-Marz-Tisch], and finite difference schemes applied to a reduced linear DAE of index 1 [@Rabier-Rh2] are known.
Advantages and disadvantages of the mentioned numerical methods are discussed in the corresponding literature [@Gear_Petz84; @Hairer-W; @HairerLR; @Knorrenschild; @Ascher-Petz; @Brenan-C-P; @Kunkel_Mehrmann; @Lamour-Marz-Tisch; @Boyar-Dan-L-Ch; @Linh-Mehrmann]. The following are disadvantages, in connection with which the new methods described in the present paper were developed:
1. Most of the mentioned methods for solving semilinear and nonlinear DAEs can be correctly applied only on a sufficiently small (local) interval of time and the calculation of the allowable length of this interval is a separate problem. This is due to the fact that the existence of an exact solution and, accordingly, the approximate solution is proved only on a sufficiently small time interval [@Hairer-W; @HairerLR; @Knorrenschild; @Ascher-Petz; @Brenan-C-P; @Kunkel_Mehrmann; @Boyar-Dan-L-Ch; @Lamour-Marz-Tisch; @Linh-Mehrmann]. However, for the analysis of the global dynamics of mathematical models, it is important to be able to investigate the behavior of the solution on an arbitrary (arbitrarily large) time interval.
2. \[disadv2\] For the convergence of the methods (in the case of an arbitrary time interval), conditions that are too restrictive for certain classes of mathematical models described by DAEs are used. These include the global Lipschitz condition and similar conditions, for example, the global condition of contractivity (the Lipschitz condition with a constant less than 1) [@Piven; @Kylikov; @Lamour-Marz-Tisch], the condition of global index 1 for a regular DAE [@Lamour-Marz-Tisch; @Gear_Petz84] and the like. These conditions are not fulfilled for mathematical models of electrical circuits with certain nonlinear parameters (for example, in the form of power functions mentioned in section \[Appl\]) on the time interval $[t_0,T]$, which is not sufficiently small, and, obviously, on the interval $[t_0,\infty )$. The intervals $[t_0,T]$ and $[t_0,\infty )$ are considered when finding numerical solutions and when studying the existence, uniqueness and Lagrange stability of exact solutions, respectively. In general, various types of differential equations with nonlinear functions which may not satisfy the global Lipschitz condition and similar conditions, for example, various classes of stochastic differential equations with non-Lipschitz (or non-globally Lipschitz) functions (see e.g. [@BenchaabaneSakth; @IzgiCetin] and references therein), arise in many applications. Besides, the methods considered above require that the nonlinear functions appearing in the equation be at least continuously differentiable in all arguments and most of the methods require the higher smoothness of the nonlinear functions.
3. \[disadv3\] For the application of many methods, additional analytical transformations and constructions are required. To use the algorithms and methods from [@Gear_Petz84], it is necessary that there exist the projection matrices or mappings which transform the DAE to a canonical or semicanonical form. It is assumed in [@Rabier-Rh2] that there exist the projection mappings with the help of which the linear DAE can be reduced to the ODE at a certain step of the reduction process. In some works, conditions for the existence of operators allowing one to reduce a semilinear DAE to a semi-explicit DAE are given, however, their construction requires additional transformations and computations (cf. [@Lamour-Marz-Tisch; @Boyar-Dan-L-Ch]). It should be noted that one can reduce the semilinear DAE to the equivalent semi-explicit DAE using the spectral projectors discussed in section \[IndexMeth\]. Further, in order to apply the aforementioned $\varepsilon$-embedding method [@Ascher-Petz; @Hairer-W; @Kunkel_Mehrmann; @Knorrenschild; @Brenan-C-P; @HairerLR], it is necessary to reduce a semi-explicit DAE to the corresponding stiff system of ODEs that requires a sufficient smoothness of the nonlinear function in the algebraic part of the DAE (the continuity of the function is not enough here).
In this paper, *we obtain numerical methods having the following advantages* in comparison with other known methods for solving equations of the type :
1. \[adv1\] Applying the obtained methods, it is possible to compute approximate solutions on any given time interval $[t_0,T]$. The theorems on the existence and uniqueness of global solutions (see section \[GlobSolv\]) and on the convergence of the numerical methods (see section \[NumMs\]) ensure the correctness and convergence of the methods. The results of the theoretical research of the global dynamics of mathematical model considered in section \[Appl\] are consistent with the results of the analysis of numerical solutions.
2. \[adv2\] The obtained methods require weaker restrictions for the nonlinear part of the equation. To prove the existence and uniqueness of global solutions and to prove the convergence of the methods, the restrictions of the type of the global Lipschitz condition are not used (we discussed these conditions above). Moreover, the methods require the less smoothness the nonlinear part of the equation than other known methods. They are applicable to the DAEs with the continuous nonlinear part which may not be continuously differentiable in $t$ (see Remarks \[remNum-meth\], \[remModNum-meth\]). This is important for applications, since such equations arise in various practical problems. In particular, the functions of currents and voltages in electric circuits may not be continuously differentiable (or be piecewise continuously differentiable) or may be approximated by functions that are not continuously differentiable. As examples, nonsinusoidal currents and voltages of the “sawtooth”, “triangular” and “rectangular” shapes [@Erickson-Maks; @Borisov-Lip-Zor] can be considered. In section \[Appl\], the examples of numerical solutions for an electrical circuit with the voltage of the triangular and sawtooth shapes, which have been obtained using the proposed methods, are given. It should be noted that currents and voltages of a more complex shape are occurred.
3. \[adv3\] The spectral projectors, with the help of which the DAE is reduced to the equivalent system of a purely differential equation and a purely algebraic equation (to the semi-explicit form), can be constructively determined by the formulas [@Rut_Vlas (5), (6)] and be numerically found using . The possibility to easily compute the projectors on a computer, using , enables to numerically solve the DAE directly in the form , i.e., additional analytical transformations are not required for the application of the developed numerical methods.
To construct the numerical methods, the (differential and algebraic) equations of the system equivalent to the DAE (see item \[adv3\] above) are approximated by using Taylor expansions and finite differences. As a result, the combined numerical methods are obtained. In method 1 (the method –), we apply the explicit Euler method to the differential equation, that is, the derivative is approximated by a forward difference. In method 2 (the method –), the derivative is approximated by a centered difference, which leads to a certain modification of the Euler method. Therefore, method 2 will also be called modified method 1. In the algebraic equation (for both methods), the Taylor expansion of the nonlinear function in one of the components of the phase variable is used, which gives a method similar to the Newton method with respect to this component. This technique of the expansion allows us to weaken the requirements for the nonlinear function and to apply the obtained methods even for the DAE with the nonlinear part continuous in $t$ (taking into account Remarks \[remNum-meth\], \[remModNum-meth\]). The consistency condition $(t_0,x_0)\in L_0$ for the initial values $t_0$, $x_0$ ensures the best choice of the initial value for the method applied to the algebraic equation. Methods 1 and 2 have the first and second orders of accuracy, respectively. It is clear that the approximations of a higher order of accuracy are used for method 2 and, accordingly, the higher smoothness of the nonlinear part of the equation is required (see Theorems \[ThNum-meth\], \[ThModNum-meth\]).
*The paper has the following structure.* In Section \[IndexMeth\], we consider the restriction on the operator coefficients of the equation (on the operator pencil) and give the corresponding definition of a regular pencil of index not higher than 1; also, we consider the method of spectral projectors for the reduction of the semilinear DAE to an equivalent semi-explicit form. In Section \[GlobSolv\], the theorems proved in earlier works [@Fil.MPhAG; @RF1], which give conditions for the existence and uniqueness of exact global solutions, are presented. In Section \[NumMs\], the two combined numerical methods for solving the semilinear DAE are obtained and their convergence is proved (note that the method from Subsection \[Num-meth\] was proposed by the author in [@Fil.num_meth] without the theorem on its convergence), as well as the important remarks on the convergence of the methods, when weakening the smoothness requirements for the nonlinear function, are given. In Section \[CompareMeth\], the comparative analysis of these methods is carried out and conclusions about the effectiveness of their application in various situations are made. In Section \[Appl\], we provide the examples demonstrating the capabilities of the developed methods and information on applied problems in which the semilinear DAEs arise.
The following notation is used in this paper: $L(X,Y)$ is the space of continuous linear operators from $X$ into $Y$, $L(X,X)=L(X)$; $\left. A\right|_X$ is the restriction of the operator $A$ to $X$; $I$ is an identity operator (matrix); $x^T$ is the transpose of $x$. Sometimes, in the paper, a function $f$ is denoted by the same symbol $f(x)$ as its value at the point $x$ in order to explicitly indicate its argument (or arguments), but it will be clear from the context what exactly is meant.
Index of the regular pencil $\lambda A+B$ and spectral projectors {#IndexMeth}
=================================================================
Consider the initial value problem , : $$\frac{d}{dt}[Ax]+Bx=f(t,x), \qquad x(t_0)=x_0.$$
It is assumed that the pencil $\lambda A+B$, where $\lambda$ is a complex parameter, is *regular*, i.e., there exists a number $\lambda_0$ such that $\det(\lambda_0 A+B)\not = 0$. Further, we assume that there exist constants $C_1,\, C_2 >0$ such that $$\label{index1}
\left\|(\lambda A+B)^{-1}\right\| \le C_1\quad \text{ for all }\; |\lambda|\ge C_2. $$ The condition [@Rut_Vlas] means that either the point $\mu = 0$ is a simple pole of the resolvent $(A+ \mu B)^{-1}$ (this is equivalent to the fact that $\lambda = \infty$ is a removable singular point of the resolvent $(\lambda A+B)^{-1}$), or $\mu = 0$ is a regular point of the pencil $A+ \mu B$ (i.e., there exists a resolvent $(A+\mu B)^{-1}$ at the point $\mu =0$ and, hence, the operator $A$ is nondegenerate).
If $A$ is degenerate and the point $\mu = 0$ is a simple pole of the resolvent $(A+ \mu B)^{-1}$, i.e., is fulfilled, then we will say that $\lambda A+ B$ is a regular *pencil of index 1*.
If $A$ is nondegenerate, i.e., $\mu = 0$ is a regular point of the pencil $A+ \mu B$, then we will say that $\lambda A+ B$ is a regular *pencil of index 0*.
Thus, if $\lambda A+B$ is a regular pencil and is fulfilled, then we will say that $\lambda A+B$ is a regular *pencil of index not higher than 1* (i.e., of index 0 or 1).
In the general case, according to [@Vlasenko1 section 6.2], the maximum length of the chain of an eigenvector and adjoint vectors of the matrix pencil $A+\mu B$ at the point $\mu = 0$ is called the index of the matrix pencil $\lambda A+B$.
Various notions of an index of the pencil, an index of a DAE and their relationship with the mentioned notion of the pencil of index 1 are considered in [@Fil.MPhAG Remark 2.1].
For the regular pencil $\lambda A+B$ satisfying there are the two pairs of mutually complementary spectral projectors [@Rut_Vlas (5), (6)] $$\label{Proj.1}
\begin{split}
&P_1 =\frac{1}{2\pi i}\, \oint\limits_{|\lambda |=C_2 }(\lambda A+B)^{-1}\, A\, d\lambda,\quad P_2 =I-P_1, \\ &Q_1 =\frac{1}{2\pi i}\, \oint\limits_{|\lambda |=C_2 }A\, (\lambda A+B)^{-1}\,d\lambda,\quad Q_2 =I-Q_1,
\end{split}$$ which decompose the space ${{\mathbb R}^n}$ into direct sums of subspaces $$\label{Proj.2}
{{\mathbb R}^n}=X_1 \dot{+}X_2,\quad {{\mathbb R}^n}=Y_1 \dot{+} Y_2,\quad X_j =P_j {{\mathbb R}^n},\quad Y_j =Q_j {{\mathbb R}^n},\quad j=1,2,$$ such that the operators $A$, $B$ map $X_j$ to $Y_j$ and the induced operators $A_j {=\nobreak\discretionary{}{\hbox{$\mathsurround=0pt =$}}{}} \left. A\right|_{X_j}\colon X_j \to Y_j$, $B_j =\left. B\right|_{X_j}\colon X_j \to Y_j$ (${j=1,2}$) are such that $A_2=0$ and there exist the inverse operators $A_1^{-1} \in L(Y_1,X_1)$, $B_2^{-1} \in L(Y_2,X_2)$ [@Rut_Vlas sections 2, 6]. The projectors are real (since $A$ and $B$ are real) and have the properties $$AP_1 =Q_1 A=A,\qquad AP_2 =Q_2 A=0,\qquad BP_j =Q_j B,\qquad j=1,2.$$
Using the spectral projectors, we can also obtain the auxiliary operator $G\in L({{\mathbb R}^n})$ [@Rut_Vlas Sections 2, 6], $$\label{Proj.G}
G=A +BP_2 = A+Q_2 B,\quad GX_j =Y_j,\quad j=1,2,$$ which has the inverse operator $G^{-1}\in L({{\mathbb R}^n})$ with the properties $$G^{-1} AP_1 =P_1,\quad G^{-1} BP_2 =P_2,\quad AG^{-1} Q_1 =Q_1,\quad BG^{-1} Q_2 =Q_2.$$
Obviously, the projectors can be calculated by using residues: $$\label{ProjRes}
P_1 = \mathop{Res }\limits_{\mu =0}\left(\frac{(A+ \mu B)^{-1} A}{\mu} \right),\quad P_2 =I-P_1,\quad
Q_1 =\mathop{Res }\limits_{\mu =0}\left(\frac{A(A+ \mu B)^{-1}}{\mu} \right),\quad Q_2 =I-Q_1.$$
With respect to the decomposition any vector $x\in {{\mathbb R}^n}$ can be uniquely represented in the form $$\label{xr}
x=x_{p_1} +x_{p_2},\qquad x_{p_1} =P_1 x\in X_1,\quad x_{p_2} =P_2 x\in X_2.
$$
Applying the spectral projectors $Q_1$, $Q_2$ to the DAE and taking into account their properties, we obtain the equivalent system $$\label{systDAE1}
\begin{split}
\frac{d}{dt} [AP_1 x]+BP_1 x & = Q_1 f(t,x), \\
BP_2 x & = Q_2 f(t,x).
\end{split}$$ In the equations of the system we restrict the operators to the subspaces $X_1$, $X_2$ from . Taking into account the invertibility of the induced operators $A_1$, $B_2$ and the representation , the system can be rewritten in the form $$\begin{aligned}
\frac{d}{dt} x_{p_1}+ A_1^{-1} B_1 x_{p_1}& = A_1^{-1} Q_1 f(t, x_{p_1} +x_{p_2}), \\
x_{p_2} & =B_2^{-1} Q_2 f(t, x_{p_1} +x_{p_2}).\end{aligned}$$ Thus, the spectral projectors allow one to reduce the original semilinear DAE to the equivalent system of purely differential and purely algebraic equations (to the semi-explicit form).
Since the projectors are easily computed with the help of on a computer, and, consequently, the operator $G$ are easily computed, to construct the numerical methods it is more convenient to use $G^{-1}\in L({{\mathbb R}^n})$. Using $G^{-1}$, we can write the system (equivalent to the DAE ) as $$\label{prThreg1}
\begin{split}
\frac{d}{dt} [P_1 x]+G^{-1} BP_1 x & = G^{-1} Q_1 f(t,P_1x+P_2x), \\
P_2 x & = G^{-1} Q_2 f(t,P_1x+P_2x).
\end{split}$$ Thus, the possibility to easily compute the spectral projectors enables to numerically solve the DAE directly in the form , i.e., to apply the developed numerical methods it is not necessary to carry out additional analytical transformations.
The existence and uniqueness of global solutions {#GlobSolv}
================================================
Recall that a function $x \in C([t_0,t_1),\, {{\mathbb R}^n})$ is a solution of the initial value problem , on the interval $[t_0,t_1)$ ($t_1\le \infty $) if $Ax \in C^1([t_0,t_1),{{\mathbb R}^n})$, $x$ satisfies the equation on $[t_0,t_1)$ and the initial condition . The solution is global if it exists on the whole interval $[t_0,\infty )$.
To formulate the theorems given below, we need the following definition.
[@RF1; @Fil.MPhAG] An operator function (a mapping) $\Phi \colon D\to L(W,Z)$, where $W$, $Z$ are $s$-dimensional linear spaces and $D\subset W$, is called *basis invertible* on an interval $[\Hat w, \Tilde w]$, where $\Hat w,\, \Tilde w\in D$, if for some additive resolution of the identity $\{\Theta _k\}_{k=1}^s$ in the space $Z$ (see [@Fil.MPhAG Definition 2.2] or [@RF1 Definition 2]) and for any set of vectors $\{w^k\}_{k=1}^s \subset [\Hat w, \Tilde w]$ the operator $\Lambda {=\nobreak\discretionary{}{\hbox{$\mathsurround=0pt =$}}{}}\sum\limits_{k=1}^s \Theta _k \Phi (w^k) {\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} L(W,Z)$ has the inverse operator $\Lambda ^{-1} \in L(Z,W)$.
In this paper, we use for convenience the term “interval $[\Hat w, \Tilde w]$” instead of the term “convex hull $conv\{\Hat w, \Tilde w\}$ of vectors $\Hat w,\, \Tilde w$” used in the papers [@RF1; @Fil.MPhAG], taking into account that $conv\{\Hat w, \Tilde w\}=[\Hat w, \Tilde w]=\{\alpha \Tilde w + (1-\alpha)\Hat w \mid \alpha \in [0,1]\}$.
Note that the property of basis invertibility does not depend on the choice of an additive resolution of the identity in $Z$. Also note that if the operator function $\Phi $ is basis invertible on $[\Hat w, \Tilde w]$, then it is invertible on $[\Hat w, \Tilde w]$, i.e., for each point $w^*\in [\Hat w, \Tilde w]$ its image $\Phi (w^*)$ under the mapping $\Phi $ is an invertible continuous linear operator from $W$ into $Z$. The converse statement is not true unless the spaces $W$, $Z$ are one-dimensional (see [@Fil.MPhAG Example 2.1]).
Below we use the projectors $P_1$, $P_2$, $Q_1$, $Q_2$ (see or ) and the subspaces $X_1$, $X_2$, $Y_1$, $Y_2$ (see ) discussed in section \[IndexMeth\]. Recall that $L_0 = \{(t,x) \in [0,\infty) \times {{\mathbb R}^n}\mid Q_2[Bx-f(t,x)] = 0\}$ .
\[ThRF1\] Let $f\in C([0,\infty)\times {{\mathbb R}^n}, {{\mathbb R}^n})$, $\displaystyle\frac{\partial f}{\partial x} \in C([0,\infty)\times {{\mathbb R}^n}, L({{\mathbb R}^n}))$, and $\lambda A+B$ be a regular pencil of index not higher than 1. Assume that for any $t\ge 0$ and any $x_{p_1}\in X_1$ there exists $x_{p_2} \in X_2$ such that $$\label{soglreg2}
(t,x_{p_1}+x_{p_2})\in L_0,$$ and for any $x^i_{p_2} \in X_2$ such that $(t_*, x_{p_1}^* + x^i_{p_2}) \in L_0$, $i=1,2$, the operator function $$\label{funcPhi}
\Phi \colon X_2 \to L(X_2,Y_2),\quad \Phi (x_{p_2})=\left[\frac{\partial Q_2 f}{\partial x} (t_*,x_{p_1}^* +x_{p_2})-B\right] P_2,$$ is basis invertible on $[x^1_{p_2},\,x^2_{p_2}]$. Assume that there exists a self-adjoint positive operator $H\in L(X_1)$ and for each $T>0$ there exists a number $R_T >0$ such that $$\label{ineqreg}
(H P_1 x,G^{-1}Q_1f(t,x))\le 0\quad \text{ for all } (t,x)\in L_0 \text{ such that } 0\le t\le T,\; \|P_1 x\| \ge R_T.$$ Then for each initial point $(t_0,x_0)\in L_0$ there exists a unique global solution $x(t)$ of the initial value problem , .
The proof of the theorem is carried out in the same way as the proof of Theorem 1 from [@RF1]. If the DAE has the regular pencil of index 0 ($A$ is nondegenerate), then it can be reduced to an ordinary differential equation. In this case, $Q_2 = P_2 = 0$, $Q_1 = P_1 = I$ and , are absent [@Fil].
\[conseq-ThRF1\] Assume that in Theorem the projection $Q_1 f$ admits the representation $$\label{consRF1-1}
Q_1 f(t,x)=S_1(t)P_1 x+\psi (t,x) + \Pi (x)e(t),$$ where $S_1\in C([0,\infty),L(X_1, Y_1))$, $\psi \in C([0,\infty)\times {{\mathbb R}^n}, Y_1)$, $\displaystyle\frac{\partial \psi }{\partial x} \in C([0,\infty)\times {{\mathbb R}^n}, L({{\mathbb R}^n},Y_1))$, $e\in C([0,\infty),{{\mathbb R}^n})$, $\Pi\in C^1 ({{\mathbb R}^n},L({{\mathbb R}^n},Y_1))$ and there exist numbers $C,\, r >0$ such that $$\| \Pi (x)\| \le C\quad \text{ for all } \| P_1 x\| \ge r.$$ Then Theorem remains valid if instead of the following condition is satisfied: $$\label{consRF1-3}
(HP_1 x,G^{-1} \psi (t,x))\le 0\quad \text{ for all }(t,x) \in L_0 \text{ such that } 0\le t\le T,\; \| P_1 x\| \ge R_T.$$
The proof of the corollary is similar to the proof of Theorem 1 from [@RF1].
[@Fil.num_meth; @Fil.MPhAG] A solution $x(t)$ of the initial value problem , is called *Lagrange stable* if it is global and bounded, i.e., the solution $x(t)$ exists on $[t_0,\infty )$ and $\mathop{\sup }\limits_{t\in [t_0,\infty )} \| x(t)\|<+\infty$.
*The equation is Lagrange stable* if every solution of the initial value problem , is Lagrange stable.
The theorem on the Lagrange stability of the semilinear DAE [@Fil.MPhAG] is given below.
\[Th\_Ust1\] Let $f\in C([0,\infty)\times {{\mathbb R}^n}, {{\mathbb R}^n})$, $\displaystyle\frac{\partial f}{\partial x} \in C([0,\infty)\times {{\mathbb R}^n}, L({{\mathbb R}^n}))$, $\lambda A+B$ be a regular pencil of index not higher than 1 and be fulfilled. Let for any $x^i_{p_2} {\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} X_2$ such that $(t_*, x_{p_1}^* + x^i_{p_2}){\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} L_0$, $i=1,2$, the operator function be basis invertible on $[x^1_{p_2},\,x^2_{p_2}]$. Assume that for some self-adjoint positive operator $H \in L(X_1)$ and some number $R>0$ there exist functions $k\in C([0,\infty),{\mathbb R})$, $U{\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} C((0,\infty), (0,\infty))$ such that $\displaystyle\int\limits_c^{+\infty} \frac{dv}{U(v)} =+\infty $ $(c>0)$ and $$\label{Lagr1}
\big(HP_1 x,G^{-1}[-BP_1 x +Q_1 f(t,x)]\big)\le k(t)\, U\big({\textstyle \frac{1}{2}}(HP_1x,P_1x)\big)\, \text{ for all } (t,x)\!\in\! L_0 \text{ such that } \| P_1 x\|\ge R.$$ Then for each initial point $(t_0,x_0)\in L_0$ there exists a unique global solution $x(t)$ of the initial value problem , .
If, additionally, $$\int\limits_{t_0}^{+\infty} k(t)\, dt <+\infty,$$ and there exists $\Tilde{x}_{p_2} \in X_2$ such that for any $x_{p_2}^*\in X_2$ satisfying $(t_*, x_{p_1}^* + x_{p_2}^*)\in L_0$ the operator function is basis invertible on $(\Tilde x_{p_2},x_{p_2}^*]$, and $$\label{LagrA1}
\mathop{\sup }\limits_{t\in [0,\infty ),\: \| x_{p_1}\|\le M} \|Q_2 f(t,x_{p_1}+ \tilde{x}_{p_2})\| < +\infty,\quad M>0\; (M\in {\mathbb R}),$$ then for the initial points $(t_0,x_0)\in L_0$ the equation is Lagrange stable.
The proof is carried out in the same way as the proof of Theorem 3.1 from [@Fil.MPhAG].
\[RemDiff\] The solution $x(t)$ of the initial value problem , is such that $P_1 x(t)\in C^1([t_0,\infty), X_1)$ and $P_2 x(t)\in C([t_0,\infty),X_2)$. If in Theorems \[ThRF1\], \[Th\_Ust1\] $f\in C^m([0,\infty)\times {{\mathbb R}^n}, {{\mathbb R}^n})$, $m\in {\mathbb N}$, then the solution $x(t)$ is such that $P_1 x(t)\in C^{m+1}([t_0,\infty), X_1)$ and $P_2 x(t)\in C^m([t_0,\infty),X_2)$.
The theorem on the Lagrange instability of the semilinear DAE , which gives conditions for the existence and uniqueness of solutions with a finite escape time (the solutions are blow-up in finite time) [@Fil.MPhAG Definition 2.4], is also proved in [@Fil.MPhAG] (Theorem 4.1). Thus, this theorem gives the conditions under which the initial value problem , does not have global solutions.
The combined methods and their convergence {#NumMs}
==========================================
We denote by $z=x_{p_1}$, $u=x_{p_2}$ the components of a vector $x=z+u\in {{\mathbb R}^n}$ ($z$, $u$ are the projections of the vector $x$ onto subspaces $X_1$, $X_2$ from ). We seek a solution of the initial value problem , on an interval $[t_0,T]$. Introduce the uniform mesh $\omega_h=\{t_i=t_0+ih,\; i=0,...,N\}$ with the step size $h=(T-t_0)/N$. The values of an approximate solution of the problem , at the points $t_i$ are denoted by $x_i = z_i+u_i$, $i=0,...,N$ ($z_i=P_1 x_i$, $u_i=P_2 x_i$).
Initial values $z_0$, $u_0$ are chosen so that the consistency condition $B u_0 = Q_2 f(t_0,z_0+u_0)$, equivalent to the condition $(t_0,x_0)\in L_0$ of Theorems \[ThRF1\], \[Th\_Ust1\], is satisfied.
Recall that the spectral projectors $P_1$, $P_2$, $Q_1$, $Q_2$ and the auxiliary operator $G$ are easily computed by the formulas and .
Method 1 {#Num-meth}
--------
\[ThNum-meth\] Let the conditions of Theorem or Corollary or the conditions of part of Theorem be satisfied. Let, additionally, $f\in C^1([t_0,T]\times {{\mathbb R}^n}, {{\mathbb R}^n})$, and the operator function $\Phi $ where $t_*=t$, $x_{p_1}^*=z$, $x_{p_2}=u$ be invertible for any points $(t,z+u)\in [t_0,T]\times {{\mathbb R}^n}$. Then the method $$\begin{aligned}
&x_0 = z_0 +u_0, \label{met1}\\
&z_{i+1}=(I-h G^{-1} B) z_i+h G^{-1} Q_1 f(t_i,z_i+u_i), \label{met2}\\
&u_{i+1}=u_i-\bigg[I - G^{-1}{\displaystyle \frac{\partial Q_2 f}{\partial x}} (t_{i+1},z_{i+1}+u_i)\bigg]^{-1}\! \big[u_i- G^{-1}Q_2f(t_{i+1},z_{i+1}+u_i)\big], \label{met3} \\
&x_{i+1}=z_{i+1}+u_{i+1},\quad i=0,...,N-1, \label{met4}\end{aligned}$$ approximating the initial value problem , on $[t_0,T]$, converges and has the first order of accuracy $\mathop{\max }\limits_{0\le i\le N} \|z(t_i)-z_i\|=O(h)$, $\mathop{\max }\limits_{0\le i\le N} \|u(t_i)-u_i\|=O(h)$.
By virtue of the theorem conditions, there exists a unique global (exact) solution $x(t)$ of the initial value problem , such that $z(t)=P_1 x(t){\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} C^2([t_0,T],X_1)$ and $u(t)=P_2 x(t)\in C^1([t_0,T],X_2)$.
The equation is equivalent to the system which, with the new notations, is written in the form $$\begin{aligned}
& \frac{dz}{dt} +G^{-1} Bz =G^{-1} Q_1 f(t,z+u), \label{num1}\\
& u =G^{-1} Q_2 f(t,z+u). \label{num2}\end{aligned}$$ Using the Taylor formula, we obtain $$\begin{aligned}
&\frac{dz}{dt}(t)=\frac{z(t+h)-z(t)}{h}+O(h), \label{Taylor1}\\
\begin{split} G^{-1} Q_2 f(t+h,z(t + h)+u(t + h)) = G^{-1} Q_2 f(t + h,z(t + h) + u(t)) +\\
+G^{-1} \frac{\partial Q_2 f}{\partial x}(t+h,z(t+h)+u(t)) [u(t+h)-u(t)]+O(h). \end{split} \label{Taylor2}\end{aligned}$$ From the theorem conditions and $$-G^{-1}\bigg[\frac{\partial Q_2 f}{\partial x}(t,z+u)-B\bigg]P_2 = \bigg[I - G^{-1} \frac{\partial Q_2 f}{\partial x} (t,z+u)\bigg]P_2=\bigg[I - G^{-1} \frac{\partial Q_2 f}{\partial x} (t,z+u)\bigg]\bigg|_{X_2}$$ it follows that there exists the inverse operator $\bigg[I - G^{-1} \displaystyle\frac{\partial Q_2 f}{\partial x} (t,z+u)\bigg]^{-1}\in L(X_2)$ for any points $(t,z+u)$ from $L_0$ and $[t_0,T]\times {{\mathbb R}^n}$. Using , and taking into account the equivalence of the equation and the system , , we can write the problem , at the points of the introduced mesh $\omega_h$ as $$\begin{aligned}
& x(t_0) = z(t_0)+u(t_0),\quad z(t_0) = z_0,\; u(t_0)=u_0, \label{exact1}\\ & z(t_{i+1})=(I-h G^{-1} B) z(t_i)+h G^{-1} Q_1 f(t_i,z(t_i)+u(t_i))+O(h^2), \label{exact2}\\
\begin{split} u(t_{i+1}) = u(t_i)- \bigg[I- G^{-1} \frac{\partial Q_2 f}{\partial x}(t_{i+1},z(t_{i+1})+u(t_i))\bigg]^{-1}\! \big[u(t_i)- \\
-G^{-1}Q_2 f(t_{i+1},z(t_{i+1})+u(t_i))\big]+ O(h),
\end{split} \label{exact3} \\
& x(t_{i+1})=z(t_{i+1})+u(t_{i+1}),\quad i=0,...,N-1, \label{exact4}\end{aligned}$$ where follows from $$\begin{split}
&u(t_{i+1})= G^{-1}Q_2 f(t_{i+1},z(t_{i+1})+u(t_i))+{} \smallskip \\
&{}+ G^{-1} \frac{\partial Q_2 f}{\partial x} (t_{i+1},z(t_{i+1})+u(t_i))[u(t_{i+1})-u(t_i)]+O(h). \label{exact3*}
\end{split}$$ The corresponding numerical method takes the form – [@Fil.num_meth], where is obtained from $$u_{i+1} = G^{-1}Q_2 f(t_{i+1},z_{i+1}+u_i) + G^{-1} \frac{\partial Q_2 f}{\partial x}(t_{i+1},z_{i+1} + u_i)[u_{i+1}-u_i]. \label{met3*}$$
Since $\frac{\partial f}{\partial x}(t,x)$ is continuous on $[0,\infty)\times {{\mathbb R}^n}$, then there exists the constant $$\label{M1}
M_1=\mathop{\sup }\limits_{0<\theta_1 <1} \bigg\|G^{-1} \frac{\partial Q_1 f}{\partial x}(t_i,x_i + \theta_1(x(t_i)-x_i))\bigg\|$$ such that (by the formula of finite increments) $$\|G^{-1} Q_1[f(t_i,z(t_i)+u(t_i))- f(t_i,z_i+u_i)]\|\le M_1 \big(\|z(t_i)-z_i\|+ \|u(t_i)-u_i\|\big).$$ Then using , we obtain the estimate $$\label{num3}
\|z(t_{i+1})- z_{i+1}\|\le \big(\|I-h G^{-1} B\| + h M_1\big) \|z(t_i)- z_i\|+ h M_1\|u(t_i)-u_i\|+ O(h^2).$$
Denote $\varepsilon ^z_{i+1} = \|z(t_{i+1})-z_{i+1}\|$, $\varepsilon ^u_{i+1}= \|u(t_{i+1})-u_{i+1}\|$, $g(h) = \|I-h G^{-1} B\| +h M_1$. Then takes the form $$\label{num6}
\varepsilon ^z_{i+1}\le g(h)\, \varepsilon ^z_i+h M_1\, \varepsilon ^u_i +O(h^2).$$ It follows from the initial condition and , that $\varepsilon ^z_0= 0$, $\varepsilon ^u_0 = 0$ and $\varepsilon ^z_1= O(h^2)$. Using , we find recurrently the estimate $$\label{num8}
\varepsilon ^z_{i+1}\le h M_1 \sum\limits_{j=0}^i g^{i-j}(h) \varepsilon ^u_j+ O(h^2)\sum\limits_{j=0}^i g^j(h).$$ Since $g^j(h) \le (1+h(\|G^{-1} B\|+M_1))^j {\le\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \le$}}{}} e^{(T-t_0)(\|G^{-1} B\| + M_1)}$, $j=1,...,N$, then $$\label{num10}
\varepsilon ^z_{i+1}\le O(h) \sum\limits_{j=0}^i \varepsilon ^u_j+ O(h),\quad i=0,...,N-1.$$
Further, it follows from , that $$\begin{gathered}
u(t_{i+1})-u_{i+1}=G^{-1}\Big[ Q_2 f(t_{i+1},z(t_{i+1}) +u(t_i))
- Q_2 f(t_{i+1},z_{i+1}+u_i) +\\
+ \frac{\partial Q_2f}{\partial x} (t_{i+1},z(t_{i+1})+u(t_i)) [u(t_{i+1})-u(t_i)]- \frac{\partial Q_2f}{\partial x} (t_{i+1},z_{i+1}+u_i) [u_{i+1}-u_i]\Big] +O(h).\end{gathered}$$ Hence, we easy obtain $$\begin{gathered}
u(t_{i+1}) - u_{i+1} = \left[I-G^{-1} \frac{\partial Q_2 f}{\partial x}(t_{i+1},z_{i+1}+u_i) \right]^{-1} \Big[ G^{-1}\frac{\partial Q_2f}{\partial x} (t_{i+1},z(t_{i+1}) + u(t_i)) [z(t_{i+1})-\\
-z_{i+1}] - G^{-1}\frac{\partial Q_2f}{\partial x} (t_{i+1},z_{i+1} + u_i) [u(t_{i+1}) - u_i]+O(\|z(t_{i+1}) - z_{i+1}+u(t_i) - u_i \|)+O(h)\Big].\end{gathered}$$ Denote $$C_1 = \mathop{\sup }\limits_{0\le i\le N-1} \left\|G^{-1}\frac{\partial Q_2f}{\partial x} (t_{i+1},z(t_{i+1})+u(t_i))\right\| ,\;
C_2 = \mathop{\sup }\limits_{0\le i\le N-1} \left\|G^{-1}\frac{\partial Q_2f}{\partial x} (t_{i+1},z_{i+1}+u_i)\right\| ,$$ $$\label{C3}
C_3=\mathop{\sup }\limits_{0\le i\le N-1}\left\|\left[I-G^{-1}\frac{\partial Q_2f}{\partial x} (t_{i+1},z_{i+1}+u_i)\right]^{-1}\right\|.$$ Then $$\varepsilon ^u_{i+1}\le C_3 [ C_1 \varepsilon ^z_{i+1}+C_2\varepsilon ^u_i+ (C_2+1)O(h)+O(\varepsilon ^z_{i+1}+\varepsilon ^u_i)] .$$ Consequently, there exist constants $\alpha $, $\beta $ such that $$\label{num5}
\varepsilon ^u_{i+1}\le \alpha \varepsilon ^z_{i+1}+ \beta \varepsilon ^u_i +O(h),\quad i=0,...,N-1.$$ From we obtain that $\varepsilon ^u_1\le O(h)$ and, taking into account , $$\varepsilon ^u_{i+1}\le O(h) \sum\limits_{j=1}^i \varepsilon ^u_j+ \beta \varepsilon ^u_i +O(h),\quad i=1,...,N-1.$$ By using the method of mathematical induction, it is easy to prove that $$\varepsilon ^u_{i+1}\le O(h),\quad i=0,...,N-1.$$ Then from it follows that $$\varepsilon ^z_{i+1}\le O(h),\quad i=0,...,N-1.$$ Thus, $\mathop{\max }\limits_{0\le i\le N} \varepsilon ^u_i = O(h)$ and $\mathop{\max }\limits_{0\le i\le N} \varepsilon ^z_i = O(h)$. Hence, $\mathop{\max }\limits_{0\le i\le N} \left\| \left( \begin{array}{c} z(t_i)-z_i \\ u(t_i)-u_i \end{array} \right) \right\| {\le\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \le$}}{}} K h$, where $K >0$ is some constant, and the method – converges and has the first order of accuracy.
\[remNum-meth\] If in Theorem \[ThNum-meth\] we do not require the additional smoothness of the function $f$, i.e., we assume that $f\in C([0,\infty)\times {{\mathbb R}^n}, {{\mathbb R}^n})$ and $\displaystyle\frac{\partial f}{\partial x} \in C([0,\infty)\times {{\mathbb R}^n}, L({{\mathbb R}^n}))$, then the method – converges, but may not have the first order of accuracy: $\mathop{\max }\limits_{0\le i\le N} \|z(t_i)-z_i\|=o(1)$ (i.e., $\mathop{\max }\limits_{0\le i\le N} \|z(t_i)-z_i\|\to 0$), $h\to 0$, $\mathop{\max}\limits_{0\le i\le N} \|u(t_i)-u_i\|=o(1)$, $h\to 0$.
The proof is carried out in the same way as the proof of Theorem \[ThNum-meth\], where instead of , we use the representations $$\label{Taylor1o1}
\frac{dz}{dt}(t)=\frac{z(t+h)-z(t)}{h}+o(1),\; h\to 0,$$ $$\label{Taylor2o1}
\begin{split}
G^{-1} Q_2 f(t+h,z(t + h)+u(t + h)){=\nobreak\discretionary{}{\hbox{$\mathsurround=0pt =$}}{}} G^{-1} Q_2 f(t + h,z(t+h)+u(t))+ \\
+G^{-1} \frac{\partial Q_2 f}{\partial x}(t+h,z(t+h)+u(t)) [u(t+h)-u(t)]{+\nobreak\discretionary{}{\hbox{$\mathsurround=0pt +$}}{}}o(1),\; h\to 0.\end{split}$$
Method 2 (modified method 1) {#ModNum-meth}
----------------------------
\[ThModNum-meth\] Let the conditions of Theorem or Corollary or the conditions of part of Theorem . Let, additionally, $f\in C^2([t_0,T]\times {{\mathbb R}^n}, {{\mathbb R}^n})$, and the operator function $\Phi $ where $t_*=t$, $x_{p_1}^*=z$, $x_{p_2}=u$ be invertible for any points $(t,z+u)\in [t_0,T]\times {{\mathbb R}^n}$. Then the method $$\begin{aligned}
&x_0 = z_0 +u_0,\quad z_1=(I-h G^{-1} B) z_0+h G^{-1} Q_1 f(t_0,z_0+u_0), \label{Impmet1}\\
&u_{i+1}=u_i-\!\bigg[I - G^{-1}{\displaystyle \frac{\partial Q_2 f}{\partial x}} (t_{i+1},z_{i+1}+u_i)\bigg]^{-1}\! \big[u_i- G^{-1}Q_2f(t_{i+1},z_{i+1}+u_i)\big],\, i=0,...,N-1, \label{Impmet3}\\ &z_{i+1}=z_{i-1} + 2h G^{-1}[Q_1 f(t_i,z_i+u_i) -B z_i],\; i=1,...,N-1, \label{Impmet2}\\
&x_{i+1}=z_{i+1}+u_{i+1},\; i=0,...,N-1, \label{Impmet4}\end{aligned}$$ approximating the initial value problem , on $[t_0,T]$, converges and has the second order of accuracy $\mathop{\max }\limits_{0\le i\le N} \|z(t_i)-z_i\|=O(h^2)$, $\mathop{\max }\limits_{0\le i\le N} \|u(t_i)-u_i\|=O(h^2)$.
By virtue of the theorem conditions, there exists a unique global (exact) solution $x(t)$ of the initial value problem , such that $z(t){=\nobreak\discretionary{}{\hbox{$\mathsurround=0pt =$}}{}}P_1 x(t){\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} C^3([t_0,T],X_1)$ and $u(t)=P_2 x(t){\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} C^2([t_0,T],X_2)$.
Taking into account the smoothness of the solution and the function $f$, we obtain $$\begin{aligned}
&\frac{dz}{dt}(t)=\frac{z(t+h)-z(t-h)}{2h}+O(h^2), \label{Taylor3}\\
\begin{split} G^{-1} Q_2 f(t+h,z(t + h)+u(t + h)) = G^{-1} Q_2 f(t + h,z(t + h) + u(t)) +\\
+G^{-1} \frac{\partial Q_2 f}{\partial x}(t+h,z(t+h)+u(t)) [u(t+h)-u(t)]+O(h^2).\end{split} \label{Taylor4}\end{aligned}$$
As shown in the proof of Theorem \[ThNum-meth\], there exists the inverse operator $\bigg[I - G^{-1} \displaystyle\frac{\partial Q_2 f}{\partial x} (t,z+u)\bigg]^{-1}\in L(X_2)$ for any points $(t,z+u)$ from $L_0$ and $[t_0,T]\times {{\mathbb R}^n}$. Using , , and the system , equivalent to the equation , we can write the problem , at the points of the introduced mesh $\omega_h$ as $$\begin{aligned}
\begin{split}
& x(t_0) = z(t_0)+u(t_0),\; z(t_0) = z_0,\; u(t_0)=u_0, \\
& z(t_1)=(I-h G^{-1} B) z_0+h G^{-1} Q_1 f(t_0,z_0+u_0)+O(h^2),
\end{split}\label{Impexact1}\\
\begin{split}
u(t_{i+1}) = u(t_i)- \bigg[I- G^{-1} \frac{\partial Q_2 f}{\partial x}(t_{i+1},z(t_{i+1})+u(t_i))\bigg]^{-1}\! \big[u(t_i)- \\
-G^{-1}Q_2 f(t_{i+1},z(t_{i+1})+u(t_i))\big]+ O(h^2),\; i=0,...,N-1,
\end{split} \label{Impexact3} \\
& z(t_{i+1})=z(t_{i-1}) + 2h G^{-1}[Q_1 f(t_i,z(t_i)+u(t_i)) - B z(t_i)] + O(h^3),\; i = 1,...,N-1, \label{Impexact2}\\
& x(t_{i+1})=z(t_{i+1})+u(t_{i+1}),\; i = 0,...,N-1, \label{Impexact4}\end{aligned}$$ The corresponding numerical method takes the form –, where $z_1$ coincides with $z_1$ from and is obtained from .
Denote $\varepsilon ^z_{i+1}= \|z(t_{i+1})-z_{i+1}\|$, $\varepsilon ^u_{i+1}= \|u(t_{i+1})-u_{i+1}\|$. Obviously, $\varepsilon ^z_0= 0$, $\varepsilon ^u_0 = 0$, $\varepsilon ^z_1= O(h^2)$, and there exists the constant $M_1$ such that $$\label{epsz1}
\varepsilon ^z_{i+1}\le \varepsilon ^z_{i-1}+2h(\|G^{-1} B\| +M_1)\varepsilon ^z_i+ 2 h M_1\, \varepsilon ^u_i +O(h^3).$$
Consider the system consisting of the inequality and equality $\varepsilon ^z_i=\varepsilon ^z_i$. Denoting $$\Hat\varepsilon ^z_{i+1}=\begin{pmatrix} \varepsilon ^z_{i+1} \\ \varepsilon ^z_i\end{pmatrix},\quad e_1 =\begin{pmatrix} 1 \\ 0\end{pmatrix},\quad \Hat g(h)=\begin{pmatrix} 2h(\|G^{-1} B\|+M_1) & 1 \\ 1 & 0 \end{pmatrix},$$ we obtain the representation of this system as $$\label{hatepsz}
\Hat\varepsilon ^z_{i+1}\le \Hat g(h) \Hat\varepsilon ^z_i+ 2h M_1 e_1 \varepsilon ^u_i +e_1 O(h^3),\quad i=1,...,N-1.$$ Using , we find recurrently the estimate $$\label{hatepsz2}
\Hat\varepsilon ^z_{i+1}\le \Hat g^i(h) \Hat\varepsilon ^z_1 +2 h M_1 \sum\limits_{j=1}^i \Hat g^{i-j}(h) e_1\varepsilon ^u_j+ O(h^3)\sum\limits_{j=1}^i \Hat g^{j-1}(h) e_1.$$ Define the norm of $\Hat\varepsilon ^z_{i+1}$ as $\|\Hat\varepsilon ^z_{i+1}\|=\max\{|\varepsilon ^z_{i+1}|, |\varepsilon ^z_i|\}$. Then $\|e_1\|=1$, $\|\Hat\varepsilon ^z_1\|=\varepsilon ^z_1=O(h^2)$, $\|\Hat g(h)\|=1+2h(\|G^{-1} B\| +M_1)$, $\varepsilon ^z_{i+1}\le \|\Hat\varepsilon ^z_{i+1}\|$ and consequently $$\label{num8new}
\varepsilon ^z_{i+1}\le 2 h M_1 \sum\limits_{j=1}^i \|\Hat g(h)\|^{i-j} \varepsilon ^u_j+ O(h^3)\sum\limits_{j=1}^i \|\Hat g(h)\|^{j-1}+ O(h^2).$$ Since $\|\Hat g(h)\|^j \le e^{2(T-t_0)(\|G^{-1} B\| + M_1)}$, $j=1,...,N-1$, then $$\label{epsz2}
\varepsilon ^z_{i+1}\le O(h) \sum\limits_{j=1}^i \varepsilon ^u_j+ O(h^2),\quad i=1,...,N-1.$$
Further, it follows from , that $$\begin{gathered}
u(t_{i+1})-u_{i+1}=G^{-1} \Big[Q_2f(t_{i+1},z(t_{i+1}) +u(t_i))
- Q_2f(t_{i+1},z_{i+1}+u_i) +\\
+ \frac{\partial Q_2f}{\partial x} (t_{i+1},z(t_{i+1})+u(t_i)) [u(t_{i+1})-u(t_i)]- \frac{\partial Q_2f}{\partial x} (t_{i+1},z_{i+1}+u_i) [u_{i+1}-u_i]\Big] +O(h^2).\end{gathered}$$ Hence, we easy obtain $$\begin{gathered}
u(t_{i+1})-u_{i+1}=\left[I-G^{-1} \frac{\partial Q_2 f}{\partial x}(t_{i+1},z_{i+1}+u_i) \right]^{-1}\Big[z(t_{i+1})-z_{i+1} + \\
+O(\|z(t_{i+1})-z_{i+1}+u(t_i)-u_i \|^2+\|z(t_{i+1})-z_{i+1}+u(t_i)-u_i \|\, \|u(t_{i+1})-u(t_i) \|)+O(h^2)\Big].\end{gathered}$$ Denote $C_3$ as . It is easy to verify that $$\varepsilon ^u_{i+1}\le 3C_3 \big[\varepsilon ^z_{i+1}+O\big((\varepsilon ^z_{i+1})^2+ (\varepsilon ^u_i)^2\big)+O(h^2)\big].$$ Taking into account , we easy obtain $$\varepsilon ^u_{i+1}\le O(h) \sum\limits_{l=1}^i \varepsilon ^u_l +O(h^2)\Big(\sum\limits_{l=1}^i \varepsilon ^u_l\Big)^2+\Hat C_3 (\varepsilon ^u_i)^2 + O(h^2),$$ where $\Hat C_3 $ is some constant. From the obtained estimate, using the polynomial (multinomial) theorem, we obtain $$\label{epsu1}
\varepsilon ^u_{i+1}\le O(h) \sum\limits_{l=1}^i \big(\varepsilon ^u_l+ (\varepsilon ^u_l)^2 \big) +\Hat C_3 (\varepsilon ^u_i)^2 + O(h^2),\quad i=1,...,N-1.$$ Since $\varepsilon ^u_1\le O(h^2)$, then, by using the method of mathematical induction, it is easy to prove that $$\varepsilon ^u_{i+1}\le O(h^2),\quad i=1,...,N-1.$$ From the obtained estimate and it follows that $$\varepsilon ^z_{i+1}\le O(h^2),\quad i=1,...,N-1.$$ Thus, $\mathop{\max }\limits_{0\le i\le N} \varepsilon ^u_i= O(h^2)$ and $\mathop{\max }\limits_{0\le i\le N} \varepsilon ^z_i= O(h^2)$. Hence, the method – converges and has the second order of accuracy.
\[remModNum-meth\] Analogously to Remark \[remNum-meth\], if in Theorem \[ThModNum-meth\] $f\in C([0,\infty)\times {{\mathbb R}^n}, {{\mathbb R}^n})$ and $\displaystyle\frac{\partial f}{\partial x} \in C([0,\infty)\times {{\mathbb R}^n}, L({{\mathbb R}^n}))$, then the method – converges: $\mathop{\max }\limits_{0\le i\le N} \|z(t_i)-z_i\|{=\nobreak\discretionary{}{\hbox{$\mathsurround=0pt =$}}{}}o(1)$, $h\to 0$, $\mathop{\max }\limits_{0\le i\le N} \|u(t_i)-u_i\|=o(1)$, $h\to 0$.
The proof is carried out in the same way as the proof of Theorem \[ThModNum-meth\], where instead of , , we use the representations , $\displaystyle\frac{dz}{dt}(t)=\frac{z(t+h)-z(t-h)}{2h}+o(1)$, $h\to 0$, and .
\[globBasInv\] If the operator function $\Phi$ is basis invertible on $[x^1_{p_2},\,x^2_{p_2}]$ for any $x^1_{p_2},\, x^2_{p_2} {\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} X_2$, $t_*\in [0,\infty)$, $x^*_{p_1}\in X_1$, then the additional condition of the invertibility of $\Phi$ for any $(t,z+u)\in [t_0,T]\times {{\mathbb R}^n}$ from Theorems \[ThNum-meth\], \[ThModNum-meth\], as well as the additional condition of the basis invertibility of $\Phi$ on $(\Tilde x_{p_2},x_{p_2}^*]$ from part of Theorem , are not needed.
Comparison of methods 1, 2 {#CompareMeth}
==========================
Consider the semilinear DAE , where $$\label{two-pole}
x = \begin{pmatrix} x_1 \\ x_2 \\ x_3 \end{pmatrix},\,
A=\begin{pmatrix} L & 0 & 0 \\ 0 & C & 0 \\ 0 & 0 & 0 \end{pmatrix},\, B = \begin{pmatrix} 0 & 1 & r \\ 0 & g & -1 \\ 0 & 1 & r \end{pmatrix},\,
f(t,x)=\begin{pmatrix} e(t)-\varphi_0(x_1)-\varphi(x_3) \\ -h(x_2) \\ \psi(x_1-x_3)-\varphi(x_3) \end{pmatrix}.$$ It is assumed that $L,\, C,\, r,\, g>0$, $x\in {\mathbb R}^3$, $e \in C([0,\infty ),{\mathbb R})$, and $\varphi _0 $, $\varphi $, $\psi$, $h \in C^1 ({\mathbb R})$. The physical interpretation of this equation and the corresponding parameters, functions and variables $x_1 =I_L$, $x_2 =U_C$, $x_3 =I$ is given in section \[Appl\]. Let $L = 5\cdot 10^{-4}$, $C = 5\cdot 10^{-7}$, $r = 2$, $g =0.2$, $e(t) = \sin t$, $\varphi_0(x_1) = x_1^3$, $\varphi(x_3)= x_3^3$, $\psi(x_1-x_3)= (x_1-x_3)^3$, and $h(x_2)= x_2^3$. Obviously, the initial data $t_0 = 0$, $x_0 =(0,0,0)^T$ satisfy the consistency condition $(t_0,x_0)\in L_0$, that is, $B P_2 x_0 = Q_2 f(t_0,x_0)$. Below, it is shown how the plot of a solution $x(t) = (I_L(t),U_C(t),I(t))^T$ changes with the mesh refinement ($h=0.1,\,0.01,\,0.001,\,0.0001$).
{width="8cm"} {width="8cm"}
![The components of the numerical solution obtained by method 1[]{data-label="modSinCom"}](modSinCom3.eps){width="8cm"}
{width="8cm"} {width="8cm"}
![The components of the numerical solution obtained by method 2 (modified method 1)[]{data-label="modSinComNew"}](modSinComNew3.eps){width="8cm"}
The plots in Figure \[modSinCom\] were obtained using the method – (method 1), while the plots in Figure \[modSinComNew\] were obtained using the method – (method 2). For greater clarity, the first plots in Figures \[modSinCom\], \[modSinComNew\] (the plots of the solution component $I_L(t)$) are shown on an enlarged scale in Figure \[modSinLarge1\].
![The comparison of the first plots in Figures \[modSinCom\] (left) and \[modSinComNew\] (right)[]{data-label="modSinLarge1"}](modSinCom1+.eps "fig:"){width="7cm"} ![The comparison of the first plots in Figures \[modSinCom\] (left) and \[modSinComNew\] (right)[]{data-label="modSinLarge1"}](modSinComNew1+.eps "fig:"){width="7cm"}
The figures show that method 2 (modified method 1) converges faster to the exact solution.
However, when choosing a more suitable method for solving a certain problem, one should also take into account the quantitative characteristics of stability of the methods, namely, the values of the coefficients $h M_1$, $g(h)=\|I-h G^{-1} B\| +h M_1\le 1+h(\|G^{-1} B\| +M_1)$ in (for method 1) and $2h M_1$, $\|\Hat g(h)\|=1+2h(\|G^{-1} B\| +M_1)$ in (for method 2) and the length of the interval $[t_0,T]$ on which the computation is performed. The larger the quantitative characteristic of stability of a method (the quantities characterizing the stability of the method), the smaller step size should be chosen to achieve the required accuracy of computations. In turn, the choice of a smaller step size leads to larger roundoff errors. In this connection, to carry out a computation with a given accuracy using method 2, it may be necessary to choose a much smaller step size and consequently to spend much more time than using method 1. The differences in the mentioned coefficients from and arise because the derivative is approximated by a forward difference, obtained from the representation , in method 1 and by a centered difference, obtained from , in method 2. Similar differences arising when using such approximations for the linear differential equation $\dot{x}+Ax=f(t)$ are described in detail in [@G-R Chapter 5]. It is clear that for small values of $\|G^{-1} B\|$, $M_1$ or a sufficiently small interval $[t_0,T]$ there are no distinct differences in stability.
In particular, for the considered DAE , , the larger $L$, $C$, $r$ and the smaller $g$, $M_1$, the less differences in stability. Also, the differences in stability are small on a sufficiently small of time intervals $[t_0,T]$ and Figures \[modSinCom\], \[modSinComNew\] confirm this. If we compute the solution for the above quantities and functions on a larger time interval, then method 2 starts to lose stability. This is not visible in Figure \[mSinC1\], but this is already noticeable in Figure \[mSinC2-3\]. Therefore, it is necessary to decrease the step size, as shown in Figure \[mSinC4h2-3\].
![The component $I_L(t)$ of the solution obtained by method 1 (left) and method 2 (right); the step size $h=0.001$[]{data-label="mSinC1"}](mSinC3h_1.eps "fig:"){width="8cm"} ![The component $I_L(t)$ of the solution obtained by method 1 (left) and method 2 (right); the step size $h=0.001$[]{data-label="mSinC1"}](mSinCnew3h_1.eps "fig:"){width="8cm"}
![The components $U_C(t)$, $I(t)$ of the solution obtained by method 1 (left) and method 2 (right); the step size $h=0.001$[]{data-label="mSinC2-3"}](mSinC3h_2.eps "fig:"){width="8cm"} ![The components $U_C(t)$, $I(t)$ of the solution obtained by method 1 (left) and method 2 (right); the step size $h=0.001$[]{data-label="mSinC2-3"}](mSinCnew3h_2.eps "fig:"){width="8cm"}
![The components $U_C(t)$, $I(t)$ of the solution obtained by method 1 (left) and method 2 (right); the step size $h=0.001$[]{data-label="mSinC2-3"}](mSinC3h_3.eps "fig:"){width="8cm"} ![The components $U_C(t)$, $I(t)$ of the solution obtained by method 1 (left) and method 2 (right); the step size $h=0.001$[]{data-label="mSinC2-3"}](mSinCnew3h_3.eps "fig:"){width="8cm"}
![The components $U_C(t)$, $I(t)$ of the solution obtained by method 2; the step size $h=0.00001$[]{data-label="mSinC4h2-3"}](mSinCnew5h_2.eps "fig:"){width="8cm"} ![The components $U_C(t)$, $I(t)$ of the solution obtained by method 2; the step size $h=0.00001$[]{data-label="mSinC4h2-3"}](mSinCnew5h_3.eps "fig:"){width="8cm"}
However, for the larger value of $r=4$ (the value is doubled) and the smaller value of $g=0.1$ (the value is decreased by 2 times), method 2 is stable for the same step size and time interval that were originally (see Figure \[mSinC2new\]) and, accordingly, there are no visible differences in the stability of the methods.
![The components $U_C(t)$, $I(t)$ obtained by method 2 for the new $r$, $g$ and the original step size $h=0.001$[]{data-label="mSinC2new"}](mSinC3new3h_2.eps "fig:"){width="8cm"} ![The components $U_C(t)$, $I(t)$ obtained by method 2 for the new $r$, $g$ and the original step size $h=0.001$[]{data-label="mSinC2new"}](mSinC3new3h_3.eps "fig:"){width="8cm"}
*We can draw the following conclusions about the effectiveness of application of the methods in various situations.* When the parameters in the DAE are chosen arbitrarily, method 2 is better for computing or refining a solution on sufficiently small time intervals (or on the parts of a large interval), since it has a higher order of accuracy with respect to the step size $h$. For large time intervals, it is better to use method 1, since it is more stable (has a smaller coefficient characterizing stability). However, once again note that method 2 can be more effective on a large time interval if the parameters, which are chosen on the basis of the conditions of the problem being solved, do not have a large negative effect on the stability.
The application of the numerical methods to the analysis of mathematical models {#Appl}
===============================================================================
Differential-algebraic (descriptor, degenerate differential) equations have a wide range of practical applications. Such equations arise from the mathematical modeling of the dynamics of physical, economic, technical, ecological and other processes because of the availability of algebraic connections between the coordinates of the vectors of states of the corresponding dynamical systems. Semilinear DAEs are used in modeling the dynamics of complex mechanical and robotic systems (P.J. Rabier, W.C. Rheinboldt [@Rabier-Rh], B. Fox, L.S. Jennings, A.Y. Zomaya [@Fox-Jen-Zom], B. Siciliano, and others), interindustry balance (M. Morishima [@Morishima], S.R. Khachatryan, and others), various descriptor systems and neural networks (R. Riaza, J. Zufiria [@Riaza-Zufiria], P. Kunkel, V. Mehrmann [@Kunkel_Mehrmann], J.C. Engwerda, I.E. Wijayanti [@Engwerda-Salm-W], and others), transient processes in electrical circuits (R. Riaza [@Riaza], K.E. Brenan, S.L. Campbell, L.R. Petzold [@Brenan-C-P], R. Lamour, R. März, C. Tischendorf [@Lamour-Marz-Tisch], A.G. Rutkas, M.S. Filipkovska [@RF2], and others) and other objects and processes [@Vlasenko1].
A wide class of mathematical models of electrical circuits can be written with the help of the regular (i.e., with the regular pencil $\lambda A+B$) semilinear DAE $d/dt\,[Ax(t)]+Bx(t)=f(t,x)$ and the regular linear DAE $d/dt\,[Ax(t)]+Bx(t)=f(t)$ as a particular case (see, for example, [@Rutkas86; @Rutkas76], [@Lamour-Marz-Tisch section 3.6], [@Riaza section 5, 6]).
In section \[CompareMeth\], the mathematical model designed in [@Fil.MPhAG] for a certain nonlinear electrical circuit is presented. This model is described by the DAE with the matrices $A$, $B$ and nonlinear function $f$ of the form and the vector of variables $x= (x_1, x_2, x_3)^T=(I_L, U_C, I)^T$. Here $I_L$, $I$ are unknown currents, and $U_C$ is an unknown voltage. An input voltage $e$ is given. The remaining currents and voltages in the circuit are uniquely expressed in terms of $I_L$, $U_C$ and $I$. Also, here $L$ is an inductance, $C$ is a capacitance, $r$ is a linear resistance, $g$ is a linear conductance, $\varphi$, $\varphi _0$, $\psi$ are nonlinear resistances, and $h$ is a nonlinear conductance. Inductance is given in henries (H), capacitance is given in farads (F), resistance is given in ohm ($\Omega$), and conductance is given in siemens (S). When computations are carried out, it is assumed that inductance, capacitance and time are given in $\mu$H (microhenries), $\mu$F and $\mu$s, respectively. For example, in section \[CompareMeth\], the parameters $L = 500$, $C = 0.5$ ($L = 500\; \mu\text{H} =5\cdot 10^{-4} \text{ H}$, $C =0.5\; \mu\text{F}=5\cdot 10^{-7} \text{ F}$) are used to carry out computations and it is assumed that the numerical solutions are obtained for the initial parameters on time intervals given in $\mu$s ($1\, \mu\text{s}=10^{-6} \text{ s}$). It is easy to verify that this transition is true. It is done in order to the small values of $L$ and $C$ do not lead to large roundoff errors in the computations.
The conditions for the unique global solvability and the Lagrange stability of the DAE with for arbitrary $\varphi _0$, $\varphi $, $\psi$, $h$, $e$ are given in [@Fil.MPhAG]. Below we present some classes of functions for which these conditions are satisfied.
In [@Fil.MPhAG] it is proved that for each initial point $(t_0,x^0)$, satisfying the consistency condition $x_2^0 +r x_3^0 {=\nobreak\discretionary{}{\hbox{$\mathsurround=0pt =$}}{}} \psi(x_1^0-x_3^0)-\varphi(x_3^0)$, there exists a unique global solution of the DAE with and the initial condition $x(t_0)=x^0$ ( $x(t_0)=(I_L(t_0), U_C(t_0), I(t_0))^T$ ) for the functions of the form $\varphi_0(x_1)=\alpha _1 x_1^{2k-1}$, $\varphi(x_3)=\alpha _2 x_3^{2l-1}$, $\psi(x_1-x_3)=\alpha _3 (x_1-x_3)^{2j-1}$, $h(x_2)=\alpha _4 x_2^{2s-1}$, $k,l,j,s\in \mathbb N$, $\alpha _i >0$, $i=1,...,4$, if $j\le k$, $j\le s$ and $\alpha _3$ is sufficiently small, and for the functions of the form $\varphi _0 (x_1)=\alpha _1 x_1^{2k-1}$, $\varphi(x_3)=\alpha _2 \sin x_3$, $\psi(x_1-x_3)=\alpha _3\sin (x_1-x_3)$, $h(x_2)=\alpha _4 \sin x_2$ (instead of $\sin$ one can use $\cos$), if ${\alpha _2 + \alpha _3 < r}$. If, additionally, $\mathop{\sup }\limits_{t\in [0,\infty )}|e(t)| < +\infty$ or $\int\limits_{t_0}^{+\infty} |e(t)|\, dt <+\infty$, then for the initial points $(t_0,x^0)$ the DAE is Lagrange stable (in both cases), i.e., every its solution is bounded. These conditions are fulfilled, for example, for the sinusoidal voltage $e(t)=\beta\sin(\omega t + \theta)$, the power-law voltage $e(t)=\beta(t+\alpha)^{-n}$, the exponential voltage $e(t)=\beta e^{-\alpha t}$, and the voltage $e(t)=\beta e^{-\frac{(t-\alpha)^2}{\sigma^2}}$, $\alpha> 0$, $\beta,\, \sigma,\, \omega \in {\mathbb R}$, $n \in {\mathbb N}$, $\theta \in [0, 2\pi]$. For the voltage $e(t)= \beta(t+\alpha)^n$, $\alpha,\,\beta \in {\mathbb R}$, $n \in {\mathbb N}$, global solutions exist, but they are not bounded on the whole domain of definition.
We consider some particular cases. Below, the plots of numerical solutions, demonstrating the evolutionary properties of the mathematical model in the various choice of the linear parameters $L$, $C$, $r$, $g$, the nonlinear parameters $\varphi _0 $, $\varphi $, $\psi$, $h$ and the input voltage $e$, are presented. The parameters are given in the units mentioned above and time $t$ is given in $\mu$s.
Choose the linear parameters $L = 5\cdot 10^{-4}$, $C = 5\cdot 10^{-7}$, $r = 2$, $g =0.2$, the nonlinear resistances $\varphi_0 (x_1)= x_1^3$, $\varphi(x_3)= \sin x_3$, $\psi(x_1-x_3)=\sin (x_1-x_3)$ and conductance $h(x_2)= \sin x_2$, the input voltage $e(t)=(2 t+10)^{-2}$ and the initial data $t_0 = 0$, $x_0 =(10,-10,5)^T$. In this case the solution is Lagrange stable, i.e., global and bounded, as confirmed by the plots presented in Figure \[MM\_Ex5\_10-12\].
![The components of the numerical solution[]{data-label="MM_Ex5_10-12"}](MM_Ex5_10.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="MM_Ex5_10-12"}](MM_Ex5_11.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="MM_Ex5_10-12"}](MM_Ex5_12.eps "fig:"){width="8cm"}
If $e(t) = t^2$ and $L= 10^{-3}$, $C = 5\cdot 10^{-7}$, $r =2$, $g =0.3$, $\varphi_0(x_1) = x_1^3$, $\varphi(x_3)= x_3^3$, $h(x_2)= x_2^3$, $\psi(x_1-x_3)= (x_1-x_3)^3$, $t_0 = 0$, $x_0 =(0,0,0)^T$, then a solution is global, but not bounded. The plots for the corresponding numerical solution are presented in Figure \[MM\_Ex5\_1-3k\].
![The components of the numerical solution[]{data-label="MM_Ex5_1-3k"}](MM_Ex5_1k.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="MM_Ex5_1-3k"}](MM_Ex5_2k.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="MM_Ex5_1-3k"}](MM_Ex5_3k.eps "fig:"){width="8cm"}
For the voltage of the triangular shape (Figure \[eTriang\]) $e(t)= 50-|t-50-100 k|$, $t{\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} [100\, k, 100+ 100\, k]$, $k{\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} \{0\}\cup {\mathbb N}$, and $L=5\cdot 10^{-4}$, $C=5\cdot 10^{-7}$, $r=2$, $g=0.2$, $\varphi_0(x_1) = x_1^3$, $\varphi(x_3)= x_3^3$, $h(x_2)= x_2^3$, $\psi(x_1-x_3)= (x_1-x_3)^3$, $t_0 = 0$, $x_0 =(0,0,0)^T$, the solution is Lagrange stable, as confirmed by the plots presented in Figure \[mTriang3h\].
![The input voltage $e(t)$[]{data-label="eTriang"}](eTriang.eps){width="6cm"}
![The components of the numerical solution[]{data-label="mTriang3h"}](mTriang3h_1.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="mTriang3h"}](mTriang3h_2.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="mTriang3h"}](mTriang3h_3.eps "fig:"){width="8cm"}
For the voltage of the sawtooth shape (Figure \[ePila3\]) $$e(t)= \begin{cases} t-5\, k, & t\in [5\, k, 4+ 5\, k],\, k\in \{0\}\cup {\mathbb N}, \\ 20(k+1)-4t, & t\in [4+5\, k, 5+5\, k],\, k{\in\nobreak\discretionary{}{\hbox{$\mathsurround=0pt \in$}}{}} \{0\}\cup {\mathbb N}, \end{cases}$$ and $L = 10^{-5}$, $C =2\cdot 10^{-7}$, $r = 55$, $g =0.015$, $\varphi_0(x_1) = x_1^3$, $\varphi(x_3)= x_3^3$, $h(x_2)= x_2^3$, $\psi(x_1-x_3)= (x_1-x_3)^3$, $t_0 = 0$, $x_0 =(0,0,0)^T$, the solution is also Lagrange stable. The plots presented in Figure \[mPilaNew4h\] confirm this.
![The input voltage $e(t)$[]{data-label="ePila3"}](ePila3.eps){width="6cm"}
![The components of the numerical solution[]{data-label="mPilaNew4h"}](mPilaNew5h_1.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="mPilaNew4h"}](mPilaNew5h_2.eps "fig:"){width="8cm"} ![The components of the numerical solution[]{data-label="mPilaNew4h"}](mPilaNew5h_3.eps "fig:"){width="8cm"}
Below is the example of a Lagrange-unstable solution (Figure \[MM\_Ex6\_109-111\]), which is blow-up in finite time (the norm of the solution tends to infinity on a finite time interval). Here $L=5\cdot 10^{-6}$, $C = 5\cdot 10^{-7}$, $r = 2$, $g =0.2$, $\varphi_0(x_1) =-x_1^2$, $\varphi(x_3)=x_3^3$, $h(x_2)=x_2^2$, $\psi(x_1-x_3)=(x_1-x_3)^3$, $e(t) = 2\, \sin t$, and $t_0 = 0$, $x_0 =(1, -6.5, 1.5)^T$.
![The components of the numerical solution[]{data-label="MM_Ex6_109-111"}](MM_Ex6_109.eps "fig:"){width="5.4cm"} ![The components of the numerical solution[]{data-label="MM_Ex6_109-111"}](MM_Ex6_110.eps "fig:"){width="5.4cm"} ![The components of the numerical solution[]{data-label="MM_Ex6_109-111"}](MM_Ex6_111.eps "fig:"){width="5.4cm"}
In [@Fil.MPhAG] the plots of numerical solutions for other linear parameters, nonlinear functions and voltages are given. The analysis of the obtained numerical solutions verifies the results of theoretical studies.
Thus, the developed numerical methods allow one to carry out the numerical analysis of global dynamics for the mathematical models described by semilinear DAEs and to draw the conclusions about their evolutionary properties. Also, the developed methods have a number of advantages, which have already been mentioned in section \[Intro\].
Conclusions and outlooks
========================
To solve semilinear DAEs, the combined methods of the first and second orders of accuracy were obtained in this work. The correctness and convergence of the methods were proved. The developed methods enable to find a solution on any given time interval and require weaker restrictions for the nonlinear part of the equation than other known methods. Also, the effectiveness of the developed methods is due to the possibility to numerically find the spectral projectors using the formulas , which enables to numerically solve and analyze a semilinear DAE in the original form without additional analytical transformations. In section \[Appl\], the numerical analysis of the semilinear DAE describing the mathematical model for a nonlinear electrical circuit has been carried out. The results of the numerical analysis verify the results of the theoretical studies of global dynamics for the mathematical model. The plots of numerical solutions, demonstrating the evolutionary properties of the considered model, have been presented.
The obtained combined methods are easy to implement, effective enough, and can be applied to a study of real systems and processes which are modeled using DAEs.
In the future, it is planned to improve the methods obtained in this paper and to extend them to more general classes of DAEs. In particular, it is intended to use a predictor-corrector method to achieve a higher order of accuracy of method 1 without additional constraints. Also, it is planned to extend the obtained methods to a certain class of semilinear DAEs with a singular operator pencil. The theorems on the unique global solvability and on the Lagrange stability and instability for this class of equations were proved by the author in [@Fil.sing; @Fil.UMJ].
Acknowledgments {#acknowledgments .unnumbered}
===============
The publication is based on the research provided by grant support of the State Fund for Fundamental Research of Ukraine (project No. [Ф]{}83/82-2018).
[50]{} S.L. Campbell, Singular systems of differential equations II, Pitman, Boston, 1982.
P.J. Rabier, W.C. Rheinboldt, Nonholonomic Motion of Mechanical Systems from a DAE Viewpoint, SIAM, Philadelphia, PA, 2000. <https://doi.org/10.1137/1.9780898719536>
B. Fox, L.S. Jennings, A.Y. Zomaya, Numerical computation of differential-algebraic equations for nonlinear dynamics of multibody android systems in automobile crash simulation, *IEEE Transactions of Biomedical Engineering* 46, No. 10 (1999), 1199–1206. <https://doi.org/10.1109/10.790496> M. Morishima, Equilibrium, stability, and growth, Clarendon Press, Oxford, 1964. R. Riaza, P.J. Zufiria, Differential-algebraic equations and singular perturbation methods in recurrent neural learning, *Dynamical Systems* 18, No. 1 (2003), 89–105. <http://dx.doi.org/10.1080/1468936031000086843>
P. Kunkel, V. Mehrmann, Differential-Algebraic Equations. Analysis and Numerical Solution, European Mathematical Society, Zürich, 2006. <https://doi.org/10.4171/017>
J.C. Engwerda, Y. Salmah, I.E. Wijayanti, The Optimal Linear Quadratic Feedback State Regulator Problem for Index One Descriptor Systems, CentER Discussion Paper Series No. 2008-90 (2008), 1–19. <http://dx.doi.org/10.2139/ssrn.1288571>
R. Riaza, Differential-Algebraic Systems. Analytical Aspects and Circuit Applications, World Scientific, Hackensack, NJ, 2008. <http://dx.doi.org/10.1142/6746>
A.G. Rutkas, M.S. Filipkovska, Global solvability of the differential-algebraic equations of nonlinear electric circuits, *Zh. Obchysl. Prykl. Mat.*, No. 4 (2013), 131–142. <https://scholar.google.com.ua/scholar?oi=bibs&hl=ru&cluster=8801926006820179771> K.E. Brenan, S.L. Campbell, L.R. Petzold, Numerical Solution of Initial-Value Problems in Differential-Algebraic Equations, SIAM, Philadelphia, PA, 1996, <https://doi.org/10.1137/1.9781611971224>. R. Lamour, R. März, C. Tischendorf, Differential-Algebraic Equations: A Projector Based Analysis, in: A. Ilchman, T. Reis (Eds.), Differential-Algebraic Equations Forum, Springer-Verlag, Berlin, Heidelberg, New York, Dordrecht, London, 2013. <https://doi.org/10.1007/978-3-642-27555-5> L.A. Vlasenko, Evolution models with implicit and degenerate differential equations, Sistemnye Tekhnologii, Dniepropetrovsk, 2006. <https://scholar.google.com/scholar?cluster=16601459210830033847&hl=en&inst=8697446408056752236&oi=scholarr> A.G. Rutkas, Cauchy problem for the equation $Ax'(t) +Bx(t) = f(t)$, *Differ. Uravn.* 11, No. 11 (1975), 1996–2010. <http://mi.mathnet.ru/de2600> A.G. Rutkas, L.A. Vlasenko, Existence, uniqueness and continuous dependence for implicit semilinear functional differential equations, *Nonlinear Anal.* 55, No. 1–2 (2003), 125–139. <https://doi.org/10.1016/S0362-546X(03)00219-0>
U.M. Ascher, L.R. Petzold, Computer Methods for Ordinary Differential Equations and Differential-Algebraic Equations, SIAM, Philadelphia, PA, 1998.
E. Hairer, C. Lubich, M. Roche, The Numerical Solution of Differential-Algebraic Systems by Runge-Kutta methods, Springer-Verlag, Berlin, Heidelberg, New York, 1989. <https://doi.org/10.1007/BFb0093947>.
E. Hairer, G. Wanner, Solving Ordinary Differential Equations II. Stiff and Differential-Algebraic Problems, Springer, Berlin, 2010. <https://doi.org/10.1007/978-3-642-05221-7>.
M. Knorrenschild, Differential/algebraic equations as stiff ordinary differential equations, *SIAM J. Numer. Anal.* 29, No 6. (1992), 1694–1715. <https://doi.org/10.1137/0729096>
M.S. Filipkovska, Lagrange stability of semilinear differential-algebraic equations and application to nonlinear electrical circuits, *Journal of Mathematical Physics, Analysis, Geometry* 14, No. 2 (2018), 169–196. <https://doi.org/10.15407/mag14.02.169>. Yu.E. Boyarintsev, Methods for Solving Continuous and Discrete Problems for Singular Systems of Equations, Nauka, Novosibirsk, 1996. Yu.E. Boyarintsev, V.A. Danilov, A.A. Loginov, V.F. Chistyakov, Numerical Methods for Solving Singular Systems, Nauka, Novosibirsk, 1989. C.W. Gear, L.R. Petzold, ODE methods for the solution of differential/algebraic systems, *SIAM J. Numer. Anal* 21, No. 4 (1984), 716–728, <https://doi.org/10.1137/0721048>.
V.H. Linh, V. Mehrmann, Efficient integration of strangeness-free non-stiff differential-algebraic equations by half-explicit methods, *J. Comput. Appl. Math.* 262 (2014), 346–360. <https://doi.org/10.1016/j.cam.2013.09.072> A.L. Piven, Combined numerical method for solving a degenerate nonlinear integro-differential equation with delays, *Visn. Kharkiv. Nats. Univ. Mat. Model. Inform. Tekh. Avt. Syst. Upr.* 25, No. 1131 (2014), 168–180. <http://nbuv.gov.ua/UJRN/VKhIMAM_2014_1131_25_15> G.Yu. Kulikov, The numerical solution of the autonomous Cauchy problem with algebraic constraints on the phase variables, *Computational Mathematics and Mathematical Physics* 33, No. 4 (1993), 477–492. <http://www.ams.org/mathscinet-getitem?mr=1217952> C.W. Gear, L.R. Petzold, Differential/algebraic systems and matrix pencils, *Lecture Notes in Mathematics*, No. 973 (1983), 75–89, <https://doi.org/10.1007/BFb0062095>.
M. Hanke, R. März, C. Tischendorf, E. Weinmüller, S. Wurm, Least-squares collocation for linear higher-index differential–algebraic equations, *J. Comput. Appl. Math.* 317 (2017), 403–431. <https://doi.org/10.1016/j.cam.2016.12.017> P.J. Rabier, W.C. Rheinboldt, Finite difference methods for time dependent, linear differential algebraic equations, *Appl. Math. Lett.* 7, No. 2 (1994), 29–34. <https://doi.org/10.1016/0893-9659(94)90026-4>
A. Benchaabane, R. Sakthivel, Sobolev-type fractional stochastic differential equations with non-Lipschitz coefficients, *J. Comput. Appl. Math.* (2015). <http://dx.doi.org/10.1016/j.cam.2015.12.020>
B. İzgi, C. Çetin, Semi-implicit split-step numerical methods for a class of nonlinear stochastic differential equations with non-Lipschitz drift terms, *J. Comput. Appl. Math.* (2018). <https://doi.org/10.1016/j.cam.2018.03.027>
R.W. Erickson, D. Maksimović, Fundamentals of Power Electronics, Kluwer Academic Publishers, Boston, 2004. Yu.M. Borisov, D.N. Lipatov, Yu.N. Zorin, Electrical Engineering, BHV-Petersburg, St. Petersburg, 2012. A.G. Rutkas, M.S. Filipkovska, Extension of solutions of one class of differential-algebraic equations, *Zh. Obchysl. Prykl. Mat.*, No. 1 (2013), 135–145. <https://scholar.google.com.ua/scholar?oi=bibs&hl=ru&cluster=13372098116284618506> M.S. Filipkovska, Lagrange stability and numerical method for solving semilinear descriptor equations, *Visn. Kharkiv. Nats. Univ. Mat. Model. Inform. Tekh. Avt. Syst. Upr.* 26, No. 1156 (2015), 152–167. <http://nbuv.gov.ua/UJRN/VKhIMAM_2015_1156_26_17>
Filipkovska M.S. Continuation of solutions of semilinear differential-algebraic equations and applications in nonlinear radiotechnics, *Visn. Kharkiv. Nats. Univ. Mat. Model. Inform. Tekh. Avt. Syst. Upr.* 19, No. 1015 (2012), 306–319. <https://scholar.google.com.ua/scholar?oi=bibs&hl=ru&cluster=2989657459653377362>
S.K. Godunov, V.S. Ryabenkii, Difference Schemes. An Introduction to the Underlying Theory, North-Holland, Amsterdam, 1987.
A.G. Rutkas, A characteristic function and model of a linear pencil of operators, *Teor. funktsiy, funkts. anal. prilozh.* 45 (1986), 98–111. A.G. Rutkas, To the theory of characteristic functions of linear operators, *Dokl. Akad. Nauk. of the USSR* 229, No. 3 (1976), 101–111. M. Filipkovskaya (Filipkovska), Global solvability of singular semilinear differential equations and applications to nonlinear radio engineering, *Challenges of modern technology* 6, No. 1 (2015), 3–13. <http://www.journal.young-scientists.eu/>
M.S. Filipkovska, Lagrange stability and instability of nonregular semilinear differential-algebraic equations and applications, *Ukrain. Mat. Zh.* \[*Ukrainian Math. J.*\] 70, No. 6 (2018), 823–847. <http://umj.imath.kiev.ua/article/?lang=en&article=11273>
| {
"pile_set_name": "ArXiv"
} |
---
address: 'Joint Institute for Nuclear Research, 141985 Dubna, Russia'
author:
- 'S.G.Bondarenko, V.V.Burov, E.P.Rogochaya[^1]'
title: |
Final state interaction effects in electrodisintegration\
of the deuteron within the Bethe-Salpeter approach
---
Introduction
============
The electrodisintegration of the deuteron is a useful instrument which makes it possible to investigate the electromagnetic structure of the neutron-proton ($np$) system. Many approaches have been elaborated to describe this reaction for last 40 years [@Forest:1983vc; @Arenhovel:1982rx; @Shebeko; @Gakh:2004zq; @Jeschonnek:2008zg]. The simplest of them considered the electrodisintegration within a nonrelativistic model of the nucleon-nucleon (NN) interaction and outgoing nucleons were supposed to be free [@Forest:1983vc] (the plane-wave approximation - PWA). Those approaches were in a good agreement with experimental data at low energies. However, further investigations have shown that the final state interaction (FSI) between outgoing nucleons, two-body currents and other effects should be taken into account to obtain a reasonable agreement with existing experimental data at higher energies. Most of these effects have been considered within nonrelativistic models [@Arenhovel:1982rx; @Shebeko]. In relativistic models, FSI effects could be calculated within quasipotential approaches including the on-mass-shell nucleon-nucleon $T$ matrix [@Gakh:2004zq; @Jeschonnek:2008zg]. One of fundamental approaches for a description of the $np$ system is based on using the Bethe-Salpeter (BS) equation [@Salpeter:1951sz]. To solve this equation we use a separable ansatz [@Yamaguchi:1954mp] for the NN interaction kernel. In this case, we have to deal with a system of algebraic equations instead of integral ones [@Bondarenko:2002zz]. A separable kernel model within the BS approach was not used to describe most of high-energy NN processes for a while since calculated expressions contained nonintegrable singularities. The separable kernels proposed in [@Bondarenko:2008fp; @Bondarenko:2010qv; @Bondarenko:2008mm] make it possible to avoid those difficulties. Using these kernels FSI can be taken into account in the electrodisintegration in a wide range of energy. The electrodisintegration cross section is calculated in the present paper under different kinematic conditions (see Table \[tab:2\] below). The rank-six NN interaction potential MY6 [@Bondarenko:2010qv] is used to describe scattered and bound $^3S_1$-$^3D_1$ partial-wave states. The uncoupled partial-wave states with total angular momentum $J=0,1$ ($^1S_0$, $^1P_1$, $^3P_0$, $^3P_1$) are described by multirank separable potentials [@Bondarenko:2008mm]. The obtained results are compared with nonrelativistic model calculations [@Shebeko] where NN interactions were described by the realistic Paris potential [@Lacombe:1980dr].
The paper is organized as follows. In Sec.\[sec:2\] , the three-differential cross section of the $d(e,e^\prime p)n$ reaction is considered in the relativistic impulse approximation. The used BS formalism is presented in Sec.\[sec:3\]. The details of calculations are considered in Sec.\[sec:4\]. Then the obtained relativistic results are compared with nonrelativistic ones [@Shebeko] and experimental data of the Sacle experiment [@Bussiere:1981mv; @TurckChieze:1984fy] in Sec.\[sec:5\].
Cross section {#sec:2}
=============
When all particles are unpolarized the exclusive electrodisintegration of the deuteron $d(e,e^\prime p)n$ can be described by the differential cross section in the deuteron rest frame - laboratory system (LS) - as: $$\begin{aligned}
&&\frac{d^3\sigma}{dE_e'd\Omega_e'd\Omega_p}=
\frac{\sigma_\textmd{Mott}}{8M_d(2\pi)^3}{\frac{\pp_p^2{\sqrt s}}{\sqrt{1+\eta}|\pp_p|-E_p\sqrt{\eta}
\cos\theta_p}}
\nonumber\\
&&\times\left[ l^0_{00}W_{00}+l^0_{++}(W_{++}+W_{--})+ 2 l^0_{+-}\cos2\phi~{\rm Re}
W_{+-}\right.\nonumber\\
&&\hspace*{5mm}- 2 l^0_{+-}\sin2\phi~{\rm Im}
W_{+-}-2 l^0_{0+}\cos\phi~{\rm Re} (W_{0+}-W_{0-})\nonumber\\
&&\hspace*{5mm}\left.
-2 l^0_{0+}\sin\phi~{\rm Im} (W_{0+}+W_{0-})\right]. \label{3cross_0}\end{aligned}$$ where $\sigma_{\rm
Mott}=(\alpha\cos\frac{\theta}{2}/2E_e\sin^2\frac{\theta}{2})^2$ is the Mott cross section, $\alpha={e^2}/{(4\pi)}$ is the fine structure constant; $M_d$ is the mass of the deuteron; $q=p_e-p_e'=(\omega,\q)$ is the momentum transfer; $p_e=(E_e,\l)$ and $p_e'=(E_e',\l')$ are initial and final electron momenta, respectively; $\Omega_e'$ is the outgoing electron solid angle; $\theta$ is the electron scattering angle. The outgoing proton is described by momentum $\pp_p$ ($E_p=\sqrt{\pp_p^2+m^2}$, $m$ is the mass of the nucleon) and solid angle $\Omega_p=(\theta_p,\phi)$ where $\theta_p$ is the zenithal angle between the $\pp_p$ and $\q$ momenta and $\phi$ is the azimuthal angle between the ([**ee$^\prime$**]{}) and ([**qp**]{}) planes. Factor $\eta=\q^2/s$ can be calculated through the $np$ pair total momentum $P$ squared: $$\begin{aligned}
s=P^2=(p_p+p_n)^2
%=(M_d+\omega)^2-\q^2
=M_d^2+2M_d\omega+q^2,
\label{pair_s}\end{aligned}$$ defined by the sum of the proton $p_p$ and neutron $p_n$ momenta. The photon density matrix elements have the following form: $$\begin{aligned}
&&l_{00}^0=\frac{Q^2}{\q^2},\quad l_{0+}^0=\frac{Q}{|\q|\sqrt
2}\sqrt{\frac{Q^2}{\q^2}+\tan^2 \frac{\theta}{2}},\nonumber\\
&&l_{++}^0=\tan^2\frac{\theta}{2}+\frac{Q^2}{2\q^2},\quad
l_{+-}^0=-\frac{Q^2}{2\q^2},\end{aligned}$$ where $Q^2=-q^2$ is introduced for convenience. The hadron density matrix elements $$\begin{aligned}
W_{\lambda\lambda'}=W_{\mu\nu}\varepsilon_\lambda^\mu\varepsilon_{\lambda'}^\nu,\end{aligned}$$ where $\lambda$, $\lambda'$ are photon helicity components [@Dmitrasinovic:1989bf], can be calculated using the photon polarization vectors $\varepsilon$ and Cartesian components of hadron tensor $$\begin{aligned}
W_{\mu\nu}=\frac{1}{3}\sum_{s_ds_ns_p}\left|<np:SM_S|j_\mu|d:1M>\right|^2,
\label{ht}\end{aligned}$$ where $S$ is the spin of the $np$ pair and $M_S$ is its projection; $s_d$, $s_n$ and $s_p$ are deuteron, neutron and proton momentum projections, respectively. The matrix element $<np:SM_S|j_\mu|d:1M>$ can be constructed according to the Mandelstam technique [@Mandelstam:1955sd] and has the following form in LS: $$\begin{aligned}
&&<np:SM_S|j_\mu|d:1M>=i\sum_{n=1,2}\int\frac{d^4p\cm}{(2\pi)^4} \times\label{cur_fsi}\\
&&{\rm Sp}\left\{\Lambda({\cal L
}^{-1})\bar\psi_{SM_S}(p\cm,{p^*}\cm;P\cm) \Lambda({\cal L
})\Gamma_\mu^{(n)}(q)\right.\times \nonumber\\
&&S^{(n)}\left(\frac{K_{(0)}}{2}-(-1)^np-\frac q2\right)
\left.\Gamma^M
\left(p+(-1)^n\frac{q}{2};K_{(0)}\right)\right\}\nonumber\end{aligned}$$ within the relativistic impulse approximation. The sum over $n=1,2$ corresponds to the interaction of the virtual photon with the proton and with the neutron in the deuteron, respectively. The total $P\cm$ and relative ${p^*}\cm$ momenta of the outgoing nucleons and the integration momentum $p\cm$ are considered in the final $np$ pair rest frame - center-of-mass system (CM), $p$ denotes the relative $np$ pair momentum in LS. To perform the integration, momenta $p$, $q$ and the deuteron total momentum $K_{(0)}=(M_d,\vz)$ in LS are written in CM using the Lorenz-boost transformation ${\cal L}$ along the $\q$ direction. The $np$ pair wave function $\psi_{SM_S}$ is transformed from CM to LS by the corresponding boost operator $\Lambda$. A detailed description of $\psi_{SM_S}$, the $n$th nucleon interaction vertex $\Gamma_\mu^{(n)}$, the propagator of the $n$th nucleon $S^{(n)}$, and the deuteron vertex function $\Gamma^M$ can be found in our previous works [@Bondarenko:2002zz; @Bondarenko:2005rz].
Separable kernel of NN interaction {#sec:3}
==================================
The outgoing $np$ pair is described by the $T$ matrix which can be found as a solution of the inhomogeneous Bethe-Salpeter equation [@Salpeter:1951sz]: $$\begin{aligned}
&&T(p^{\prime}, p; P) = V(p^{\prime}, p; P) \label{t_op} \\
&&\hspace*{10mm}+ \frac{i}{4\pi^3}\int d^4k\, V(p^{\prime}, k;
P)\, S_2(k; P)\, T(k, p; P),\nonumber\end{aligned}$$ where $V$ is the NN interaction kernel, $S_2$ is the free two-particle Green function: $$\begin{aligned}
S_2^{-1}(k; P)=\bigl(\tfrac12\:{\slashed P}+{\slashed
k}-m\bigr)^{(1)} \bigl(\tfrac12\:{\slashed P}-{\slashed
k}-m\bigr)^{(2)},\end{aligned}$$ and $p~(p')$ is the relative momentum of initial (final) nucleons, $P$ is the total $np$ pair momentum.
To solve the BS equation (\[t\_op\]) partial-wave decomposition [@Kubis:1972zp] for the $T$ matrix: $$\begin{aligned}
T_{\alpha\beta,\gamma\delta}(p^{\prime},p; {P_{(0)}}) =
&&\sum_{JMab} t_{ab}(p_0',|\p'|;p_0,|\pp|; s) \times\label{bse_spd}\\
&&\hspace*{-7mm}({\cal Y}_{aM}(-{\p'})U_C)_{\alpha\beta}\otimes
(U_C {\cal Y}^{\dag}_{bM}({\pp}))_{\delta\gamma}\nonumber\end{aligned}$$ is used. Here $P_{(0)}=(\sqrt s,\vz)$ is the $np$ pair total momentum in CM, $U_C=i\gamma^2\gamma^0$ is the charge conjugation matrix. Indices $a,b$ correspond to the set $^{2S+1}L_J^\rho$ of spin $S$, orbital $L$ and total $J$ angular momenta, $\rho=+$ defines a positive-energy partial-wave state, $\rho=-$ corresponds to a negative-energy one. Greek letters $\{\alpha,\beta,\gamma,\delta\}$ in (\[bse\_spd\]) are used to denote Dirac matrix indices. The spin-angle functions: $$\begin{aligned}
&&{\cal Y}_{JM:LS {\rho}}(\pp) U_C= \label{an_pa}
\\
&& i^{L}\sum_{m_Lm_Sm_1m_2\rho_1\rho_2}C_{\frac12 \rho_1
\frac12 \rho_2}^{S_{\rho} {\rho}} C_{L m_L S m_S}^{JM} C_{\frac12
m_1 \frac12 m_2}^{Sm_S}\nonumber\\
&&\hspace*{20mm}\times Y_{L{m_L}}(\hat\pp)
{U^{\rho_1}_{m_1}}^{(1)}(\pp) {{U^{\rho_2}_{m_2}}^{(2)}}^{T}(-\pp)
\nonumber\end{aligned}$$ are constructed from the free nucleon Dirac spinors $u,~v$. It should be mentioned that only positive-energy states with $\rho=+$ are considered in this paper. Using a similar decomposition for $V$, the BS equation for radial parts of the $T$ matrix and kernel $V$ is obtained: $$\begin{aligned}
&&t_{ab}(p_0', |\p'|; p_0, |\pp|; s) = v_{ab}(p_0', |\p'|; p_0,
|\pp|; s) \label{BS_decomp}\\
&&+ \frac{i}{4\pi^3}\sum_{cd}\int\limits_{-\infty}^{+\infty}\!
dk_0\int\limits_0^\infty\! \k^2 d|\k|\, v_{ac}(p_0', |\p'|; k_0,
|\k|; s)\nonumber\\
&& \hspace*{20mm}\times S_{cd}(k_0,|\k|; s)\, t_{db}(k_0,|\k|;p_0,|\pp|;
s).\nonumber\end{aligned}$$ To solve the resulting equation (\[BS\_decomp\]) a separable ansatz [@Yamaguchi:1954mp] for the interaction kernel $V$ is used: $$\begin{aligned}
&&v_{ab}(p_0', |\p'|; p_0, |\pp|; s)=\label{ansatz}\\
&& \hspace*{10mm} \sum_{i,j=1}^N\lambda_{ij}(s) g_i^{[a]}(p_0',
|\p'|)g_j^{[b]}(p_0, |\pp|),\nonumber\end{aligned}$$ where $N$ is a rank of a separable kernel, $g_i$ are model functions, $\lambda_{ij}$ is a parameter matrix. Substituting $V$ (\[ansatz\]) in BS equation (\[t\_op\]), we obtain the $T$ matrix in a similar separable form: $$\begin{aligned}
&&t_{ab}(p_0', |\p'|; p_0, |\pp|; s)= \label{t_separ}\\
&& \hspace*{10mm}\sum_{i,j=1}^N\tau_{ij}(s)g_i^{[a]}(p_0', |\p'|)
g_j^{[b]}(p_0, |\pp|)\nonumber\end{aligned}$$ where: $$\begin{aligned}
\tau_{ij}(s)=1/(\lambda_{ij}^{-1}(s)+h_{ij}(s)),\end{aligned}$$ and $$\begin{aligned}
&&h_{ij}(s)= \label{hij}\\
&& -\frac{i}{4\pi^3}\sum_{a}\int dk_0\int \k^2d|\k|
\frac{g_i^{[a]}(k_0,|\k|)g_j^{[a]}(k_0,|\k|)}{(\sqrt
s/2-\ek+i\epsilon)^2-k_0^2}\nonumber\end{aligned}$$ are auxiliary functions, $\ek=\sqrt{\k^2+m^2}$. Thus, the problem of solving the initial integral BS equation (\[t\_op\]) turns out to finding the $g_i$ functions and parameters $\lambda_{ij}$ of separable representation (\[ansatz\]). They can be obtained from a description of observables in $np$ elastic scattering [@Bondarenko:2010qv; @Rupp:1989sg; @Bondarenko:2008mm; @Mathelitsch:1981mr; @Bondarenko:2011hs].
Final state interaction {#sec:4}
=======================
In our previous works [@Bondarenko:2005rz; @Bondarenko:2010qv] the electrodisintegration of the deuteron was considered within the plane wave approximation (PWA) when the final nucleons were supposed to escape without interaction. Even though PWA is enough to describe the electrodisintegration at low energies, it is important to take into account the final state interaction (FSI) between the outgoing nucleons. As it was shown in other works [@Shebeko], the contribution of FSI effects increases with increasing nucleon energies or/and momentum transfer. Therefore, the relativistic models of the NN interaction must be elaborated and FSI should be included into calculations to get an adequate description of the electrodisintegration. The first relativistic model based on a separable kernel approach was Graz II [@Rupp:1989sg]. However, it was impossible in principle to calculate FSI using it [@Bondarenko:2008mm]. To solve this problem new relativistic separable kernels [@Bondarenko:2010qv; @Bondarenko:2008mm] were elaborated. We apply them to the deuteron electrodisintegration including FSI in this paper.
The outgoing nucleons are described by the BS amplitude which can be written as a sum of two terms: $$\begin{aligned}
&&\psi_{SM_S}(p,p^*;P) = \psi^{(0)}_{SM_S}(p,p^*;P) \label{psi_all}\\
&&+\frac{i}{4\pi^3}
S_2(p;P)\int d^4k ~V(p,k;P)\psi_{SM_S}(k,p^*;P).\nonumber\end{aligned}$$ The first term $$\begin{aligned}
\psi_{SM_S}^{(0)} (p,p^*;P)=
(2\pi)^4\chi_{SM_S}(p;P)\delta(p-p^*)\label{wf_PWA}\end{aligned}$$ is related to the outgoing pair of free nucleons (PWA), $\chi_{SM_S}$ is a spinor function for two fermions. The second term in (\[psi\_all\]) corresponds to the final state interaction of the outgoing nucleons. It can be expressed through the $T$ matrix if we use the following relation: $$\begin{aligned}
&&\int d^4k~V(p,k;P)\psi_{SM_S}(k,p^*;P)=\\
&&\int d^4k~T(p,k;P)
\psi^{(0)}_{SM_S}(k,p^*;P)\nonumber\end{aligned}$$ and can be rewritten as $$\begin{aligned}
&&\psi_{SM_S}^{(t)}(p,p^*;P)= \label{psi_t}\\
&& \hspace*{10mm} 4\pi i S_2(p;P)
T(p,p^*;P)\chi_{SM_S}(p^*;P),\nonumber\end{aligned}$$ here $(t)$ means that this part of the $np$ pair wave function is related to the $T$ matrix. Applying the partial-wave decomposition of the $T$ matrix (\[bse\_spd\]) the expression (\[psi\_t\]) can be written as follows: $$\begin{aligned}
&&\psi_{SM_S}^{(t)}(p,p^*;P)= 4\pi i \times \label{psi_t_decomposed}\\
&&\sum_{LmJMa} C_{LmSM_S}^{JM} Y_{Lm}^*(\hat\pp^*){\cal
Y}_{aM}(\pp) \phi_{a,J:LS+}( p_0,|\pp|; s), \nonumber\end{aligned}$$ where $p^*=(0,\pp^*)$ with $|\pp^*|=\sqrt{s/4-m^2}$ is the relative momentum of on-mass-shell nucleons in CM, $\hat\pp^*$ denotes the azimuthal angle $\thetaps$ between the $\pp^*$ and $\q$ vectors and zenithal angle $\phi$. Since only positive-energy partial-wave states are considered here the radial part is: $$\begin{aligned}
\phi_{a,J:LS+}(p_0,|\pp|; s)=\frac
{t_{a,J:LS+}(p_0,|\pp|;0,|\pp^*|; s)}{(\sqrt
s/2-\ep+i\epsilon)^2-p_0^2}.\end{aligned}$$ According to definition (\[an\_pa\]) spin-angle functions ${\cal
Y}$ can be written as a product of Dirac $\gamma$ matrices in the matrix representation [@Bondarenko:2002zz] as: $$\begin{aligned}
&&{\cal Y}_{aM}(\pp)= \label{sap_np}\\
&&\frac{1}{\sqrt{8\pi}}
\frac{1}{4\ep(\ep+m)}(m+{\slashed p}_1)(1+\gamma_0){\cal G}_{aM}
(m-{\slashed p}_2),\nonumber\end{aligned}$$ matrices ${\cal G}_{aM}$ are given in Table \[tab:1\]. Decomposition (\[psi\_t\_decomposed\]) is considered in detail in [@Bondarenko:2004pn].
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
$a={\footnotesize \left\{^{2S+1}L_J^\rho\right\}^{{\phantom{1}}^{\phantom{1}}} }$ ${\cal G}_{aM}$
----------------------------------------------------------------------------------- -------------------------------------------------------------------------------
$~{^1S_0^+}^{{\phantom{1}}^{\phantom{1}}}$ $-\gamma_5$
$~^3S_1^+$ ${\slashed \xi}_{M}$
$~^1P_1^+$ $\frac{\sqrt{3}}{|\pp|}(p_1\cdot\xi_{M})\gamma_5$
$~^3P_0^+$ $-\frac{1}{2|\pp|}({\slashed p_1}-{\slashed p_2})$
$~^3P_1^+$ $-\sqrt{\frac{3}{2}}\frac{1}{|\pp|}\left[(p_1\cdot\xi_{M})-\frac{1}{2}
{\slashed \xi}_{M}({\slashed p_1}-{\slashed p_2})\right]\gamma_5$
$~^3D_1^+$ $\frac{1}{\sqrt{2}}\left[{\slashed
\xi}_{M}+\frac{3}{2}\frac{1}{\pp^2}(p_1\cdot\xi_{M})({\slashed p}_1-{\slashed
p}_2)\right]$
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
: Spin-angular parts ${\cal G}_{aM}$ (\[sap\_np\]) for the $np$ pair; $p_1=(\ep,\pp)$, $p_2=(\ep,-\pp)$ are on-mass-shell momenta, $\ep=\sqrt{\pp^2+m^2}$; $\gamma$ matrices are defined as in [@Bjorken]. []{data-label="tab:1"}
Using definition (\[psi\_all\]) and substituting (\[wf\_PWA\]), (\[psi\_t\_decomposed\]) into (\[cur\_fsi\]), the final expression for hadron current $<np:SM_S|j_\mu|d:1M>$ can be obtained. It consists of two parts. One of them: $$\begin{aligned}
&&<np:SM_S|j_\mu|d:1M>^{(0)}= \label{cur_0}\\
&&i\sum_{n=1,2}\left\{\Lambda({\cal
L}^{-1})\bar\chi_{SM_S}\left({p^*}\cm; P\cm\right)\Lambda({\cal
L})\Gamma_\mu^{(n)}(q) \right.\times \nonumber\\
&&S^{(n)}\left(\frac{K_{(0)}}{2}-(-1)^n
p^*-\frac{q}{2}\right)\left.\Gamma^{M}\left(p^*+(-1)^n\frac{q}{2};
K_{(0)}\right)\right\}\nonumber\end{aligned}$$ corresponds to the electrodisintegration in PWA. Another one: $$\begin{aligned}
&&<np:SM_S|j_\mu|d:1M>^{(t)}= \label{cur_t} \\ &&\frac{i}{4\pi^3}\sum_{n=1,2}\sum_{LmJM_JL'lm'}
C_{LmSM_S}^{JM_J} Y_{Lm}(\hat\pp^*)\times\nonumber\\
&&\int\limits_{-\infty}^\infty dp_0\cm \int\limits_0^\infty
{(\pp\cm)}^2 d|\pp\cm| \int\limits_{-1}^1
d\cos\thetap\cm \int\limits_0^{2\pi}d\phi\times \nonumber\\
&&{\rm Sp}\left\{\Lambda({\cal L }^{-1})\bar{\cal
Y}_{JL'SM_J}(\pp\cm) \Lambda({\cal L })
\Gamma_\mu^{(n)}(q)\times \right.\nonumber \\
&& \left. S^{(n)}\left(\frac{K_{(0)}}{2}-(-1)^np-\frac
q2\right){\cal Y}_{1lSm'}\left(\pp+(-1)^n\frac{\q}{2}\right)\right\}\times \nonumber\\
&& \frac {t_{L'L}^*(p\cm_0,|\pp\cm|;0,|\pp^*|; s)}{(\sqrt
s/2-\ep+i\epsilon)^2-p_0^2} \times \nonumber\\
&& g_l
\left(p_0+(-1)^n\frac{\omega}{2},\pp+(-1)^n\frac{\q}{2};K_{(0)}\right)
\nonumber\end{aligned}$$ corresponds to the process when FSI is taken into account. Here $g_l$ is a radial part of the deuteron vertex function $\Gamma^M$. The part ${\rm Sp}\left\{\ldots\right\}$ has been calculated using the algebra manipulation package MAPLE. The three-dimensional integration over $p_0\cm$, $|\pp\cm|$ and $\cos\thetap\cm$ has been performed numerically using the programming language FORTRAN.
set I [@Bussiere:1981mv] set II [@Bussiere:1981mv] [@TurckChieze:1984fy]
----------------------------- ----- -------------------------- --------------------------- -----------------------
$E_e$, GeV 0.500 0.500 0.560
$E_e^\prime$, GeV 0.395 0.352 0.360
$\theta$, ${}^{\circ}$ 59 44.4 25
$\pp_n$, GeV/$c$ min 0.005 0.165 0.294
max 0.350 0.350 0.550
$\theta_n$, ${}^{\circ}$ min 101.81 172.07 153.01
max 37.78 70.23 20.81
$\theta_{qe}$, ${}^{\circ}$ min 48.79 44.74 33.06
$\pp_p$, GeV/$c$ min 0.451 0.514 0.557
max 0.276 0.403 0.306
$\theta_{p}$, ${}^{\circ}$ min 0.622 2.54 13.86
max 51.03 54.90 140.28
$\theta_{pe}$, ${}^{\circ}$ min 49.41 47.28 46.92
max 99.81 99.64 173.35
$\sqrt s$, GeV 1.929 1.993 2.057
$\sqrt s- 2m$, GeV 0.051 0.115 0.176
$Q^2$, (GeV/$c$)$^2$ 0.192 0.101 0.038
$\omega$, GeV 0.105 0.148 0.200
$|\q|$, GeV/$c$ 0.450 0.350 0.279
: [Kinematic conditions considered in the paper. Here all quantities are in LS. In addition to those which are defined in the text, they are: angle $\theta_{qe}$ between the beam and the virtual photon; neutron momentum $\pp_n$ and angle $\theta_n$ between the neutron and the virtual photon ($\pp_p,\theta_p$ – the same for the proton); $\theta_{pe}$ ($\theta_{qe}$) – the angle between the beam and the proton (virtual photon).]{} []{data-label="tab:2"}
![Cross section (\[3cross\_0\]) depending on recoil neutron momentum $|\pp_n|$ calculated under kinematic conditions set I of the Sacle experiment [@Bussiere:1981mv]. The notations are following: MY6 (PWA) (red solid line) - relativistic calculation in the plane-wave approximation with the MY6 potential [@Bondarenko:2010qv]; MY6 (FSI) (blue dashed line) - relativistic calculation including FSI effects; NR (PWAc) (violet dotted line) - nonrelativistic calculation [@Shebeko].[]{data-label="fig:1"}](cs1.eps){width="1\linewidth"}
![The same as in Fig.\[fig:1\] but under kinematic conditions set II of the Sacle experiment [@Bussiere:1981mv]. The nonrelativistic calculation NR (FSIc) (brown dashed-dotted line) includes FSI effects.[]{data-label="fig:2"}](cs2.eps){width="1\linewidth"}
![The same as in Figs.\[fig:1\],\[fig:2\] but under kinematic conditions of the Sacle experiment [@TurckChieze:1984fy]. Two additional results are presented for comparison: MY6 (PWAc) (pink dashed-dotted-dotted line) - relativistic PWA calculation; MY6 (FSIc) (orange dashed-dotted line) - relativistic calculation with FSI effects; both obtained under current conservation condition (\[curcon\]).[]{data-label="fig:3"}](cs3.eps){width="1\linewidth"}
Results and discussion {#sec:5}
======================
The differential cross section (\[3cross\_0\]) is calculated under three kinematic conditions of the Sacle experiment [@Bussiere:1981mv; @TurckChieze:1984fy] (described in Table \[tab:2\]) and is present in Figs.\[fig:1\]-\[fig:3\]. The calculations have been performed within the relativistic impulse approximation for two different cases: when the outgoing nucleons are supposed to be free (PWA) and when the final state interaction between the nucleons is taken into account (FSI). The partial-wave states of the $np$ pair with total angular momentum $J=0,1$ have been considered. The used relativistic model consists of two parts: the separable potential MY6 [@Bondarenko:2010qv] for the bound (deuteron) and scattered $^3S_1$-$^3D_1$ states and separable potentials of various ranks [@Bondarenko:2008mm] - for all the other partial-wave states ($^1S_0$, $^1P_1$, $^3P_0$, $^3P_1$). The obtained results have been compared with the nonrelativistic model [@Shebeko].
In Fig.\[fig:1\], relativistic MY6 (PWA), MY6 (FSI) and nonrelativistic NR (PWAc) calculations under kinematic conditions [@Bussiere:1981mv] (set I) are practically coincide with each other and go close to the experimental data. The notation “c” in NR (PWAc) means that the corresponding hadron current satisfies the current conservation condition [@Forest:1983vc]: $$\begin{aligned}
q^\mu J_\mu=0. \label{curcon}\end{aligned}$$ So, it can be concluded that relativistic and FSI effects do not play an important role in description of the deuteron electrodisintegration at low energies.
Fig.\[fig:2\] considers the cross section (\[3cross\_0\]) calculated under kinematic conditions [@Bussiere:1981mv] (set II). Here FSI effects decrease the resulting cross section noticeably. A slight difference is seen between nonrelativistic and relativistic results.
The cross section under kinematic conditions [@TurckChieze:1984fy] is presented in Fig.\[fig:3\]. In this case PWA and FSI calculations differ significantly. It means that the influence of FSI on observables increases with increasing $|\pp_n|$. To investigate relativistic effects, two additional calculations MY6 (PWAc), MY6 (FSIc) have been performed when the condition (\[curcon\]) has been taken into account. The relativistic one-particle current within the impulse approximation (\[cur\_fsi\]) does not satisfy (\[curcon\]). We impose this condition on the hadron current (\[cur\_0\]), (\[cur\_t\]) as it was done in [@Shebeko] for a nonrelativistic model to compare our relativistic result with it and investigate relativistic effects. It has been done only under kinematic conditions [@TurckChieze:1984fy] since an influence of the current conservation condition becomes significant at high outgoing neutron momenta $|\pp_n|$. The comparison of MY6 (PWAc) and NR (PWAc), MY6 (FSIc) and NR (FSIc) demonstrates that relativistic effects become significant with increasing $|\pp_n|$. FSI effects decrease the obtained cross section.
As it is seen from the figures, an influence of FSI increases with increasing outgoing neutron momentum $|\pp_n|$. FSI effects are significant at outgoing neutron momenta $|\pp_n|>150$MeV/c even at low momentum transfer. It has been demonstrated that the FSI contribution always decreases the value of the cross section (\[3cross\_0\]). From present results it can also be seen that the FSI contribution relatively to the PWA one is approximately the same both for nonrelativistic and relativistic models. However, other effects, like negative-energy partial-wave states and two-body currents, should be taken into account for further conclusions. It should be also emphasized that the relativistic calculation of FSI has been performed within the BS approach using a relativistic separable potential for the first time. Recently the considered separable potentials [@Bondarenko:2010qv; @Bondarenko:2008mm] have been modified to take into account inelastic processes (production of various mesons) occurring in the NN interaction at high energies [@Bondarenko:2011hs]. However, energies of nucleons do not reach the inelasticity threshold under kinematic conditions [@Bussiere:1981mv]. The inelasticity effects are nonzero under conditions [@TurckChieze:1984fy] but they are conjectured to be small in this case.
[99]{}
T. De Forest, Nucl. Phys. [**A392**]{}, 232 (1983).
H. Arenhovel, Nucl. Phys. [**A384**]{}, 287 (1982).
A.Yu. Korchin, Yu.P. Melnik, and A.V. Shebeko, Sov. J. Nucl. Phys. [**48**]{}, 243 (1988).
G.I. Gakh, A.P. Rekalo, and E. Tomasi-Gustafsson, Annals Phys. [**319**]{}, 150 (2005).
S. Jeschonnek and J.W. Van Orden, Phys. Rev. [**C78**]{}, 014007 (2008).
E.E. Salpeter and H.A. Bethe, Phys. Rev. [**84**]{}, 1232 (1951).
Y. Yamaguchi, Phys. Rev. [**95**]{}, 1628 (1954);\
Y. Yamaguchi and Y. Yamaguchi, Phys. Rev. [**95**]{}, 1635 (1954).
S.G. Bondarenko, V.V. Burov, A.V. Molochkov, G.I. Smirnov, and H. Toki, Prog. Part. Nucl. Phys. [**48**]{}, 449 (2002).
S.G. Bondarenko, V.V. Burov, W-Y. Pauchy Hwang, and Rogochaya, E.P., JETP Lett. [**87**]{}, 653 (2008).
S.G. Bondarenko, V.V. Burov, W.-Y.Pauchy Hwang, and E.P. Rogochaya, Nucl. Phys. [**A848**]{}, 75 (2010). S.G. Bondarenko, V.V. Burov, W.-Y. Pauchy Hwang, and E.P. Rogochaya, Nucl. Phys. [**A832**]{}, 233 (2010). M. Lacombe et al. Phys. Rev. [**C21**]{}, 861 (1980).
M. Bernheim et al., Nucl. Phys. [**A365**]{}, 349 (1981).
S. Turck-Chieze et al., Phys. Lett. [**B142**]{}, 145 (1984).
V. Dmitrasinovic and F. Gross, Phys. Rev. [**C40**]{}, 2479 (1989).
S. Mandelstam, Proc. Roy. Soc. Lond. [**A233**]{}, 248 (1955).
S.G. Bondarenko, V.V. Burov, E.P. Rogochaya, and A.A. Goy, Phys. Part. Nucl. Lett. [**2**]{}, 323 (2005);\
S.G. Bondarenko, V.V. Burov, E.P. Rogochaya, and A.A. Goy, Phys. Atom. Nucl. [**70**]{}, 2054 (2007). J.J. Kubis, Phys. Rev. [**D6**]{}, 547 (1972). S.G. Bondarenko, V.V. Burov, K.Yu. Kazakov, and D.V. Shulga, Phys. Part. Nucl. Lett. [**1**]{}, 178 (2004). G. Rupp and J.A. Tjon, Phys. Rev. [**C41**]{}, 472 (1990).
L. Mathelitsch, W. Plessas, and W. Schweiger, Phys. Rev. [**C26**]{}, 65 (1982).
S.G. Bondarenko, V.V. Burov, and E.P. Rogochaya, preprint 1106.4478 \[nucl-th\], doi:10.1016/j.physletb.2011.10.009; preprint 1108.4170 \[nucl-th\], doi:10.1016/j.nuclphysbps.2011.10.081.
J.D. Bjorken and S.D. Drell, Relativistic Quantum Mechanics and Relativistic Quantum Fields. McGraw-Hill, New York (1964).
[^1]: e-mail: [email protected]
| {
"pile_set_name": "ArXiv"
} |
Let $M^\prime$ be a pseudo-Riemannian manifold of dimension $n>1$. The fourth order operator defined below (equation (\[three\])) bears some analogy with the second-order shift of the Laplace–Beltrami operator ${\mathcal{L}}$ by a multiple of the scalar curvature $R$, namely $$\label{one}{\mathcal{L}}+\frac{n-2}{4(n-1)}R$$ defined by [Ø]{}rsted, and studied by others in special cases. The relation satisfied by operators of the form (\[one\]) is the following: if $M$ is a manifold of dimension $n>1$ and $g_1$ and $g_2$ are two metric tensors on $M$ that satisfy $g_1=p^2g_2$ for some positive function $p$, then $$p^{(n+2)/2}\Big({\mathcal{L}}_{g_1}+\frac{n-2}{4(n-1)}R_{g_1}\Big)\phi
=\Big({\mathcal{L}}_{g_2}+\frac{n-2}{4(n-1)}R_{g_2}\Big)(p^{(n-2)/2}\phi)$$ for all smooth functions $\phi$, where ${\mathcal{L}}_{g_i}$ and $R_{g_i}$ are the Laplace–Beltrami operator and scalar curvature determined by the pseudo-Riemannian metric $g_i$, for $i=1,2$. It is not necessary that the scalar curvatures be constants. We will refer to the multipliers $p^{(n-2)/2}$ and $p^{(n+2)/2}$ as the “initial” and “final” multipliers, respectively.
Note that when $n=2$ (and only in this case), the initial multiplier is $1$, the multiplicative part of the operator (\[one\]) is $0$, the equation ${\mathcal{L}}_g\phi=0$ is independent of the metric $g$, and thus these solutions are simply related, modulo topology, to the harmonic $1$-forms $\omega$ satisfying $d\omega=\delta\omega=0$, merely by exterior differentiation.
The quartic operator defined below has an entirely comparable list of special features in the physical case of dimension $n=4$. Its solutions in this case are related to the conformally invariant class of harmonic differential forms, which being forms of half the dimension of the manifold, are the two-forms; if the metrics $g$ have Lorentzian signature the equations of harmonicity are the Maxwell equations.
It seems reasonable to conjecture the existence of analogous operators of arbitrary higher even order; indeed, powers of the flat wave operator in Minkowski space were shown to have a conformal covariance property in Jakobsen’s thesis and work of Jakobsen–Vergne.
To define the quartic operator, we must assume that the dimension of the manifold is neither $1$ nor $2$. Given a pseudo-Riemannian metric $g$ on $M$, with associated Laplace–Beltrami operator ${\mathcal{L}}_g$, scalar curvature $R$, Ricci tensor components $R_{ij}$ (tensorial notation referring to coordinates $x^1,x^2,\dots,x^n$), metric components $g_{ij}$ and covariant derivatives $T_i=\nabla_{\frac{\partial}{\partial x^i}}$, define $$\begin{gathered}
{\mathcal{Q}}(g)\phi=({\mathcal{L}}_g)^2\phi+
T_j\left[\left(-\frac{4}{n-2}R^{ij}+\frac{n^2-4n+8}{2(n-1)(n-2)}g^{ij}R\right)
\frac{\partial}{\partial x^i}\phi\right]\nonumber\\
\phantom{{\mathcal{Q}}(g)\phi=}{} +\left[\frac{n-4}{4(n-1)}({\mathcal{L}}_gR)
-\frac{n-4}{(n-2)^2}R^{ij}R_{ij}+
\frac{(n-4)(n^3-4n^2+16n-16)}{16(n-1)^2(n-2)^2}R^2\right]\phi\label{three}\end{gathered}$$ for scalar functions $\phi$ (the usual summation convention being followed).
Let $M$ be an arbitrary smooth manifold of dimension $n>2$, and assume given two pseudo-Riemannian metrics $g_1$ and $g_2$ on $M$ related by $g_1=p^2g_2$ for a positive function $p$. Then for all scalar functions $\phi$, $$\label{four}
p^{(n+4)/2}{\mathcal{Q}}(g_1)\phi={\mathcal{Q}}(g_2)(p^{(n-4)/2}\phi).$$ When $n=4$, $${\mathcal{Q}}(g)\phi={\mathcal{L}}_g{}^2\phi
-2T_j\left[\left(R^{ij}-\frac{1}{3}g^{ij}R\right)\frac{\partial}{\partial x^i}\phi
\right],$$ and the kernel of this operator is independent of the metric $g$, and also invariant under the ordinary pointwise action of the conformal group $(M,g_1)$ (or equivalently $(M,g_2)$) on functions.
- - - - - - - - - - - - - - -
In the special case of a conformally flat four-dimensional manifold $M$ of Lorentzian signature, we can sketch the relation to Maxwell’s equations alluded to above. For such an $M$, any smooth local conformal transformation extends uniquely to $\widetilde{M}={\mathbb R}\times S^3$. But now any global distribution two-form solution of Maxwell’s equations on $\widetilde{M}$ is automatically invariant under the central transformation of $\widetilde{M}$ sending $(t,p)$ into $(t+\pi,\mbox{ antipodal point to }p)$, as shown in the work of I.E. Segal. The associated compact quotient manifold of $\widetilde{M}$, denoted $\overline{M}$, is casually identifiable with the conformal compactification of four-dimensional Minkowski space. Now set $g_c=dt^2-ds^2$ (the Einstein metric on ${\mathbb R}\times S^3$), and set $X_0=\partial/\partial t$ and $\Delta$ the Laplacian on $S^3$. Then $${\mathcal{Q}}(g_c)=(X_0{}^2-\Delta)^2+4X_0{}^2,$$ and $$p^4{\mathcal{Q}}(g_c)={\mathcal{Q}}(g_f)=({\mathcal{L}}_{g_f})^2,$$ where $g_f=(dx^0)^2-(dx^1)^2-(dx^2)^2-(dx^3)^2$ is the flat metric, and we have $p^2g_f=g_c$. (Regarding $S^3$ as ${\mathrm{SU}}(2)$, one can take $p(t,u)=\frac{1}{2}\cos t+\frac{1}{4}{\mathrm{tr}}(u)$ ($t\in {\mathbb R}$, $u\in{\mathrm{SU}}(2)$).)
${\mathcal{Q}}(g_c)\phi=0$ has an infinite-dimensional space of smooth solutions on $\overline{M}$ (all solutions on $\widetilde{M}$ are lifted up from $\overline{M}$), and this equation satisfies Huygens’ principle. This space of solutions is the unique conformally invariant closed subspace of the space of all smooth fields on $\overline{M}$ whose elements are determined by Cauchy data of a finite number of $X_0$-derivatives on a space-like surface.
Any two-form solution of Maxwell’s equations on $\overline{M}$ is the exterior derivative of a $1$-form $A$ on $\overline{M}$ satisfying $d\ast_cA=0$ and $A(X_0)=0$ (gauge conditions; $\ast_c$ is the $\ast$-operator determined by $g_c$). The space of $1$-forms $A$ on $\overline{M}$ satisfying $d\ast_c dA=0$ (Maxwell’s equations) and $d\ast_cA=0$ and $A(X_0)=0$ is not conformally invariant, but its direct sum with the conformally invariant space of exterior derivatives of functions $\phi$ satisfying ${\mathcal{Q}}(g_c)\phi=0$ (or this space augmented by $dt$, namely the corresponding space of closed $1$-forms that are locally exterior derivatives of functions $\phi$ satisfying ${\mathcal{Q}}(g_c)\phi=0$), is conformally invariant.
Moreover, all the conformally invariant closed subspaces of the real smooth $1$-forms on $\overline{M}$ satisfying Maxwell’s equations are generated by the above-noted conformally invariant subspaces and the obvious one of all closed $1$-forms on $\overline{M}$.
| {
"pile_set_name": "ArXiv"
} |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.